METHODS OF NEGOTIATION RESEARCH
International Negotiation Series
Vol. 1 I.W. Zartman (ed.) Negotiating with Terrorists
Vol. 2 P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research
Methods of Negotiation Research edited by
P. Carnevale and C.K.W. de Dreu
MARTINUS NIJHOFF PUBLISHERS LEIDEN/BOSTON
This volume is reprinted from the journal International Negotiation, Volume 9 (3) and 10 (1), 2004–2005
ISBN 10 90 04 14858 2 ISBN 13 978 90 04 14858 1 © 2006 by Koninklijke Brill NV, Leiden, The Netherlands. Koninklijke Brill NV incorporates the imprints Brill, Hotei Publishers, IDC Publishers, Martinus Nijhoff Publishers and VSP. http://www.brill.nl All rights reserved. No part of this publication may be reproduced, translated, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission from the Publisher. Authorization to photocopy items for internal or personal use is granted by Brill Academic Publishers provided that the appropriate fees are paid directly to The Copyright Clearance Center, 222 Rosewood Drive, Suite 910, Danvers, MA 01923, USA. Fees are subject to change. PRINTED AND BOUND IN THE NETHERLANDS
CONTENTS Contributors ................................................................................................
ix
Methods of Negotiation Research: Introduction ...................................... Peter J. Carnevale and Carsten K.W. De Dreu
1
Methods of Negotiation Research II .......................................................... Peter J. Carnevale and Carsten K.W. De Dreu
5
Chapter 1 The Joys of Field Research ........................................................................ James A. Wall, Jr. Chapter 2 How Much Do We Know About Real Negotiations? Problems in Constructing Case Studies .................................................... David Matz
7
23
Chapter 3 Studying Negotiations in Context: An Ethnographic Approach .............. Ray Friedman
39
Chapter 4 The Problem-Solving Workshop as a Method of Research .................... Ronald J. Fisher
49
Chapter 5 Time-Series Designs and Analyses ............................................................ Daniel Druckman
61
Chapter 6 Social Research and the Study of Mediation: Designing and Implementing Systematic Archival Research ............................................ Jacob Bercovitch Chapter 7 Reflections on Simulation and Experimentation in the Study of Negotiation .................................................................................................. Jonathan Wilkenfeld
79
93
vi
CONTENTS
Chapter 8 Quantitative Coding of Negotiation Behavior ...................................... Laurie R. Weingart, Mara Olekalns, and Philip L. Smith
105
Chapter 9 The Use of Questionnaires in Conflict Research .................................. Aukje Nauta and Esther Kluwer
121
Chapter 10 A Multilevel Approach to Investigating Cross-National Differences in Negotiation Processes .............................................................................. Xu Huang and Evert Van de Vliert
135
Chapter 11 Methodological Challenges in the Study of Negotiator Affect ............ Bruce Barry and Ingrid Smithey Fulmer
149
Chapter 12 Comparative Case Studies ...................................................................... I. William Zartman
165
Chapter 13 Discourse Analysis: Mucking Around with Negotiation Data ............ Linda Putnam
177
Chapter 14 Field Experiments on Social Conflict .................................................... Dean G. Pruitt
193
Chapter 15 Laboratory Experiments on Social Conflict .......................................... Peter J. Carnevale and Carsten K.W. De Dreu
211
Chapter 16 Managing Conflict in the Literature: Meta-analysis as a Research Method .................................................................................................... Alice F. Stuhlmacher and Treena L. Gillespie Chapter 17 When, Where and How: The Use of Multidimensional Scaling Methods in the Study of Negotiation and Social Conflict .................. Robin S. Pinkley, Michele J. Gelfand, and Lili Duan
227
239
CONTENTS
Chapter 18 Markov Chain Models of Communication Processes in Negotiation .............................................................................................. Philip L. Smith, Mara Olekalns, and Laurie R. Weingart Chapter 19 All that Glitters is Not Gold: Examining the Perils of Collecting Data on the Internet ................................................................................ Yeow Siah Cha
vii
257
273
Chapter 20 The Method of Experimental Economics .............................................. Rachel Croson
289
Chapter 21 Empirical Research in Law and Negotiation ........................................ Rebecca Hollander-Blumoff
307
Chapter 22 Methodologies for Studying Personality Processes in Interpersonal Conflict .............................................................................. Lauri A. Jensen-Campbell and William G. Graziano Chapter 23 The Heart of Darkness: Advice on Navigating Cross-Cultural Research .................................................................................................. Catherine H. Tinsley Chapter 24 Disparate Methods and Common Findings in the Study of Negotiation .............................................................................................. Carsten K.W. De Dreu and Peter J. Carnevale Index ........................................................................................................
323
341
351
361
Contributors Bruce Barry: Owen Graduate School of Management, Vanderbilt University, Nashville, TN 37203 USA E-mail:
[email protected] Jacob Bercovitch: Department of Political Science, University of Canterbury, Private Bag 4800, Christchurch, New Zealand E-mail:
[email protected] Peter J. Carnevale: Department of Psychology, New York University, 6 Washington Place, Room 577, New York, NY 10003 USA E-mail:
[email protected] Rachel Croson: The Wharton School, University of Pennsylvania, 567 Huntsman Hall, 3730 Walnut Street, Philadelphia, PA 19104–6340 USA E-mail:
[email protected] Carsten K.W. de Dreu: Organizational Psychology, University of Amsterdam, Roetersstraat 15, 1018 WB, Amsterdam, The Netherlands E-mail
[email protected] Daniel Druckman: Institute for Conflict Analysis and Resolution (ICAR), George Mason University, Fairfax, VA 22030 USA E-mail:
[email protected] Lili Duan: Department of Psychology, University of Maryland, College Park, MD 20742 USA E-mail:
[email protected])
x
CONTRIBUTORS
Ronald J. Fisher: International Peace and Conflict Resolution Program, School of International Service, American University, 4400 Massachusetts Avenue, NW, Washington, DC 20016 USA E-mail: rfi
[email protected] Ray Friedman: Owen Graduate School of Management, Vanderbilt University, Nashville, TN 37203 USA E-mail:
[email protected] Michele J. Gelfand: Department of Psychology, University of Maryland, College Park, MD 20742 USA E-mail:
[email protected] Treena L. Gillespie: Department of Management, California State University-Fullerton, P.O. Box 6848, Fullerton, CA 92834–6848 USA E-mail:
[email protected] Rebecca Hollander-Blumoff: New York University School of Law, Lawyering Program, 245 Sullivan Street, New York, NY 10012–1301 USA E-mail:
[email protected] Xu Huang: Department of Management & Marketing, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong E-mail:
[email protected] Esther Kluwer: Department of Social and Organisational Psychology, Utrecht University, P.O. Box 80, 140 NL 3508 TC Utrecht, The Netherlands E-mail:
[email protected] David Matz: Graduate Programs in Dispute Resolution, University of Massachusetts Boston, 100 Morrissey Boulevard, Boston, MA 02125 USA E-mail:
[email protected]
CONTRIBUTORS
xi
Aukje Nauta: TNO Work and Employment, P.O. Box 718, 2130 AS Hoofddorp, the Netherlands E-mail:
[email protected] Mara Olekalns: Melbourne Business School, University of Melbourne, 200 Leicester Street Carlton VIC 3053 Australia E-mail:
[email protected] Robin L. Pinkley: Director of the American Airlines Center for Labor Relations and Conflict Resolution, Edwin L. Cox School of Business, Southern Methodist University, Fincher Building, P.O. Box 750333, Dallas, TX 75275 USA E-mail:
[email protected] Dean G. Pruitt: Institute for Conflict Analysis and Resolution, George Mason University, 9006 Friars Road Bethesda, Maryland 20817 USA E-mail:
[email protected] Linda L. Putnam: Department of Communication, Texas A&M University, 4234 TAMU, College Station, TX 77843–4234 USA E-mail:
[email protected] Chayeow Siah: Department of Social Work and Psychology, National University of Singapore, 11 Law Link, Singapore 117570 E-mail:
[email protected] Philip L. Smith: Department of Psychology, University of Melbourne, Victoria, 3010 Australia E-mail:
[email protected] Ingrid Smithey Fulmer: Eli Broad Graduate School of Management, Michigan State University, East Lansing, MI 48824 USA E-mail:
[email protected]
xii
CONTRIBUTORS
Alice F. Stuhlmacher: Department of Psychology, DePaul University, 2219 N. Kenmore Avenue, Chicago, IL 60614 USA (E-mail:
[email protected]) Catherine H. Tinsley: The McDonough School of Business, Georgetown University, Washington, DC 20057 USA E-mail:
[email protected] Evert van de Vliert: Social and Organizational Psychology, University of Groningen, Grote Kruisstraat 2/I, 9712 TS Groningen, The Netherlands E-mail:
[email protected] James A. Wall, Jr.: College of Business, University of Missouri, Columbia, 506 Cornell Hall, Columbia, MO 65211–2600 USA E-mail:
[email protected] Laurie R. Weingart: David A. Tepper School of Business Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA 15213 USA E-mail:
[email protected] Jonathan Wilkenfeld: Department of Government and Politics, University of Maryland, College Park, Maryland 20742 USA E-mail:
[email protected] I. William Zartman: School of Advanced International Studies, the Johns Hopkins University, 1740 Massachusetts Ave, Washington, DC 20036 USA E-mail:
[email protected]
Methods of Negotiation Research: Introduction PETER J. CARNEVALE and CARSTEN K.W. DE DREU
Main Entry: method Function: noun 1: a way, plan, or procedure for doing something 2: orderly arrangement from The American Heritage® Dictionary of the English Language, Fourth Edition A close look at the many methodological practices in the study of negotiation reveals a simple fact: there is no one best way, no one best plan, no single orderly arrangement that best produces understanding about negotiation. Indeed, the cornucopia of methods is impressive – as is the strength of the field. There are historical case studies, laboratory experiments, survey studies, archival data analysis, mathematical modeling – the diversity of method is extraordinary, but perhaps not surprising. After all, negotiation and social conflict span all levels of society, including interactions between nation states, small groups and organizations, people in close relationships, and even children on a playground. Its study reflects work in fields as diverse as political science, psychology, law, economics, communication, organization behavior, and anthropology. This diversity of method is clearly seen in the pages of International Negotiation, and in the many other journals that publish original research in the field. Yet practitioners and scholars, on occasion, may wonder if a particular research technique is remote and only distantly relevant. Practitioners in particular may not be inclined to appreciate the minutiae of method that often occupy the attention of their academic siblings. In as much as International Negotiation seeks to involve and support all aspects of a diverse audience, a special focus on matters of method is highly appropriate and desirable. This special issue of International Negotiation (as well the next issue, Vol. 10, no. 1) contains original essays on the topic of methods of negotiation research. We present here a focused thematic effort that reviews the state-of-the-art on research method in negotiation. Our goal in putting these special issues together is to provide a series of presentations that span both traditional and International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 1–3 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
2
PETER J. CARNEVALE AND CARSTEN K.W. DE DREU
innovative methods, common and less than common, all that move the field forward. With these articles, we make the point that there is a wealth of methodological tools that negotiation and conflict researchers have at hand, and each has strengths and weaknesses. Our specific objectives include the following: provide an introduction to a variety of methods and their utility; identify issues and controversies with various methods; increase the accessibility of works in one empirical domain to readers in another, and thus broaden the scope of research and theory; improve communication between domains so the collective enterprise is improved; provide stimulus for yet unknown approaches and procedures that further contribute to the validity and vitality of research in this domain; and stimulate the application of a method used in one domain into another domain. Another goal that we have for these special issues is to extract valuable insights about conflict phenomena from data collected with diverse methods in different settings. Since there are synergies among the methods (multi-method approaches are greater in their impact than the sum of the parts), a synthesis is needed. To that effect, at the conclusion, we present a review article that highlights common features of effects in negotiation that have been obtained with diverse methodological tools. For example, many studies show that a forgiving strategy in negotiation can be exploited, and a tough strategy can backfire by producing a competitive response, and this has been obtained in both laboratory studies of university students and in studies of international disputes. There are of course many examples of such cross-domain consistency in effects, for example in work on aggression (Anderson & Bushman 1997). Our concluding article reflects Don Campbell’s notion that validity is achieved through triangulation, that is, by using a variety of methodological approaches and procedures. In navigation, triangulation is the technique where two visible points are used to determine the location of a third point. Applied to validity and reliability of measurement, triangulation is the use of multiple, different indicators in such a way that errors can be excluded and underlying constructs can be identified (Campbell & Fiske 1959). The concept applies quite well not only to measurement, but to all aspects of method. When two or more methods or data sources converge on a construct, we have greater assurance that our conclusions are not driven by an error or artifact of any one procedure. Each method has strengths and weaknesses, and to the extent they do not overlap, we can stand on more solid ground with conclusions about theory.
METHODS OF NEGOTIATION RESEARCH: INTRODUCTION
3
The Author’s Task Each author was asked to write a paper about a specific methodological domain, one in which they are an expert. They also were asked to provide a general introduction and overview of the method as they understand and employ it, to identify its strengths and weaknesses, give noteworthy examples of it, and identify its potential. But mainly they were asked about their own experience in exercising that particular method, to write also about aspects of the methods they usually do not report in the method section of their research. We wanted the papers to come alive with the personal experience of conducting research in the given domain – the trials and tribulations and the tricks of the trade, so to speak, covering things like gut feelings about training coders for content analysis, implicit theories about which order of questionnaires works best, etc. We believe we have succeeded. Our hope is that this collection of essays will inspire new and established researchers alike to broaden and enhance their methodological practices. Additionally, we hope that this will stimulate new as yet unknown approaches and procedures that further contribute to the validity of research in negotiation and social conflict. Many people helped make this method special issue possible. We are grateful to the following individuals who provided reviews: Andy Schotter, Kathleen O’Connor, Andrea Hollingshead, Jim Wall, Dean Pruitt, Bianca Beersma, Ray Friedman, Ching Wan, Kees van Veen, Eric van Dijk, Bernard Nijstad, and Peter Molenaar.
References Anderson, C.A., & Bushman, B.J. (1997). “External validity of ‘trivial’ experiments: The case of laboratory aggression,” Review of General Psychology 1:19–41. Campbell, D.T., & Fiske, D.W. (1959). “Convergent and discriminant validation by the multitrait-multimethod matrix,” Psychological Bulletin 56:81–105.
Methods of Negotiation Research II PETER J. CARNEVALE and CARSTEN K.W. DE DREU
One aim of these special issues of International Negotiation on negotiation research methods – the 13 papers found here and the 12 in the previous issue (Vol. 9, no. 3) – is to illustrate the diverse methods being used today in the behavioral study of negotiation and social conflict. The 25 papers cover a lot of ground: general methodological techniques and approaches – field research, case studies, laboratory work, and so on – and some cover relatively narrow domains or statistical techniques, for example, multidimensional scaling as applied to negotiation data. The depth and breadth of the articles, we believe, reflect well on the application of the scientific approach to understanding conflict and negotiation. But space limitations led us to leave some areas out; we hope the reader will forgive us our omissions. One difficulty in putting this collection together was just how to organize the papers. As you can see, the collection is quite diverse with contributions from a broad array of scholars. At first, our inclination was to order the papers randomly – to avoid the imposition of an artificial scheme that would be more a procrustean bed than a way to facilitate analysis and thought. But a note from Dan Druckman led us to impose an order, or at least a rough order. He noted that there are three groupings of the papers, those that deal with experimentation of one sort or another (in the laboratory or in the field), analysis methods (for example, Markov analysis or discourse analysis), and “other settings” (e.g., cross-cultural, law, personality, internet). So, with thanks to Dan, this is how the papers are ordered within each issue for the most part. Our hope is that this collection of essays will inspire new and established researchers alike to broaden and enhance their methodological practices. Additionally, we hope that this will stimulate new and, as yet, unknown approaches and procedures that further contribute to the validity of research in negotiation and social conflict. We especially like the idea of papers that combine different methods (for example, a lab study together with an archival data study). As we mention in our concluding article, we believe it is triangulation and convergence of evidence that is the key to progress in the field. We wish to thank the many people who helped with this effort: Herb Kelman who inspired us early on in a conversation at the IACM meeting in International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 5–6 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
6
PETER J. CARNEVALE AND CARSTEN K.W. DE DREU
San Sebastian, and Bert Spector who guided the entire effort. We are grateful to the many people who helped with reviews, including Bianca Beersma, Eric van Dijk, Ray Friedman, Andrea Hollingshead, Peter Molenaar. Bernard Nijstad, Kathleen O’Connor, Dean Pruitt, Andy Schotter, Donna Shestowsky, Kees van Veen, Jim Wall, Ching Wan, and several others who indicated a preference for remaining anonymous.
The Joys of Field Research JAMES A. WALL, JR.
A previous mission had been a disaster. My father’s B-17 squadron had put up 13 planes. Over Germany, the fighters came out of the sun and in 10 seconds eight planes – 72 airmen – went down, “like burning leaves.” The current mission had gone better; fewer comrades had been lost; and Dad’s bomber was somewhere back over England, but in a cloud bank, isolated from the other planes. Lost. And only one person knew they were lost, the navigator. As for the others, the crew was celebrating a completed mission, and the pilot was focused on the low fuel gauge. The compass wasn’t working; neither was the radio (B-17s seemed to be allocated to crews in alphabetical order and with a pilot named Wallace and a navigator named Wall, this crew usually drew a decrepit bomber). The pilot called back, “Navigator, report our current position, we have to land now!” “Give me a minute,” Dad said. As he stalled, the clouds parted for a moment to reveal the plane was directly over the base. “Bank right and take her down; we’re home”. Then, in relief, he threw up. Because of several experiences like this, my father does not like being lost, but his firstborn does. I enjoy being lost in the woods, mountains, caves and on rivers, lakes, and oceans (okay, maybe not on the ocean). I also like being lost when I conduct field research. It, along with mistakes, scientism, the participants, “stuckness,” and discovery are what I refer to as the “joys of field research.”
Lost When I refer to being lost in field research, I mean physically lost as well as conceptually lost. My first recognition that I liked “lostness” came on a blazing August afternoon in Nanjing during the mid-1980s. My translator and I were pedaling up and down countless hills through a laocoönian tangle of streets in the city, lost. We were seeking an inner-city community – a “street” – whose mediator had agreed to an interview, but we were hopelessly lost. I do not believe we ever found that mediator. Rather, my translator International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 7–21 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
8
JAMES A. WALL, JR.
probably bribed a local mediator to talk with me. The entire search was, in retrospect, a great deal of fun. On a cooler day, a dozen or so years later, I was lost in Canada. I had driven across the U.S.-Canadian border on an expressway, traveling to Kingston, to find some Canadian peacekeepers. On the return trip, I could not find my way back into the U.S. “This is stupid,” I thought. If I came across the border on a four-lane highway, surely I can find my way back. But I could not, and as I spent hours exploring the back roads of Canada, I found it was enjoyable and concluded this was a fine physical representation – a metaphor – for being lost in field research. Usually, a field researcher is to some extent lost. You’re not always certain what the process is that you’re studying. An independent variable seems to be missing. There are a confusing number of dependent variables. Cause and effect are difficult to tease out. Results and conclusions are frequently contradictory. As for personal examples of being lost, my past is replete with them, but a current one is that I am lost as I consider the effects of conflict in business-to-business e-commerce. But I am having fun.
Mistakes Mistakes are somewhat akin to being lost. To me, conducting field research can be likened to rowing a single, a skiff. You are rowing backwards; therefore, you cannot see where you are going. If you attempt to overcome this inconvenience by looking over your shoulder, you will flip, because the boat is only ten inches wide, and very unstable. So in order to row well, you relax, look straight ahead (backwards), focusing on your stroke, your legs, and your speed. Conse-quently you run into obstacles such as rocks, boats, stumps, and shores only to find yourself in the cool water. That is, you make mistakes. It is a lot of fun. In field research, mistakes are not only fun; they also are instructive. A heart transplant surgeon emphasized this to me when I was interviewing leaders for my book Bosses (Wall 1986). “In my field,” he noted, “we obviously try to avoid as many mistakes as possible. But any mistake I make with one patient makes me a better surgeon with the next.” I agree. Each specific mistake teaches an individual lesson; moreover, mistakes in general are quite helpful because they eradicate – or at least lower – researcher hubris. In field research, arrogance is the researcher’s worst enemy. If you are overly confident, if you know that you are investigating the correct
THE JOYS OF FIELD RESEARCH
9
problem, from the correct perspective, with the correct procedure, with the correct sample, at the correct time – you will probably find yourself with a year’s worth of useless data. A few mistakes, early in the research, will bring you back to reality and reduce the long run costs. Mistakes also teach the field researcher that “good enough” is “good enough.” When the Russian general Kalashnikov was recovering from battle wounds and designing the AK-47, he remembered a local saying, “Perfection is the enemy of good enough.” Therefore, he designed his automatic rifle to be only good enough. It is not extremely accurate. Its bullets do not go as far as those of some other machines. The muzzle velocity is relatively slow. When it is fired, it sounds like a child rumbling through the kitchen pots and pans. But it always fires, under almost any condition; seldom does it jam; rust and abuse does not stop it or even slow it down, all because Kalashnikov designed its parts to fit together good enough, not precisely. Why did Kalashnikov design the AK-47 this way? I honestly do not know, but my guess is that he did not sit in a hospital room thinking, “I want to build a weapon that will fire under almost any condition and can easily be operated by a technological dummy.” My guess is that mistakes had taught him a lesson. Probably he, or his troops, had experienced weapons that did not fire when they became dirty or rusty or frigid – weapons that may have been too complicated to quickly disassemble and reassemble. For field researchers, mistakes usually have the same effect as the soldiers’ experience in that they teach the researcher not to be overly precise. I learned this lesson in a humorous experience. When in Nanjing, I found there was one mediator per 1000 citizens; therefore, in a city of two million there was a large potential sample. Awed by the number, my aspirations soared. As they did, I decided to precisely measure the techniques employed by Chinese mediators. I knew that self-reports by the mediators would be good enough, since this was the first empirical study of Chinese mediation. But I wanted precision. Accompanied by two translators and a pack of Marlboros (as a gift to the mediator), I begin my first observation. We – mediator, two disputants, two translators and I – sat at a very short, rough table. Initially, it went well; yet, a couple dozen faces soon peered in the open windows. As the mediator noted these, his voice grew louder. Then people started climbing through the windows and sitting on the floor; we were up to 25 spectators. Rising to the occasion, the mediator began mediating with gusto. As the floor space became inadequate, observers sat in the windows and three decided to join us at the table. Now the mediator stood, pointed his hand, addressed the audience and
10
JAMES A. WALL, JR.
raised his shrillness about three octaves. After two hours of this, I expressed my appreciation, delivered the Marlboros, mounted my bike and muttered, “Self-reports should be good enough for this project.” Finally, and most importantly, mistakes emphasize the value of laboratory procedures. That is, mistakes make it quite obvious that a lot of variance is occurring, and you had better take it seriously. Therefore, you learn/relearn to keep the study as simple as possible. Control as many factors as possible and randomize the rest. Match within conditions and obtain as large a set of observations as possible.
Scientism Perhaps the major contribution and invigorating joy from field research is that it allows the researcher to engage in “scientism” which is a combination of empiricism and reason (Shermer 2002). When describing this term or process, Michael Shermer notes that it “embraces (these) twin pillars of a philosophy of life appropriate for an Age of Science”. He holds that scientism is/was practiced by Jacob Bronowski, Stephen Hawking, Carl Sagan, Stephen Jay Gould, and Jared Diamond, some of my intellectual heroes. While I doubt if I adequately understand this process or practice it as well as the above intellectuals, I do find it quite useful to focus, first, on my method of study and its results (empiricism), then shift to reasoning and return to the results. My best example of this centers upon the number 1.30. In a study on community mediation in India, I was examining the mediation techniques used by a formal mediation group, the panchayat (group of five). One of the techniques was a dictated agreement point, and the average use for this technique was 1.30. That is, this mediation group, on average, dictated about one and a third agreement points (or concessions) per mediation. Initially, I recorded this figure in a table, compared it to the average usage of individual mediators – an elder – in India and moved on with my writing. Then I started to use some reasoning. The average is 1.30 per mediation. If there are two or more disputants in each mediation, should not the average number of directives be about 2.00, with one side being told what to do and likewise, the other? In some cases, the panchayat could have issued one ruling that covered both parties (e.g., farmer A is to quit throwing dead rats into farmer B’s irrigation ditch), but still the 1.3 seemed too low. As I thought about this finding, it occurred to me that ideas about the settlement might be coming from the disputants as well as from the mediator – that self-direction might be supplementing mediator direction. So I examined
THE JOYS OF FIELD RESEARCH
11
the behaviors of the disputants instead of focusing solely on the mediation techniques. When I returned to the empirical arena, to read what the disputants had done, I found they occasionally proposed their own concessions so that panchayat dictates were not necessary. These, in eight of the panchayat proceedings, seemed to fall into two distinct categories. In four of these, the concessions were totally self-imposed: • A father paid a self-imposed fine of 2100 rupees to the temple because of his son’s indiscretions and said that God has brought his son to the right path. • An uncle, beaten by his nephew, paid 1100 rupees to the temple and gave a prayer that the nephew will be prosperous and have a noble life. • An employer, whose employee had stolen four pumps from farmers, reinstalled the pumps as a self-imposed penalty. And in four others, the disputants increased the penalty imposed upon them by the panchayat: • After being found guilty, a man paid an extra 500 rupees to the village temple. • After being told to give 25 percent of his land to the temple, a disputant did so and also agreed to build the entire temple. • After a judgment was made against him, a disputant thanked the panchayat for its speed and justice; then he gave 5000 rupees to the temple. • After being told to clean the temple every morning for a week, a poor man did so for three weeks. The number of voluntary concessions did not totally solve the mystery, but it did reveal an important facet of panchayat mediation. In many disputes, it is strictly inquisitional: it hears evidence, makes inquiries, and decides. But it can also (separately or in concert with its inquisition) provide a platform and motivation for the disputants – who wish to attain/maintain good relations with the panchayat or who seek to gain face in public – to settle the dispute themselves. Somewhat ironically, this powerful body empowers the parties to settle their own disputes.
Participants For me, the greatest joy in field research comes from the interaction with the participants, that is, with the individuals being studied. Here there is a brio because the participants are excited that you are studying them and what they
12
JAMES A. WALL, JR.
do. You are interested! And you perhaps can help them improve. I vividly recall a conversation with an American appeals judge, who for years had been mediating in settlement conferences prior to trial. (A settlement conference is a meeting in which the judge, the plaintiffs’ attorney and the defendant’s attorney prepare the case for trial. Here points of agreement are noted, disagreements are acknowledged, points of law are discussed. If the judge wishes, he or she can attempt to mediate the case.) He noted he had been taking notes on how he mediated his cases for the last two years. His goal was to determine how his efforts affected the plaintiffs’ and defendants’ decisions on whether or not to settle out of court. When I told him that was exactly what I was studying, he beamed and devoted the rest of the day to delineating his mediations from the last few months. As we discussed these, he gave his opinion as to what he felt worked and what did not. Then he asked me for my opinion. Being asked for your opinion, to evaluate the performance of the people you are interviewing – judges, peacekeepers, CEOs, e-commerce directors, imams, panchayat leaders – and who usually are performing quite well is extremely invigorating. Even though this interaction occurred more than a decade ago, I still recall the emotional boost it gave me. A few months ago, a similar question from an experienced mediator produced the same effect. This mediator, whom I was shadowing, had mediated over 2000 cases and was currently fighting an uphill battle in a case where a 16 year-old girl had stepped into the side of a moving – speeding? – truck. About four hours into the ordeal, the mediator detoured me into a storage room and asked, “How am I doing?” Such interactions are not only invigorating, they also provide valuable information about the processes under study. For example, in a recent study of peacekeeping, Dan Druckman and I had predicted that a time constraint – in the form of impending darkness – would have an effect on peacekeepers’ mediations in a role-playing case presented to them. (Concisely stated, the prediction was that the time constraint would reduce the amount of the intervention.) After a dozen or so interviews, I concluded that the time constraint/ darkness manipulation was having no effect. So at the end of each interview I began to ask the peacekeepers (in the time constraint condition) if the looming darkness had any effect on their mediations. Some replied affirmatively, which was consistent with the theory and our previous field observations. Others said there was no effect because they and their subordinates were trained to operate 24/7 and were outfitted with night-vision equipment to do so. Most informative was the third group, who held that darkness would elim-
THE JOYS OF FIELD RESEARCH
13
inate the need to mediate. One peacekeeper from this group noted that darkness, rain, cold, wind and snow reduce the pressure. When these come, disputants go home and there is no conflict to mediate.
Stuckness Somewhere among the pages of Zen and the Art of Motorcycle Maintenance (Pirsig 1974), Robert Pirsig discusses the value of being “stuck.” Stuckness, he notes, is the beginning of true insight because when you are stuck, and realize that this is the case, you will begin to analyze and understand the situation so as to get unstuck. This stuck-then-understand process is superior to erroneously concluding that you are not stuck and advancing in the wrong direction. Guided by Pirsig’s logic, I have attempted to develop a taste for stuckness. While it has improved my field research and while I have frequently been able to tolerate the condition, it does not bring me joy. This might seem odd – if not a bit inconsistent – for a fellow who likes to be lost. Consider though, when you are lost, there is a dynamic flow. You are moving around, looking, thinking, testing, and experimenting. Admittedly, these movements are accompanied by a touch of fear as well as slight chill or coolness and an odd feeling in the stomach. But overall, it is an enjoyable process. By contrast, stuckness brings me a sense of immobilization, of going nowhere as when I get my Forerunner stuck in deep snow or mud. And the emotion? It is one of impatience, irritation, and internal roasting. The difference here is worth explaining; therefore, let me give you a mundane comparison and then one from field research. First, the mundane. A few weeks ago my wife and I locked ourselves out of our house (the responsibility for this blunder will forever be disputed) which is rather burglar-proof. When I realized we were on the outside looking in, I felt lost – not stuck – because I knew that I would eventually find a way into the house. Feeling rather cool/calm but somewhat stupid, I enjoyed thinking about the problem from different perspectives. Initially, I thought like a thief; then I took the perspective of a mountain climber. Subsequently, I played John Nash and thought, “take the option that keeps the joint costs for you and your wife as low as possible.” As I was thinking this, I also approached the problem as a carpenter, who has a lot of wood, tools, and some skill. A strategy came to me: I cut a very large piece
14
JAMES A. WALL, JR.
of wood from the doorframe (large pieces are easier to replace than are small ones) near the lock. This gave me direct access to the bolt which I slid over with a long putty knife. Later I repaired the doorframe. Total cost = 0. The time spent, including repair, was one hour. Compare this to an example of being stuck. One spring morning when I was a graduate student in Durham, I decided to mow the yard prior to taking a hike with my wife and our Irish Setter. First, I drove to get some fresh gasoline; secondly, I took out the grinder to sharpen the blade and picked out a 1/2-inch wrench to take off the blade. Then I started to pull the lawnmower out from under the house. Or I should say, that I started looking for the lawn mower, because it was not there. I was stuck; my face was hot and my feet felt stuck to the ground. Despite all of Pirsig’s advice – stare at the problem, push thoughts out of your mind, start over, go back to basics, write down your ideas, take a break, drink a cup of coffee, spend additional time thinking about the problem – I could not get unstuck nor develop a positive approach to the problem. So I quit looking and was irritable for the rest of the day. (A month or so later I discovered the mower in the garage of my wife’s mentor.) Now to an analogous comparison in field research. For me, the sensations of being lost and stuck both center upon the same set of variables: the techniques used by U.S. mediators as they handle disputes. As for being lost – and content – I feel this way when I broach the general question of what determines these techniques; that is, what are the independent variables that affect or determine which techniques (dependent variables) are used by U.S. mediators. Drifting, with this question in mind, I can contently review the literature, shadow mediators, interview disputants, ask mediators to recall mediations and discuss the question openly with professional and informal mediators around the country. For about five years, I have been lost but it has been great fun and currently I have some tentative answers. The first one is that we have in the United States high inter-mediator variation in the use of the techniques because there is no model of mediation in the U.S. that guides or reduces the variation among mediators. This contrasts with mediation in China where there is PLA-dictatorial model; in Korea, where there is a Confucian model; in Malaysia, which has an imam (Muslim) model; and in India where there is a panchayat (formal group of five) model. The second conclusion is that mediator power is important, because it affects how the disputants treat the mediator, rather than how the mediator treats the disputants. Finally, there is the role of reinforcement and superstition. Because about 80 percent of the mediations in the U.S. result in agree-
THE JOYS OF FIELD RESEARCH
15
ment, mediators here are reinforced for whatever approach they take. Therefore, they continue to employ the same techniques from mediation to mediation (i.e., there is low intra-mediator variation.) In addition, the mediators superstitiously associate the techniques they use with the successful settlement of the dispute. This combination of reinforcement and superstition gives rise to the various “styles” that mediators report and champion. Turning from “lost” to “stuck,” I find myself in the latter state when I treat the mediators’ techniques as the independent variable and attempt to investigate its effects on other variables (e.g., on rate of agreement). Specifically, I am most stuck and irritated that I have been unable to relate any single technique or group of techniques (e.g., a strategy) to disputant agreement. It is generally acknowledged that mediation (versus no mediation) does result in a higher level of settlements; yet, no one – including me – has been able to determine which techniques are responsible for the success. Here, I am stuck, and irritated. Having expressed my ire with stuckness, I will come full circle to admit that it does have its benefits in field research. It reduces one’s hubris. It makes one think; snuffs out mediocre ideas; eliminates false starts; and along with mistakes and being lost, it often underpins “discovery,” a major joy of field research.
Discovery Field research furnishes insights, perhaps because it allows me to observe people, places, events, and solid objects in five dimensions: length, width, height, time, and gravity. I cannot explain why insights come from observing concrete phenomena, but I can give a parallel example. There is a problem I give my class. They have an empty corked wine bottle, with a dime inside it. Their task is to remove the dime without damaging the bottle or removing the cork. When I give this problem to the class in written form, very few can solve it quickly. But when I haul in a lot of wine bottles with dimes in them, most students quickly “see” the answer. (The answer is at the end of the article.) For me, a similar process takes place in the field. I can, for example, review the mediation literature (e.g., Wall 1981; Wall & Lynn 1993; Wall, Stark & Standifer 2001), think about mediation while holed up in my basement office, or conduct laboratory studies about it (e.g., Arunachalem, Wall & Lytle 2000). Yet, I feel my best insights come when I am in the field. There I glean insights as I discuss their mediations with the mediators themselves,
16
JAMES A. WALL, JR.
ask mediators to recall their mediations, observe mediations, review maps of locations where mediations took place, review court records of the results of mediations, disassemble a peacekeeper’s M-16, fiddle with his machine gun, look through night-vision goggles and hear one mediator being called “judge” while another is called “Bob.” The insights that come to me seem to be of two sorts: the crisp isolated insights and the more expanded patterns. One of my wife’s deductions provides a superb example of the former. She is a biochemist – a genetic engineer – who is very smart, even though she occasionally locks the keys in the house. In the lab, she works with a sulfatereducing bacterium, which has the nasty habit of devouring metal. It chews away metal only in the absence of oxygen, which is good for all the steel above ground. On the other hand it is very bad news for stainless steel cooling towers, and for subterranean water and oil pipes. While trying to find a way to eradicate this irritating characteristic from the sulfate reducer, my wife considered removing the gene responsible for the behavior from the microbe’s DNA. Good idea, but not the most impressive insight. Here is the insight: while laboring at this task, she reasoned, instead of trying to eliminate this corrosion affinity, why not transfer it to a bacterium that attaches itself to a metal we want to corrode, such as uranium. (When uranium is corroded, it becomes insoluble and therefore, does not enter the water supply.) Of course I am jealous of this insight because all of mine pale by comparison, but one I have had is that effective leaders are those who protect their subordinates from distractions, time-demands, and even danger from others. Another – unre-lated – insight is that the orders under which U.S. peacekeepers work screen out many of the techniques they would normally use. However, two of the orders – to prevent violence and to stay out of civilians’ problems – are so opposed to each other (i.e., often preventing violence entails becoming involved) that they give peacekeepers a great deal of flexibility in their approaches. Turning to patterns of discovery, they usually require more time to emerge and initially they are disheveled, but this state enhances the joy when a pattern finally comes into focus. It’s somewhat like watching the crystallization of a supersaturated solution or watching ice forming a glaze on your windshield as you drive down the expressway. Ronda Callister and I recently noted a pattern as we were attempting to study community mediation in Thailand. A study in Malaysia (Wall & Callister 1999) was quite successful because we had clear cut differences between the mediations of the imams (Muslim
THE JOYS OF FIELD RESEARCH
17
mosque leaders) and those of the ketua kampungs (village mayors). With the Malaysian model as a starting point, we attempted to compare the mediations of the temple monks in Thailand to that of the Malaysian village mayor or a respected elder. Unfortunately, the work with the monks was not proceeding smoothly. The data were odd and we were confused as to why. (I think Ronda was lost and I was stuck.) Then a pattern emerged for Ronda as she pieced together many of our observations: The monks seemed uncooperative or confused. Some insisted they did not mediate. Our interviewer seemed to be making no progress in asking people about disputes they took to the monks or in obtaining reports from the monks. The pattern/insight was that villagers were not taking important disputes to the monks because they did not respect them. In Thailand, most, if not all, young men are expected to be monks for a certain period of time and some of them continue as monks for a lifetime. As a result, monks are not generally perceived as wise, or as ones who should be asked to mediate a dispute. While the above pattern is simple and unwanted, the one currently forming in a study of legal mediation has more positive aspects. It is more complex because a chain of causation is involved, and it is beneficial in that it charts out the direction for future research. My beginning hypothesis in the study of attorneys and judges as mediators was that judges would mediate differently than attorneys. Specifically, they would reduce the clients’ perceived BATNAs (i.e., the outcomes from a trial) by citing their own experience on the bench and they would use their status to control attorney behavior. The emerging pattern, however, seems to have the causation coming from the opposite direction, from the clients’ and attorneys’ behaviors toward the judges. I noted that prior to any of the judge’s actions, the clients and attorneys paid substantial deference to him, in that they addressed him as judge (not by his first name), waited for him to shake hands, observed where he sat, and were quiet when he spoke. Occasionally the clients even asked the judge for his opinion as to what they should do. Over coffee breaks the attorneys told the judge how helpful he was in a previous case. These types of behavior by the clients and attorneys tended to shape the judge’s behavior, rather than vice versa. The specific pattern seems to be that the clients seek guidance from the judge, who supplies it in a professional way. In turn, the clients instruct the attorneys, who – after their own input – make their offers to the judge. Observing this causal chain, the judge realizes the client is in the driver’s seat; therefore, when an impasse is reached, or the concession-making slows, the judge addresses the clients. A related causation exists in attorney mediations, but the pattern and tone are different. Here the client views the mediating attorney as a lawyer who at
JAMES A. WALL, JR.
18
best is neutral, but who teams up with the opposing lawyers against him or her. Therefore, the client does not want the mediator’s advice and their attorneys must proffer the offers and concessions. Noting this, the mediator concludes the attorney is in the driver’s seat and addresses him or her.
Conclusion A balanced article would follow the discussion of the joys of field research with a delineation of the costs, subsequently compare the two, and announce the net benefits or costs. As you might expect, this is not to be a balanced article. I do not make cost-benefit analyses when considering field research, and I am in good company: Michelangelo didn’t consider the costs and benefits of sculpturing “David;” rather he took a chisel to the marble and chipped away everything that did not look like David. Bruce Springsteen does not think like an accountant when he chooses to belt out, “This Gun’s for Hire.” Al Pachino probably did not conduct a cost-benefit analysis as he starred in “A Scent of a Woman.” Bill Cosby and Robin Williams certainly do not choose to continue in stand-up comedy routines because of the net benefits. And John LeCarre, just as Stephen King, writes because of the passion for the art, not because the benefits outweigh the costs. Likewise, I have a passion for field research. When I first begin a project, I think about it continually – day and night – with a resultant number of sleepless nights. Then I begin to investigate to subject, losing more sleep because of my excitement. Excitement grows as I fiddle with a real-life phenomenon that I can see, hear, smell, and feel, interacting with participants who are similarly excited that I am studying them. They want a better understanding of their activities; they want to improve them and usually they want me to be successful. This initial excitement, cultivated by the interaction with the participants, tends to flow toward a passion as my waywardness, mistakes, and stuckness give way to discovery and progress. Yet, experience has shown me that this zeal for field research – like most passions – must be guided with several cautions so that it serves research rather than making the researcher its victim. With this goal in mind I proffer the following admonitions: 1) Do not get excited for the sake of excitement; because if you do, you will waste a great deal of time and energy. When I was a graduate student, I asked my mentor, Stacy Adams, how a social scientist maintains excitement about social phenomena for a lengthy period of time. The leathery wrinkles of his right cheek indicated that I had asked the wrong question and his response verified
THE JOYS OF FIELD RESEARCH
19
my conclusion. “I’m very surprised you asked that question,” were his exact words.
I had posed it because we, at the time, were studying personal space, which was a dull topic to me. Concluding that I had asked the wrong question which perhaps indicated some laziness on my part, I knew I had to offset these with a “born again” excitement in the project. So I became excited and then wasted about a year on a dull topic instead of moving to boundary role and negotiation research, where I had a truly exciting dissertation (Wall & Adams 1974). 2) The second admonition is to avoid believing that a topic is important because you have a passion for it. Cognitive dissonance explains why each of us believes there is high value in what we do. Yet it can lead us to self-delusion; that is, we can be studying something that is unimportant, but we think it is important.
Let me be more precise. Studying unimportant phenomena is fine, because there is a possibility that the unimportant will become important. Are we not glad, for example, that Flemming’s mentor did not tell him to quit fooling around with mold and to move on the something bigger? Also, it is acceptable to study unimportant phenomena when it is fun to do so. To me, this perquisite – along with not having to maintain a fixed schedule in the office – is a major luxury of our trade. Personally, I enjoy studying unimportant phenomena, such as the electron-microscope images of the protein crystals that my wife has isolated from some of her bacteria. The crystals are bright orange and more interestingly have the same angles as a large calcite crystal on my desk. I enjoy studying the interaction between cats and snakes. A female cat, I have noted, will cautiously avoid a snake or a rope that moves like a snake; however, she will very contently observe her kittens’ playing with a snake. I enjoy studying BATNA levels. In tinkering with these, I have found that negotiators do not abandon a negotiation when the BATNA exceeds the probable net outcome of the negotiation. Rather, the BATNA level must be about 11/3 times the negotiation outcomes. Returning to my admonition, from these odd examples, I contend it is appropriate to be excited about unimportant phenomenon because it is fun and may eventually be fruitful. But it is inappropriate to believe that a phenomenon is important simply because you are excited about it. To engage in such delusion sacrifices one’s professional objectivity. 3) As a third suggestion: Abandon some projects. Admittedly, I am better at advising this than doing it, but I know that for some field research projects there comes a time for abandonment, because, for example, the phenomenon cannot be understood or a process cannot be divided into manageable parts. It is too
JAMES A. WALL, JR.
20
complex or so simple that it’s trivial. The process unfolds so quickly that the independent variables and dependent variables seem meshed. The participants are uncooperative. There are an inadequate number of research sites. You run out of money, or the project quits being fun.
How do you do it? My approach is to put all my notes, articles, books, and data on the project into some Xerox boxes, walk to the nearest dumpster, toss them in and go for a long hike. Two activities allow you to make such a clean break. One is a back-up field project. Always have two or more projects going at once so that you can abandon one – without remorse or panic – when it is necessary. The second activity is teaching. This is not only a service to students and probably your major contribution to society. It also provides you with emotional backing when you must abandon a field project. 4) As a final area for caution, do not get so excited about the subject matter that you are sloppy in writing it up for publication. Because most readers, reviewers and editors are not as excited about your project as you are, they need to be convinced. Their conversion requires that you get their attention, explain why the phenomena under study is important, present lucid details on what you did find and tie up the package neatly at the end. Moreover, remember that writing is – as Stephen King puts it – “work.” Set aside adequate time for writing because it takes ample time. Then focus, revise your work, have young colleagues review it, and then revise, revise, revise. Lay it aside for a month, then revise, revise, revise.
In closing, I shift from the four cautions back to my major theme: Field research not only allows scholars to contribute to our knowledge pool; it can also be a lot of fun. So, relax – or get excited – and enjoy it.
References Arunachalem, Vairam, Lytle, Ann and Wall, James (2001). “An evaluation of two mediation techniques, negotiator power, and culture in negotiation.” Journal of Applied Social Psychology, 31: 951–980. Pirsig, Robert (1974). Zen and the Art of Motorcycle Maintenance. London: Head. Sherman, Michael (2002). “The shamans of scientism.” Scientific American, June, p. 35. Wall, James (1981). “Mediation: An analysis, review, and proposed research.” Journal of Conflict Resolution, 25: 157–180. Wall, James (1986). Bosses. Lexington, Mass.; D.C. Heath. Wall, James and Adams, Stacy (1974). “Some variables affecting a constituent’s evaluations of and behavior toward a boundary role occupant.” Organizational Behavior and Human Performance, 11: 390–408. Wall, James and Callister, Ronda (1999). “Malaysian community mediation.” Journal of Conflict Resolution, 43: 343–365.
THE JOYS OF FIELD RESEARCH
21
Wall, James and Lynn, Ann (1993). “Mediation: A current review.” Journal of Conflict Resolution, 36: 160–194. Wall, James and Druckman, Daniel (2003). “Mediation in peacekeeping missions.” Journal of Conflict Resolution, 47: 693–705. Wall, James, Stark, John and Standifer, Rhetta (2001). “Mediation: A current review and theory development.” Journal of Conflict Resolution, 45: 370–391.
How Much Do We Know About Real Negotiations? Problems in Constructing Case Studies DAVID MATZ
How do we know what occurs in a real negotiation? We know that laboratory-generated data about negotiations are at some remove from what occurs in the field (Pruitt 1981). But do we know enough about behavior in the field to understand the size of that remove? It is common understanding that our methods of gathering field data on any topic do not promise complete accuracy (Bernard 1994), and that we must be content with something less. It is the point of this note to suggest that gathering case study data about negotiation presents problems even beyond the usual, and that as a result our teaching and theory building may be working with some faulty materials. Most of this essay is a summary of how these questions came most forcefully to my attention, and how I have sought to answer them. In the fall of 2000, the Israelis and the Palestinians had just returned from the Camp David II mediation which had not produced an agreement. In September, the second Intifada began as the two sides were still trying to negotiate. Prime Minister Barak’s government was disintegrating, leading him in December to call for new elections. His opponent emerged as Ariel Sharon who promised an end to the negotiations and the concessions that Barak had been offering. By January, three weeks before the election, Sharon was far ahead in the polls. The Israelis and Palestinians decided to try one more round of negotiation. Each side sent a full negotiating team (without the head of government from either side) to Taba, an Egyptian resort. The negotiation ended six days later with a joint statement to the press which said that “The sides declare that they have never been closer to reaching an agreement and it is thus our shared belief that the remaining gaps could be bridged with the resumption of negotiations following the Israeli election” (Enderlin 2003). At that point there were nine days left before the election. As things stood there was no practical chance that Barak would win that election, and I was astounded. How could the parties walk away from the negotiating table if they were really making such progress? Convinced that the newspaper accounts could not be accurate I decided to use an upcoming sabbatical in Israel to learn what actually happened at Taba. What I learned was indeed at odds with International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 23–37 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
24
DAVID MATZ
the published reports. The parties in fact were much closer to agreement than reported, and many of the negotiators believed that with four more days of negotiating a framework agreement on most – though not all – of the major issues could have been reached. The newspaper account of modest progress had thus muted what, in my research, became the central question: Who ended the negotiation and why? Elsewhere, I have written about the actual results of the negotiation and my conclusions about what ended it (Matz, 2003a, 2003b). Here I want to focus on some of the problems I encountered in learning about those results. My research began eight months after Taba ended. There were by then two memoirs and several newspaper interviews and analyses by and with mostly Israeli participants (Beilin 2001; Sher 2001); there was also one scholarly account, done by an Israeli with strong contacts among the Palestinians (Klein 2001). There were twenty eight participants at Taba; I interviewed seventeen, of whom 11 were Israeli. The Palestinians were harder to reach because some were traveling out of the region, and because the politics in the fall of 2001 made travel difficult. (One telephone interview proved to be an unsatisfactory solution.) Interviewees from both sides were equally forthcoming about issues in the negotiation process, but the Palestinians were much more circumspect when asked about the larger context. This imbalance of sources certainly made me wary of imparting bias to my results, but my practice of crosschecking within teams and between teams gives me some assurance that this weighting did not itself lead to errors in my narration or interpretation.
Why It Is Hard To Learn What Really Happens In Negotiations Almost all negotiations experience the tension between publicity and secrecy. Occasionally, as at Oslo, the entire process including its very existence was secret until it is completed. This is rare. Nonetheless, negotiators most often work to keep most of their moves out of the public eye during the negotiation period itself, and sometimes permanently. Privacy is an essential component of good negotiating (Pruitt 1999). The tension exists because of the pressure applied by a negotiator’s constituencies or audiences, groups that may hold a range of views about the negotiation, from full support to full hostility. Privacy blunts the capacity of these constituents to shoot down a component of an agreement before it can be presented in the flattering light of a full agreement. Privacy, moreover, allows negotiators to explore in a tentative, testing manner, what the other side will offer and accept. It allows either side to hint at a willingness to change position, without committing itself to such a change.
HOW MUCH DO WE KNOW ABOUT REAL NEGOTIATIONS?
25
Privacy thus enables the parties to use, even create, their own subtle communication process. Taba negotiators told me that “in four eyes” (i.e. in private conversation) they knew what the other side was willing to do, though there was no specific sentence uttered by the party in question to which such an inference could be attached. In an interview in Haaretz eight months after Taba, Shlomo Ben Ami (Israeli Foreign Minister under Barak) described a meeting with Abu Ala (chief Palestinian negotiator at Taba), using telling phrases to explain how he had inferred receptivity among the Palestinians: “they showed readiness,” “the feeling was,” “it could be assumed” (Ben Ami in Haaretz 2001)). Though this process of inference is vulnerable to misunderstanding and duplicity, exploration without commitment is nonetheless essential for good negotiating. And only privacy enables this to occur. One result of this secrecy is that scholars and journalists are dependent in large measure on what the negotiators are willing to tell. And negotiators have a number of incentives not to tell the public the truth. In addition to the usual problems of faulty memory, self serving perspectives, and scores to settle, recollections of negotiations raise special problems. Whatever the collaborative energy in a negotiation, no such process can be free of competition. It may be that people drawn to negotiation have already a well developed competitive sense, and it may be that the negotiation process (with its inevitable supply of omniscient Monday-morning-quarterbacks) further exaggerates that sense. (Perhaps, the act of memoir writing enhances the competitive spirit yet further.) Whatever the cause, a perusal of negotiator memoirs makes clear that competition does not end with the negotiation (see Kissinger 1982). Negotiator’s memoirs often exhibit an apparent need to show superiority to the other side and frequently to one’s own teammates; to show that one was not duped, and to show that one did the best that anyone could. These incentives exist with all negotiations, but they are especially strong when the negotiation, as at Taba, is one chapter in an on-going process. Participants’ efforts to communicate what happened in a negotiation are not made easier by the subtlety of the forms of communication inherent in the artof ne gotiation, as suggested above. Body language, tone of voice, pacing, energy level, the impact of humor all have much to do with negotiating success, but they are extremely difficult for authors to recapture after the event. The bold judgment (“He never wanted a deal.”) or the quoted dialogue (“When he said that to me, I told him that. . . .”) have an impact on the reader that can overshadow what may have been a more significant flow of communication at the table. On some occasions the competition between negotiating teams is subordinated to the competition between the negotiators on the one hand and the audiences each negotiator needs to face on the other. Whatever the degree of
26
DAVID MATZ
hostility and distrust between two sides, negotiators need each other and this often creates strong bonds of allegiance between them. One upshot can be a collaborative falsification to the public. Oslo is of course one such instance in which both sides conspired to hide from, and lie to, the press. Another occurred at the press conference closing Taba where the leadership of both sides agreed explicitly to tell the press, falsely, that the sides would not budge on the Palestinian commitment to a right of return and the Israeli resistance to it. In reality, they had reached most of an agreement. Attempting to learn what actually occurred in a negotiation is influenced by the availability of writings produced during the negotiation. Sometimes there are position papers and proposals, drafts and redrafts. But often these do not exist or are not available. At Taba there was a marked aversion to putting anything in writing. As negotiating progress was made, the idea of drafting proposals was proposed and was usually shot down, by both sides at different times. But some writings were produced. Since the location of particular boundaries was a central part of the negotiation, the use of maps would seem to have been inevitable. But which maps were put on the table has been a matter of assertion and denial. Though there has been much talk of various maps, only half of one Palestinian map has so far “as of 2001” been allowed into the public light (Klein 2001); the Israelis put one map on the table, though this has been disputed by the Palestinians (Enderlin 2003). One group of negotiators at Taba – those dealing with the refugee questions, – did produce a great deal of writing, mainly in the form of proposals and counter proposals, some drafted by one side, some drafted jointly. There would appear to have been at least four such documents, and there are suggestions that there were yet more. Of these, only one has appeared in print, showing up in Le Monde, eight months after the negotiation ended (2001). In short, scholars and journalists have no direct access to what occurs in a negotiation, important documents may be misleading or missing, and the eye witness participant/reporters have numerous incentives to distort their reports. Of course, cross checking of sources (interviews, documents, memoirs) when there is a high level of consistency does allow one to develop a reasonably reliable picture of at least some portions of the negotiation. But inconsistencies in the sources are frequent, and thus unsurprisingly one is rarely free of the need to use a standard of plausibility as well as one’s imagination. Let me now give an account of one episode in the Taba negotiation in which inconsistency of reports raised questions, and then describe how I have tried to determine what really happened.
HOW MUCH DO WE KNOW ABOUT REAL NEGOTIATIONS?
27
Crosschecking and Detective Work: Some Examples The Taba negotiation began on a Sunday evening, and for two days things went very well. On Tuesday afternoon however, two Israeli restaurant owners visited a Palestinian town and were killed by Palestinians. Prime Minister Barak then made the decision to recall the cabinet ministers on the negotiating team. Gilad Sher (Barak’s chief of staff, and one of the Israeli heads of the negotiation delegation) describes a dinner on Wednesday evening at which he, two other members of the Israeli delegation (both former members of Israeli security services), Abu Ala (head of the Palestinian delegation) and Mohammed Dahlan (head of Gaza security and member of the Palestinian delegation) attended (2001). The dinner occurred at a fish restaurant 10 miles south of Taba. Sher makes clear that no one else, other than some Egyptian coordinators, knew of the dinner, and no one (other than Barak and Arafat) was told of the dinner after it occurred. Sher reports that the purpose in arranging the dinner was to learn if Arafat was prepared to reach an agreement. The Israelis offered Abu Ala an airplane to fly immediately to Arafat for an answer. According to Sher, Abu Ala answered: “The master of the house does not want an agreement.” Reading of this dinner I was puzzled. Sher presented himself as having seen Taba, from the time it was first proposed, as a somewhat empty negotiating effort (too little too late), at best an opportunity to “freeze dry” the positions of the parties for use in later, presumably much later, negotiations. With such a pessimistic view, why call a melodramatic meeting? In our interview Sher sloughed off any inconsistency and said it had made good sense to have the dinner. A week or so later I interviewed Israel Hasson, one of the Israeli former security officials present at the dinner. Even before my first question, Hasson volunteered to tell me of the dinner, but never mentioned Sher’s presence, though I asked him directly about it. Hasson described the dinner in almost the identical words used by Sher. When this difference in attendance surfaced some weeks later Hasson called to say that he, Hasson, had initiated the dinner, and after Abu Ala had answered as quoted above, Abu Ala asked to have Sher join them. Hasson then contacted Sher, who drove down and had the same dialog again. The issue of who was present at the dinner is not in itself important for the issues raised by the Taba negotiation, but this conflict in the testimony does support skepticism about some of the versions of what actually happened at this dinner. In an interview, Abu Ala recalled the dinner but could not recall who was there. When asked about the “master of the house” comment he said that Arafat had told him the night before Taba began that if Abu Ala could reach
28
DAVID MATZ
an agreement and would recommend it, he, Arafat, would sign it. Since one of the core questions at Taba and throughout all Israeli-Palestinian negotiation – does Arafat want a deal? – is raised by this event, the question of what Abu Ala said at the dinner justifies some scrutiny. There are, I think, three possible interpretations of these accounts. One is, that the Israelis invented Abu Ala’s admission altogether in order to blame Arafat should the negotiation fail. A second is that, the Israelis may have asked a question about Arafat’s intent, the Palestinian leader, a master of the shrug, may have responded with a physical gesture, and the Israelis may have interpreted this to mean Arafat would sign nothing. And third is that, the story occurred just as dramatically as the Israelis told it. What to do with the various contradictions? Though in retrospect, Sher presents a blasé account of his own expectation of Taba and of his own performance there, on Monday evening Sher saw that some agreements at Taba were possible. He called Barak and told him so. (On Tuesday some Palestinians were also seeing agreement possibilities and called their Ramallah headquarters asking for additional staff.) On Wednesday, following the killings, Barak left open whether he would be willing to re-start the talks. But for Barak to end the talks would have left the blame for termination solely on him (though he of course could in turn blame the terrorists and thus Arafat). As such, an “admission” from Abu Ala that Arafat would accept no agreement would be handy to balance the blame. An alternative – but not inconsistent – Israeli goal for the dinner would have been to see if the professional intensity of the Palestinian negotiators at Taba (commented on by several Israeli negotiators at the time) reflected a directive from Arafat to reach a deal. However, asking Abu Ala the question directly would have risked falling into a trap of their own devising: if the answer had been that Arafat was open to a deal that week, any failure in the negotiation would have fallen more heavily on Barak. It is possible that the Israelis did raise the question of Arafat’s intent, though perhaps indirectly. What is not plausible is that Abu Ala, with Dahlan sitting there, would answer it in the blunt, explicit way he is quoted. As a constant in Israeli-Palestinian negotiations has been the need to blame the other side for any failure, how likely is it that the head of the Palestinian delegation in the middle of the negotiation would tell major players on the other side that his chief wants no agreement? What could he gain by such an admission? Though tensions within the Palestinian team were so intense that anyone at some moment might have blurted out anything about anyone even Arafat, I think it unlikely that Abu Ala, a very experienced professional, would do so. Also, why would Sher and Hasson keep such a striking admission secret from the Israeli negotiating team for the rest of the
HOW MUCH DO WE KNOW ABOUT REAL NEGOTIATIONS?
29
week, and from the press even after Taba was over? Considering the importance of blame in the negotiations, this silence is striking. The story showed up in public only in Sher’s memoirs published 8 or 9 months after Taba. My view is that at the dinner in the desert the Israelis opened the topic of Arafat’s intentions and Abu Ala responded with a shrug. Since Taba ended without a mutual blaming orgy – as there had been after Camp David – this story stayed in Sher’s notes until he came to write his memoir. In the writing, as is not uncommon, the narrative became less ambiguous than the original event, i.e. the shrug became a speech. With what clarity Sher conveyed the story to Barak at the time of Taba is uncertain, though Barak’s failure to mention it to me when it would have been useful (see below) suggests that Sher did not relate it to Barak in as stark a way as he did in his memoirs. I would like now to turn to a more central concern of my research. There are two generally accepted and inconsistent understandings about where the parties stood when Taba ended. According to one, they made progress and ran out of time; according to the other, no progress occurred because the only goal of Taba negotiators (shared by both sides) was to make Barak look good for some voters in the coming election. My understanding, based on interviews and on piecing together the various accounts, is far more optimistic than either story. I asked the negotiators: “If you had four more days in which to negotiate, could you have reached a framework agreement on your topic?” On all issues, except two very important ones, the answer was affirmative. (The two exceptions were control of the holy places in Jerusalem, and the process for ending the conflict.) This assessment was created by matching each side’s estimation, issue by issue, of what a settlement could look like.1 It was also supported, if obliquely, by the joint zeal of the parties in keeping secret the documents and maps produced at the negotiation. Such systematic secrecy can be explained only by assuming that these papers contained serious concessions. Even I, however, optimist-in-chief for the Taba negotiation, have to acknowledge several qualifications. A negotiator’s inference from the behavior of his opposite number that an agreement was possible leaves open the possibility – as had happened before Taba – that the negotiator was reading the signals incorrectly, that his opposite number would change his mind, or that the other side was intentionally deceptive. Even if the signals were read correctly, no one on either side had, when the negotiation ended on Saturday, yet pulled together all the potential agreements and reviewed the entire package. This process would have been full of politics, drama, and guessing about what the other side would do. In short, it would have been unpredictable. Perhaps most importantly, no one believed that an agreement on the holy places was possible, and this failure was so significant that it could have
30
DAVID MATZ
derailed everything else. Nonetheless, I believe the evidence shows that the parties at Taba were much closer to an agreement than is generally believed, and that a framework agreement was possible with a few more days of negotiating. Since the record is consistent that the negotiation ended on Saturday at the initiative of the Israelis with the uncomplaining concurrence of the Palestinians, the question is: Why?
Inferences About Intentions Indeed, in studying negotiation, the question we frequently ask is “Why?” We all want mightily to know an actor’s intention. It is a base for measuring a negotiator’s success. More generally, it is a major source from which we create meaning and the foundation for understanding what might otherwise seem an arbitrary set of negotiation moves. Knowing a negotiator’s intention also goes a long way toward satisfying our need for good guys and bad guys, a matter not to be taken lightly as many a negotiation is assessed as a play of moralities. The conundrum arises because rarely can we answer the “why.” The question indeed is an invitation into quicksand. Intentions are always many, multileveled, morphing, floating, dissolving. A serious reflection to determine our own intention for one (relatively) simple task (e.g. writing an article) is itself a humbling experiment. Moreover, there is the evidence problem. Negotiators, like all public players, have considerable incentive to describe (in memoirs, interviews) their motives and goals as virtuous and reasonable. In the case of negotiators we again have some special problems. The significant role in negotiation of ambiguity, hints, feints, and various kinds of pretence, make the development of patterns of behavior into evidence very problematic. Even a negotiator with some limited means of testing for his/her opponent’s intention is often unsure. It is, indeed, inherent in all negotiating that each side is suspicious of the other, and reads the evidence with a worry about being duped. One might think that long standing knowledge of one’s opposite number would make the decoding of intent more plausible. But when Yossi Beilin, among the most pro-negotiation of the Israeli leadership, entered the first plenary session at Taba, and faced people he had long known and long worked with, he nonetheless wondered whether “the people before us (are) terror operatives or those trying to prevent it” (Beilin 2001). Or, we can take an example from the refugee negotiations – an issue which was discussed with a considerable fund of trust between the parties. Nevertheless, the Israelis wondered whether the Palestinian insistence on a reparation plan that put
HOW MUCH DO WE KNOW ABOUT REAL NEGOTIATIONS?
31
the emphasis on individual Palestinian applications for compensation wasn’t “really” a way to keep the refugee question open “forever.” Thus there is a deep conflict between our passion to know intention and the near impossibility of reliably knowing it. The common way out of this dilemma for scholars is to ignore it: impute an intention, offer no (or some highly selected) support, and move on. This approach can be seductive, satisfying the reader’s desire to know the inside story; but it generally substitutes narrative skill for evidence in the effort to present a persuasive account. To take a prominent example, Robert Malley and Hussein Agha in the summer of 2001 wrote a now much cited article analyzing the Camp David II mediation. Malley was present at Camp David as staff to President Clinton, and Agha is a long time consultant to the Palestinian leadership. In the course of a six page essay, they inform the reader no less than twelve times about what Arafat wanted, what Barak was thinking, or what either was driven by. They offer little evidence for their mind-reading, but because Malley was present and Agha is well connected their assertions about intent are passing via footnotes into scholarship on the subject. Recognizing the conundrum did not exempt me from it: I wanted to know why Barak and Arafat ended a promising negotiation four or five days before they had to. I began by interviewing Barak. I explained my purpose and finally asked: With an agreement at least possible and with at least four days left to negotiate, why did you end the negotiation on Saturday? He was equally direct: “Because that was the day the Palestinians brutally killed two Israelis and further negotiation was impossible.” When I pointed out that there were no terrorist killings on Saturday or Friday, he said, “Oh, right, but that was the day that Arafat made that horrendous speech calling us fascists, and there was no way to negotiate with someone who spoke like that.” When I pointed out that that speech was made the day after Taba ended, he said, “It doesn’t make any difference why I ended it. It had to end because it wasn’t going anywhere.” When I told him that my piecing together of interviews and memoirs suggested that a framework agreement was at least possible with a few more days of negotiating, Barak simply denied it. This of course may represent precisely how he saw things on the last day of the negotiation. In the swirl of an election campaign, with a somewhat loosely managed negotiation, and with little effort made to coordinate the various negotiators working on different issues, it is possible that Barak did not know how close the negotiators were to reaching agreement. Indeed it is possible that no single person at the negotiation was aware of this larger picture either. In any event, Barak rebuffed further effort to discuss what he knew at the time. It was significant that Barak did not take the easy, conventional way out, saying that Arafat
32
DAVID MATZ
wanted no deal. That he did not mention the Wednesday night meeting in the desert is a hint, as suggested above, that he had not been aware of it at the end of Taba, and perhaps, if he had not read Sher’s memoir, he was not aware of it at all. Believing that there was more to learn about Barak’s thinking than I could get directly from him, I decided to ask some experts. I began a tour of people who claimed to be Barak insiders, people who knew his mind. From these sources I gathered thirteen explanations for why he ended the Taba negotiation. • His wife did it. She never liked the idea of Taba and finally persuaded Barak that it was a political mistake. • His political advisor, James Carville, did it. He convinced Barak that an agreement at Taba would cost him any chance at the election. • Even before Taba began, he had lost the conviction that a historic breakthrough was possible. He instead focused on negotiation only as an electoral ploy. • He had a history of getting cold feet when an agreement started to look possible. That’s what he did with Syria and he did it several times with the Palestinians. Put more positively, when he got very close to an agreement, the concessions made him worry deeply for the future of Israel. • His characteristic pattern as prime minister had been the zigzag: move toward a goal and then away from it. Who knows why? • He really wanted an agreement with the Palestinians but was so maladroit as both politician and negotiator that he missed opportunity after opportunity. (This item has many examples and subcategories.) • He had wanted an agreement at Taba, but having learned (on Wednesday night) that Arafat didn’t want a deal, he changed his focus to one of making the world understand Arafat’s real intent. • He decided that no deal was possible at Taba and that negotiations would have to continue later, before and after the election. • Ben Ami (Israeli Foreign Minister) and Abu Ala (Head of the Palestinian delegation) agreed at Taba that any further progress in negotiating would require breakthroughs at a leader-to-leader summit, and this advice convinced Barak. • His negotiators at Taba were too desperate to achieve an agreement and were giving away the store. They were making deals neither he nor the Israeli public could or should accept. • Individual Israeli negotiators were making what amounted to private deals and he was not sure what they were. He felt himself to be losing control of the process, worried that he would be trapped with whatever they agreed
HOW MUCH DO WE KNOW ABOUT REAL NEGOTIATIONS?
33
to, and thus ended the negotiation to avoid being responsible for a situation he had not anticipated and could not control. • The Israeli public would never accept any deal from Taba and a negative popular vote would clearly present Israel as the reason for no peace. • If the Israeli public approved an agreement by a small margin, it would create a storm of opposition from the nationalist right wing that could lead Israel into civil war. What does one do with this motivational smorgasbord? Each has at least some inherent plausibility, each can claim someone up close who says its true, and each has some smattering of facts that can be adduced to support it. Selecting a few elements of the context in which Barak made his decision may help. We know that Barak expressed no optimism about sending a team to Taba in the first place. He seems never to have given his negotiating team one clear mandate, and his various utterances were thus interpreted by different negotiators according to their own predispositions. Pulling back his main negotiators on Tuesday suggests a softness in his commitment to reach an agreement at Taba; still, though he could easily have let the negotiation end right there, he let the team return to Taba on Thursday. Many of the negotiators at Taba were important to his election campaign and, being at Taba, were not able to carry out their campaign functions. Some of those negotiators recalled the pressure to “get back to work.” Pulling this together into an interpretation of Barak’s thinking on that Saturday, I would conclude that he just wore out. His optimism and drive to reach an agreement were not enough to overcome the pessimism and skepticism all around him, and in him. Put differently, nearly everything on the above list weighed on him. Perhaps we should not choose from the list, but rather see it as an array of lures beckoning Barak not to negotiate further at Taba. Taken together (even where they are inconsistent with each other) they describe a climate resisting further negotiation, impacting his viewpoint and thus his decision. In this climate it would have taken a leader of extraordinary strength, willing to take larger risks, with something larger than Barak’s tiny and diminishing base of popular support, to continue the negotiation and push for agreement. It is thus not surprising that Barak would see it as prudent to adopt the conventional campaign stance: assert that he had done everything possible for peace. About Arafat’s intentions I have less evidence. Abu Ala says that Arafat was willing to sign an agreement. Arafat sent a good negotiating team, a team that clearly thought agreements were possible and that worked diligently throughout the week, a team that tried to keep the Israelis at the table on Tuesday
34
DAVID MATZ
afternoon when Barak was calling his cabinet ministers back from the negotiation. On the other hand, we have the puzzling account of the Wednesday night dinner that might suggest Arafat’s lack of commitment. And we have, finally, the ready acquiescence by the Palestinian leadership on Saturday to the Israeli decision to end the negotiation, suggesting that the Palestinians too – whether reflecting Arafat or not – did not choose to negotiate further. Arafat’s vicious speech in Switzerland the day after Taba ended attacked Barak and the Israelis, and made no mention of Taba. When I asked Palestinians close to Arafat what they made of this confusing picture and why he seemed willing to end the negotiation, they professed no insight into his intentions. As Arafat is a master of the masked motive, I take their confession seriously and can only echo it.
Conclusions One goal for this essay has been to identify special problems in understanding what happens in a negotiation. The imputation of motive is one concern. Another is the necessity of comparing many sources (interviews, memoirs, documents) to learn what happened, of cautiously accepting a consensus of sources when there is one, and of assessing inconsistent accounts by explicitly stated tests of plausibility. Any account that seems to rely on one source, or that does not identify differences among sources is, I would suggest, suspect on those grounds alone. These are hardly novel insights, but they have not necessarily been the rule in studying negotiations. Let me close with two examples, from other negotiations, of (mis)understandings that have become part of our scholarly lore. Social Conflict (Rubin, Pruitt, and Kim 1994), in its analysis of “interests,” cites the example of Henry Kissinger mediating the provision of food and medicine to Egyptian troops surrounded by Israeli troops after the Yom Kippur War. Kissinger in this account “after careful analysis. . . . concluded that Israel wanted actual control of the road, whereas Egypt wanted only the appearance that Israel did not control it. . . . A bridging solution was found that involved stationing Israeli soldiers unobtrusively on the sides of the road (so that they actually controlled it) and having UN checkpoints on the road itself (to give the impression of international control).” The authors cite Golan (1976) but Golan actually provides a rather different picture, especially when supplemented by the report of that negotiation done by the the Insight Team of the London Sunday Times (1974). First, on a minor matter, Kissinger’s “careful analysis” came after he was told by an Israeli aid that the solution
HOW MUCH DO WE KNOW ABOUT REAL NEGOTIATIONS?
35
lay in separating the issue of control of the road from allowing aid through to the Egyptian army (Golan 1976). More importantly, this part of the negotiation was, as described by Golan and The Times, a four way power fight. The parties were Israel, Egypt, the US, and the UN, and the topic was the control of the road to the Egyptian soldiers. Egypt gave in early in the process, recognizing as Golan saw it, the limits of its power. Kissinger first proposed that the UN control the road, and Israel objected. Then he proposed one UN inspection station and Israel agreed. Kissinger later converted that to three UN stations and pressed Israel very hard to take it, though there is no indication that Sadat sought this (Brecher and Geist 1980).2 Furthermore, as the checkpoints were being constructed there were fist fights between Israeli and UN soldiers over their construction (Sunday Times 1974). The result of these various tussles was that Israelis were in the UN checkpoints inspecting whether the material being sent to the Egyptian soldiers was only humanitarian in nature. Moreover, according to Kissinger what made these concessions effective was that, as part of the package the Israelis achieved the return of their prisoners (Kissinger 1982). In summary, the “bridging solution” gave Israel everything it had originally sought, gave Kissinger a deal that allowed the process to go forward and gave him much of the credit for achieving it, gave the UN a hollow role that might charitably be called symbolic, gave Sadat not so much a fig leaf as the barely visible appearance of a fig leaf, and can, with no undue cynicism, be seen as a conspiracy among the Egyptian, Israeli, and US governments to fool the Egyptian people. The term “bridging solution” suggests that each side used problem solving technique to satisfy it’s “interests.” I think it accurate to say that this “bridging solution” was a thin cover that reflected perfectly the power relations of the parties. To focus now on a later chapter in the aftermath of the Yom Kippur War, the 1978 Camp David negotiations dealt with the Israeli withdrawal from the Sinai. In their account in Getting To Yes, Fisher, Ury, and Patton (1981), contrast the parties’ opening positions (Egypt wanted all the Sinai returned and Israel wanted to keep a part of it) with the negotiated result (Egypt got full sovereignty, but with limits placed on where Egypt could put any weaponry). This development is used to illustrate the value for seeking a settlement of “interests”, Israel’s being security, Egypt’s being sovereignty. The difficulty here is that the basic principles that satisfied these interests (limits on the use of weapons, demilitarized zones, U.N. control of yet other areas) were already in place on the ground when Camp David began. It was the formula the Egyptians and Israelis had agreed to with Henry Kissinger as mediator in 1974. Significant modifications of the plan were indeed negotiated at Camp David, including moving the lines separating the parties, substitution of an
36
DAVID MATZ
international force for the UN, and most contentiously the time table for phasing out this plan. But the concepts were already in the note books of President Carter’s mediating team before the summit began, and it is difficult to believe that the two sides had not themselves anticipated the Carter team proposal in their own planning for the negotiation.3 Again, this was not problem solving and an exploration of interests. It was taking an obvious solution staring the parties in the face and modifying it to fit the situation. These examples, and the Taba account more generally, raise the question of what we know about what goes on in real negotiations. They suggest that we may well know less than we have assumed. As a result we may need to review some of our theory to see how much of it may be based on such assumptions.
Notes 1. The negotiation concerning Maale Adumim provides an illustration of how this matching process can be done. Maale Adumim is a large Israeli settlement (25,000 people) situated east of Jerusalem which blocks much north-south traffic for a new state of Palestine. The Palestinians had been intensely fearful that a Palestinian state would be divided by Israeli lands and roads, leaving it without a coherent land mass, and Maale Adumim raised this specter. Nonetheless, at one point, the Palestinian negotiators conceded that Maale Adumim would be annexed to Israel as part of the overall agreement. Israel argued for more area around the settlement, and connection to other settlements, and the Palestinians then rescinded their concession. “Unofficially” however, the Palestinians acknowledged that Maale Adumim could be part of the annexation if the parties could work out the scope of the surrounding area. Thus, when the negotiation ended, the official Palestinian position refused Israeli annexation of Maale Adumim, but unofficially negotiators on both sides saw the trade that would have reversed that position (access roads for the Palestinians – which the Israelis would grant – and limited growth space for Maale Adumim – which the Palestinians would grant). 2. Brecher confirms Kissinger’s own interests in getting the results he thought best for the US. 3. Compare Quandt, W. (1986) Camp David: Peacemaking and Politics. Washington D.C.: Brookings Institution (p. 382) with Kissinger’s Years of Upheaval (p. 839.) Supplemented with an interview with Herman Eilts, then U.S. Ambassador to Egypt, who was present at Camp David.
References Asfour, Hanan (2001). “Ben Ami’s Occupation Syndrome.” Haaretz Magazine, October 19. Beilin, Yossi (2001). Manual For a Wounded Dove. (Hebrew) Tel Aviv. Miskal-Yidioth Ahronoth Books and Chemed Books. Ben Ami, Shlomo (2001). “End of a Journey.” Haaretz Magazine, September 14.
HOW MUCH DO WE KNOW ABOUT REAL NEGOTIATIONS?
37
Bernard, H. Russell (1994). “Methods Belong To All Of Us,” in Robert Borofsky, editor, Assessing Cultural Anthropology. New York: McGraw-Hill. Brecher, Michael, with Geist, Benjamin (1980). Decisions in Crisis: Israel 1967 and 1973. Berkeley, California: University of California. Eldar, Akiva (2001a). “How to Solve the Palestinian Refugee Problem.” Haaretz, May 29. Eldar, Akiva (2001b). “Sir, If You Please, Tear Down Your House.” Haaretz, May 31. Enderlin, Charles (2003). Shattered Dreams. Paris. The Other Press. Fisher, Roger and Ury, William and Patton, Bruce (1991). Getting To Yes. New York. Penguin. Golan, M (1976). The Secret Conversations of Henry Kissinger. New York. Quadrangle. Holbrooke, Richard (1999). To End a War. New York. The Modern Library. Insight Team of The Sunday Times (1974). The Yom Kippur War. Garden City, New York. Doubleday. Kissinger, Henry (1982). Years of Upheaval. Boston, MA. Little Brown and Co. Klein, Menachem (2001). Shattering a Taboo: The Contacts Toward A Permanent Status Agreement in Jerusalem 1994–2001. (Hebrew.) Jerusalem: The Jerusalem Institute for Israel Studies. Malley, Robert and Agha, Hussein (2001). The New York Review of Books. (August 9) Matz, David (2003a). “Trying To Understand The Taba Talks.” Palestine-Israel Journal of Politics, Economics and Culture 3. Matz, David (2003b). “Why Did Taba End?” Palestine-Israel Journal of Politics and Culture 4. Pruitt, Dean (1981). Negotiating Behavior. New York. Academic Press. Pruitt, Dean (1999). “Secrecy, Ripeness, and Working Trust at Oslo,” in Deborah Kolb, editor, Negotiation Eclectics. Cambridge, MA. PON Books. Quandt, William (1986). Camp David: Peacemaking and Politics. Washington, D.C. Brookings Institution. Rubin, Jeffrey, Pruitt Dean, and Kim, Sung Hee (1994). Social Conflict: Escalation, Stalemate, and Settlement, second edition. New York. McGraw-Hill. Schiff, Ze’ev (2001). “What Was Obtained at Taba Regarding Palestinian Refugees.” Haaretz, September 12. Sher, Gilad (2001). Just Beyond Reach. (Hebrew) Tel Aviv. Miskal-Yidioth Ahronoth Books and Chemed Books.
Studying Negotiations in Context: An Ethnographic Approach RAY FRIEDMAN
Ethnographic research methods have provided powerful insights into the way negotiations and dispute resolution occurs. It was through the use of ethnographic methods that Walton and McKersie (1965) developed their insight that there are four dimensions of labor-management negotiations: integrative bargaining, distributive bargaining, attitudinal restructuring, and intra-organizational bargaining. In my work, ethnographic observations led me to see the importance of rituals in labor negotiations, and the way those rituals derive from the role structure within which negotiators have to work (Friedman 1994; Friedman & Gal 1991). In a study of mediation, participant observation enabled Kolb (1983) to see how mediators brought particular frames to their work, rather than choosing frames based on the particular situation. Close observations of mediators led Silbey and Merry (1986) to distinguish between bargaining and therapeutic styles of mediation. Merry’s (1989) use of anthropological reports allowed her to distinguish between the legalistic model of American mediators and the more socially-embedded structure of mediation common to non-industrial societies. This approach to negotiation research stands in sharp contrast to the more frequently-used experimental method for studying negotiation. While experiments are designed to isolate specific phenomenon of interest and “control” all other factors, ethnographic research looks at behavior in the total negotiation context. While experiments can be done with naïve negotiators, ethnographic research targets experienced negotiators who carry with them communal norms, ongoing relationships, and detailed knowledge about the institutional context. While experiments are done in brief periods of time, ethnographic research follows real negotiations over natural periods of time – often weeks or months. Experimental research can teach us a great deal about, for example, psychological biases (Neale & Bazerman 1985), the effects of personality (Barry & Friedman 1998), or the influence of social motivations (De Dreu, Weingart, and Kwon 2000). But what we can not learn from these experiments is how experienced negotiators and mediators act in real, complex situations. Nor do experiments, as Hopmann (2002) points out, “capture International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 39–48 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
40
RAY FRIEDMAN
some of the most important, albeit subtle aspects of negotiation [such as]. . . . identify formulation (p. 73).” For five years, I studied labor negotiations by sitting in negotiation sessions, attending caucuses on both sides, debriefing lead bargainers after negotiation sessions, interviewing participants, and following the negotiations over weeks and months (Friedman 1994). This process has several significant disadvantages – it takes a great deal of time (in classic studies, such as Whyte [1934] and Suttles [1968] researchers spend years living in the communities they studied), and does not produce data of a type preferred by many journal editors. Nonetheless, ethnographic research can generate unique insights into negotiation that can not be gained in any other way. Consider, for example, the issue of constituent pressures. We know from experimental research that when negotiators are observed by and accountable to constituents they are more aggressive in their negotiation style (Carnevale 1985), at least within the U.S. (Gelfand & Realo 1999). But the relationship between constituents and negotiators is far more complex and subtle. Lead bargainers can control external pressures through appropriate choice of team members. One union negotiator reported that he puts his most difficult constituents (what he calls the “bomb tossers”) on the committee: They are the guys that stand in the back of the meeting and they throw the bomb out there that gets everybody riled up and then they sneak out of the meeting when it comes to the strike vote. You grab them and you put them on your team and you put them on the committee, put them on the hot spot. Get the guys with the biggest mouth. Put them on the committee; put them on your side.
Another union negotiator talked about the need to continually test her ability to control her side through “saber rattling”: It is not rattling sabers to scare an employer. It is doing two things: one is testing my organization to make sure I have the power. I want to know if I have the power or not. My bottom line is going to be affected by how much membership support there is for those issues. The other reason is to remind the employer that this is not my agenda, it is our member’s agenda.
The relationship between a negotiator and his or her constituents requires careful management and relationship building. As one other negotiator put it: You have to feel it. Each committee is different. Have they worked with me before? A lot of it is trust. If they trust you, then they trust you to lead them. You find that most rank and file people don’t want to say anything at the table, so you give them all the freedom they want and you tell them that they are encouraged to speak and most won’t. Yet if I am talking about a custodian issue, it is far more effective hearing it from the custodian than it is for me.
STUDYING NEGOTIATIONS IN CONTEXT
41
Yes, constituent pressures do affect negotiators, but that is just the starting point. By studying real negotiators in context, we see how they manage those pressures. Similarly, we know that relationships affect negotiators, with different tactics being used for friends and strangers (Sondak, Neale and Pinkley 1999). We must bear in mind again that relationships between negotiators are very complex. It is not just that you care more about the other side when you know them well, but also that you have more tactical options if you know the other side well. One negotiator used his relationship with the other side to reign in an overly aggressive lawyer: “John – a manager – called, and he said ‘this is b.s.’ I said ‘I know, but no one told the lawyer that.’ He said, ‘go back to work and I’ll settle the rest.’ I did. Then we got one of our best contracts.” If there is a strong relationship, there is more trust in both the validity of the proposals made and less need for formal contracting. Two negotiators explained: Summers – he knows me, I know him. Whatever he can extract from the company he puts on the table as a final proposal. This is a guy I would trust implicitly. We will not bull shit each other. Friedrich and Lambert’s CEO is Tom O’Brian. If he says this is what we agreed on, I’d sign it tomorrow. If I have a question about a clause I can call him and he’ll fix it. He has integrity.
Relationships can be a source of leverage, and provide alternative avenues for communication. One union lawyer reported, discussing a particular management labor lawyer, said “I have only dealt with him myself one or twice I think. He tends to be very legalistic actually. But the nice thing is he has a client who is open to a getting-to-yes style. So you can kind of get around Fred sometimes by convincing his principal. Fred is much more likely to say don’t do that, it will get you in trouble.” Yes, relationships matter, but what is also important is how negotiators assess the quality of relationships, the fact that relationships with the other side may include both good and bad relationships, and how negotiators depend on and use good relationships in negotiations. As these examples show, we gain insights from ethnographic research that can not be obtained through experimental methods. Before proceeding, I should be clear about what is distinctive about ethnography. Ethnography has four features (Friedman & McDaniel 1998). First, ethnography requires direct observation of the activity being studied – in the case of negotiations, sitting in on some or all of the preparation meetings, negotiations, and caucuses. Second, it requires that the observation be in a natural setting so that the researcher can “uncover special languages, unique and
42
RAY FRIEDMAN
peculiar problems, and, more generally, distinct patterns of thought and action (Van Maanen 1983: 13).” Natural observation also allows the researcher to see how people manage in situations where there are multiple, concurrent pressures. Third, ethnography gives prominence to the words, interpretations, and experiences of the people studied. Ethnographers want this information because it allows them to “examine the cultural knowledge, behavior, and artifacts that participants share and use to interpret their experiences (Schwartzman 1993).” Fourth, observations of how people behave, and how they interpret their experiences, are shared with the reader. The goal is to enable the reader to experience the situation much as the negotiator does. That being said, ethnography is not at all the most common approach to the study of negotiations. Ethnographic methods are very time consuming, and require a different set of research and data analysis skills. The rest of this article will be devoted to explaining some of the challenges in studying negotiations ethnographically. What I will report is my own personal experiences and research strategies.
Research Steps and Strategies Gaining Access The main problem with gaining access is that fact that you, the researcher, pose risks for the negotiators. Typically, information is tightly controlled in labor negotiations (and, indeed, most negotiations; see Hopmann [2002] who explains that there is often a legal responsibility to maintain confidentiality in negotiations). If you were to reveal to the other side what you hear in caucuses, you could cause enormous damage to the negotiations. While revealing information will certainly not be a researcher’s intention, there are usually social opportunities (such as when everyone gathers to get coffee before negotiations begin). On such occasions, an inadvertent comment could reveal something, or one’s body movement and facial expression during negotiations might reveal information. Also, the very act of having an observer present may change behavior, making people self-conscious or hesitant to say what they want. For these reasons it takes a tremendous amount trust to gain access to a negotiation. Negotiators will want reassurances about confidentiality and one’s credibility as a researcher. Further complicating the situation is the fact that, in labor negotiations at least, many people are involved. On several occasions, a lead bargainer approved my study, but I had to explain myself all over again when I first met
STUDYING NEGOTIATIONS IN CONTEXT
43
additional members of the bargaining team. In two cases, a union negotiator approved my presence and presumed that since they had the right to form their own bargaining team they also had the right to bring whomever they wanted. The company’s lawyers did not agree and insisted that I not be allowed to return. In another case, the company’s CEO approved the project but would not force any individual to participate. In that case, one plant manager refused to let me enter, fearing that my asking questions about a previous bitter negotiation would increase anger among employees. Entry into negotiations is never really settled – it requires constant renegotiation. Given natural sensitivities to the flow of information, access will usually be on one side or the other (I was able to access both side’s caucuses in only one case). Therefore, it is important to access multiple negotiations. Only then will you be able to see both sides of the negotiation process. Being Unobtrusive Once negotiations begin, great care must be taken not to attract too much attention. On the one hand, it is impossible not to develop relationships with people. You are with them for weeks or months, and they will want to get to know you and ask your opinion and advice. Indeed, one of the really fun parts of ethnographic research is that you encounter very interesting people. Also, you do want to keep track of how they see the situation, what their goals are, and how they interpret negotiating tactics on both sides. Still, the more you interact, the more you risk becoming an actor rather than an observer. If you take stands on issues or strategies, participants may not be as open and accepting of your presence. Another way that you may get unwarranted attention is through note-taking. It is already an intrusion for you to be present, but when they see you suddenly whip out your pen and pad and start scribbling, that reminds them that they are being watched. Some ethnographers prefer to wait until after a meeting before taking notes, or go to the backroom frequently to write down quick reminders. I prefer to take notes at the bargaining table, but I try to take notes continuously – even if nothing is really note-worthy at the moment. If I am seen as just writing all the time, them my note taking gets less attention. What to Observe During observation, the researcher monitors two parallel levels of behavior. At one level, I try to record what happened, including what was said and
44
RAY FRIEDMAN
what proposals were made. In some cases this is factual, while in other cases it is interpretive. Fred may tell the other lead negotiator that the company can not provide a certain level of benefits, but it is also important to note the intensity of this comment (is it matter-of-fact, or with great passion?), who Fred is looking at when he is making that comment (is it just the other lead bargainer, or each member of the union team?), and the reaction of the other team (do they appear surprised or angry?). At the second level, I am recording my questions and analysis. Why is Fred making this comment now? It seems that he is trying to change the topic because he wants to avoid another difficult issue for the moment. He is suddenly taking a new approach that is different than his plan expressed in caucus – maybe the union’s prior comment really was a surprise? So, notes can be very extensive, and one eye is always on data gathering while the other is on interpretation. Tracking What You Observe The biggest challenge of ethnographic research is to keep track of the data you collect and the ideas you develop. Unlike experimental research, where data is clean, structured, and fairly limited, ethnographic research produces endless amounts of data. It is critical to carefully note the day, time, and place for all notes, and to fill out your notes right away. As you take notes, you will inevitably have sentences that are not complete, or ideas that are scribbled quickly. At the time, you fully understand what you are writing, but a week or month later it may be incomprehensible. Therefore, each night go back to your notes and fill them in. Also, unlike experimental research where the theory is known before data is collected, in ethnographic research you are building theory from your observations. This means that, with each new day of study, you may have extended or developed your tentative ideas, or created new ones. Spend some time each day actually writing out your evolving conceptual ideas. One method that I like is to go over notes (or transcripts, if I was lucky enough to be able to use a tape recorder) and write down the concept or idea related to that part of my notes. This helps me to see what concepts are emerging, and how best to interpret the data. Later, I reorganize the data around concepts. For example, for my book I had one file filled just with comments related to “trust.” This eventually came to include about 100 quotes and observations, each of which had a code identifying which negotiation or interview it came from, and what page in my notes it came from. Writing was often based around these conceptually-organized files (most computer programs designed for qualitative analysis, such as Aquad, have functional equivalents of these files).
STUDYING NEGOTIATIONS IN CONTEXT
45
The Scope of the Study Nearly all of the focus in traditional experimental studies of negotiation is on the actual process of the two parties meeting. For labor negotiations, or other large-scale negotiations, much of what matters happens away from the bargaining table. In preparation for negotiations, people may meet for months collecting information about constituent goals and desires, deciding between priorities, and developing a bargaining strategy. In several cases, I was able to begin attending meetings long before the negotiations actually began. There is a limit to how much you can cover, but “if possible” it is advisable to get involved in the negotiation process as early as possible. During negotiations, there are frequent caucus discussions and meetings between individual bargainers. These should also be covered. In addition to caucuses, I have been able to glean much information simply by walking to the parking lot with negotiators after the sessions were over. Perhaps there had been a key phone call between the lead bargainers that morning or a conflict between key people in management that was affecting who in the bargaining team had more power and influence. Getting this type of information requires getting people to trust you and simply being there enough that you get to hear what is going on. I also gathered a great deal of information from interviews. While I could see what was being said during a negotiation session or caucus, I could not necessary discern the strategies and intentions of the parties. In one-on-one meetings with key people, I could ask why a certain approach was taken, or what the goal was of making a given statement within a caucus meeting. I could ask how they interpreted the other side’s behavior, and what concerns they had that shaped what their plan was for the next day. I could hear about coalitions within a bargaining team, and the effect they might have on support for key issues. These personal interviews were invaluable as a complement to actual field observations. While I tried to get greater depth of understanding from additional meetings with key negotiators, I also tried to get greater breadth of understanding by meeting, if possible, with constituents. I tried to meet with workers and union members, or with managers who were not at the bargaining table. How did they view the negotiation process? What signals did they see? How strongly did they feel about key issues? What was the reputation of the bargaining committee members? As you can see, ethnographic research requires a great deal of time. It reaches out, if possible, to include all the parties relevant to a negotiation and all the places where decisions are made and coalitions built, not just acrossthe-table bargaining. In also can extend out over weeks or months. This can
46
RAY FRIEDMAN
be a logistical problem, since real world negotiations do not happen within the confines of university teaching schedules or summers. Also, since access can be problematic, you may only be able to access a negotiation in a different city, which makes the logistics even harder (and the costs even greater). Reporting Results Unlike more traditional research methods, there are few established conventions for reporting ethnographic research results. Van Maanen (1988) says that much of it comes down to an ability to tell a compelling story. A compelling story both makes the situation clear to the reader and makes the reader more likely to trust the findings – a good story is hard to tell if there is not real, in-depth knowledge of and understanding of a situation. As Golden-Biddle and Locke (1993) put it, one must convince the reader that the research was authentic, and that the interpretations are plausible. Validity depends not on Alphas or Chi-squares, but in-depth explanation and clarity of writing. The challenge is finding a way to use observations, quotes, and stories in such a way that it leads towards a conceptual argument. Good ethnographic work should lead to new theories. In their book, Walton and McKersie (1965) used their field observations to build a conceptual model of negotiation processes, and connected their ideas to social science theory. In my book, I focused on the ways role constraints affected attempts to change how people negotiate. Kolb’s (1983) study builds towards a model explaining factors that influence mediators’ choice of mediation strategy. Note, however, that ethnographic work tends to be seen most often in books, not articles. It often takes more room to provide the data for an ethnographic study than for a statistical study, and many journals and their reviewers seem to be less accepting of ethnographic data than survey data. There has been some publishing of ethnographic studies of organizational issues in top journals (see, e.g., Barley 1996), but I do not know of a single ethnographic study of negotiation in either OBHDP or JAP, two of the leading journals for negotiation research. Scholars typically write books, or look for alternative outlets for their work.
Conclusion Traditional research methods for the study of negotiation can be very powerful, but that research has been criticized for “decontextualizing” conflict (Barley 1991). Ethnographic research offers an alternative approach which can provide insights into the types of complex situations that negotiators really
STUDYING NEGOTIATIONS IN CONTEXT
47
face. This approach is not easy – it can be more time consuming, costly, and difficult than other research methods – but the payoff comes from the way these in-depth studies challenge scholars to develop new ideas and theories, based on what really happens in negotiations rather than on the logical next step in a series of experiments. Hopefully, this article will serve to encourage others to venture down this path, and provide them with a few more ideas for managing their ethnographic study of negotiations.
References (* included general references for ethnographic research that are not cited) Barry, B. and Friedman, R. (1998) “Bargainer characteristics in distributive and integrative negotiation,” Journal of Personality and Social Psychology, 74(2): 345–359. Barley, Stephen R. (1991) “Contextualizing conflict: Notes on an anthropology of dispute and negotiation.” In Max Bazerman, Roy Lewicki and Blair Sheppard (Eds.) Handbook of Research on Negotiation, v. 3, Greenwich, CT: JAI Press, 165–199. Barley, Stephen R. (1996) “Technicians in the workplace: Ethnographic evidence for bringing work into organization studies.” Administrative Science Quarterly, 41(3): 404–441. Carnevale, Peter J. (1985) “Accountability of group representatives and intergroup relations.” In E.J. Lawler (Ed.), Advances in Group Processes, 227–248, Greenwich, CN: JAI Press. DeDreu, Carsten, Laurie Weingart, and Seungwoo Kwon. (2000) “Influence of social motives on integrative negotiation: A meta-analytic review and test of two theories.” Journal of Personality and Social Psychology. 75(5): 889–905. Friedman, R. (1994). Front Stage, Backstage: The Dramatic Structure of Labor Negotiations. Cambridge: MIT Press. Friedman, R. and Gal, S. (1991) “Managing around roles: Building groups in labor negotiations,” Journal of Applied Behavioral Science, 27(3): 356–378. Friedman, R. and McDaniel, D. (1998) “In the Eye of the Beholder: Ethnography in the Study of Work.” In R. Callus, G. Strauss, and K. Whitfield (Eds.) Researching the World of Work: Strategies, Methods, and Critical Views. Ithaca: ILR Press, 113–126. * Garfinkel, Harold. 1967. Studies in Ethnomethodology. Englewood Cliffs, N.J. Gelfand, M.J. and Realo, A. (1999) “Individualism-collectivism and accountability in intergroup negotiations.” Journal of Applied Psychology, 84(5), 721–736. * Glaser, B. G. and Strauss, A.L. (1967). The Discovery of Grounded Theory. New York: Aldine. Golden-Biddle, K. and Locke, K. (1993) An Appealing work: An investigation of how ethnographic texts convince. Organization Science. 4(4): 595–616. * Hammersley, M. and Atkinson, P. (1995) Ethnography: Principles in Practice. London: Routledge. Hopmann, P.T. (2002) “Negotiating data: Reflections on the qualitative and quantitative analysis of negotiation processes.” International Negotiation, 7:67–85. Kolb, D. (1983) The Mediators. Cambridge, MA: The MIT Press. Merry, S.E. (1989) “Mediation in nonindustrial societies.” Pages 68–90 in Kressel & Pruitt, Eds., Mediation Research: The Process and Effectiveness of Third Party Interventions, San Francisco: Jossey-Bass.
48
RAY FRIEDMAN
Neal, M.A. and Bazerman, M.H. (1985) “The effects of framing and negotiator overconfidence on bargaining behaviors and outcomes.” Academy of Management Journal, 28(1): 34–49. * Schutz, A. (1967) The Phenomenology of the Social World. Evanston: Northwestern University Press. Silbey, S.S. and Merry, S. 1986. “Mediator settlement strategies,” Law and Policy, 8: 7–32. Suttles, G.D. (1968) The Social Order of the Slum, Chicago: University of Chicago Press. Sutton, R. (1994) The Virtues of Closet Qualitative Research. Stanford University: Draft. Schwartzman, H. (1993) Ethnography in Organizations. Newbury Park, CA: Sage. Sondak, H., Neale, M.A., and Pinkley, R.L. (1999) “Relationship, contribution, and resource constraints: Determinants of distributive justice in individual preferences and negotiated agreements.” Group Decision and Negotiation, 8(6): 489–510. * Van Maanen, J. (1988) Tales of the Field. Chicago: The University of Chicago Press. Van Maanen, J. (1983) “The Fact of Fiction in Organizational Ethnography.” Pp. 37–56 in J. Van Maanen, ed., Qualitative Methodology. Beverly Hills, CA: Sage. Walton, R.E. and McKersie, R. (1965) A Behavioral Theory of Labor Negotiations: An Analysis of a Social Interaction System. New York: McGraw-Hill. Whyte, W.F. (1943) Street Corner Society, Chicago: University of Chicago Press.
The Problem-Solving Workshop as a Method of Research RONALD J. FISHER
Introduction The problem-solving workshop is one of the best known methods in the rapidly developing, interdisciplinary field of conflict resolution. Indeed, the problem-solving workshop was one of the first innovations that helped to define and operationalize the field in the 1960s. The genesis of the approach at the intercommunal and international levels is attributed to John Burton and his colleagues at University College London who drew upon social casework, human relations, intergroup problem solving in organizational settings, and other practices of the time (Mitchell 2001). The definition and characteristics of the method were initially captured by Burton (1969), Fisher (1972), and Kelman (1972) among others. Simply put, the problem-solving workshop involves bringing together unofficial yet influential representatives of parties (states or groups) involved in destructive conflict. The parties are brought into intense, face-to-face discussions, facilitated by a third party team, that focus on an analysis of the conflict and possible options for de-escalating and re-solving it. Initially, the approach was seen primarily as a prenegotiation method, which could help prepare the way for productive negotiation. It has since become increasingly clear that problem-solving workshops can make contributions at all stages of conflict resolution, as proposed by Kelman and Cohen (1976). The problem-solving workshop was intended by most proponents as a method of practice, and not as a method of research. However, some of the early writings in the field identified its potential role in learning about the phenomenon of conflict and in testing concepts and models of conflict causation, escalation, and resolution. Subsequently, the problem-solving workshop has also been identified as an action research intervention, and as a component of a wider program of action research on violent and protracted conflict. Unfortunately, the use of the problem-solving workshop as a method of
Author’s Note: The author wishes to thank Herbert Kelman and an anonymous reviewer for helpful comments on an earlier draft of this article. International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 49–59 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
50
RONALD J. FISHER
research has been severely limited, and the potential of the approach in this domain has not been realized. This article will identify the ways that the problem-solving workshop can serve as a research method, and identify the issues that must be addressed to increase its usage and utility in this direction.
The Workshop Method The initial conceptualization by Burton (1969) identified the method as “controlled communication.” This was in order to emphasize the necessity of the third party influencing the interaction of the participants toward an open, analytical discussion of their conflict, leading to increased mutual understanding, followed by the joint creation of ideas for resolution, which would be injected into official negotiations. The label of “problem solving” came in later in order to distinguish the new approach from traditional diplomatic practices and to better capture the nature of the enterprise (de Reuck 1974). The term “problem-solving workshop” has been elucidated most clearly and consistently by Kelman, who served as a third party panel member in one of Burton’s first workshops, and went on to develop his own approach of “interactive problem solving” (Kelman 1972, 1986). My own contribution was to define this method as a form of “third party consultation” in order to emphasize the central role of the intermediary in organizing and facilitating the problem-solving discussions in an impartial, nondirective, noncoercive, and nonevaluative manner (Fisher 1972). Regardless of the label applied, the workshop method evidences a number of essential characteristics (Kelman 1972; Kelman & Cohen 1976, 1986). A small group of individuals from both sides (usually three to six) are invited by a third party team (usually three to five) to engage in low risk, noncommittal, off-the-record discussions over a period of three to five days in a neutral and secluded setting conducive to a relaxed atmosphere and devoid of intrusions. The participants are typically influential individuals in their communities who are not in official policy-making roles, but have access to the political leadership. Some variations involve officials, but in a private, unofficial capacity. The role of the third party is to facilitate the discussions in an impartial manner and to suggest conceptual tools that might be useful to the participants in analyzing their conflict. The objective is to create an informal atmosphere in which participants can freely express their views while respecting those of the other side. The atmosphere also encourages them to move from adversarial debate to a joint analysis of the conflict and the creation of problem solutions that might help address it. Following agreement on ground
THE PROBLEM-SOLVING WORKSHOP AS A METHOD OF RESEARCH
51
rules, the third party provides a rough agenda for the sessions, starting with an initial exchange of perceptions. The third party then moves on to an analysis of the attributions, interests, and needs underlying incompatible positions and escalatory interactions. Next, the participants engage in application and development of insights and models of understanding and the creation of ideas for peacebuilding and resolution. Finally, the third party urges participants to consider the constraints and resistances to these options. In other words, the workshop follows the sequence of problem solving (with recycling as required), which is common in many forms of human endeavor. Casting the workshop as part of a research enterprise with an academic base was part of the initial rationale that invited the participants into the role of conflict analyst and collaborator in problem solving (Burton 1969; Kelman 1972). Subsequent to the initial explications, more detailed and systematic guides have been developed for organizing and facilitating problem-solving workshops (Burton 1987; Mitchell & Banks 1996). What has been largely ignored in the press to develop the theory of practice for conducting workshops is their investigative potential for validating and creating theoretical notions about protracted and violent conflicts, and for operationalizing a form of action research with implications for social change.
A Method for Applying and Generating Theory The academic base of early problem-solving workshops was compatible with the organizers casting the work as a research enterprise, in which the third party would facilitate analytical discussions focusing on the nature of the conflict in question (Burton 1969; Kelman 1972). To assist in the analysis, the third party would suggest concepts, models and research results from the scholarly literature on conflict for the parties to apply to their joint problem, thus hopefully illuminating some of the counterproductive processes that had brought them to their present predicament. Inherent in this analysis would be a field testing of the theoretical notions applied, allowing the third party to judge their validity and scope of applicability. Thus, for example, in Burton’s early Cyprus workshop, the third party injected the conflict spiral model of escalation to help illuminate the development of the conflict. However, one participant questioned its validity in explaining the conflict, because the parties did not possess nuclear weapons, as was the case with the superpower escalation on which the model was developed (Kelman & Cohen 1976). In a later workshop dealing with the same conflict organized by the author (Fisher 1997), the concept of the security dilemma was applied to the conflict with
52
RONALD J. FISHER
the general agreement of the participants that it was helping to fuel the arms buildup on the island. Thus, a generic third party intervention identified by Kelman and Cohen (1976) is the provision of theoretical inputs to enrich the analysis, a function identified as diagnosing the conflict in my model of third party consultation (Fisher 1972). Moreover, one of the objectives in this initial model was the study of conflict, by which the members of the third party team could learn about the nature of escalated and destructive conflict, partly as it is portrayed in the analysis and interaction of the participants. This learning can include the development of new understandings or new applications of concepts from other domains to the study of conflict. For example, in one of the Cyprus workshops I organized, a member of the third party team developed the notion of an identity dilemma, which the Greek and Turkish Cypriot participants appeared to be exhibiting in their discussions. While there was a joint appreciation of the importance of supporting and exhibiting their shared Cypriot identity, movement in this direction put them into potential conflict with their Greek or Turkish identity. Thus, if becoming more Cypriot meant becoming less Greek or Turkish, this left them vulnerable to charges by nationalists in both communities that they were compromising their primary cultural identification. On a grander scale, it appears that the motivation to apply and develop human needs theory, which arose mainly in international development, to the study of conflict occurred because participants in workshops repeatedly revealed such needs when asked to engage in conflict analysis. In other words, when asked to go beneath the surface positions and interests in the conflict, participants would typically move to a discussion of needs for security, recognition of identity, distributive justice, and so on that paralleled the typologies of basic human needs developed in sociology and elsewhere. Thus, the frustration of these fundamental requirements for human development and their related fears (e.g., over a lack of security), came to center stage in the theoretical analyses of protracted or deep-rooted, i.e., intractable, conflicts offered by theorists such as Burton (1990) and Azar (1990). Human needs theory is now regarded as an essential element of conflict theory and a common tool for conflict analysis in problem-solving workshops. Problem-solving workshops have been seen from the beginning as a useful complement to negotiations between conflicting parties, assuming that transfer effects occur between the unofficial and official tracks. Kelman and Cohen (1976) initially outlined the ways in which workshops could make contributions at all stages of negotiation – from encouraging parties to enter negotiations, to working on issues that are hampering negotiations, to discussing long
THE PROBLEM-SOLVING WORKSHOP AS A METHOD OF RESEARCH
53
range issues that need to be addressed once a settlement is reached. Within this context, workshops have been regarded primarily as a prenegotiation supplement that aids parties in making the cognitive shift that moves them into negotiation and helps prepare them for successful negotiations by improving relationship qualities such as understanding and trust (Fisher 1989). The experience of the past forty years is summarized by Kelman (2002): Binding agreements can be achieved only through official negotiations. The very binding character of official negotiations, however, makes it very difficult for certain other things to happen in that context – such as the exploration and discovery of the parties’ basic concerns, their priorities, their limits. This is where problem-solving workshops – precisely because of their nonbinding character – can make a special contribution to the larger process of negotiation and conflict resolution. . . . What we try to facilitate is not the process of negotiation itself but communication that helps the parties overcome the political, emotional, and at times technical barriers that often prevent them from entering into negotiations, from reaching agreement in the course of negotiations, or from changing their relationship after a political agreement has been negotiated (p. 83).
What occurs in problem-solving workshops is partly an analysis of the cognitive and emotional barriers and resistances that hamper parties from entering into negotiations or negotiating effectively when they do so. Black and white mirror images, selective and distorted perceptions, self-serving attributions, blame, hostility, and mistrust lie embedded in the parties’ thinking and their relationship. Surfacing and working through these residues of destructive conflict allows the participants to experience and to communicate with others a different and more complex picture of the other party and the relationship. This revised view then supports negotiation as a preferred strategy in comparison to unilateral attempts to prevail and punish. Therefore, the problemsolving workshop provides a unique and powerful tool for learning about the many resistances parties develop to negotiation as well as about how to reduce these barriers so that cooperative action can occur. While there is some literature on the resistances that hamper negotiation (e.g., Ross & Stillinger 1991), there has been little inductive theorizing based on the problem-solving experience. A valuable exception occurs in the work of Kelman on the IsraeliPalestinian conflict, which has documented the resistances to negotiation in that conflict over time, and proposed strategies and principles for dealing with them, a number of which were represented in the Oslo Accords (Kelman 1978, 1982, 1987).
54
RONALD J. FISHER
Workshops as a Form of Action Research If the problem-solving workshop is cast as a method of social research, it should be that of action research. Pioneered by Kurt Lewin (1948) and others, this research method engages the social scientist as a collaborator with concerned citizens in the process of investigating and addressing a social problem in their environment. Action research is seen as a cyclical process in which information is gathered about the problem situation, a social intervention is planned and carried out to address the problem, and further information is then collected to evaluate the intervention and plan future ones (Fisher 1982). This form of research is carried out in field settings and is directed toward social change, which the researcher and his/her collaborators deem is desirable in order to address the identified problem. To illustrate how the problem-solving workshop serves as a form of action research, I will first draw on my experience in organizing a series of workshops on the Cyprus conflict in the early 1990s through the now defunct Canadian Institute for International Peace and Security (Fisher 1997). CIIPS became interested in the conflict through Canada’s longstanding contribution to peacekeeping on the island, and organized a project consisting of a series of academic style seminars that discussed the many issues that made movement toward a settlement difficult. At the last seminar, interest was shown by influential Cypriots in official roles for a follow-up project that would take a problem-solving approach to the conflict. To assess interest in possible workshops and to initiate a project if interest was shown, I made three site visits to the island and met with a wide range of concerned and influential Cypriots on the two sides of the Green Line. There was generally wide support for such an unofficial, low-risk, exploratory venture, and tacit approval from the two leaderships was forthcoming. Further discussions worked to clarify expectations about the workshop and to develop a written outline that was shared with gatekeepers and potential participants. A “pilot workshop” was held in Canada with leaders of the Greek and Turkish Cypriot communities who maintained close contact with the situation and in some cases had connections to the two leaderships. This workshop served to build the credibility and experience of the third party team and provided some valuable analysis of the conflict. During the later site visits, participants were selected and invited with the assistance of key associates in each community, and the location and timing of the event was decided. The main intervention of the project was a four-day workshop held in the neutral location of England with four influential but unofficial participants from each side working with a four-member third party team. Given the intractable
THE PROBLEM-SOLVING WORKSHOP AS A METHOD OF RESEARCH
55
nature of the dispute, the focus was on the issues that seemed to be sustaining the conflict, both in terms of making it very difficult to resolve and in terms of being resistances in the negotiation process. The agenda also included the underlying needs and fears of the two communities, acknowledgements and assurances that would assist in moving into a shared future, the principles and qualities of a renewed relationship, and the development of peacebuilding activities that would support a shift from peacekeeping to peacemaking. By all accounts, the workshop intervention was a success in terms of both process and outcomes. As a result of the workshop, a series of cross-line meetings and joint committees of business leaders were initiated, a bicommunal art exhibit was launched for a period of several years, and the seeds were sown for the organization of a bicommunal committee to organize peacebuilding activities, including training in conflict resolution (Diamond & Fisher 1995). The third party was established as an acceptable and competent facilitator of intercommunal cooperation, and two further workshops were held before funding dried up in the aftermath of the closure of CIIPS. To evaluate the effects of the workshop as a social intervention, a variety of indictors were used. Closing comments and post-workshop questionnaires evidenced high levels of satisfaction and perceived utility. Participants indicated that they had increased their understanding of the perspectives, rationales, and sensitivities of the other community, and that the third party team had facilitated the interaction and injected new ideas that helped them better understand the difficult issues in the conflict. Pre and post-workshop questionnaires indicated that the increased understanding and respect led to increased involvement in intercommunal contacts. Follow-up interviews one and a half years later demonstrated that participants had maintained positive evaluations of the workshop, and that most had continued their engagement in bicommunal peacebuilding activities. In addition, six of the eight reported that they had conveyed a positive evaluation of the workshop project to their respective leaderships. This outcome was likely critical in garnering continuing support for the workshop project. Conceiving of the workshops as a form of action research was useful in conceptualizing the project and in understanding its place in the ongoing context of the conflict. It also assisted in supporting the importance of evaluation, both formal and informal, at each step of the way, so that future actions could be planned in a collaborative and effective manner. For example, discussions at the main workshop and site visits following it were instrumental in deciding to hold further workshops focusing on the role of education in helping to maintain the conflict and its potential for helping to de-escalate it and rebuild the relationship between the two communities (Fisher 1997).
56
RONALD J. FISHER
The stance taken here is strongly supported by the work of Kelman, who describes problem-solving workshops as the central tool in his program of action research on the Arab-Israeli conflict (Kelman 1979). In this context, the goal of the workshops is to understand and overcome the psychological barriers to change, to create new possibilities for negotiation, and to contribute to reaching a settlement at the political level. According to Kelman: Our work can best be described as an action research program, which combines and integrates efforts at conflict resolution in the Middle East with unique opportunities to observe and learn about the dynamics of the Middle East conflict and of international conflict in general. Our action is designed to contribute to conflict resolution by creating opportunities for communication between influential Arabs and Israelis under the auspices of our group of social scientists (1979, p. 104).
In addition to workshops, the program of action research involves site visits to the region, interviews with a variety of individuals, organizing and attending conferences, and other activities to learn about the conflict and to establish the action researcher role. Although the program involves both action and research, Kelman (1986) notes that the action requirements have taken precedence over the research requirements. In other words, the procedural and ethical necessities of organizing, implementing, and evaluating workshops are given priority, because these ensure the viability of workshops and their contribution to conflict resolution. For example, Kelman precludes manipulating workshop procedures in order to test the effects on participants. Nonetheless, the research aspect is critical to the enterprise. As Kelman observes: . . . it is our role as researchers that provides the rationale and legitimacy of our action involvement and that allows representatives of conflicting parties to interact with each other under auspices in ways that deviate from the norms generally governing their relationship . . . Thus our action requires involvement in a research program just as our research requires involvement in an action program. The unique advantage of this type of research is that it provides us the opportunity to make rich, detailed observations of ongoing processes of conflict and conflict resolution, which would not be accessible to us unless we were engaged in an action program (1986, p. 297).
The outcome of this form of action research for Kelman and his colleagues is that it produces learning that support negotiations in a variety of ways and at all stages of the negotiation process. For example, in 1986, Kelman noted that Israeli and Palestinian participants had gained insight into each others’ perspectives (concerns, areas of flexibility, constraints, priorities), had learned that there was someone to negotiate with, and had learned of positive changes in the adversary and how they could promote further change through their
THE PROBLEM-SOLVING WORKSHOP AS A METHOD OF RESEARCH
57
own actions. These kinds of lessons are clear contributors to the prenegotiation process that helped pave the way to the Oslo accords (Kelman 1995).
Concluding Issues It is unfortunate that the problem-solving workshop is not used more systematically as a method of research to study the complex phenomenon of social conflict. As the field has developed, the predominant institutional base of conflict resolution work has shifted from academia to non-governmental organizations (Fisher, in press). While this shift entails advantages in terms of greater flexibility and variety of funding sources, it does tend to support practitioners who have less interest and expertise in the application and induction of theoretical ideas about social conflict. Thus, the use of the method to study conflict is compromised, although some practitioners demonstrate commendable scholarship in inducing theories of practice. Saunders and his colleagues, for example, have produced a very useful model of the problem-solving stages of sustained dialogue, based on the experience of the Dartmouth Conference task force on regional conflicts (Chufrin & Saunders 1993; Saunders 1999). More recently, Saunders and Slim (2002) have articulated a detailed description of the role of the third party moderator who organizes and leads such sessions. Unfortunately, such work does not substitute for helping to build a theory of understanding about social conflict based on the cumulative experience of problem-solving workshops. Casting the problem-solving workshop as an action research intervention highlights the inadequate attention that is typically devoted to the evaluation of such interventions (Fisher, in press; Rouhana 2000). Most documentation of workshops is anecdotal and self-serving, and more complex research techniques beyond case study description are typically not applied to evaluate the process or the outcomes of workshops. Practitioners surely must attempt to develop an awareness of the effectiveness of their work, based on feedback from participants and others involved in the conflict, but the application of more objective and comprehensive measures is woefully absent. A fuller application of the action research model would encourage third party intervenors to use a variety of indicators to gauge the effectiveness of their work. There are of course many practical and ethical reasons for constraining the intrusion of evaluation procedures on workshop practice, but with more creativity and investment, the action research model could be better implemented. Such developments would allow the workshop method to develop its full potential as a method of research in addition to one of practice.
58
RONALD J. FISHER
References Azar, Edward E. (1990). The Management of Protracted Social Conflict. Hampshire, UK: Dartmouth Publishing. Burton, John W. (1969). Conflict and Communication: The Use of Controlled Communication in International Relations. London: MacMillan. Burton, John W. (1987). Resolving Deep-Rooted Conflict: A Handbook. Lanham, MD: University Press of America. Burton, John W. (ed.) (1990). Conflict: Human Needs Theory. New York: St. Martin’s Press. Chufrin, Gennady I. and Saunders, Harold H. (1993). “A Public Peace Process.” Negotiation Journal 9, 2:155–177. de Reuck, Anthony V.S. (1974). “Controlled Communication: Rationale and Dynamics.” The Human Context 6, 1:64–80. Diamond, Louise and Fisher, Ronald J. (1995). “Integrating Conflict Resolution Training and Consultation: A Cyprus Example.” Negotiation Journal 11, 3: 287–301. Doob, Leonard W. (ed.) (1970). Resolving Conflict in Africa: The Fermeda Workshop. New Haven, CT: Yale University Press. Fisher, Ronald J. (1972). “Third Party Consultation: A Method for the Study and Resolution of Conflict.” Journal of Conflict Resolution 16, 1:67–94. Fisher, Ronald J. (1982). Social Psychology: An Applied Approach. New York: St. Martin’s Press. Fisher, Ronald J. (1989). “Prenegotiation Problem-Solving Discussions: Enhancing the Potential for Successful Negotiation.” In Janice G. Stein, editor, Getting to the Table: The Process of International Prenegotiation. Baltimore, MD: Johns Hopkins University Press. Fisher, Ronald J. (1997). Interactive Conflict Resolution. Syracuse, NY: Syracuse University Press. Fisher, Ronald J. (In Press). “Interactive Conflict Resolution.” In I. William Zartman, editor, Peacemaking in International Conflict (rev. ed.). Washington, DC: United States Institute of Peace. Kelman, Herbert C. (1972). “The Problem-Solving Workshop in Conflict Resolution.” In Robert L. Merritt, editor, Communication in International Politics. Urbana: University of Illinois Press. Kelman, Herbert C. (1978). “Israelis and Palestinians: Psychological Prerequisites for Mutual Acceptance.” International Security 3, 1:162–186. Kelman, Herbert C. (1979). “An Interactional Approach to Conflict Resolution and its Application to Israeli-Palestinian Relations.” International Interactions 6, 2:99–122. Kelman, Herbert C. (1982). “Creating the Conditions for Israeli-Palestinian Negotiations.” Journal of Conflict Resolution 26, 1:39–75. Kelman, Herbert C. (1986). “Interactive Problem Solving: A Social-Psychological Approach to Conflict Resolution.” In William Klassen, editor, Dialogue Toward Inter-Faith Understanding. Jerusalem: Ecu-menical Institute for Theological Research. Kelman, Herbert C. (1987). “The Political Psychology of the Israeli-Palestinian Conflict: How Can We Overcome the Barriers to a Negotiated Solution?” Political Psychology 8, 3:347– 363. Kelman, Herbert C. (1995). “Contributions of an Unofficial Conflict Resolution Effort to the Israeli-Palestinian Breakthrough.” Negotiation Journal 11, 1:19–27. Kelman, Herbert C. (2000). “The Role of the Scholar-Practitioner in International Conflict Resolution.” International Studies Perspectives 1, 3: 273–288.
THE PROBLEM-SOLVING WORKSHOP AS A METHOD OF RESEARCH
59
Kelman, Herbert C. (2002). “Interactive Problem Solving as a Tool for Second Track Diplomacy.” In John Davies and Edy Kaufman, editors, Second Track/Citizens’ Diplomacy: Concepts and Techniques for Conflict Transformation. Lanham, MD: Rowman & Littlefield. Kelman, Herbert C. and Cohen, Stephen P. (1976). “The Problem-Solving Workshop: A SocialPsychological Contribution to the Resolution of International Conflict.” Journal of Peace Research 13, 2:79–90. Kelman, Herbert C. and Cohen, Stephen P. (1986). “Resolution of International Conflict: An Interactional Approach.” In Stephen Worchel and William G. Austin, editors, Psychology of Intergroup Relations (2nd ed.). Chicago, IL: Nelson-Hall. Lewin, Kurt (1948). Resolving Social Conflicts. New York: Harper. Mitchell, Christopher (2001). “From Controlled Communication to Problem Solving: The Origins of Facilitated Conflict Resolution.” International Journal of Peace Studies 6, 1:59–67. Mitchell, Christopher and Banks, Michael (1996). Handbook of Conflict Resolution: The Analytical Problem-Solving Approach. New York/London: Pinter. Ross, Lee and Stillinger, Constance (1991). “Barriers to Conflict Resolution.” Negotiation Journal 7, 4:389–404. Rouhana, Nadim N. (2000). “Interactive Conflict Resolution: Issues in Theory, Methodology, and Evaluation.” In Paul C. Stern and Daniel Druckman, editors, International Conflict Resolution after the Cold War. Washington, DC: National Academy Press. Saunders, Harold H. (1999). A Public Peace Process: Sustained Dialogue to Transform Racial and Ethnic Conflicts. New York: St. Martin’s Press. Saunders, Harold H. and Slim, Randa M. (2002). “Moderating Sustained Dialogue in Tajikistan.” Paper presented at Conducting Dialogues for Peace: A Symposium on Best Practices, United States Institute of Peace, Washington, DC, November. Walton, Richard E. (1970). “A Problem-Solving Workshop on Border Conflicts in Eastern Africa.” Journal of Applied Behavioral Science 6, 5:453–489.
Time-Series Designs and Analyses DANIEL DRUCKMAN
Negotiation is a dynamic process that can be analyzed as a time-series of events, decisions, or moves that occur through the course of a defined time period. The changes take the form of trends that can be charted for different parties making offers and counter-offers in a negotiation or for different nations responding to each other’s policy initiatives. The family of time-series techniques allows for a variety of types of negotiation process analyses include that: describing patterns of reciprocity among parties, forecasting concessions or decisions, evaluating impacts of mediation and other conflictsettlement interventions, estimating probabilities of future events such as the likelihood of coups or wars, changes in policies corresponding to historical time periods, and paths taken by negotiators or dialogue participants from initial conditions through a stream of decisions or actions leading to outcomes. These examples are discussed in this article. Each application illustrates how time-series analysis contributes in important ways to our understanding of conflict management or resolution processes and outcomes by capturing the dynamics of unfolding processes. The time series usually consists of a sequence of events accumulated over a relatively long period of time in the context of a case. It has several features including, (a) the analysis of variation in chronological events that occur within cases referred to as diachronic variance, (b) a focus on trends that may reveal patterns or shapes of change, (c) the opportunity to compare trends for two or more cases with the same or a different number of data points, and (d) the frequent use of regression and correlational statistics, taking into account the correlations that exist among the data points themselves, referred to as autocorrelation. This type of analysis is similar in many ways to measuring subjects’ responses over time in experiments. A difference, however, is that within-subject responses (or repeated measures) are part of an experimental design; unlike case time-series, these data are analyzed as part of the error-term in the analysis of variance. Several types of time-series designs are presented in this article, including examples of both quantitative and qualitative analyses. I focus attention primarily on the way the analysis was done in the context of the overall study International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 61–78 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
62
DANIEL DRUCKMAN
design. For technical details on calculations for many of these techniques, see the chapters on forecasting in Frei and Ruloff (1989). The first section shows how time-series data can be used for post-diction with correlation and regression techniques. The next section presents the idea of interrupted time-series analysis. The following section on Bayesian approaches to analysis includes examples of the way that the calculations are done. The final section provides examples of several qualitative time-series analyses of conflict-resolving processes. Each of these applications addresses a research problem in the analysis of negotiation and related interactions. Together, the examples call attention to the range of applications afforded by time-series techniques.
Post-dicting Events with Time Series Data Perhaps the most popular application of time-series techniques is to predict outcomes retrospectively. For negotiation analysis, the value of this approach is to understand the processes that preceded – or led to – a known outcome. Several examples of this type of analysis are presented in this section. When an analyst is interested in predicting known outcomes in completed or historical cases, he or she is doing post-diction, which is sometimes referred to as retro-diction. This type of analysis is performed for several reasons. One is that an investigator is interested in learning about the antecedents of particular events such as crises. Another is that an analyst desires to recreate a path of events leading to an outcome. A third reason is to evaluate the plausibility of alternative explanations or theories for the occurrence of an event. For each of these reasons, it is advantageous to know – or to have control over – the outcome. The question of interest is, what process or pattern unfolded prior to the event of interest? This question can be addressed from either an inductive or a deductive perspective. An example of each approach to post-diction – on the same research issue – is presented. A negotiation occurred in 1975 between the governments of Spain and the United States to renew the military base rights granted by Spain to the U.S. during the previous five-year period. These talks were regarded by U.S. policymakers to be routine. Previous agreements provided the framework for the terms of a new agreement. However, the negotiation process defied these expectations. It was not at all smooth. Spain’s demands were extraordinary – including a demand to be admitted to NATO – and the U.S. delegation was not prepared for them. A bilateral agreement was reached in 1976, a year and a half after the talks began. The impasse during this period was punctuated with several crises including walk-outs by the delegations. Three research questions
TIME-SERIES DESIGNS AND ANALYSES
63
were asked: Was there a pattern in the events that preceded the crises? How were the impasses resolved? Was there a pattern in the events that occurred following impasse resolution? Time-series analysis was used to address the first and third questions. Statements made by both delegations during the sessions were coded into categories reflecting hard and soft negotiating rhetoric. The categories were from a coding system known as bargaining process analysis (BPA). Designed originally by Walcott and Hopmann (1978), the BPA is intended to facilitate analyses of verbal behavior in international negotiation. Ratios indicating the percentage of the total statements made by a delegation during a session (total of 22) within a round (total of 10) were calculated and summed for an aggregate score for each delegation by session. The aggregate scores were then plotted in two dimensions, percent hard statements (plotted on the vertical dimension as the sum of retractions, commitments, threats, and accusations) and sessions (the horizontal dimension). The resulting trends provided a time series for each delegation. By placing each of the six crisis sessions on the graph, it was possible to analyze patterns of verbal behavior before and after the event. Correlational analyses revealed distinct patterns of behavior before and after the crises. Succinctly stated, the pattern took the form of a gap between the delegations in percent hard during the session before the crisis. The gap was closed by the “softer” delegation. It increased the number of hard statements, matching the “harder” delegation, resulting in mutual hardness and an impasse. This pattern was discovered through inductive analysis rather than as a test of an hypothesis. It became known as comparative responsiveness and formed the basis for further research with multiple cases to be discussed below. It also corroborated a pattern of reciprocity found in earlier bargaining research with children (Druckman & Bonoma 1976). (The base-rights study also investigated the way these crises were resolved and the patterns of behavior that followed resolution; for details see Druckman 1986.) A companion study reported by Stoll and McAndrew (1986) investigated patterns of responsiveness through ten years (1969–1979) of talks between the Soviet Union and the U.S. on strategic arms limitations (SALT). Performed with a deductive approach, these investigators evaluated the fit of three alternative models of reciprocity, referred to as directional, trend, and comparative reciprocity. Each of these models predicts the reactions to concessions made by the other negotiating team. The directional model specifies that negotiators will either match (a concession for a concession) or mismatch (a retraction in response to a concession) the other’s move. The trend model indicates that negotiators will either increase or decrease their concession in response to a
64
DANIEL DRUCKMAN
previous increase or decrease by their opponent. The comparative model posits that negotiators will compare their own with the other’s moves made in the previous round, increasing or decreasing their concessions in response to a higher or lower concession made by the opponent in the previous round. This study illustrates a use of time-series analysis for evaluating different assumptions about the way information is processed and used during negotiation. Several time series were used to assess these models. These included the round-by-round moves made by the Soviet and US delegations during SALT I (November 1969–May 1972), at Vladivostok (June 1972–November 1974), in SALT II (January 1977–June 1979), and across the entire set of rounds from 1969 to 1979. The results showed that the comparative model provided the best fit to the concession data: In technical terms, the highest correlations between the expected and actual responses from one round to the next occurred for the comparative model in 5 of the 8 analyses (four data sets for each national delegation). These results are consistent with those obtained in the base-rights study. A key difference between the studies, however, is between an inductive discovery (Druckman 1986) and a deductive evaluation of alternative hypotheses (Stoll & McAndrew 1986). The deductive time-series work was extended to an evaluation of additional models with more cases of international negotiation. In this study, ten models of reciprocity were evaluated with time series of round-by-round moves in seven negotiations. The directional model was compared to five versions of the trend and four versions of the comparative model: The five trend models consisted of different lags (varying in terms of the number of previous rounds included) and weightings for more recent and more distant rounds; the four comparative models consisted of different combinations of the comparative and trend models. The negotiation cases varied in terms of the length of the time series from 8 rounds (the Mutual and Balanced Force Reductions between NATO and the Warsaw Pact) to 22 rounds (post-war disarmament talks between the Soviet Union and the U.S.). The actual moves made by the delegations from one round to the next were correlated with the moves made in the same or earlier rounds by the opposing delegation. The data used for the calculations of correlations were specified by each model: for example, response to the immediately previous concession for the directional model, an increase or decrease in concession from t – 2 (two rounds earlier) to t – 1 (one round earlier) is correlated with the change in the other’s concession from t – 1 to t for the trend model. Goodness-of-fit for the various models was evaluated by the relative size of the correlations. The strongest correlations occurred for the comparative model (9 of 14 correlations were significant). Thus, the comparative model produced a better fit to the data than the 9 alternative models, including the various
TIME-SERIES DESIGNS AND ANALYSES
65
combinations of comparative and trend models. These time-series findings corroborate and attest to the generality of the results obtained in the two 1986 studies described above. They tell us that professional negotiators are attentive and react to any difference that exists between the moves made by their own and the other delegation. For most of the negotiating delegations in this sample, the reactions follow a comparison of moves made in the immediately previous round. (For details, see Druckman & Harris 1990.) Another example of post-diction illustrates how multiple and partial correlations are used together to detect how national representatives respond to each other in a multilateral negotiation over conventional troop reductions. The time series consisted of eight rounds of statements made by each of seven national delegations. The delegations were divided into two alliances, a Western and Eastern bloc. The statements were coded according to the BPA categories. These statistical techniques allow an analyst to gauge more precisely the influence between any pair or trio of nations controlling for the effects of any other national delegation or delegations in the talks. We discovered, for example, that the effect of moves made by the leader of one alliance had a stronger impact on the moves made by the leader of the other alliance in the next round when the influence of a key ally in the former alliance was removed. This ally attenuated the lagged correlation (from moves made in round t – 1 to those made in round t) between the alliance leaders. In technical terms, a bivariate lagged correlation of .47 increased to .63 when the ally’s influence was controlled through the use of a multiple-partial correlation. The importance of these sorts of analyses is that they parse out influences between pairs of nations in highly interdependent interactive systems like multilateral negotiations. They allow an analyst both to infer causation through the use of lagged correlations and to reduce spuriousness of relationships with partial correlations. (For more on the use of these techniques for inferring causation, see Blalock 1960.) A slight variant on the theme of post-diction is interrupted time series analyses. This approach adds an element of “control” to the time series. By doing so, an analyst can make a somewhat stronger case for causal inference. An example of a recent analysis serves to illustrate this feature in the next section.
Interrupted Time-Series Analysis The variables used in the correlational analyses above are measured as changes in postures or concessions over time. The computations illustrate the use of correlational techniques, including ways of attempting to infer
66
DANIEL DRUCKMAN
causation. Another way of evaluating a time series is to examine the impact of decisions or events that occur at a particular juncture in a sequence of moves made by the parties. For example, an external event may occur abruptly at the beginning of a middle round of the talks. Assuming that the event is attended to by the parties, it interrupts the course of the negotiation. The question of interest is, what impact does the event have on the negotiation process? It is answered by comparing the trend of concessions prior to the event with those that occurred following it. The comparison is evaluated with significance tests rather than with correlational techniques. A data set of events collected during a six-year period of the conflict between Armenia and Azerbaijan over the territory known as Nagorno-Karabakh provides an example of the way an interrupted time-series design is implemented. The case provides an opportunity to examine whether various attempts to mediate a serious conflict were effective. Hypotheses about timing and readiness were examined by the investigators. A total of 3,856 events spread over the six-year period were coded on a scale ranging from most peaceful (+3) to most violent (–3). Average monthly scores were calculated for analyses of trends before and after each of six attempts to mediate the conflict. In addition, the impact of another event, intensive warfare occurring between April 1993 and February 1994, was examined by comparing the average violence scores six months before and six months after the event. Thus, seven types of interruptions or interventions were analyzed; six mediations and the military combat. This approach to analysis follows the logic of experimentation rather than modeling or regression. Regarding both the mediations and combat as “treatments,” similar to independent variables, the conflicting parties’ behavior is compared from before to after the treatments. The data set then consists of dependent variables – averaged violence scores calculated on a scale ranging from +3 (action taken toward peace) to –3 (violent actions) – evaluated in a sequence of beforeand-after quasi-experimental designs without control groups (see Cook & Campbell 1979). Since the mediations and combat are treatments, rather than dependent variables, neither the mediators’ behavior nor the combat casualties during the period of warfare were included in the events data set. Significance tests were used to assess the changes from before to after an intervention or event (mediation, combat) in a manner similar to the way data would be analyzed from a laboratory experiment. A difference however is that it is possible to establish control groups in laboratory experiments. The results showed that changes in violent behavior, assessed by the various before-to-after comparisons, occurred only as a result of the combat.
TIME-SERIES DESIGNS AND ANALYSES
67
None of the six mediations was effective in changing the parties’ behavior. The mediators did not create opportunities for settlements by reducing the violence and bringing about the conditions for a ceasefire. On the other hand, the combat did provide the condition for reduced violence and a ceasefire agreement. Escalation proceeded to the large-scale use of military force, resulting in a mutually-hurting stalemate followed by a period of de-escalation (Zartman 2000). Reasons for this outcome are discussed in Mooradian and Druckman (1999). Although this study provided insights into a question that has both theoretical and practical implications, it is a weak experimental design. The differences found between the mediated interventions and the combat suggest that the latter had a stronger impact on violent behavior than the former. It does not suggest that escalation (combat) caused de-escalation (reduced violence). A causal interpretation of these findings would be more plausible if a comparable case without the interruptions (mediations, combat) was included in the design; for example, de-escalation occurs only following combat in timeseries that compare cases with and without combat that includes casualties. By limiting the analysis to the single case of Nagorno-Karabakh, these sorts of inferences are unwarranted. By extending the analysis to another case, they become more plausible. This extension is made in the context of focused or matched case comparisons. (See, for example, Faure 1994 for discussion of the approach.) The idea that interruptions are treatments suggests that they have certain characteristics. These include being relatively abrupt, although not necessarily sudden, relevant to an ongoing process, attended to by the parties involved in the process, and hypothesized to have an impact on that process. Other kinds of interruptions may consist of worker strikes, government or company collapse, a natural disaster such as an earthquake, the insertion of a peacekeeping force, a summit conference among national leaders, an influx of foreign economic aid, or an entirely new insight into the source of a conflict. The key point made in this section is that planned or unplanned events can interrupt an ongoing process, turning it towards or away from agreements. The impacts of the interruptions can be evaluated with before and after quasiexperimental designs, allowing causal inferences to be made. This discussion concludes our treatment of quantitative retrospective analyses. We turn now to an approach that examines the way changing situations can alter the estimates of probabilities for future (unknown) events.
68
DANIEL DRUCKMAN
Bayesian Analysis Another type of analysis that captures change over time, invented by Reverend Thomas Bayes in the eighteenth century, is based on assumptions about how the past and the present can be used to estimate probabilities about the occurrence of future events. Bayes’ theory of probability is described in an essay published posthumously in a 1764 issue of the Philosophical Transactions of the Royal Society of London. It was sent to the Royal Society by Richard Price who wrote: I now send you an essay which I have found among the papers of our deceased friend Mr. Bayes, and which, in my opinion, has great merit. In an introduction which he has writ to this Essay, he says, that his design at first in thinking on the subject of it was, to find out a method by which we might judge concerning the probability that an event has to happen, in given circumstances, upon supposition that we know nothing concerning it but that, under the same circumstances, it has happened a certain number of times, and failed a certain number of times (for more on Bayes see the web page, “Bayes Ways to Gaze into the Future . . . and how it pays” (at www.vma.bme.hu/mathhis/Mathematicians/ Bayes.html).
The analysis consists of evaluating alternative hypotheses about the occurrence (Hi) or non-occurrence (Hj) of an event, for example, whether or not a coup will occur within a defined period such as one year. The first estimates consist of initial probabilities that a coup will (p[H1]) or will not occur (p[H2]). These probabilities can be estimated from statistics on coups that have occurred in the country over the last 30 years. The next estimate is referred to as a conditional probability and consists of the chances that an event (coup or no coup) will occur if a particular symptom is present (p[EHi]). Examples of symptoms related to coups, which are listed in Frei and Ruloff (1989), are cabinet reshuffling (.4 that coup will occur, .3 that it will not occur if there is reshuffling), call for free elections (.6 that a coup will occur, .5 that it will not occur), and purges (.1 and .8 respectively). The estimates for symptoms often come from experts or from expert panels. When the conditional probabilities are taken into account, a revised initial probability results (p[HiE]) as calculated by the following formula: P(HiE) = P(Hi).P(EHi)/ S (P[Hi],P(EHi])
By inserting the initial (.25, .75) and conditional probabilities for reshuffling (.4, .3) into this formula and performing the calculations, the result is .31 that a coup will occur in the next year and .69 that it will not occur. Note that the initial probability for a coup, based on past evidence, increases somewhat when the symptom of reshuffling is taken into account (from .25 to .31).
TIME-SERIES DESIGNS AND ANALYSES
69
Similar calculations can be made for each of the other symptoms, free elections and purges. An example more closely related to negotiation is the chance that a peace agreement will hold through time. Suppose that past experience suggests that peace agreements have held for at least five years in one-third of the cases examined; thus, p(Hi) is .33 and p(Hj) is .67. Suppose also that an expert panel decided that there are five factors that affect the longevity of agreements. They assigned conditional probabilities to the factors as follows: factions of spoilers (.1 supporting Hi; .6 supporting Hj), schisms within the parties (.2 and .4), regime continuity (.5 and .3), international pressure (.7 and .2), and availability of arms (.2 and .5). The effect of the symptom referred to as spoilers is to reduce the chances of sustaining the agreement from .33 (initial probability) to .08 (revised probability). Schisms reduce the chances for longevity from .33 to .21. Regime continuity increases the chances from .33 to .46. International pressure has the strongest positive impact on longevity, from .33 to .63. And, arms availability reduces the chances from .33 to .17. When the conditional probabilities of all five symptoms are combined in the formula the revised probability is .30, roughly the same as the initial probability of .33; apparently the symptoms are off-setting with two favoring longevity – regime continuity and international pressure – and three jeopardizing the agreement – spoilers, schisms, and availability of arms. Bayesian analysis is an example of time-series analysis that focuses on changes in the probabilities of events (revised probabilities) due to circumstances. It can also be understood as an alternative to regression for forecasting events. The beta weights that derive from a regression analysis are estimates of the relative importance of variables – which may be symptoms in a Bayesian analysis – in predicting a dependent variable (an event in Bayesian terms). When used for forecasting, the regression equation is a linear extrapolation of past trends, considered as the initial probabilities or “priors” in a Bayesian analysis. Unlike regression, Bayesian forecasts are updated probabilities for events that take both past (initial probabilities or “priors”) and present (conditional probabilities of symptoms) into account. The updating can be continuous as a result of monitoring situations on a regular basis. Expert judgments can also be refined as new information becomes available to analysts. Regression is backward-looking in the sense of relying on a past which cannot be changed, although memories of the past are often adjusted. Bayesian analysis is an attempt to capture present events or processes, focusing attention on changes which have implications for forecasts. Despite its appeal, Bayesian techniques have rarely been used in addressing problems in negotiation or conflict resolution.
70
DANIEL DRUCKMAN
A key point made in this section is that it is important to monitor situations in order to detect possible changes that have implications for future events or states. Bayesian techniques provide a systematic framework, based on probabilities, for performing these kinds of analyses. They take into account the ongoing situational changes ignored by techniques that rely only on the past for forecasting the future. This concludes our discussion of quantitative timeseries analysis. We turn now to a variety of qualitative analyses performed to depict changes more difficult to capture with numerical codes.
Qualitative Time-Series Analysis Qualitative approaches can provide insight into patterns of interaction over time that are difficult to measure with precision. An example is Lepgold and Shambaugh’s (1998) analysis of Sino-American relations from 1969 to 1997. These researchers attempted to explain the causes and consequences of reciprocal exchanges between the two nations during this period. Their typology of four distinct patterns of exchanges was used to depict changes in the bilateral relationship. The types were defined in terms of expectations about time horizons and the degree to which national decision-makers anticipate a reliable stream of benefits from the relationship. Various time periods within the three decades were depicted in terms of one of four cells: short or long time horizons and low or high expectations of benefits. The results showed that when American and Chinese perceptions of their time horizons and the likelihood of reliable streams of benefits diverged, the country that had a longer time horizon or perceived more benefits from the relationship was able to drag out negotiations until a preferable outcome had been achieved. These variables are also shown to influence bargaining strategies, for example whether the parties engage in specific reciprocity (exchanges are precisely matched in terms of equivalence or contingency), diffuse reciprocity (exchanges are not precisely matched), or a mixed strategy. The time series in this study consists of the three decades of interactions between these governments. The analysis consists of distinguishing periods within the time series in terms of the four-fold typology and related reciprocity strategies. It sheds light on the sources and consequences of various types of expectations by each party over time. Leng’s (1998) analysis of 12 recurring militarized crises between post World War II rivals (U.S. and Soviet Union, India and Pakistan, Israel and Egypt) provides an example of a mixed qualitative-quantitative time series design. The qualitative analyses consisted of characterizing each of four crises within
TIME-SERIES DESIGNS AND ANALYSES
71
each rivalry in terms of the dominant influence strategy used by the two nations. The strategies were bullying, reciprocating, trial and error, and stonewalling. For example, the four crises in the U.S.-Soviet relationship with corresponding strategies and outcomes are as follows: Berlin Blockade, 1948–49 (Soviet bullying, U.S. reciprocating leading to a U.S. victory); Berlin Wall, 1961 (Soviet bullying, U.S. reciprocating leading to a stalemate); Cuban Missile, 1962 (Soviet trial and error, U.S. bullying leading to a U.S. victory), and Middle East Alert, 1973 (Soviet trail and error, U.S. reciprocating leading to a U.S. victory). The other two rivalries relied primarily on bullying strategies leading usually to wars or, in one crisis over Kashmir, stalemate. Based on these patterns, Leng concluded that reciprocating strategies are the most effective means of promoting cooperation in militarized crises. These findings are consistent with quantitative time-series analyses, calculated as weekly average hostility scores during the period of the crisis, also performed by the author on these three rivalries, as well as with the author’s earlier findings based on a larger sample of crises occurring between 1816 and 1980 (Leng 1993). These are particularly interesting findings in light of the selection of cases. They challenge a prevailing assumption that the most effective means of prevailing in interstate militarized disputes is through the use of escalating coercion. They also reinforce the value of combined analyses for bolstering confidence in findings obtained by only one approach. Process Tracing A qualitative analogue to time-series analysis is known as process tracing. This is a technique that consists of searching an historical record of events for evidence about whether a postulated process did or did not occur. The record is usually, but not always, developed from bounded cases such as the periods of crisis used in Leng’s analysis or from negotiation or mediation interactions. The analysis is facilitated when the choice of events is guided by a framework as in Druckman’s (2001) study of turning points. In that study, a path of events that occurred in each of 34 cases of negotiation was developed in terms of a three-part framework: precipitants, departures, and consequences. Precipitants are regarded as causes of changes in the negotiation process; they can be external or internal to the process. An example of an external precipitant in the talks over strategic arms limitations (SALT I) is when the Soviet Union’s nuclear arsenal approached parity with the United States. Departures are changes that occur in the process; they may be relatively abrupt or non-abrupt and are regarded as turning points. In reaction to the Soviet nuclear arsenal development (precipitant), domestic pressure increased
72
DANIEL DRUCKMAN
for a bilateral agreement to suspend further weapons development. This was a rather abrupt departure from a public that placed little pressure on the US administration to seek an agreement. Consequences can be escalatory, leading away from agreements, or de-escalatory, leading toward agreements. The deescalatory consequence of the departure in this case was that bilateral talks between these nations began, leading to a treaty in 1972. The path is depicted in the following form: SALT I: external precipitant → abrupt departure → de-escalatory consequence
Another example of process tracing comes from the global environmental talks leading to the signing of the Montreal Protocol in 1987. The procedural or internal precipitant was a succession in the presidency of the conference from the United Kingdom to Belgium. This led to a departure in process taking the form of European Community support for a U.S. plan. The consequence was an agreement on banning the production of certain types of CFCs. The path takes the following form: Montreal Protocol: procedural precipitant → non-abrupt departure → de-escalatory consequence
These kinds of paths were constructed for each sampled case of international negotiation in three issue areas, security cases (13), trade cases (10), and political/environmental cases (11). For most cases, paths were constructed for early, middle, and late time periods in the talks; several turning points occurred in most cases. The case paths were then aggregated for each type of negotiation, security, trade, and political, resulting in a “typical” path. A typical security cases’ path, which includes longer-term consequences (at time t + 1) takes the following form: Security cases: external precipitants → abrupt departure in the process → deescalatory consequences at time t → de-escalatory consequences at time t +1.
Another application of process tracing comes from my research on situational levers of flexibility. Using experimental techniques, an attempt was made to identity the situational variables that had the strongest impact on flexibility in each of four stages of a simulated environmental conference on ozone depletion. Paired-comparison judgments made by the negotiators provided weights for each of several aspects of the situations manipulated in the experiment. (Guilford [1954] provides a description of the paired-comparison procedure in chapter 7.) The paths consisted of the aspect (or variable) with the largest weight. Separate paths were constructed for the sessions that produced an
TIME-SERIES DESIGNS AND ANALYSES
73
agreement and those that concluded with an impasse. An example of an agreement path is as follows: Friendly relations (pre-negotiation stage) → peripheral location for the talks (setting-the-stage) → limited media coverage (bargaining stage) → limited media coverage (endgame stage).
An example of a stalemate path is as follows: Being the delegation’s primary representative (pre-negotiation stage) → central location for the talks (setting-the-stage) → wide media coverage (bargaining stage) → wide media coverage (endgame stage).
Note that the critical variables in these paths are location and media coverage. They appear in stages two, three, and four of both paths. The variables identified in each stage influence a process toward an outcome. Together, these influences are considered to be trajectories (or levers) toward eventual agreement or stalemate. (See Druckman 1993; and Druckman & Druckman 1996, for details.) The negotiation or related process can be thought about as a kind of backward tracing from a particular event or outcome through a series of earlier decisions or events arranged chronologically. This can be done with documented cases by developing a chronology of events. Examples of chronologies can be found in the Pew-sponsored case studies on international affairs (1999); a detailed chronology of a negotiation that occurred in 1986 between the Philippines Aquino regime and an insurgent group is shown in Druckman and Green (1995). It can also be done with observational data by developing a category system for verbal statements and nonverbal behavior as these occur through the course of an interaction. Carstarphen’s (2003) analysis of a dialogue about racial prejudice provides an illustration. The dialogue occurs among a group of men from different backgrounds. It is presented in the wellknown documentary film, “The Color of Fear.” She used these interactions to investigate shifts that occur at the individual, relational, and group levels of analysis. The focal event of interest was a shift in relationships at the group level. Moving backward from this outcome, she captured the unfolding group dynamics at each of several points in time. Those dynamics take the form of the following path, culminating in the focal event. Lee Mun Wah’s question to David (D): ‘What keeps you from believing?’ → D’s lack of understanding produces an impasse in the discussion → facilitator intervention → D reflects on his own experience and resistance to change → D’s reframing → D’s new self awareness → D’s acknowledgement of his own prejudices and that racism exists → other members’ changed
74
DANIEL DRUCKMAN
attitudes toward D → African-American member acknowledges D’s acknowledgement → group shift in D’s relationship with the other members (focal event). This path illuminates the group’s reaction to a “deviant” member, the role played by the facilitator, change at both emotional and cognitive levels, and the role of acknowledgement in changing relationships within the group. It consists of sequential observations or a chain of events, including a conflictresolution intervention, that lead to a particular outcome. It is a more complex, process-oriented, time series of causation than many correlational analyses involving only a few variables. (For more on the methodology of process tracing see George and Bennett [2004] and Stern and Druckman [2000].) Another example of process tracing comes from Pruitt’s (2004) analysis of the negotiation process between Israel and the PLO in Oslo. He traced the process through four stages by depicting a chain of communication among various official and unofficial actors. Stage I depicts the chain before the talks began as the mediator, Larson, got both sides to the table. It looks like this: I. Arafat ← → Mediator ← → Israeli Professors ← → Israeli diplomat
Stage II is the situation that obtained during the first five sessions. The Israeli Foreign Minister, Peres, was added at one end of the chain and the mediator was dropped in the middle. The mediator sat outside the meeting room ready to provide assistance without becoming involved in the discussions. II. Arafat ← → PLO delegates ← ← → Israeli Foreign Minister
→ Israeli Professors ← → Israeli diplomats
Stage III is the situation that obtained in the last seven sessions, when Israeli diplomats took over from the professors, who stayed on as advisors. III. Arafat ← → PLO delegates ← Foreign Minister ← → Rabin
→ Israeli diplomats ← → Israeli
Stage IV is a final telephone conversation between Arafat and Peres, in which the details of the agreement were hammered out. They did not however talk directly but through interpreters. IV. Arafat ←
→ Israeli Foreign Minister ← → Rabin
The progression from Stage I to IV shows how intermediaries drop out of a chain as optimism grows. The progression from Stages I to III shows how unofficial (Track II) diplomacy involving the mediator and professors, was replaced by official (Track I) diplomacy, involving government officials on both sides.
TIME-SERIES DESIGNS AND ANALYSES
75
The process tracings illustrated here have several features in common. Each depicts a linear process through chronological time. They all proceed toward an outcome that serves to conclude a process. And, each involves a relatively small number of actors or parties. Of interest is the question whether the technique can be used to depict non-linear interactions among many actors in situations where the process does not conclude with a single outcome. An example is the problem of coordination among different intervenors (or NGOs) working to reduce conflict in the same region. Data collected by Allen Nan (1999) on the activities of many organizations operating in three former Soviet republics make evident the difficulty of constructing paths to outcomes. For this problem, multiple paths involving various types of intervenors and interventions converged and diverged toward or away from reduced levels of conflict: There were too many actors, activities, and varieties of coordination and complementarity to discern a path proceeding from initial to end states. This sort of complexity is captured better by frameworks that show the interplay among the actors, activities, and processes, including recursive relationships between them. (See, for example, Allen Nan 1999: Figure 8.8.1.) Both process tracing and framework construction are examples of ways to envision information to enhance understanding of social processes. A variety of qualitative time-series analyses were illustrated in this section. They included the use of typologies for depicting relationships and influence strategies used at different points in time, tracing of processes that occur in simulated and actual negotiations, and extended interactions between multiple parties or organizations through the course of staged peace processes or nonlinear activities intended to address ongoing conflicts. All of these examples illuminate the value of using frameworks or category systems for longitudinal analysis. They capture changes through time that would be missed by crosssectional (same point in time) comparisons and complement the quantitative analyses discussed earlier. This concludes our treatment of the family of timeseries approaches. We turn now to a summary of the applications discussed and an idea for combining several of the techniques.
Conclusions Several types of time-series designs were discussed in this article. When a researcher is interested in explaining an historical outcome, he or she may analyze the course of events or processes leading to that outcome. The question of interest is whether the known outcome can be “predicted” from information about those processes. Another purpose served by these analyses is to
76
DANIEL DRUCKMAN
evaluate alternative explanations for a process that occurred in historical cases. The availability of detailed information about these processes enables a researcher to code the events or moves made by the actors. Examples from correlational studies of negotiation show how these analyses are done. Sophisticated analyses that use lagged, partial, and multiple correlations provide a basis for inferring causation from time-sequenced events. Time-series data are shown also to be used in quasi (or field-) experimental evaluations of interventions. Impacts of interventions such as mediation can be assessed with before-and-after comparisons of trends. Although these are often weak designs, especially because they do not include control groups, they have the advantage of focussing an analyst’s attention on events that can alter a process in important ways. The process-altering features of particular events are also the focus of Bayesian analysis. This form of forecasting entails monitoring situations on a regular basis. It highlights the role played by indicators of key events such as coups, wars, or the unraveling of a peace agreement. Like regression, it takes the past into account. Like interrupted time series, it pays attention to key events. Unlike both of those approaches, however, Bayesian analysis revises probabilities based on past events with information from indicators or “symptoms” in the present that may signal a possible future event. Alternatives to the precise coding needed to perform the quantitative analyses of time series data are qualitative approaches. These approaches are particularly useful for analyzing complex interactions among many parties or settings where detailed information is not available. Theoretically-inspired typologies or frameworks replace coding systems as analytical tools. They have been shown to distinguish among patterns of exchange between nations, bargaining and influence strategies, and the kinds of events and decisions that precede turning points in negotiation or shifts in group processes as well as the consequences of those changes. When it is feasible to depict events in the form of a traced path in chronological time, these analyses suggest causal processes involving many more elements than is usually measured by correlation analyses. It would seem that these advantages of qualitative time-series analysis complement those of the quantitative techniques. Further analytical specificity can be achieved by combining the various time-series techniques within a study. With regard to a particular negotiation, the analytical challenge is to forecast an event – such as tabling a key proposal – that can alter the process (Bayesian techniques), evaluate the impact of the proposal on that process (interrupted time series), and, then devise a path of the events and decisions that led to its occurrence or to an outcome that emanated from it (process tracing). This multi-method analysis leads to
TIME-SERIES DESIGNS AND ANALYSES
77
conclusions about the likelihood that the event will occur, its impact on a process, and the way it emerged from a sequence of prior events. These are the facets of analysis that contribute to the goal of explaining the way that negotiation processes unfold toward outcomes. (See Irner 2003 for an example of a multi-method analysis of negotiation process-outcome relationships.)
References Allen Nan, S. (1999). “Complementarity and coordination of conflict resolution efforts in the conflicts over Abkhasia, South Ossetia, and Transdniestria.” Unpublished doctoral dissertation, George Mason University. Blalock, H.M. (1960). Social Statistics. New York: McGraw-Hill. Carstarphen, N.M. (2003). “Shift happens: Transformations during small group interactions in protracted social conflicts.” Unpublished doctoral dissertation, George Mason University. Cook, T., and Campbell, D.T. (1979). Quasi-Experimentation: Designs and Analysis Issues for Social Research in Field Settings. Boston: Houghton Mifflin. Druckman, D. (2001). “Turning points in international negotiation: A comparative analysis,” Journal of Conflict Resolution 45:519–544. Druckman, D. (1993). “The situational levers of negotiating flexibility,” Journal of Conflict Resolution 37:236–276. Druckman, D. (1986). “Stages, turning points, and crises: Negotiating military base rights, Spain and the United States,” Journal of Conflict Resolution 30:327–360. Druckman, D., and Bonoma, T.V. (1976). “Determinants of bargaining behavior in a bilateral monopoly situation II: Opponent’s concession rate and similarity,” Behavioral Science 21:252–262. Druckman D., and Druckman, J.N. (1996). “Visibility and negotiating flexibility,” Journal of Social Psychology 136:117–120. Druckman D., and Green, J. (1995). “Playing two games: Internal negotiations in the Philippines.” In I.W. Zartman (Ed.) Elusive Peace. Washington, DC: Brookings. Druckman, D., and Harris, R. (1990). “Alternative models of responsiveness in international negotiation,” Journal of Conflict Resolution 34:234–251. Faure, A.M. (1994). “Some methodological problems in comparative politics,” Journal of Theoretical Politics 6:307–322. Frei, D., and Ruloff, D. (1989). Handbook of Foreign Policy Analysis. Boston, MA: Martinus Nijhoff. George, A.L., and Bennett, A. (2004). Case Study and Theory Development. Cambridge, MA: MIT Press. Guilford, J.P. (1954). Psychometric Methods. New York: McGraw-Hill. Irner, C.G. (2003). The promise of process: Evidence on ending violent international conflict. Unpublished doctoral dissertation, GeorgeMason University. Leng, R.J. (1998). “Reciprocity in recurring crises,” International Negotiation 3:197–226. Leng, R.J. (1993). Interstate Crisis Behavior, 1816–1980: Realism versus Reciprocity. Cambridge: Cambridge University Press. Lepgold, J., and Shambaugh, G. (1998). “Rethinking the notion of reciprocal exchange in international negotiation: Sino-American relations, 1969–1997,” International Negotiation 3:227–252.
78
DANIEL DRUCKMAN
Mooradian, M., and Druckman, D. (1999). “Hurting stalemate or mediation?: The conflict over Nagorno-Karabakh, 1990–95.” Journal of Peace Research 36:709–727. Pew Case Studies in International Affairs (1999). “The ISD compendium of case study abstracts and indexes,” the Institute for the Study of Diplomacy. Edmund A. Walsh School of Foreign Service, Georgetown University. Pruitt, D.G. (2004). “Escalation, readiness for negotiation, and third party functions.” In I.W. Zartman and G.O. Faure (Eds.) Escalation and Negotiation. Cambridge, UK: Cambridge University Press. Stern, P.C., and Druckman, D. (2000) “Evaluating interventions in history: The case of international conflict resolution,” International Studies Review 2:33–63. Stoll, R.J., and McAndrew, W. (1986). “Negotiating strategic arms control, 1969–1979.” Journal of Conflict Resolution 30:315–326. Walcott, C., and Hopmann P.T. (1978). “Interaction analysis and negotiating behavior.” In R.T. Golembiewski (Ed.) The Small Group in Political Science. Athens, GA: The University of Georgia Press. Zartman, I.W. (2000). “Ripeness: The Hurting Stalemate and Beyond.” In P.C. Stern and D. Druckman (Eds.) International Conflict Resolution after the Cold War. Washington, DC: National Academy Press.
Social Research and the Study of Mediation: Designing and Implementing Systematic Archival Research JACOB BERCOVITCH
Introduction How exactly do we explain politically significant events? How best to make sense of complex patterns of behavior in international relations? Scholars of political science have wrestled with these issues and debated the merits of case studies versus statistical studies, experimental studies versus field studies for many years now. The debate has been as passionate as it was lively and acrimonious. Each side has held to its opinions with remarkable ferocity. Each has believed that the path to truth was with their adherents only. Each was convinced of the irrelevance of the other. Nowhere was this schism more apparent than in the study of conflict resolution in general and mediation in particular. When it came to conflict resolution there was a widespread feeling that only by gaining an intimate and personal knowledge about a conflict, can one say anything meaningful about its management. Conflict was seen as so aberrant a phenomenon that only an in-depth study of a selected case could be expected to yield any insights. Any attempt to go beyond the single case and the discursive, qualitative method was to be frowned upon as an example of scholarly irresponsibility. Nowhere was this logic more evident than in the study of negotiation and mediation. The result of such a belief was that, despite its popularity, longevity, ubiquity, and importance, we seem to know far less about mediation than we imagine, and what we know is mostly of an ideographic nature. Systematic, archival analyses of mediation cases in the real world have been all too rare. Mediation has for too long been studied through single cases or quasirealistic frames. Little wonder that the phenomenon remained so poorly understood for so long. The difficulties and obstacles of this form of research meant that scholars often looked for answers where “there was some light” (e.g., experimental studies) rather than where the real phenomenon of mediation was buried. I strongly believe that we should not succumb to the allure of obtaining easy yet not always realistic data. We should instead work with International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 79–92 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
80
JACOB BERCOVITCH
reports of real events, by real mediators and negotiators, in real conflict situations, and use that data in a systematic fashion to develop useful theories, and answer real questions on the process and outcomes of mediation. To do otherwise would be to mistake our wishful thinking for a complex reality. Mediation is practiced in a variety of fields, ranging from domestic disputes through legal conflicts and community disputes to international conflicts. Thus, mediation may be studied by sociologists, psychologists, political scientists, anthropologists, economists, and others. These scholars borrow concepts and methods from each other. Whichever approach or discipline one adopts, the study of mediation is something of a research puzzle for which we must find less than perfect answers through the analysis of data. Some of the questions that make up the mediation puzzle include the following: • When should a mediator enter a dispute? Should one wait till a certain threshold of violence has been crossed, or is it better to intervene before the parties engage in openly hostile behavior? • Should a mediator be, neutral (could a mediator be neutral)? Would a biased mediator be acceptable to the parties under some conditions? Would a biased mediator be successful? • What exactly do mediators do when they intervene in a dispute? Are there some recognizable patterns of behavior, and are such patterns associated with any measure of effectiveness? How do mediators even choose one strategy over another? Do they have an identifiable decision calculus? • Which disputes are more amenable to mediation? • What factors or conditions affect the process and outcome of mediation? Can any of these be changed so as to help mediation? • Are some mediators more likely to achieve their objectives than others? Are factors such as personal attributes, rank, position and prestige related at all to mediation outcomes, and if so how? • Do mediated outcomes endure? Do they satisfy the parties’ basic needs, or do they merely provide an interlude for the parties to regroup, marshal resources and engage in further aggression?
Experimental and Case Study Approaches How best can we organize our research efforts, answer these questions, and resolve the puzzle? Several approaches may be identified. A significant part of the literature dealing with mediation and purporting to answer these questions has been largely descriptive, relying mostly on single cases. The best exemplars of this approach (e.g., Princen 1992; Touval 1982; Touval and
SOCIAL RESEARCH AND THE STUDY OF MEDIATION
81
Zartman 1985) used analytical frameworks and examined their chosen cases with great care. However, it was, and is, difficult to draw general inferences from these studies, no matter how carefully crafted they may have been. Descriptive approaches make good sense of idiosyncratic situations, they encourage policy makers and practitioners to publish their memoirs, but they do have too wide a diversity of focal points, and it is impossible to generalize from them or assess their validity. All that we can know is what happened in the specific case and why mediation failed or succeeded. To answer the questions above, we need to treat mediation as a class of public events, not as a unique set of circumstances. An approach that has attempted to answer broad questions and deal with issues of causes, effects, and outcomes in a large number of cases is the experimental approach. Experimental approaches have universal rather than particular goals. Experimental approaches, frequently conducted in laboratory settings, test for causal relations. They permit one to manipulate values of independent variables and determine how each affects the dependent variable. Above all, they offer the possibility of replication and reinterpretation. Some of the best studies in this tradition (e.g., Brookmire & Sistrunk 1980; Carnevale 1986; Carnevale et al. 1989; Kochan & Jick 1978; Pruitt and Johnson 1970. For a fuller review see Rubin 1980) exemplify how a phenomenon can be studied scientifically, and done so in ingenious ways using exciting research designs and challenge some of our basic notions about the process. Experimental approaches represent a systematic way of collecting information about the world. We usually know how the data is recorded and generated. We know the hypotheses tested and the causal inferences made. We know that when more data is needed, further experiments may be undertaken to improve data quality. What we do not know is how we can overcome the sense of artificiality that inevitably characterizes these studies. In other words, the data is real, but the real world is so very different form the data. To conduct an experiment with undergraduate psychology students in any university on the validity of threats under conditions of multilateral deterrence may tell us very little indeed about the behavior of policy-makers in hostile countries with nuclear weapons at their disposal. There is a considerable leap of faith required of us if we are to presume that the micro processes observed by social psychologists, under quite well controlled situations, resemble the stressful, often chaotic, and usually unpredictable political environment of international relations. Experimental approaches offer the prospect of excellent data collection methods, but their reliability and external (not internal) validity is at best doubtful.
82
JACOB BERCOVITCH
Archival Data Analysis; Some Possible Benefits The reality of negotiation and mediation is made up of numerous particular events; particular negotiators, particular individual mediators, particular conflict situations and so forth. What we are trying to do is to go beyond the particulars of each case, describe these events systematically, search for the best data or information, and offer knowledge that can help us achieve a better understanding of some general acts pertaining to human behavior in conflict. As political scientists we are trying to study real events, be aware of the relationship between them, be as precise and realistic as possible in describing them, and hopefully be in a position to say something meaningful about the class of events that is of interest to us. To do so, I strongly believe we should rely only on available data, both public and private, systematically collected, as the only form of data that give us the ability to get as close as possible to a first hand account by the parties directly involved. In comparison with available archival or observational data, other forms of data seem manufactured at best. Available data may be in several categories. It may be in the form of public and official records and documents, newspaper accounts, private memoirs, or social science data archives. Studying puzzles by relying on available, usually archival, data is the most unobtrusive, non reactive way of getting information. This form of research is particularly suited to the study of mediation. It has a number of advantages of other ways of colleting and interpreting data: • Archival research emphasizes naturalness of the setting, behavior, and event. Archival sources provide researchers with the best, perhaps the only, way of studying past occurrences. There is simply no substitute for naturalness in the design of research. • Archival data enable a researcher to study contextual information and features of larger social units (experiments of field studies rarely study the group as the unit of analysis). • Archival data is spontaneous. By this I mean it was produced without subjects or participants being requested to do so. • Archival sources provide high quality information that can be easily summarized and expressed in terms of different values or variables. • Archival sources permit us to conduct research over long spans of time. The analysis of archival data is well suited to studies of trends, change, and traditions. Archival research permits systematic studies of processes. • The use of archival data may permit us to study large samples of cases. Using archival sources one may study thousands of families, hundreds of conflicts and numerous mediators. The larger the sample of cases, the greater our confidence in our findings and interpretations.
SOCIAL RESEARCH AND THE STUDY OF MEDIATION
83
• Archival research is quickly available and easily accessible, thus saving much on cost, time, and personnel in research. I believe the systematic use of archival data offers the best opportunity to study mediation in the real world. Archival research offers a flexible and powerful form of research with the strongest measure of external validity. In the sections below I hope to show how I approach the puzzle of mediation and try and answer the basic questions in the field by working with a very large data set of mediation events.
Describing the Data Set Before embarking on archival research it is important that we have strong theoretical framework and a proper research design. What I propose to do is to examine all cases of international conflict since 1945, and all the cases of international mediation in these conflicts. The project is an extension of my early work on theories of international mediation. With this project I use extensive archival sources, take the theory as my starting point and engage in a thorough data search and empirical analysis. The project is not predicated on some inductive grounds of empiricism, but on a theoretical understanding of mediation and other forms of conflict management mechanisms. I am guided throughout by concepts, ideas and theories, even though much of the work is empirical in nature. Prompted by dissatisfaction with previous studies on mediation which have rested mostly on ideographic, normative, or experimental approaches, this project was established with the aim of utilizing archival data to demonstrate the soundness of our theoretical framework. The theoretical underpinning of the project are to be found in the so called contingency framework developed by Bercovitch (1986). At its simplest, the framework is a dynamic, timedependent theory that regards the outcome of mediation or negotiations as dependent, or contingent, upon a number of contextual, process and personal variables. These variables have different values and the manner of their interaction affects how mediation is conducted, and how successful it is going to be. I believe we can now answer many of the questions about mediation and move closer toward solving the puzzle of mediation.
Defining the Relevant Population The first challenge in undertaking archival research is to identify the population of cases to be studied. To start with, we examined existing sources of
84
JACOB BERCOVITCH
information on international conflicts, (e.g., Small & Singer 1982: Butterworth 1976: Tillema 1991; Wallensteen 2002). International conflicts are relatively rare events, but most of these publicly available sources had different numbers of conflicts since 1945. Where each project uses different dimensions to define conflict, different thresholds (that is, number of fatalities incurred before an event can be described as an international conflict), and focuses on different categories of conflict (e.g., within or between states), neither comparability nor generalizations are possible. An approach to data collection on international conflict which does not define conflict primarily in terms of violence, and is more concerned with conflict termination than with its causes, was clearly needed. We decided, at the very outset of the project, to dispense with the notion of a threshold of fatalities. A situation could be one of conflict even in the absence of 100 or 1000 combat fatalities. We define conflict operationally as a state of organized hostilities and exchange of threats between two or more states which may or may not result in violent beahvior and fatalities. With this definition in mid, we scanned numerous archival sources (e.g. existing data, newspapers accounts, World Factbooks, etc.) and identified 309 conflicts from 1945 to 1995 (see Bercovitch & Jackson 1997). Each conflict was studied for its history, duration, issues at stake, intensity, and manner of termination. Information about the parties in each conflict (e.g., nature of political system, power resources, etc.) was also collected. Our interest of course has been with the occurrence, trends and characteristics of mediation events in the real world. Very little systematic information was ever collected on mediation events. Such information that does exist contains very few numbers of cases (e.g., Northedge & Donelan 1971), or deals mostly with mediation and peacekeeping by the United Nations (e.g., Zacher 1979). Here there was a real need to collect information from scratch. Having thus identified our population of conflicts, we studied each and every one of them in terms of the conflict management methods used in these conflicts and the outcomes of each method. Each conflict management method was studied for such dimensions as timing, manner of performance, and impact. Of the 309 conflicts, 190 experienced some form of formal mediation. Some conflicts, such as the one in the Middle East or the conflict in the former Yugoslavia (186 different mediation efforts from 1989–1995) experienced many mediation efforts by different mediators. Other conflicts had very few, if any, mediation efforts. In these 190 conflicts we had a total of 2194 actual international mediation cases, (this does not include offers of mediation offers that were rejected). A total of 559 different mediators were involved in these
SOCIAL RESEARCH AND THE STUDY OF MEDIATION
85
efforts. These ranged from individuals to groups, to regional organization, officials of a state or representatives of international organizations. Each mediation effort was studied in terms of many independent variables to capture the features of their identity, composition, resources, strategy etc. Table 1 below gives some idea of the range and distribution of conflict management events in international conflict. Table 1. Conflict management activities; 1945–1995 No conflict management activity in 61 out of 309 international conflicts (20%) Conflict Management Strategy Mediation Negotiation Arbitration Referral to an International or Regional Organization Multilateral Conference Total Number of Conflict Management Strategies
Number of times used 2194 (58.7%) 1249 (33.4%) 29 (0.8%) 166 (4.4%) 38 (1.0%) 3737 (100%)
Information on each conflict management activity was collected from four main archival sources dealing with events data; The New York Times Index, The Times Index (London), Reuter’s Online News Service, and Keesing’s Archives (renamed Keesing’s Record of World Events.) These news sources are widely regarded as offering the best and most comprehensive information on political events since 1945. Our quest for reliable archival data was supplemented by searching other public documents such as personal memoirs of leading politicians and mediators, organizational publications on mediation, and other data sources.
Coding the Data Once we have identified our archival sources and information, we need to delineate which precise aspects are of interest to us. Basically, we collected data that pertained to six main dimensions of conflict and its management:
86
JACOB BERCOVITCH
• start and end dates of the mediation attempt; • details of those present at the attempt (e.g., names of disputing parties – negotiators, names of third parties – mediators, rank and organizational position held); • the location of the attempt (e.g., neutral territory, in disputed territory or a disputant’s territory); • strategies perceived to be in use by the mediator (e.g., communication-facilitation, procedural or directive strategies); • the outcome of each mediation attempt (e.g., agreements, cease-fires, full settlements and the durability of such outcomes); and, • numerous contextual observations (e.g., presence of ongoing hostilities, system era, dispute issues and the disputant’s political environment). All these aspects have been coded into a total of 219 independent variables (see Table 2 below on how the independent variables were organized) thus providing a us with an original and much needed resource for analyzing all aspects of the process and outcomes of international mediation. Objectivity has been maintained, as much as possible, by comparing data from one source with several other sources. As Bercovitch and Houston explain, “data were collected and cross checked over four main sources and between coders and achieved an intercoder reliability across the variables of 95%” (Bercovitch & Houston 2000: 185). The vast majority of variables coded are, by nature, objective and gathered from factual observations e.g., dates, times, attributed organizational ranks and calculated disputant power ratios. Probably the three most criticized variables are mediator strategy, conflict issues, and outcome (success or failure). While inter-coder reliability brings consistency to the overall coding of the data, these variables may require further evaluation. Table 2 . Categories and Listing of Independent Variables Characteristics of the Dispute
Characteristics of the Parties
Conflict Management Variables
Dispute Start/End Date Duration
Party Initiator in Previous Disputes Identity Party A & B
Start/End Date Type, Negotiation, Mediation, Referral to INO, Arbitration/ adjudication, multilateral conference
Fatalities
Party Alignment
Third Party Identity
Dispute Intensity
Member of UN – Party A & B
Mediator Rank
Highest Action
Power A & B (GDP, Military budget etc.)
Strategies-Primary
Hostility Level
Parties Previous Relations
Supplementary Strategies
SOCIAL RESEARCH AND THE STUDY OF MEDIATION
87
Table 2 (cont.) Characteristics of the Dispute
Characteristics of the Parties
Conflict Management Variables
Reciprocity
History Disputes with Other Parties
Previous Relationship of Parties With Mediator
System Period Dispute Start
Party Political System, Openness/Democracy, Closedness/Autocracy, Polity Type and Change
Previous Attempts Med/Neg this Dispute
Geographic Region
Leadership change during the dispute
Previous Attempts this Mediator
Geographical Proximity of Parties
Number Parties Supporting the Disputants
Timing (grouped)
Issue One
Characteristics of Previous Disputes between the parties – Initiator, Issues, Balance of Power, Highest Hostility Level, Outcome, Temporal Proximity to previous dispute, United Nations Operations
Timing (raw)
Issue Two
Dispute Phase Conflict Management
Issue Three
Initiated by
Final Outcome
Environment
Re-emergence of Dispute
Outcome
Type of Conflict
Hostilities Reported
Number Mediation Efforts
Durability Outcome
UN Involvement
Rank Negotiator Party A
Type of Superpower Involvement
Rank Negotiator Party B
UN Peacekeeping/ Sanctions/Embargos
Number of Mediators Acting
Enduring Rivalries
Functional Mediator Identity
Characteristics of Ethnic Conflict – Ethnic Groups, Ethnic War, Revolutionary War, Abrupt/Disruptive Regime Change, Genocide/ Politicide, Ethnic Issues, Origin and Internationalization of Ethnic Rivalries
Mediation Type
Dispute Initiator
Mediator Experience
Type of Power Initiator
UN Mediator
88
JACOB BERCOVITCH
The original intention of the project was to build up as complete a data-set of mediation in international conflicts as is possible, using many archival resources. Subsequently the project sought to encompass not just mediation but other forms of conflict management such as: negotiation, arbitration, multilateral conferences, and referrals to international organizations. Information was collected on (a) factors relating to the dispute, parties and issues in conflict, (b) factors relating to mediation performance and identity, and (c) factors describing the consequences or outcomes of mediation. Initially the project begun in the mid 1980’s with an examination of 72 international conflicts and 210 cases of mediation (see Bercovitch 1986). Today the project contains significant information on 309 international conflicts and 3737 cases of conflict management.
Some Findings Making use of available data, we now have close to one million observations on mediation and other forms of conflict management over a fifty year period. Acquiring knowledge about any form of social relations may not be easy. Acquiring knowledge about a process that is usually conducted behind closed doors, where stakes and loss of face may be high, is particularly difficult. In trying to understand the reality of mediation, we do, I believe, encounter fewer difficulties in constructing and validating hypotheses with this form of research than any other form. We are less likely to err with our inferences and interpretations when our knowledge is grounded in incidents and records of real human behavior. A good research design and systematic use of archival data (which after all are explicit and public) maximize internal and external validity and our ability to demonstrate real patterns, replicate them, or refute them. Better data are almost always a precursor to better theories. As social scientists we usually have to make the best of what data we have. In the case of mediation and negotiation, I doubt we have better data than that which may be obtained through archival research. Once colleted and codified, the data are arranged in the form of a large matrix, with rows containing information about our units of analysis, mediation events, and the columns with different bits of information about each of the independent variables. At this stage we can engage in a variety of quantitative or qualitative procedures to examine distribution of the data, relationships between two or more variables, and develop causal inferences and models, and answer the questions or puzzles posed above. Some of the most important findings of the project are highlighted below:
SOCIAL RESEARCH AND THE STUDY OF MEDIATION
89
Nature of Conflict Many studies posit a relationship between a given conflict and mediation outcomes. We found that conflict intensity impacts adversely on mediation effectiveness (Bercovitch 1986). High fatalities and high intensity conflicts are not really amenable to mediation. Issues in conflict showed a marked effect on mediation, with tangible issues such as resources and territory showing a much higher likelihood of successful mediation than intangible issues, such as identity or sovereignty (Bercovitch & Houston 1996). Conflict complexity, examined as numbers of issues per conflict, also showed a strong relationship with ineffective mediation. Nature of the Parties In a series of studies we were able to show that who the parties are has a great impact on how a conflict is likely to be mediated. Democratic parties are more likely to use mediation and achieve a positive outcome than nondemocratic parties (Bercovitch & Houston 1996). Similar patterns were found between economically developed countries and mediation. Domestic observance of human rights and respect for liberty correlate very highly with successful mediation, as does the level of internal homogeneity (Bercovitch & Houston 2000). Parties of equal power resources or small states make the most of mediation services offered to them (Bercovitch 1986). Mediation is not a method suited for conflicts involving superpowers. Nature of Mediation and Mediator In a series of studies we were able to show that mediator neutrality, that old chestnut in the literature, is far less important than mediator ability, experience and resources (Bercovitch & Houston 2000). We were also able to show that the best time to initiate mediation is some months after a conflict had manifested itself (Bercovitch 1986). The most prominent mediators in international conflicts are not regional or international organizations, but states, and larger states at that. The most successful mediators are not private individuals but high ranking mediators (Bercovitch & Schneider 2000). A neutral environment is highly conducive to successful mediation, as is the use of an active intervention strategy (Bercovitch & Houston 2000). This kind of archival analysis, based on a sound theoretical design, in which the data has been carefully and consistently checked, where we deal with a large number of case studies, and where we try and offer as complete a description of each case by examining as many explanatory variables as we
90
JACOB BERCOVITCH
can, helps us considerably in trying to resolve the puzzles of mediation. Archival research, supplemented by detailed case studies, provides a sound recipe for empirical research into the conditions and manifestations of mediation.
Conclusion The maturity of an academic discipline is based on the rigor of its findings, the number of its practitioners, its capacity to impact, and its acceptance of different paradigms and research methods. Conflict and its resolution encompass an array of methods, findings, and paradigms that are not easily grouped into a single set of beliefs or theories. At times we may be critical of each other’s ideas, approaches, or findings, but we can still engage in a creative dialogue, and learn much from our diverse perspectives. Such is the strength and vibrancy of our discipline. With this in mind, my objective here was to argue, not from dogma but from disciplined thought, the advantages of analyzing mediation and negotiation by using available, public information. The sources of such information may include vital national statistics, UN documents, data archives, newspapers, letters, and survey reports. Archival research is best suited to the analysis of large and complex structures. It permits us to identify a whole range of mediation events and specify all the factors and dimensions which might affect its manifestation and outcomes. Archival research is also cumulative. Much of what is reported above is built upon foundations laid by other scholars. Each succeeding phase of research builds a larger data set, analyzes events using ever more sophisticated computer programs, and increases our understanding of how mediation works and how to improve it. The study of conflict resolution in general and mediation in particular is an intellectual endeavor that tries to describe and understand some complex events in the real world. As a field of study, mediation has developed over the last three decades diverse philosophical underpinnings and different frameworks of analysis and methods, all designed to capture the range, complexity and diversity of mediation. Yet every scholar, method, or approach labors under some limitations of knowledge, insights, or mistakes. No person, or team of scholars, nor any approach can explain or control the numerous factors involved in mediation, ranging from individual variables to systemic considerations. This is why we have to see the many approaches not as competing but as complementary paradigms, looking at the same problem from different vantage points. Archival research tells us much about the structure of mediation. To understand something about the parties’ decisional calculus and choices of behavior we need to pose different questions and use different
SOCIAL RESEARCH AND THE STUDY OF MEDIATION
91
methods. All research contains elements of compromise and creative intuition. It is time scholars from different traditions look seriously for that second dimension of research. The present volume makes a step in that direction.
Acknowledgements I would like to thank Judith Fretter and Robert Trappl for all their help. This article has benefited considerably from the comments of Peter Carnevale and an anonymous reviewer.
References Bercovitch, J. (1986). “International Mediation: A Study of Incidence, Strategies and Conditions of Successful Outcomes,” Cooperation and Conflict 21(3):155–168. Bercovitch J., and Houston A. (1996). “The Study of International Mediation: Theoretical Issues and Empirical Evidence.” In J. Bercovitch (Ed.) Resolving International Conflicts. Boulder. Westview Press. Bercovitch, J., and Jackson, R. (1997). International Conflict: Encyclopedia of Conflicts and their Management, 1945–1995. Washington, D.C.: Congressional Quarterly. Bercovitch, J., and Houston, A. (2000). “Why Do They Do It Like This? An Analysis of the Factors Influencing Mediation Behavior in International Conflicts,” Journal of Conflict Resolution 44:170–202. Bercovitch J., and Schneider, G. (2000). “Who Mediates: The Political Economy of International Conflict Management,” Journal of Peace Research 37:145–165. Brookmire, D., and Sistrunk, F. (1980). “The effects of perceived ability and impartiality of mediator and time pressure on negotiation,” Journal of Conflict Resolution 24:311–27. Butterworth, R.L. (1976). Managing Interstate Conflict, 1945–1974. University Center for International Studies, University of Pittsburgh, PA, USA. Carnevale, P. (1986). “Strategic choice in mediation,” Negotiation Journal 2:41–56. Carnevale, P.J., Conlon, D., Hanisch, K., and Harris, K. (1989). “Experimental research on the strategic choice model of mediation.” In K. Kressel and D.G. Pruitt (Eds.) Mediation Research: The Process and Effectiveness of Third Party Intervention. San Francisco: JosseyBass, Inc. Kochan, T.A., and Jick, T. (1978). “A theory of public sector mediation process,” Journal of Conflict Resolution 22:209–240. Nortehdge, F., and Donelan, M. (1971). International Disputes: Political Aspects. London: Europa Publications. Pruitt, D.G., and Johnson, D.F. (1970). “Mediation as an aid to facesaving in negotiation,” Journal of Personality and Social Psychology 14:239–246. Princen, T. (1992). Intermediaries in International Conflict. New Jersey: Princeton University Press. Rubin, J.Z. (1980). “Experimental research on third party intervention in conflict,” Psychological Bulletin 87:379–391. Small, M., and David Singer, J.D. (1982). Resort to Arms Beverly Hills: Sage.
92
JACOB BERCOVITCH
Tillema, H.K. (1991). International Armed Conflict Since 1945. Boulder: Westview Press. Touval, S. (1982). The peace brokers – Mediators in the Arab-Israeli conflict 1948–1979. New Jersey: Princeton University Press. Touval, S., and I.W. Zartman, (1985). (Eds.) International Mediation in Theory and Practice. Boulder: Westview Press. Wallensteen, P. (2002). Understanding Conflict Resolution. London: Sage. Zacher, M. (1979). International Conflict and Collective Security. New York: Praeger.
Reflections on Simulation and Experimentation in the Study of Negotiation1 JONATHAN WILKENFELD
For almost 20 years, the International Communications and Negotiations Simulations (ICONS) project has been at the forefront of delivering network-based distributed foreign policy-oriented simulations in a variety of educational settings (Starkey & Blake 2001; Starkey & Wilkenfeld 1996; Wilkenfeld & Kaufman 1993).2 Those of us involved in the development of that simulation approach were almost totally absorbed with the technical and pedagogical issues, and it was many years before we began to think of the use of this type of simulation for pure research purposes (exceptions to this included the work of Torney-Purta [1992, 1996, 1998]). Building upon experience with the ICONS Project, my recent work has increasingly involved the use of simulation and experimental approaches for the study of foreign policy decision-making in general, and negotiation and mediation in particular. It is my adventures, and in many cases misadventures, in this new realm that I want to focus on in this article. I begin with some background on experimentation in political science in general, and in international politics in particular. This will be followed by the story of how I became involved with this approach in the first place. Finally, in order to examine what experimentation can and cannot do for us, I will present some preliminary findings that extend our knowledge on one particular phenomenon in international politics, the process of mediation in international crises.
Experimental Design in Political Science While experimentation remains largely outside the mainstream of international relations research, and of political science in general, efforts have been made in recent years to bolster the acceptance of experiment-based research designs within the field. Kinder and Palfrey (1993: 1) offer a treatise on why experimentation should be integrated into political science as a way to “supplement, not replace, traditional empirical methods.” What is it that experiments offer political scientists? Experimentation allows researchers to isolate a crucial International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 93–103 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
94
JONATHAN WILKENFELD
independent variable by providing the means to control extraneous factors. It allows analysts to look at the effect of that variable by comparing the results of those who were exposed to the variable with the results of an unexposed control group (Campbell & Stanley 1963). Experimentation provides researchers a precise view of a certain phenomenon, a certain relationship. It is this quality that makes experimentation valuable: It is the most reliable means of examining and understanding the nature of a causal relationship, of demonstrating the internal validity of a theory in political science (Laponce 1972; Bositis 1990; McGraw 1993). Experiment-based research provides insights into relationships that are difficult, if not impossible to observe, in any other way. Snyder (1963: 7) argues that the insights and depth of understanding which experimentation can uniquely provide make it an important heuristic tool in the “discovery phase of sciencebuilding.” Because the results of experiments can enlighten analysts’ understanding of how and why a relationship exists and functions, this type of research allows for the refinement of general or preliminary theories in a way distinct from examinations of large-n (quantitative) or small-n (qualitative) empirical data (Brody 1969; Kinder & Palfrey 1993). Without the empirical data that generate initial hypotheses and theories, there is no starting point for experimental research. But experimental research allows for the continued evolution and the “better articulation” of existing theories (Franklin 1996). Some political scientists have come to appreciate the potential benefits of experimentation and have applied it to their research. Bositis records that while only nine articles using experimental research designs appeared in major political science journals between 1924 and 1950, 104 such articles appeared from 1980 to 1989 alone (Bositis 1990: 65). McGraw (1993), too, sees experimentation as a methodology “on the rise” in political science, especially in the areas of public opinion, collective action, public policy, and – relevant to our study – of decision-making. In international politics, the origins of experimental simulation work can be traced to the early work of Geutzkow and his associates beginning in the late 1950s (Guetzkow et al. 1963). His InterNation Simulation set a precedent – all too often forgotten by scholars of international relations – for how experimentation can be used to inform and improve international relations.
The Crisis and Negotiation (CAN) Research Group About 15 years ago, I began a lengthy collaboration with Sarit Kraus, a computer scientist specializing in Distributed Artificial Intelligence (DAI). (I had
REFLECTIONS ON SIMULATION AND EXPERIMENTATION OF NEGOTIATION
95
barely heard of Artificial Intelligence at that point). Kraus had been working on developing an automated agent capable of participating with other automated agents or with human players in the Diplomacy board game. With a strong interest in modeling the negotiation process from a bargaining and game theoretic perspective, she was directed to me by her associates at the University of Maryland Institute for Advanced Computer Studies, some of whom had heard of my work on ICONS. It was apparent from the beginning that although we had very different training, we had a joint interest in better understanding how humans (in my case) and machines (in Kraus’s case) negotiated in difficult (crisis) situations in order to reach mutually beneficial agreements. Whereas my objective was to understand the conditions under which negotiations ended without resort to violence, Kraus’s objective was to understand how intelligent agents (machines) could be endowed with negotiation skills such that they could share scarce resources (tools like printers, Internet access, etc.) or cooperate to perform tasks. In both cases, negotiations are employed in situations where time is critical to some or all parties, resources are scarce, and tasks may require cooperative behavior. We wanted to minimize the amount of time spent on negotiation, and maximize the likelihood of reaching mutually beneficial outcomes (Kraus et al. 1995; Kraus 2001; Wilkenfeld et al. 1995). Our experimental work has been characterized by several features that tend to set it apart from other experimental work in foreign policy analysis (see Geva and Mayhar 1997; Mintz, Geva, Redd, and Carnes 1997; Tetlock & Belkin 1996). First, we have focused exclusively on behavior in crisis (Brecher & Wilkenfeld 2000), with a particular emphasis on the effects shortness of time and high level of threat have on crisis dynamics and outcomes. Second, our subjects had access to an elaborate Decision Support Systems (DSS) which, when properly used, could provide information on the utilities associated with various outcomes and hence enhance the subjects’ ability to maximize expected utility, even under conditions of incomplete information.3 Third, all communications among the participants have taken place in controlled network environments, allowing for detailed analysis of the types of tools they consulted as well as the content and form of their interactions during the negotiation. Together, these features provide a rich environment in which experiments can be run and their results assessed.4 At the core of our experimental work to date is the development of a strategic model of negotiation with an accompanying decision support system (DSS) based on three different scenarios: a hostage crisis involving an Indian plane hijacked by Sikh separatists and forced to land in Pakistan (see Kraus, Wilkenfeld and Zlotkin 1995, Wilkenfeld, Kraus and Holley 1999), a fishing
96
JONATHAN WILKENFELD
dispute based on the Canada-Spain dispute of 1995 (Wilkenfeld et al. 1999 and forthcoming), and the Ecuador/Peru Border Dispute of 1981 (Wilkenfeld et al., forthcoming, and below).
Experiments in the Study of Mediation in International Crises Bercovitch, Anagnoson, and Wille (1991) define mediation as “a process of conflict management where disputants seek the assistance of, or accept an offer of help from, an individual, group, state, or organization to settle their conflict or resolve their differences without resorting to physical force or invoking the authority of the law” (1991: 8). Bercovitch and Langley (1993) note that this behavioral definition is most useful because of its emphasis on the key components of mediation – the disputants, the third party, and the specific conflict resolution context. While there is no general consensus, the literature on mediation seems to have converged on three basic roles that mediators can play in a negotiation, and most authors agree that the roles change as the mediator and disputants gain information and skill (Princen 1992: 65). The first role is of the mediator as communicator or facilitator (Touval & Zartman 1985; Burton 1984; Hopmann 1996), serving as a channel of communications between two parties when a stalemate has been reached or when communications have broken down and face-to-face negotiations are not possible. This type of mediation is also referred to as third-party consultation (Fisher 1972; Keashly & Fisher 1996; Kelman 1992) good offices, or process facilitation (Hopmann 1996). The second role defined by Touval and Zartman is mediator as formulator. Here the mediator provides a substantive contribution to the negotiations in order to assist the conflicting parties in conceiving of a way to resolve their dispute when the parties reach a rigid impasse in the negotiation process. Hopmann (1996) suggests that this role is particularly effective when the parties’ emotions are running high. He highlights certain processes that the mediator as formulator may use to propose solutions: asking the parties to brainstorm, suggesting that issues be fractionated or linked together, inventing new solutions, and so forth. The third role is that of the mediator as manipulator. In addition to formulating solutions, this type of mediator uses its domestic, regional, or international position and its leverage – “resources of power, influence, and persuasion” – to manipulate the parties into agreement (Touval & Zartman 1985: 12). The mediator as manipulator becomes a party to the solution, if not the dispute itself,5 and often needs to augment the appeal of its solution by using carrotand-stick measures (adding and subtracting benefits to/from the proposed
REFLECTIONS ON SIMULATION AND EXPERIMENTATION OF NEGOTIATION
97
solution) (Touval & Zartman 1985, Zartman & Touval 1996). Hopmann indicates that only a powerful mediator can play this role, and he cites two modes of mediator involvement as manipulator: Influencing the direction of the negotiation process by raising the costs of disagreement and the rewards for agreement; Manipulating the international environment to push the parties toward agreement. From the experimental perspective, the question to be explored is how different types of mediation, as opposed to no mediation, impact the utility maximizing behavior of negotiators, the pace of crisis abatement, and the satisfaction with mediation.6
Experimental Findings on Mediation in International Crises The experimental environment in which mediation in international crisis was studied consists of a generalized decision support system for individual decision-makers, a crisis simulation scenario, and a communications system for both negotiators and mediators.7 A scenario loosely based on the Ecuador/Peru border dispute of 1981 was developed for the simulation. This case, part of an ongoing protracted conflict that began in 1935 and ultimately terminated in 1998 produced a total of five international crises.8 It involved an historical instance of successful mediation by the Organization of American States (OAS), which called upon the United States, Argentina, Brazil, and Chile to intervene. Four possible outcomes were built into the simulated negotiation: territorial division, cease fire, arbitration, and status quo. In part, we sought to establish the reliability of our experimental environment through our ability to replicate in broad general terms the outcome of this historical case. 212 students enrolled in international relations courses at the University of Maryland during 2000 and 2001 participated in simulations of a one-on-one crisis negotiation between Ecuador and Peru. Half of the students negotiated on behalf of Peru while the others represented Ecuador.9 The experimental research uncovered a relationship between mediation and the achievement of agreement, while also revealing that mediation leads to greater satisfaction with the crisis outcome. While we consider these effects of mediation to be consistent with the goals of crisis management and with the negotiators’ expectations of the effect of mediation, our subsequent analyses of mediation style demonstrate that it is necessary to attach some caveats to a general endorsement of mediation in crisis management. While both facilitative and manipulative mediation are conducive to generating agreements, and agreements that are considered satisfactory to negotiators, only manipulative mediation has a positive effect on the level of benefits
98
JONATHAN WILKENFELD
associated with crisis termination and on the duration of a crisis. Data from the Ecuador/Peru simulations indicate that only manipulative mediation meets the negotiators’ expectations of leading them to a more beneficial outcome than they could have otherwise secured, while facilitation may actually lower average benefits – a situation that could lead to discontent among disputants and, possibly, recurrent crises. In addition, the more rapid conclusion of a crisis – brought about only by manipulation – is an essential component of crisis management, given the relationship between prolonged crisis negotiations and the likelihood of escalating violence and war. We conclude that manipulative mediation, as compared with facilitation or no mediation, is an effective means of crisis management. Our endorsement of manipulation must be tempered, though, by at least two factors that need to be explored more rigorously in the future. First, our analysis has only looked at the two extreme styles of mediation: facilitation and manipulation. It may be the case that formulative mediation, which falls between these two extreme styles, is an even more effective management tool under certain circumstances. A second concern with manipulative mediation is about both the short – and long-term implications of solutions developed and forced upon parties by an outsider, fearing that these situations can lead to feelings of alienation and resentment of the mediator, of the process of negotiation, and even of the other parties involved in the negotiation (Princen 1992; Kelman 1992; Keashly & Fisher 1996). While our findings on negotiator satisfaction do not reveal resentment of manipulation vis-à-vis facilitation, greater consideration of the unintended consequences that could accompany the adoption of manipulative mediation as a crisis management tool is necessary before we can conclude that manipulation is an appropriate – perhaps the most appropriate – means of managing international crises. Our current findings, though, do indicate that manipulation shows potential as a key approach to mitigating the violence and instability associated with international crises.10
Prospects and Limitations for Experimental/Simulation Work in the Study of Negotiation Foreign policy analysis is plagued by a combination of relatively small n = s and relatively large numbers of variables potentially contributing to the foreign policy decision-making process. On the surface, experimental work can successfully address both of these issues. With a more or less unlimited supply of willing undergraduates as subjects, our n is virtually limitless. By care-
REFLECTIONS ON SIMULATION AND EXPERIMENTATION OF NEGOTIATION
99
fully constructing our simulation models and controlling our experimental environments, we can limit the number of variables we have to consider in any given experimental run. If this is the case, why are there not more eager scholars flocking to the simulation laboratories, and why are those of us who are there having so much trouble? First, I would argue that the transition from simulation environment, whose primary application in political science is for instruction and training, to the experimental laboratory, is not a simple one. First and foremost, most of us lack any real training in experimental design (how many graduate programs in political science offer courses in this area?). What to the experimental psychologist seems natural requires of the political scientist a very real investment of time and energy, and a lot of trial and error. Thus, even with the prospects of great advancements in knowledge seemingly at our fingertips, too few of us venture into the unknown terrain of the experimental laboratory. Moreover, once in the lab, we are faced with a host of other seemingly insurmountable obstacles. Consider our subjects: typically college freshmen (a few months out of high school), or, if we are lucky, upperclassmen. Cluttering their brains on any particular day are the aftermath of the previous evening’s indiscretions and plans for tonight’s, swarming hormones, extreme hunger, thirst, and a strong need for a cigarette (if not worse), not to mention their at best neutral positions – if not downright hostility – toward having to participate in the experiment in the first place. Combine that with their relatively uninformed position on the nature of the international system in general, and foreign policy decision-making in particular, and you are lucky if you can get anything that resembles replicable results.11 That is not to say that real-world decision-makers do not have extraneous things on their minds, but at least we know that some of their attention is focused on policy issues. Even if these two problems could somehow be neutralized, the issue of the simulation itself remains. We know from the simulation literature that a model should be detailed enough to represent the important aspects of the reality it is meant to represent, but not so detailed as to overwhelm the participant with information (even though we know that that is exactly what happens to real-world decision-makers, particularly in crisis). This turns out to be an extremely difficult undertaking. In our own case, we have some evidence to suggest that in a previous round of experiments involving a model of a fishing dispute roughly resembling the dispute between Canada and Spain in 1995, we not only packed the model with excessive detail, but then superimposed an elaborate mediation process. The results showed that the subjects were not coping, and were therefore making decisions based on only partial use of the decision support tools they had available to them. If you add to all this the
100
JONATHAN WILKENFELD
expense and time involved in the design and execution of a series of experiments, it is not surprising that relatively few political scientists venture into this domain. Given this rather bleak picture, where do I see simulation and experimental design fitting into the array of approaches now dominating the study of foreign policy analysis and negotiations? At the level of political psychology, there are simply some characteristics of decision-makers and decision-making that we are never going to be able to study systematically with anything approaching a large enough n to allow for careful statistical work. It is here, under controlled laboratory environments, that we can begin to make progress in understanding how decision-makers cope with and process information in crisis versus non-crisis environments, how third parties can impact the course of and outcomes to contentious situations, how actor attributes such as power (of various sorts) and willingness to compromise affect their negotiation strategies and behavior, and the circumstances under which violent strategies are chosen. While nothing can substitute for the data we can glean from in-depth comparative case studies, the experimental environment – when properly specified – can help us fill in the gaps in our knowledge and ultimately allow us to generalize. But in order to do this right, we must begin to properly train the next generation of political scientists in simulation and experimental techniques. This is not everyone’s cup of tea, nor should it be. It does offer an important alternative to the cross national and case study approaches that dominate the field of foreign policy analysis today. But we lack the experimental infrastructure, the culture of the laboratory, and perhaps most importantly, the channels by which experimental results can become important sources of information as we seek to better understand the behavior of political decision-makers. And perhaps most importantly, we need to be able to demonstrate more convincingly that results obtained in experimental simulation environments can help us better understand the behavior of real decision-makers in critical situations. Much work remains to be done on this front.
Notes 1. This work has been supported by the National Science Foundation under grant IIS0208608, and the US Institute of Peace (under grants SG-35–95 and SG-52–00). Parts of this paper are drawn from Wilkenfeld (2002) and Wilkenfeld, Young, Asal, and Quinn (forthcoming). 2. For additional information on the ICONS simulations, see http://www.icons.umd.edu. 3. For a discussion of decision support systems in general, and their application to negotiation in particular, see Wilkenfeld et al. 1995.
REFLECTIONS ON SIMULATION AND EXPERIMENTATION OF NEGOTIATION
101
4. For details on our experimental procedures, see Santmire et al. (1998). 5. Hopmann (1996) qualifies the notion of the manipulative mediator being a party to the dispute by stating that this mediator Aalmost@ assumes the role of a disputant in the negotiation process. 6. In the simulations conducted thus far, only facilitation and manipulation are examined. 7. The experimental environment is described more fully in Santmire et al. (1998), Wilkenfeld et al. (1995, 2001). 8. For a discussion of these crises in detail, see Brecher and Wilkenfeld (2000). The scenario and decision support systems, developed by Kathleen Young, David Quinn, and Chris Frain, are available from the author upon request. A more extended discussion of this research project appears in Wilkenfeld, Young, and Quinn (2001). 9. For details on the experimental procedures, see Wilkenfeld, Young, Asal, and Quinn (forthcoming). 10. For a more extended discussion of these experiments on mediator style in international crises, see Wilkenfeld, Young, Asal, and Quinn (forthcoming). 11. I am reminded of one of the very first works I became familiar with in simulation, the study by Hermann and Hermann (1967), which used the Inter-nation Simulation (INS) to attempt to replication the conditions leading to the outbreak of World War I. They checked for seemingly everything, yet curiously they couldn’t get war to break out. As it turned out, the student playing one of the key roles was an avowed pacifist, and simply wouldn’t go to war under any circumstances.
References Bercovitch, J.J., Anagnoson, T., and Wille, D.L. (1991). “Some conceptual issues and empirical trends in the study of successful mediation in international relations,” Journal of Peace Research 28 (1): 7–17. Bercovitch, J., and Langley J. (1993). “The nature of the dispute and the effectiveness of international mediation,” Journal of Conflict Resolution 37(4): 670–99. Bositis, D.A. (1990). Research Designs for Political Science: Contrivance and Demonstration in Theory and Practice. Carbondale: Southern Illinois University Press. Brecher, M., and Wilkenfeld, J. (2000). A Study of Crisis. Ann Arbor, MI: University of Michigan Press, paperback with CD-Rom edition. Brody, R. (1969). “The Study of International Politics qua Science: The Emphasis on Methods and Techniques.” In K. Knorr and J.N. Rosenau (Eds.) Contending Approaches to International Politics. Princeton, NJ: Princeton University Press. 110–128. Burton, J.W. (1984). Global Conflict: The Domestic Sources of International Crisis. Brighton, Sussex, Great Britain: Wheatsheaf Books Ltd. Campbell, D.T., and Stanley, J.C. (1963). Experimental and Quasi-Experimental Designs for Research. Chicago: Rand-McNally. Fisher, R.J. (1972). “Third Party Consultation: A Method for the Study and Resolution of Conflict,” Journal of Conflict Resolution 16:1:67–94. Franklin, A. (1996). Experiment, Right or Wrong. New York: Cambridge University Press. Geva, N., and Mayhar, J. (1997). “The Cognitive Calculus of Decisions on the Use of Force.” Paper presented at the annual meeting of the American Political Science Association, Washington, DC.
102
JONATHAN WILKENFELD
Guetzkow, H., et al. (1963). Simulations in International Relations: Developments for Research and Teaching. Englewood Cliffs, NJ: Prentice-Hall. Hermann, C., and Hermann, M. (1967). “An Attempt to Simulate the Outbreak of World War I,” American Political Science Review 61:406–416. Hopmann, P.T. (1996). The Negotiation Process and the Resolution of International Conflicts. Columbia, SC: University of South Carolina Press. Keashly, L., and Fisher, R.J. (1996). “A Contingency Perspective on Conflict Interventions: Theoretical and Practical Considerations.” In J. Bercovitch (Ed.) Resolving International Conflicts: The Theory and Practice of Mediation. Boulder, CO: Lynne Rienner Publishers. 235–261. Kelman, H.C. (1992). “Informal Mediation by the Scholar/Practitioner.” In J. Bercovitch and J.Z. Rubin (Eds.) Mediation in International Relations. New York: St. Martin’s Press. 64–96. Kinder, D., and Palfrey, T. (1993). “On Behalf of an Experimental Political Science.” In D. Kinder and T. Palfrey (Eds.) Experimental Foundations of Political Science. Ann Arbor: University of Michigan Press. 1–39. Kraus, S. (2001). Strategic Negotiation in Multiagent Environments. Cambridge: MIT Press. Kraus, S., Wilkenfeld, J., and Zlotkin, G. (1995). “Multiagent Negotiation under Time Constraints” Journal of Artificial Intelligence 75:297–345. Laponce, J.A. (1972). “Experimenting: A Two-Person Game between Man and Nature.” In J.A. Laponce and P. Smoker (Eds.) Experimentation and Simulation in Political Science. Toronto, ON: University of Toronto Press. 3–15. McGraw, K. (1993). “Political Methodology: Research Design and Experimental Methods.” In R.E. Goodin and H.D. Klingemann (Eds.) A New Handbook of Political Science. New York: Oxford University Press. 769–786. Mintz, A., Geva, N., Redd, S., and Carnes, A. (1997). “The Effect of Dynamic and Static Choice Sets on Political Decision Making,” American Political Science Review 91:553–566. Princen, T. (1992). Intermediaries in International Conflict. Princeton, NJ: Princeton University Press. Santmire, Tara, Wilkenfeld, J., Kraus, S., Holley, K., Santmire, Toni, and Gleditsch, K. (1998). “The Impact of Cognitive Diversity on Crisis Negotiations,” Political Psychology 19(4):721–748. Snyder, R. (1963) “Some Perspectives on the Use of Experimental Techniques in the Study of International Relations.” In H. Guetzkow et al. (Eds.) Simulations in International Relations. Englewood Cliffs, NJ: Prentice-Hall. 1–23. Starkey, B. and Blake, E. (2001). “Simulation in International Relations Education,” Simulation and Gaming 32(4): 537–551. Starkey, B. and Wilkenfeld, J. (1996). “Project ICONS: Computer-assisted Negotiations for the IR Classroom,” International Studies Notes 21(1):25–29. Tetlock, P., and Belkin, A. (Eds.) (1996). Counterfactual Thought Experiments in World Politics: Logical, Methodological, and Psychological Perspectives. Princeton, NJ: Princeton University Press. Torney-Purta, J. (1992). “Cognitive Representations of the Political System in Adolescents: The Continuum from Pre-Novice to Expert,” New Directions for Child Development 56:11–24. Torney-Purta, J. (1996). “Conceptual Change Among Adolescents using Computer Networks and Peer Collaboration in Studying International Political Issues.” In S. Vosniadou et al. (Eds.) International Perspectives on the Design of Technology Supported Environments. Hillsdale, NJ: Lawrence Erlbaum. 203–219. Torney-Purta, J. (1998). “Evaluating Programs Designed to Teach International Content and Negotiation Skills,” International Negotiation: A Journal of Theory and Practice 3(1): 77–97.
REFLECTIONS ON SIMULATION AND EXPERIMENTATION OF NEGOTIATION
103
Touval, S., and Zartman, I.W. (1984). “Introduction: Mediation in Theory.” In S. Touval and I.W. Zartman (Eds.) International Mediation in Theory and Practice. Boulder, CO: Westview Press. 7–17. Wilkenfeld, J., and Kaufman, J. (1993). “Political Science: Network Simulation in International Politics,” Social Science Computer Review 11(4):464–476. Wilkenfeld, J., Kraus, S., and Holley, K. (1999). “The Use of Decision Support Systems in Crisis Negotiations.” In Kent, A. (Ed.) Encyclopedia of Library and Information Science. New York: Marcel Dekker. Wilkenfeld, J., Kraus, S., Holley, K., and Harris, M. (1995). “GENIE: A Decision Support System for Crisis Negotiations,” Decision Support Systems 14:369–391. Wilkenfeld, J., Kraus, S., Santmire, Tara, and Frain, C. (1999). “The Role of Mediation in Conflict Management: Conditions for Successful Resolution.” Paper presented at the annual meeting, International Studies Association. Wilkenfeld, J., Young, K., Asal, V. and Quinn, D. (Forthcoming). “Mediating International Crises: Cross-National and Experimental Perspectives,” Journal of Conflict Resolution. Wilkenfeld, J., Young, K., and Quinn, D. (2001). “Mediating International Crises in the 20th Century.” Paper presented at the International Studies Association Meetings Hong Kong. Zartman, I.W., and Touval, S. (1996). “International Mediation in the Post-Cold War Era.” In C.A. Crocker and F.O. Hampson with P.A. all, Managing Global Chaos: Sources of and Responses to International Conflict. Washington, D.C.: United States Institute of Peace Press. 445–461.
Quantitative Coding of Negotiation Behavior LAURIE R. WEINGART, MARA OLEKALNS and PHILIP L. SMITH
We have spent most of our academic careers studying negotiation processes directly. Why? Because we are crazy? Because no one else will do it? Perhaps, but we like to think we study negotiation processes because the approach provides unique insights into how negotiations unfold. Direct examination of negotiation processes provides information about what negotiators actually do, rather than what they planned to do or what they thought they did. The resulting data can be used to capture general strategies employed by negotiators (through frequency analysis), how they employed those strategies (through sequential analysis), and when they did so (through phase analysis). Besides gaining a better understanding of the negotiation process itself, this information can be used to predict whether negotiators reach agreement as well as the quality of their agreements, and to determine how negotiators can be better trained to effectively negotiate. Much of our research has been carried out using an experimental paradigm designed to test the relationships between the negotiation context, negotiation behaviors, and outcomes. We videotape dyads and groups negotiating in roleplay exercises. The videotapes then become our primary source of data. In this article we review the steps that extract usable data from these videotapes and the lessons we have learned along the way. In this article, which builds on previous discussions of studying group (Weingart, 1997) and negotiation (Brett, Weingart, and Olekalns 2004) processes, we review our experience working with one large negotiation dataset, Towers Market II, to illustrate two steps within the larger research process: developing a coding scheme and coding the data. We then go on to discuss some of the issues that need to be resolved before data analysis begins.
Coding Scheme Development The first step in studying negotiation processes is to determine what types of behavior are theoretically relevant. This step is not to be taken lightly, as it has a profound influence on one’s ability to test and support hypotheses. International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 105–119 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
106
LAURIE R. WEINGART, MARA OLEKALNS AND PHILIP L. SMITH
For example, in our early research, we somewhat surprisingly found little or no relationship between information-sharing and quality of agreements (Weingart, Thompson, Bazerman, and Carroll 1990). We realized that this may have occurred because we had not distinguished between different types of information. Subsequent studies proved us correct: Priority information exchange positively influenced the quality of agreements whereas information about preferences and positions negatively (or neutrally) influenced them (Hyder, Prietula, and Weingart 2000; Olekalns & Smith 2000; Olekalns, Smith and Walsh 1996; Weingart, Hyder, and Prietula 1996). In this section, we consider the choices a researcher interested in negotiation processes must make in developing a coding scheme. In making these choices, researchers need to consider three questions: Should the scheme focus on task or relational aspects of process? Should it be theory – or datadriven? What steps should we take to ensure reliability? For more general discussions regarding how to construct a coding scheme see Folger, Hewes, and Poole (1984), Hewes (1979), Poole, Folger, and Hewes (1987) and Trujillo (1986). Level of coding When coding negotiations, we can focus on two aspects of behavior: their impact on outcomes and their impact on the underlying relationship. These aspects of behavior are not mutually exclusive. Rather, the issue is one of relative emphasis. Whereas coding schemes that evolved from negotiation theory emphasize outcomes, those that evolved from communication theory emphasize the relational aspects of strategy choice. Ultimately, there is considerable overlap between the kinds of tactics that researchers working from these two perspectives code for. The approaches differ in that coding schemes based on negotiation theory have been developed bottom up, with functions inferred from tactics, whereas schemes developed based on communication theory have been developed top-down, with functions driving the identification of tactics. The more important decision, for researchers, is which level of analysis best suits their needs and how extensive their list of tactics should be. Coding schemes stemming from negotiation theory have their origins in work by Pruitt and colleagues in the 1970s and 1980s (e.g., Carnevale, Pruitt, and Seilheimer 1981; Pruitt 1981; Pruitt & Lewis 1975) and Bales’ work in the 1950s (Bales 1950). These schemes focus on the substantive aspects of negotiation: value claiming and value creation. Such schemes code for specific tactics such as threats, offers, questions regarding preferences and prior-
107
QUANTITATIVE CODING OF NEGOTIATION BEHAVIOR
ities, creative solutions, and information seeking. These specific behaviors can, however, be linked to one of two strategic orientations – integrative or distributive – and so convey implicit information about the underlying relationship. The scheme that we developed for our Towers Market research is an example of this type of coding scheme (see Table 1 for the list of coding categories). Table 1. Towers II Negotiation Behaviors in Strategy Clusters Strategy
Behaviors included in strategy cluster
Integrative Information
States issue preferences States issue priorities Asks questions about preferences Asks questions about priorities Off-task comments (relationship building)
3 4 8 9 33
Create Value
Makes multi-issue offer Shows insight Notes general differences Notes general similarities Makes positive comments Suggests compromise Suggests package trade-offs Other process suggestions
2 13 17 18 21 24 25 30
Distributive Information
Facts Asks for bottom line Asks about others’ substantiation (attack arguments) Asks miscellaneous task related questions Notes differences Negative reactions Suggests discuss one issue
7 10 11 12 15 22 23
Claim Value
Makes single issue offer Refers to bottom line Substantiates Position Refers to mutual interests to influence other party Threats Refers to power Suggests creative solutions to meet own interests
1 5 6 14 19 20 31
Push to closure
Time checks Notes similarities
29 16
Process management
Suggests using reciprocity Suggests vote Suggests to move on
26 27 28
a
Numbers correspond to those in Figure 1.
Category Numbera
108
LAURIE R. WEINGART, MARA OLEKALNS AND PHILIP L. SMITH
Figure 1. Towers II Correspondence Analysis Results 2.5
Distributive information 15
2
23 16
1.5
Push to closure
12
7,10
1
29
22 11
0.5 17
21
0 –0.5
24 30 18
13
–1
Create Value
–1.5 –2
20 2 25
19
6,1,5 31
14
Claim Value
33,9 4,3 Integrative 8 information
28
26 27 Process Management
–2.5 –3 –2.5 a
–2
–1.5
–1
–0.5
0
0.5
1
1.5
2
Numbers correspond to those in Table 1.
Communication theory based schemes start by considering the relational functions served by communication: either to highlight similarities and bring people together or to highlight differences and to push people apart. In the context of negotiations, this has led to the identification of three categories of negotiation behaviors, integrating, attacking and defending, and a set of specific tactics that fit within these categories (e.g., Donohue, Diez and Hamilton 1984; Putnam & Jones 1982). Unlike early schemes derived from negotiation theory, those derived from communication theory code not just for substantive tactics but also for relational tactics (e.g., showing support for the other party). In addition, communication researchers hypothesize that communication can serve two functions: one is to provide a response to the other party; the other is to cue the other party’s subsequent behavior. Thus, it is possible to code any speaking turn twice, thereby capturing both aspects of communication (see, for example, Donohue et al. 1984; Olekalns & Smith 2000). Such an approach is especially useful when speaking turns are long and incorporate several different tactics.
QUANTITATIVE CODING OF NEGOTIATION BEHAVIOR
109
Developing and refining the coding scheme Coding schemes can be either theoretically derived (e.g., Weingart et al., 1996) or developed using a more grounded approach, derived from observation of negotiations themselves (refer to Putnam’s chapter in same volume). However, the distinction between traditional theory-driven approaches and more data-driven approaches is necessarily blurred in practice. Human behavior is typically too complex for a researcher to be able to anticipate all relevant behaviors without some direct experience of how people interact. Coding schemes and their associated manuals should be considered living documents, open to revision and clarification during the development process, but then locked in when coding begins. We have used a hybrid strategy of coding scheme development in order to capitalize on the benefits of both approaches. For example, when we initially developed the Towers Market coding scheme (Weingart, Bennett, and Brett 1993) we used negotiation theories and textbooks as well as existing coding schemes to identify theoretically important negotiation behaviors (e.g., Lewicki, Saunders, and Minton 1985; Pruitt 1981; Pruitt & Carnevale 1993; Walton & McKersie 1965). These yielded a set of task-specific behaviors such as information exchange, offers, tradeoffs across issues, and argumentation. Because we were also interested in how group members managed the process, we added several different types of process statements. After developing the categories, we tested them out on some sample transcripts. Some categories worked, others did not, so we altered the coding scheme accordingly. The next step involves refining the coding scheme. This is best done by having your coders independently code sections of transcripts and then checking for ease of usage and agreement. This will help reveal categories that are too broad (when types of behavior that seem quite different are similarly classified), and others that are too narrow or too similar to other categories (when deciding between two categories for a given behavior is repeatedly a problem). Relevant questions include: Can coders differentiate between the categories when applying them? Are there categories for all relevant behaviors? Can the scheme be applied reliably (discussed below)? If not, is it because of problems in training the coders or in the design of the scheme? Coding data In parallel with developing this coding scheme we developed a coding manual, that is, a set of application rules/guidelines for the coders, including identification of the unit to be coded (thought, sentence, speaking turn), category
110
LAURIE R. WEINGART, MARA OLEKALNS AND PHILIP L. SMITH
labels, definitions, and rules of thumb for distinguishing between categories and using context (i.e., statements surrounding the unit of interest) to interpret meaning. Our coding manuals typically include general rules about unitizing and coding as well as definitions of content categories, examples, and exceptions to the rules. This manual is used to train coders and to identify problems in applying the coding scheme. In particular, if disagreements about the application of a particular code appear to be haphazard and difficult to resolve, then the problem may reside in the design of the coding scheme itself. Categories might overlap or be missing. These definitional issues need to be resolved before coding proper begins. There are two discrete steps in coding process data: 1) unitizing the data, or identifying the unit of analysis to be coded, and 2) content coding the data, or applying labels to units as a means of capturing what occurred. Next, we discuss issues of unitizing and unitizing reliability, followed by a discussion of content coding reliability and validity. Unitizing Coders must first identify the unit of analysis for coding. This can vary from a single utterance to an interaction between two individuals. Speaking turns, that is, units that include all actions and/or statements made by an individual while he or she holds the floor, are often used when studying negotiation because they allow us to examine how one negotiator responds to another. Speaking turns can be subdivided into acts, that is, identifiable ideas or thoughts, which can be used when interested in the content of the dialogue said rather than the interactive nature of the negotiation. In the Towers II data, we unitized at the level of acts, that is, we coded each identifiable idea, but also retained information about speaking turns to maximize our flexibility. Note that one speaking turn can be made up of several thought units. Decisions regarding choice of unit type must be linked to the research question being asked, with special emphasis on the appropriate level of analysis and to the question of where in action and speech relevant meaning resides. Meaning can be lost if we select too small a unit, because the individual statements convey a different meaning to that conveyed by a speaking turn. We can also add redundancy as a result of separating immediate restatements, that which simply repeat previous tactics rather than adding new information. In contrast, by selecting too large a unit, information can be lost. If multiple categories of statements are made during a speaking turn, the researcher must decide which code best represents the behavior within a given unit. Dominance schemes, which identify the kinds of behavior that are expected to have the greatest impact on the interaction, can be developed to
QUANTITATIVE CODING OF NEGOTIATION BEHAVIOR
111
provide rules of thumb for these decisions (e.g., Weingart et al. 1993). Alternatively, the first or the last code within the unit might be retained. Regardless of the approach that is used, the risk of losing valuable information remains. Coders must meet two criteria of reliability: unitizing reliability and interpretive reliability. In both cases, the level of reliability places an upper bound on a coding scheme’s predictive ability. Unitizing reliability, the degree of agreement regarding identification of the units to be categorized, is needed to ensure that multiple coders of the same interaction will be coding from the same set of units. Unitizing reliability is most typically assessed using Guetzkow’s U, which calculates the difference between the number of units identified by an independent coder and the “true” number of units (the average of two coders’ estimates) using a simple equation U = (O1 – O2)/(O1 + O2 )
where U * 100 is the percent of disagreement and O is the observer or coder (Folger et al. 1984; Guetzkow 1950). For example, if coder 1 identified 75 units and coder 2 identified 85 units in the same transcript, Guetzkow’s U = (85 – 75)/(85 + 75) = .062, meaning that there was a 6% discrepancy between the number of units identified by either coder and the “true” number of units. Interpretive reliability refers to consistency in applying labels to the units and is typically measured across coders. That is, multiple coders independently code the same text, and their assignment of categories is compared for consistency. Thus, consistency of interpretation is assessed in terms of the reliable application of interpretive rules. It is also useful to obtain consistency measures within coders across time to check for drift in application of the coding scheme. The most common and complete global measure is Cohen’s Kappa (1960), which determines the level of agreement corrected for agreements due to chance. Cohen presents the following formula to calculate Kappa: Kappa = (P' – PC)/(1 – PC)
where P' is the observed percentage agreement among coders and PC is the proportion of chance agreement. Assuming that each unit to be coded has the same probability of accurate classification, then the probability of chance agreement (PC) is 1/k, where k is the number of categories in the coding scheme (Folger et al. 1984). For example, if there are 5 categories and the coders agreed on 75% of the codes assigned (P' = .75), then Kappa = (.75 – .2)/(1 – .2) = .6875. Additional details for how to calculate Kappa can be found in Cohen’s original article (1960). More detailed discussions of Kappa can be found in Fleiss (1971) and Folger et al. (1984).
112
LAURIE R. WEINGART, MARA OLEKALNS AND PHILIP L. SMITH
Our experience suggests that, to ensure high interpretive reliability, it is best to conduct unitizing and coding in separate passes through the data. Doing both simultaneously may result in a disagreement about what code to apply because coders have unitized (identified the speaking turns) slightly differently rather then because they are using the codes differently. This might occur because some coders treat back channels (“uh-huh,” “okay,” or “I see”) as substantive breaks in the flow of conversation and so unitize around them whereas others do not. This problem can be avoided by completing the unitizing first, checking for unitizing reliability, and then content coding from the same set of pre-unitized tapes or transcripts. Interpretive validity Interpretive validity refers to the degree to which a coding scheme taps into the information it was designed to obtain. Different approaches have been used to validate coding schemes, depending on the goals of the research. Poole and McPhee (1994) identify three types of claims that might be made in an attempt to demonstrate validity of a coding scheme. These claims are linked to the source of interpretation of the actions being studied: 1) observerprivileged: where the researcher seeks to explain interaction from the outside without reference to subjects’ perspectives, 2) generalized subject-privileged: focusing on shared meanings for members of a culture, and 3) restricted subject-privileged: with the goal of identifying idiosyncratic meanings for people in a particular group or relationship. As one moves from observer to restricted subject-privileged approaches, more input from participants and more intimate knowledge about the work group (as might be experienced by a group member) is required.
Issues to Resolve before Analyzing your Data So far, we have provided a “how to” for coding data. Having gotten this far, the obvious question is “what next?” Before one can start analyzing the data, there are still a number of issues that need to be resolved. How this is done will be determined, in part, by the kinds of research questions that one is asking. In this section, we consider three related issues: aggregating data, level of analysis and time segmentation. Aggregating data The first issue to consider is whether to focus on individual behaviors or to aggregate over individuals and focus on the dyad or group. Is it theoretically
QUANTITATIVE CODING OF NEGOTIATION BEHAVIOR
113
important to know how many times each negotiator in a dyad or group uses a specific tactic (e.g. making a demand) or can the hypotheses be tested by knowing, overall, the number of times that demands were made in a negotiation? In our research, we have largely focused on behaviors aggregated to the level of the dyad or group. This has enabled us to provide a global description of regularities in negotiators’ behaviors and the relationships between the negotiating context, negotiators’ behavior, and their outcomes. It allows us to ask whether, in general, integrative outcomes are associated with different kinds of strategies rather than distributive outcomes or impasses. It also enables us to question whether, in general, cooperatively-motivated negotiators behave differently from individualistically-motivated negotiators. Our experiences suggest that while this level of analysis captures many important aspects of negotiators’ behavior there are circumstances under which an individual level analysis is important, for example when negotiators differ in terms of power (Giebels, DeDreu, and Van de Vliert 2000) or social motives (Weingart, Brett, and Olekalns 2002). In our research, we have manipulated the composition of negotiating dyads and groups. At the level of the dyad, this means that two negotiators can have similar attributes, for example, both are cooperatively-motivated, or they can have mixed attributes, that is, one negotiator is cooperatively motivated while the other is individualistically motivated. An interesting question is how dyad or group composition affects negotiation processes: Do cooperatively-oriented negotiators behave the same way regardless of the characteristics of the other negotiator or do they adapt their behaviors? To answer questions such as this, it is necessary to shift from analyzing dyadic or group behaviors to individual behaviors within those groups. Independent of the content of the coding scheme, researchers need to decide how detailed their coding scheme should be. The answer to this question again depends on the level of your theory – whether global or specific behavior is of interest. If your hypotheses are about integrative or distributive behavior generally, you might want to classify comments using those broad categories. We find it is more effective to use more detailed coding categories and then aggregate these categories to the more general level during the analysis phase. This approach makes it less likely that two functionally different behaviors will be subsumed under one code. However, the more detailed the coding scheme the more likely we are to see coding errors and low reliability. Increasing the number of categories also means that the frequency of each category decreases, resulting in sparse data sets that create problems for most statistical methods. There are several approaches for aggregating tactics into more general groupings. You may choose to adopt a theory-driven approach, aggregating
114
LAURIE R. WEINGART, MARA OLEKALNS AND PHILIP L. SMITH
behaviors that theories classify as integrative or distributive. A pitfall for this kind of approach is that it neglects actual usage and so may group behaviors that are not associated with each other in actual negotiations. Empirical analysis thus provides an important test of our theoretical assumptions about the function and use of specific tactics. For example, we have found that the use of positional information (i.e., information about what outcome a negotiator desires on a given issue) clusters with a focus on offers and concessions, whereas argumentation clusters with threats and demands (Olekalns & Smith, 2003). In a more theoretically-driven approach, we likely would have combined positional information and argumentation into one category (distributive information exchange) and offers, demands, and concessions into a second category (offer management). To empirically determine clusters of tactics, we use correspondence analysis (Greenacre 1993). This analysis establishes relationships between tactics, based on their frequency of use. It is conceptually similar to factor analysis, but developed specifically for use with count (frequency) data. Figure 1 shows the output from a correspondence analysis of Towers II data (the mapping of data points in the two-dimensional space) with six clusters identified and interpreted. There are no hard and fast rules for the interpretation of this output. Visual inspection determines clusters based on (a) physical proximity and (b) thematic fit. Some categories were mapped in unexpected places. We had to go back to the transcripts and consider what negotiators were actually doing to help us decide how to cluster those behaviors. For example, in Figure 1, asking questions about priorities (category 9), and references to power (category 20) were mapped near one another but negotiation theory would suggest the former is an integrative behavior and the latter a distributive behavior. Rather than put these two categories of behavior in the same cluster, we looked for other proximal behavior that might cluster with each theoretically. As a result we were able to identify theoretically meaningful and proximal clusters of negotiation behavior. Levels of analysis The second issue concerns the level of analysis. In analyzing your data, you can frame your hypotheses in terms of the frequency with which strategies are used, how they are sequenced or how strategies evolve over time. Each of these levels of analysis answers slightly different questions. In this section, we provide a brief overview of the issues associated with each level of analysis (see Brett, Weingart and Olekalns, 2004; Weingart & Olekalns, 2004). Our other article, to be published in the next issue of International Negotiation
QUANTITATIVE CODING OF NEGOTIATION BEHAVIOR
115
(10-1), considers the modeling of sequences in greater detail (Smith, Olekalns and Weingart). These three approaches should not be considered as mutually exclusive. Rather, each provides different kinds of information about negotiation processes and helps build a more complete picture of what negotiators do (e.g., Olekalns & Smith 2000). Looking at the frequency with which negotiators use specific strategies is by far the most common approach to analyzing negotiators’ interactions. Using this approach, one simply counts the number of times that a given strategy (for example, a threat), is used. Focusing on the frequency with which strategies are used tells us about the global approach used by the negotiator(s). The major shortcoming of this approach is that it reflects a static view of negotiation processes: counting the number of times that types of strategies occur does not tell us about how negotiations unfold over time. Therefore, while we can draw very broad conclusions about the relationship between external factors and strategy choices, as well as between strategy choices and outcomes, we cannot learn about the patterning of strategies over time. Analyzing frequencies helps us to answer questions such as: Does the negotiating context affect the kind of strategies that are used? Are specific strategies linked to different kinds of outcomes? There are two approaches that allow us to better capture the dynamic nature of negotiations. The first focuses on how one negotiator responds to another, that is, on how strategies are sequenced. Studying sequences allows us to examine patterns in negotiators’ interactions at two levels: sequence structure and sequence content. For example, we ask whether negotiators match (reciprocate) or mismatch (fail to reciprocate) each other’s behaviors. This tells us whether negotiators are attempting to sustain or redirect the prevailing strategic orientation. A second layer of information is provided by examining the content of these sequences. Analyzing content tells us about the causes and consequences of matching, and helps us to understand how negotiators break off distributive spirals or derail integrative processes. This approach also helps us to develop more complex models of negotiators’ interactions: It provides insights into how negotiators constrain each other’s actions and into how they direct the negotiation towards obtaining their preferred outcomes. One important insight that we have gained from this type of analysis is that sequences that are influential in determining the outcome of negotiations often incorporate strategies that are not frequently used in their own right. This level of analysis thus complements rather than duplicates frequency level analyses. Our paper on Markov chain analyses provides an introduction to analyzing strategy sequences (Smith, Olekahns, and Weingart, forthcoming).
116
LAURIE R. WEINGART, MARA OLEKALNS AND PHILIP L. SMITH
Although sequences incorporate a temporal dimension, the focus on immediate actions and reactions means that the analysis is of short-term effects. Analyzing negotiation phases also captures the temporal aspect of negotiations. However, the time frame for examining the consequences of strategies is a longer-term one, allowing consideration of the ebb and flow of the negotiation over time. There are two approaches to analyzing negotiation phases. The first, a stage model approach, treats negotiations as divisible into discrete time segments and considers how the frequency of different strategies changes across segments. The second approach, called an episodic approach, looks for naturally occurring phases: Phases are clearly identifiable interaction sequences with explicit beginnings and ends (Baxter 1982). They are defined on the basis of ‘runs’ of strategies (i.e., the continuous use of a single strategy) with transitions between phases identified on the basis of interruptions to extended runs of the same strategy. In the following section we consider the methodological and empirical issues associated with adopting either of these approaches. Comparing sequence and phase analyses can highlight negotiations in which these temporal aspects of negotiation work together and reinforce each other and those in which they work in opposition. Time segmentation Researchers who want to examine negotiation phases are faced with a choice between a stage model and an episodic model. The principal advantage of a stage approach is that it ensures comparability across research; its principal disadvantage is that, by arbitrarily determining when stages begin and end, it is likely to miss naturally occurring shifts in the negotiation process. If one chooses to work within a stage approach, one still needs to make several decisions. The first is how many stages in which to divide the negotiation. Negotiation theory identifies anywhere between three and 12 stages. In our research, we have used six stages as the number of segments that provides a reasonably fine-grained analysis of negotiation while maintaining a reasonable number of observations within each segment (e.g., Olekalns, Smith and Walsh 1996). The second is how to determine these stages. You can do this either on the basis of time (divide each negotiation into, for example, five minute intervals) or on the basis of speaking turns (for example, if the negotiation is made up of 300 speaking turns, then each 50 speaking turns become a stage). The principal disadvantage of a time-based segmentation is that negotiations do not all take the same amount of time. You may thus find yourself comparing strategies that occur in the 10–15 minute interval and which represent the midpoint for one negotiating dyad but the endgame for another. We there-
QUANTITATIVE CODING OF NEGOTIATION BEHAVIOR
117
fore use segmentation based on speaking turns. We prefer this approach as it means that we are comparing comparable segments of the negotiation: midpoint to midpoint and endgame to endgame.
Conclusion In this article, we have provided an overview of the decisions that researchers need to make when they embark on an analysis of negotiation processes. Whether you choose to use an existing coding scheme or to develop one that meets your needs, you will need to make decisions about the level at which to code data, whether to aggregate tactics into broader strategies and on the level at which to analyze your data. You will need to decide whether to focus on what strategies negotiators use, how strategies combine to facilitate or inhibit goal attainment, or how negotiations move from beginning to end. We have found that each level of analysis brings with it unique insights into the negotiation process. We would be less than honest if we did not conclude by saying that this is a labor-intensive and time-consuming process. However, we believe that those of you who persevere will find the richness of your data to be its own reward.
References Bales, R.F. (1950). Interaction process analysis: A method for the study of small groups. Cambridge, MA: Addison-Wesley. Baxter, L.A. (1992). “Conflict management: An episodic approach,” Small Group Behavior 13:23–42. Brett, J.M., Weingart, L.R., and Olekalns, M. (2004). “Baubles, Bangles and Beads: Modeling the Evolution of Negotiating Groups over Time.” In M.A. Neale, E.A. Mannix, and S. Blount-Lyons (Eds.) Research in Managing Groups and Teams: Time in Groups (Volume 6, pp. 39-64). New York: Elsevier Science. Carnevale, P.J.D., Pruitt, D.G., and Seilheimer, S.D. (1981). “Looking and competing: Accountability and visual access in integrative bargaining,” Journal of Personality and Social Psychology 40:111–120. Cohen, Jacob (1960). “A coefficient of agreement for nominal scales.” Educational and Psychological Measurement, 20, 37–46. Donohue, W.A., Diez, M.E., and Hamilton, M. (1984). “Coding naturalistic negotiation interaction,” Human Communication Research 10:403–425. Fleiss, J.L. (1971). “Measuring nominal scale agreement among many raters,” Psychological Bulletin 76:378–382. Folger, J.P., Hewes, D.E., and Poole, M.S. (1984). “Coding social interaction.” In B. Dervin and M. Voight (Eds.) Progress in communication sciences (Vol. 4, pp. 115–161). Norwood: Ablex.
118
LAURIE R. WEINGART, MARA OLEKALNS AND PHILIP L. SMITH
Giebels, E., De Dreu, C.K.W., and Van de Vliert, E. (2000). “Interdependence in negotiation: Effects of exit options and social motive on distributive and integrative negotiation,” European Journal of Social Psychology 30:255–272. Greenacre, M.J. (1993). Correspondence analysis in practice. London: Academic Press. Guetzkow, H. (1950). “Unitizing and categorizing problems in coding qualitative data,” Journal of Clinical Psychology 6:47–58. Hewes, D.E. (1979). “The sequential analysis of social interaction,” Quarterly Journal of Speech 65:56–73. Hyder, E.B., Prietula, M.J., and Weingart, L.R. (2000). “Getting to best: Efficiency versus optimality in negotiation,” Cognitive Science 24(2):169–204. Lewicki, R.J., Saunders, D.M., and Minton, J.W. (1985). Negotiation. Irwin: Boston. Olekalns, M., and Smith, P.L. (1999). “Social value orientations and strategy choices in competitive negotiations,” Personality and Social Psychology Bulletin 25:657–668. Olekalns, M., and Smith, P.L. (2000). “Negotiating optimal outcomes: The role of strategic sequences in competitive negotiations,” Human Communication Research 24:528–560. Olekalns, M., and Smith, P.L. (2003). “Testing the relationships among negotiators’ motivational orientations, strategy choices, and outcomes.” Journal of Experimental Social Psychology. Olekalns, M., Smith, P.L., and Walsh, T. (1996). “The process of negotiating: Strategies, timing and outcomes,” Organizational Behavior and Human Decision Processes 67:61–77. Poole, M.S., Folger, J.P., and Hewes, D.E. (1987). “Analyzing interpersonal interaction.” In M.E. Roloff and G.R. Miller (Eds.) Interpersonal processes: New directions in communication research (pp. 220–256). Newbury Park, CA: Sage. Poole, M.S., and McPhee, R.D. (1994). “Methodology in interpersonal communication.” In M. Knapp and G.R. Miller (Eds.) Handbook of interpersonal communication. Newbury Park, CA: Sage. Pruitt, D.G. (1981). Negotiation behavior. Academic Press: New York. Pruitt, D.G., and Carnevale, P.J. (1993). Negotiation in social conflict. Pacific Grove: Brooks/ Cole. Pruitt, D.G., and Lewis, S.A. (1975). “Development of integrative solutions in bilateral negotiation,” Journal of Personality and Social Psychology 31:621–633. Putnam, L.L., and Jones, T.S. (1982). “The role of communication in bargaining,” Communication Monographs 3:262–282. Smith, P.L., Olekalns, M., Weingart, L.R. (in press). Markov claim Models of communication processes in negotiation. International Negotiation. Trujillo, N. (1986). “Toward a taxonomy of small group interaction coding systems,” Small Group Behavior 17:371–394. Walton. R.E., and McKersie, R.B. (1965). “A behavioral theory of labor negotiations,” McGraw-Hill: Boston. Weingart, L.R. (1997). “How did they do that? The ways and means of studying group processes.” In L.L. Cummings and B.M. Staw (Eds.) Research in organizational behavior (Vol. 19 pp. 184–239). Greenwich: JAI. Weingart, L.R., Bennett, R.J., and Brett, J.M. (1993). The impact of consideration of issues and motivational orientation on group negotiation process and outcome. Journal of Applied Psychology 78(3):504–517. Weingart, L.R., Brett, J.M., and Olekalns, M. (2002). Conflicting social motives in negotiating groups. Paper presented at the 62nd Annual meeting of the Academy of Management, Denver, Colorado.
QUANTITATIVE CODING OF NEGOTIATION BEHAVIOR
119
Weingart, L.R., Hyder, E.B., and Prietula, M.J. (1996). “Knowledge matters: The effect of tactical descriptions on negotiation behavior and outcome,” Journal of Personality and Social Psychology 70:1205–1217. Weingart, L.R., and Olekalns, M. (2004). “Communication Processes in Conflict and Negotiation.” In M. Gelfand and J. Brett (Eds.) The Handbook of Negotiation and Culture: Research Perspectives. Palo Alto, CA: Stanford University Press. Weingart, L.R., Thompson, L.L., Bazerman, M.H., and Carroll, J.S. (1990). “Tactical behavior and negotiation outcomes,” The International Journal of Conflict Management 1:7–31.
The Use of Questionnaires in Conflict Research AUKJE NAUTA and ESTHER KLUWER
Conflict is an important theme to study, both in organizations and in close relationships. In organizations, conflict may hinder productivity and job satisfaction. In close relationships, conflict can be a threat to relationship satisfaction and even to the endurance of the relationship. It is therefore important to study conflict empirically by gathering data on its appearance, causes and consequences, and on the emotional, cognitive, motivational and behavioral aspects that accompany it. These insights have both theoretical and practical value. For example, they can be used to teach people how to handle conflicts better and prevent conflicts from escalating into destructive outcomes. Unfortunately, it is far from easy to study conflict empirically. It is especially difficult to organize settings in which conflicts are acted out naturally, either in field settings or in the laboratory. Field settings are often too threatening for participants and too uncontrollable for researchers, whereas laboratory settings have limits regarding external validity (i.e., the extent to which experimental results can be generalized to real-life settings; Cook and Campbell 1979). In addition, observing real conflict is time-consuming and expensive. An alternative to studying real-life conflict is the use of questionnaires. Questionnaires – whether administered as paper-and-pencil surveys or as (telephone) interviews – can be useful alternatives for laboratory or observational studies on conflict when researchers know how to avoid possible pitfalls. Sometimes they may even be the only way to demonstrate certain phenomena. For instance, Pruitt, Peirce, McGillicuddy, Welton, and Castrianno (1993) used questionnaires to measure how people in a court-based mediation center felt about the quality of the agreements just after a mediation session as well as 6 months later. They found out that long-term success of the mediation (compliance, improved relations, and absence of new problems) was determined by the degree to which participants had perceived the mediation as fair just after the session. However, the quality of agreements appeared not to be related to long-term success. In order to demonstrate effects like these, questionnaires are required because attitudes such as procedural fairness and assessments of relationships can best be assessed through self-reports.
International Negotiation Series 2: P. Carnevale and C.K.W de Dreu (eds.) Methods of Negotiation Research, 121–133 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
122
AUKJE NAUTA AND ESTHER KLUWER
Many books and articles have been written about conducting questionnaire research (Belson 1981; De Vaus 1991; Dillman 1978; Kidder & Judd 1986; Schuman & Scott 1987; Shaughnessy, Zechmeister & Zechmeister 2000; Tanur 1992). For example, Schaughnessy and colleagues (2000) write that the wording of questions can be a threat to validity, because how a question is phrased has implications for how that question is answered. Good questionnaire items are short (20 or fewer words), simple, direct, clear, specific, and familiar to all respondents. Furthermore, they do not involve leading, loaded or double-barreled questions. Other issues that Schaughnessy and colleagues (2000) deal with is questionnaire design (e.g., attractiveness), ordering of questions (e.g., interesting questions first), and enhancing response rate in a mail survey (e.g., mail one or more follow-up questionnaires). Despite the attention for questionnaire research, books and articles on general research methods pay little attention to the specific challenges encountered when studying conflict. In this article, we will therefore discuss the strengths and weaknesses of using questionnaires in conflict research. We do not aim at completeness in summing up all the challenges. Instead, we will restrict ourselves to some strengths and weaknesses that we have encountered when performing our own research on conflict in organizations and in close relationships. The main focus is on measuring conflict behavior, although there are many other relevant aspects to conflict, such as conflict issues, causes, and outcomes (De Dreu, Harinck & Van Vianen 1999; Van de Vliert 1998), as well as the motives, emotions, and cognitions that accompany conflict (De Dreu & Carnevale, 2003). We believe conflict behavior is a key factor of the conflict process because it largely determines the (positive or negative) consequences of conflict. At the same time, conflict behavior may be one of the most difficult conflict features to measure through questionnaires, because questionnaires by definition measure perceptions of current or past behavior instead of the actual behavior itself. We will address five issues concerning the use of questionnaires in conflict research. The first is that conflict is a highly sensitive topic, which has consequences for the acquisition of participants and the response to surveys. The second issue concerns the validity of conflict behavior questionnaires. Third, we will discuss self-serving bias (i.e., the tendency to perceive oneself as more positive than the opponent), which poses a specific threat to validity that deserves special attention. Fourth, we will address the fact that conflict surveys usually entail correlational, and often cross-sectional, designs that prohibit conclusions about causality. The fifth and last issue concerns the fact that data on conflict behavior are by definition dyadic in nature; we discuss
THE USE OF QUESTIONNAIRES IN CONFLICT RESEARCH
123
the practical and methodological problems that go along with this. We will illustrate these issues with our own research on conflict in organizations and in close relationships and include recommendations for doing questionnaire research.
Sensitivity Conflict is a sensitive subject. People generally do not like to admit that they have conflicts with others because it is associated with negative consequences, as Nauta experienced when interviewing employees in organizations about conflict and negotiation (Nauta & Sanders 2000, 2001; Nauta, De Vries, and Wijngaard 2001). Because of this sensitivity, people are often reluctant to participate in research on conflict. For example, Nauta (2003) tried to recruit four different Dutch organizations for a study on organizational determinants and consequences of conflict. Although several organizations showed an interest in the project, many organizations ultimately refused to participate due to their fear of launching discussions and dissatisfaction among their employees. Instead of speaking about conflict in the introduction of the study, Nauta therefore chose to frame the research more positively by stressing how organizations could improve interpersonal and inter-team collaboration. This prevented participants from experiencing some of the unpleasant connotations that may accompany conflict such as “fighting” and “violence” and “aggression,” facilitating the decision to participate in the study. The fact that conflict is a sensitive topic may cause another research problem as well: Response may be low, due to the fear of participants that their answers become public. For example, in the same research project described above, Nauta (2003) performed a survey in a medium-sized organization that measured several biographic variables like gender, age, education, and organizational unit. Some participants remarked that their anonymity was not guaranteed by asking these specific background questions. For other participants, this may have been a reason not to return the questionnaire, because of their fear of being recognized. Perhaps people who experienced relatively frequent and severe conflicts were less likely to return the questionnaire, resulting in a selective sample of participants. When using a questionnaire to study organizational conflict, it is therefore important to ask as few demographic questions as possible. Also, researchers must promise (and hold their promise) that individual data will never be made available to the management of the organization. Moreover, careful attention must be paid to the introduction of the research and further communications
124
AUKJE NAUTA AND ESTHER KLUWER
about the research. Researchers must be clear about the goals and expected results of the research, in order to motivate people to participate. Nevertheless, questionnaires can also have an important advantage when studying a delicate subject. For example, in research on close relationships, questionnaires are often chosen deliberately to study topics that are sensitive, such as relationship problems, sex, and conflict (e.g., Kluwer, Heesink, and Van de Vliert 1996). People generally feel more comfortable filling out an anonymous questionnaire rather than talking face-to-face to an interviewer or a researcher about these topics.
Validity The second issue is whether questionnaires produce valid measures of conflict behavior. In other words, do we really measure what we want to measure (i.e., construct validity; Cook & Campbell 1997)? Van de Vliert (1997) and De Dreu et al. (2001) discussed the validity and reliability of several questionnaires that measure the five-part typology of conflict behavior as described by Blake and Mouton (1964), Thomas (1992), Rubin, Pruitt, and Kim (1994) and many others. This typology distinguishes between forcing (trying to force a solution upon the other party that meets own goals and interests, but not those of the other party), yielding (accepting what the other wants and reaching a solution that meets other’s but not own goals and interests), avoiding (withdrawing from the conflict, either temporarily or definitely), problemsolving (actively searching for a solution that meets both own and other goals and interest), and compromising (striving for an even distribution of the pie, without trying to enlarge it). Throughout the years, several questionnaires have been developed to measure these types of conflict behavior. The oldest is the Conflict Management Survey (Hall 1969). Although the CMS is still very popular in conflict management training settings, researchers have criticized and replaced it with other measures because of disappointing psychometric qualities (Landy 1978; Thomas & Kilmann 1978; Shockley-Zalabak 1988). Substantial improvement was reached by the development of the Rahim Organizational Conflict Inventory (ROCI-II). However, this scale still lacks optimal psychometric properties (see De Dreu et al. 2001; Rahim & Magner 1995). Other instruments worth noting are the Organizational Communication Conflict Instrument (OCCI; Putnam & Wilson 1982) and the Thomas-Kilmann Conflict Mode Instrument (TKI; Thomas & Kilmann 1974). The TKI is, like the CMS, often used for individual purposes. It is available on the internet, where individuals can fill
THE USE OF QUESTIONNAIRES IN CONFLICT RESEARCH
125
it out for free and directly receive the results on their preferred mode of conflict management, compared to a reference base. A questionnaire that is widely used in the Netherlands is the Dutch Test for Conflict Handling (DUTCH; Euwema & Van de Vliert 1990; Janssen & Van de Vliert 1996; Van de Vliert 1997; see Table 1). De Dreu et al. (2001) showed that self-ratings on the DUTCH in a role-playing experiment correlated highly with both ratings of the opponent and observer ratings. However, the correlations for avoiding behavior were not significant. Avoiding behavior is difficult to observe, especially in the laboratory, because participants are not likely to avoid conflict when they are asked to act out a conflict. In addition, avoiding is ambiguous behavior open to multiple attributions. For example, someone who consistently downplays the importance of the conflict issue may do this in order to avoid the issue and to reduce interaction to a minimum. The opponent, however, may perceive such behavior as a cunning way to get one’s way, to buy time, and to impose one’s will on others (i.e., forcing). Perhaps avoidance – more than any other conflict behavior – involves acts that are difficult to judge, which in turn makes accurate understanding of underlying intentions more important. Because individuals have better knowledge about their own intentions than opponents and neutral observers do, convergence between self-reports and other-reports of avoiding is likely to be low. Nevertheless, it is still a question whether conflict questionnaires such as the DUTCH really measure behavior in all circumstances. In the study by the Dreu et al. (2001), the DUTCH was administered immediately after the roleplay was over. Participants thus reflected upon the behavior they had just exhibited. However, in field studies, the DUTCH is generally used to measure the behavior one ‘usually’ or ‘generally’ shows when engaged in conflict. In this setting, the DUTCH appears to measure a general intention to certain conflict behavior or even a general style, instead of actual behavior. A possible solution to this problem is the critical incident method (Flanagan 1954; see also Kluwer 2000), in which participants are asked to recall a recent conflict and answer questions about this particular incident, including behavioral measures. This yields a more valid measure of actual conflict behavior and participants find it easier to answer questions about their behavior in a specific real-life situation than questions about their general behavioral style (see for example, Kluwer 2000). Nevertheless, one must realize that the answers may be influenced by the specific context in which the critical incident took place. Another option is to use vignettes or scenarios. For example, Kluwer, De Dreu and Buunk (1998) asked participants to read a scenario, in which they
126
AUKJE NAUTA AND ESTHER KLUWER
Table 1. The Dutch Test for Conflict Handling (DUTCH) When I have a conflict at work, I do the following: Yielding I give in to the wishes of the other party. I concur with the other party. I try to accommodate the other party. I adapt to the other parties’ goals and interests. Compromising I try to realize a middle-of-the-road solution. I emphasize that we have to find a compromise solution. I insist we both give in a little. I strive whenever possible towards a fifty-fifty compromise. Forcing I push my own point of view. I search for gains. I fight for a good outcome for myself. I do everything to win. Problem-solving I examine issues until I find a solution that really satisfies me and the other party. I stand for my own and other’s goals and interests. I examine ideas from both sides to find a mutually optimal solution. I work out a solution that serves my own as well as other’s interests as good as possible. Avoiding I avoid a confrontation about our differences. I avoid differences of opinion as much as possible. I try to make differences loom less severe. I try to avoid a confrontation with the other. Note. Items could be answered on a five-point scale (1 = not at all, to 5 = very much). Items are translated from Dutch and are usually presented in a random order.
described a conflict between the participant and an opponent. After reading the scenario, participants were asked to what extent they would use cooperative tactics (e.g., try to collaborate in each other’s interest) and competitive tactics (e.g., show distrust). Although these are still intentions to evince one type of conflict behavior or another rather than actual conflict behavior, those intentions are likely to be associated with real behavior because participants had to imagine themselves in a specific situation rather than a generalized conflict situation. Furthermore, the use of vignettes enables researchers to manipulate certain characteristics of the conflict. For example, in their research on relationship conflict over the division of labor, Kluwer (1998) and
THE USE OF QUESTIONNAIRES IN CONFLICT RESEARCH
127
Kluwer, Heesink, and Van de Vliert (2000) used vignettes to manipulate the topic of the conflict (housework, paid work, or child care) and each partner’s role in the conflict (complainant versus defender of the status quo; see Table 2 for an example). Table 2. Example of a Conflict Scenario (Kluwer, Heesink, and Van de Vliert 2000) “You are dissatisfied with the time your spouse spends on household tasks. For example, you are dissatisfied with how much attention your spouse pays to housework, how often your spouse tidies the house, cleans, does the dishes, does groceries, or the way your spouse carries out chores, etcetera. In other words, you want a change in the time your spouse spends on housework. However, your spouse is satisfied with the situation as it is.”
Self-serving and Social Desirability Biases Whenever researchers measure conflict behavior, chances are high that they will find some sort of self-serving bias, that is, a tendency to see one’s own conflict behavior as more constructive and less destructive than the conflict behavior of one’s opponent (De Dreu, Nauta & Van de Vliert 1995; Kluwer et al. 1998; Nauta and Van Sloten 2003). Although it is an interesting theoretical phenomenon, self-serving bias is a threat to validity, because participants report biased perceptions of their own and their opponent’s behaviors. Self-serving bias may also imply that the measures are influenced by a general tendency towards social desirability, which is more likely to occur when one rates one’s own behavior than when one rates other’s behavior. This may also occur when measuring conflict frequency, for example in close relationships. People are generally reluctant to admit that they have relationship conflicts, because it might imply that their relationship is in trouble. In addition, happy couples tend to distort their appraisal of their relationship in positive ways (e.g., Murray & Holmes 1997). Hence, people may not admit that they have conflicts with their partner due to social desirability biases or they may not report conflict due to positive illusions they hold about their relationship. For both reasons, measures of absolute conflict frequency are likely to be unreliable. A possible solution to biased measurements of conflict behavior and conflict frequency is to measure these concepts from different perspectives: Self reports, opponent reports and observer reports of conflict behavior and conflict frequency. When interrater reliability is high, it can be assumed that the average of these three perspectives is an accurate (intersubjective) measure.
128
AUKJE NAUTA AND ESTHER KLUWER
Cross-sectional Data Conflict data that are gathered through questionnaires are often correlational in nature, which prohibits conclusions about causality. For example, Nauta, De Dreu and Van der Vaart (2002) examined whether social value orientations predicted the degree to which employees paid attention to the goals of their colleagues working in other departments, which in turn would lead to problem-solving behavior during interdepartmental conflicts. Although they used multiple methods (i.e., questionnaires and oral interviews), it was still difficult to conclude whether a pro-social orientation indeed led to a higher concern for others’ goals which in turn resulted in problem-solving behavior. They could not rule out that a fourth variable (e.g., altruism or differences in the tendency to give socially desirable answers) explained why respondents scored either high or low on pro-social orientation, goal concerns, and problem-solving behavior. A solution to the problem of causality is to use experimental designs. When this is not possible in the field, quasi-experimental designs as described by Cook and Campbell (1979) (e.g., the use of vignettes or scenarios) are a good alternative. A general recommendation is to use multiple data collection methods (e.g., both observation and self-report) or longitudinal designs. However, many practical arguments that have to do with time, energy, and money hinder researchers from using these designs. Still, we recommend the use of at least one extra measurement method. For example, in their study on conflicts about working relationships between employees and their superiors in ten different organizations, Nauta and Van Sloten (2004) used questionnaires, together with data from personnel files about absenteeism. Moreover, they distributed surveys among both employees and their superiors, which enabled them to aggregate data of superiors about determinants of conflict behavior to the organizational level, and then examine whether these organizational variables (like communication culture, perceived common ground) predicted the conflict behaviors of employees.
Dyadic Data The last issue that we address concerns the dyadic nature of conflict data. By definition, conflict is an interpersonal concept: Conflict is experienced when one feels frustrated or hindered by some other party (Van de Vliert 1998). It is therefore interesting to study both sides of the conflict. However, in organizations, it is often difficult to couple data of two conflict parties while at the same time guaranteeing anonymity. Therefore, studies of dyadic conflict
THE USE OF QUESTIONNAIRES IN CONFLICT RESEARCH
129
require careful introduction, implementation and reporting to the organization. When studying close relationships, coupling data is not only much easier but also more necessary. Partners in romantic couples are not randomly paired and they will be similar or at least influence each other on various dimensions. Therefore, we cannot simply perform statistical analyses on groups of husbands and wives because this violates the assumption of independent observations. In this case, researchers have to deal with a methodological problem concerned with the dependency of data (cf. Kenny & Kashy 1991; Kenny 1995, 1996). Kenny (1996) presents three models of non-independence that may be considered when choosing a strategy for the statistical analysis of couple data. The first assumes a partner effect: A characteristic of partner A influences another characteristic of partner B. For example, partner A’s conflict behavior influences partner B’s report of relationship satisfaction and vice versa. The second model assumes mutual influence: A characteristic of partner A influences the same characteristic of partner B. For example, partner A’s reported conflict behavior influences partner B’s reported conflict behavior and vice versa. The third model assumes common fate: Partner A and B are exposed to the same causal factor. For example, the mutual conflict behavior of the couple influences their mutual report of relationship satisfaction (see Kenny & La Voie 1985). The choice of the appropriate model depends on the theory that is used and the particular model determines the type of statistical analysis that is performed (Kenny 1996). Kluwer, Heesink and Van de Vliert (1997) used a different method in a study on conflict interaction patterns in close relationships. They measured both spouses’ perceptions of several conflict interaction patterns that occurred between husband and wife (as opposed to individual behaviors). Positive intraclass correlations indicated that the dyads, rather than the individual dyad members, accounted for significant variation on the marital interaction variables (Kenny & La Voie 1985). They then computed the mean of the husband’s and the wife’s scores (Acitelli & Antonucci 1994; Christensen & Heavey 1993; Kluwer et al. 1996). In this case, there is no need to use the models of non-independence that assume two hierarchical units of analysis (individuals nested in dyads.) Instead, the couple is the only unit of analysis. Measures of joint conflict interaction patterns are couple constructs, as opposed to husbands’ and wives’ reports of their own, individual conflict behaviors. This approach improves the reliability or “robustness” of conflict measures, because assessments stem from two sources instead of one. Although computing the mean of both conflict parties’ perceptions of conflict interaction patterns can provide a valid measure of what goes on between the parties, we emphasize that this technique does not approach reality any more
130
AUKJE NAUTA AND ESTHER KLUWER
than observing actual interaction. It simply measures both parties’ joint perceptions regarding interaction patterns that occur during conflict. A weakness of this approach is that information about spouses’ individual perceptions is lost by aggregating to the dyad level. One way to use information at both the individual level and the dyadic level is to use multi-level analysis (Raudenbush & Bryk 1986; Snijders & Bosker 1999). With multi-level analysis, it is possible to examine the impact of variables at both the dyadic and the individual level upon a dependent variable at the individual level. Moreover, multi-level analysis takes into account that individuals within dyads are interdependent. Multi-level analysis is comparable to regression analysis: Its goal is to build a model that expresses how the dependent variable can be explained by multiple independent variables. An important difference, however, is that the variance in the dependent variable is split up into variance at the individual level and variance at the dyadic (or group) level. This variance is explained by both variables at the group level (e.g., dyadic means on conflict behavior) and variables at the individual level (e.g., individual measures of conflict behavior). In sum, we recommend that, when it is theoretically necessary to couple data of both sides of a conflict, researchers should use analytic techniques that take into account the nested structure of the data.
Conclusions In this article, we have described some of the issues that researchers may come across when they examine conflict through the use of questionnaires. We discussed the sensitivity of conflict as a research topic which may hinder the recruitment of participants. Nevertheless, questionnaires may sometimes be more suitable to measure touchy subjects than more confrontational research methods such as face-to-face interviews. Furthermore, we discussed the validity of conflict questionnaires and the fact that conflict behavior is easily confounded with behavioral intentions. Specifically, we noted that questionnaire measures of conflict behavior are likely to be biased, because participants tend to give self-serving and/or socially desirable answers. In addition, we discussed the correlational nature of most questionnaire research designs. Finally, we described that conflict data are often dyadic in nature, which leads to specific methodological problems due to the dependency of dyad members. We would like to conclude with a few recommendations to conflict researchers who want to use questionnaires. First, conflict research generally needs
THE USE OF QUESTIONNAIRES IN CONFLICT RESEARCH
131
careful introduction, implementation and reporting to participants. In order to get enough participants, it may be helpful to avoid using the word “conflict” and stress that one wants to examine correlates of collaboration and interpersonal relationships. Second, validity problems may be solved by using existing conflict questionnaires and by using multiple data sources (e.g., self reports combined with opponent reports and/or observer reports). Not only are these measures of behavior less confounded with intentions, they are also less biased due to self-serving or social desirability tendencies. A third and last recommendation is to couple data only when theoretically necessary and use appropriate techniques to analyze dependent data. Although there are issues to be taken into account when using questionnaires, we believe that questionnaires can be a very appropriate method to study conflict. It enables researchers to approach large samples, both in the field and in the laboratory, and when administered appropriately, it is possible to gather reliable and valid data on conflict in real-life settings.
References Acitelli, L.K., and Antonucci, T.C. (1994). “Gender differences in the link between marital support and satisfaction in older couples,” Journal of Personality and Social Psychology 67:688–698. Belson, W.A. (1981). The design and understanding of survey questions. Aldershof: Gower. Blake, R., and Mouton, J.S. (1964). The Managerial Grid. Houston, TX: Gulf. Christensen, A., and Heavey, C.L. (1993). “Gender differences in marital conflict: The demand/withdraw interaction pattern.” In S. Oskamp and M. Constanzo (Eds.), Gender issues in contemporary society. London: Sage. 113–141. Cook, T.D., and Campbell, D.T. (1979). Quasi-experimentation. Design and analysis issues for field settings. Boston: Houghton Mifflin. De Dreu, C.K.W., Evers, A., Beersma, B., Kluwer, E.S., and Nauta, A. (2001). “A theory based measure of conflict management strategies in the work place,” Journal of Organizational Behavior 22:645–668. De Dreu, C.K.W., and Carnevale, P.J. (2003). “Motivational bases of information processing and strategy in conflict and negotiation.” In M.P. Zanna (Ed.), Advances in Experimental Social Psychology. Vol. 35. New York: Academic Press 235–291. De Dreu, C.K.W., Harinck, F., and Van Vianen, A.E.M. (1999). “Conflict and performance in groups and organizations.” In C.L. Cooper, and I.T. Robertson (Eds.), International Review of Industrial and Organizational Psychology. Vol. 14. Chichester, UK: Wiley. 376–405. De Dreu, C.K.W., Nauta, A., and Van de Vliert, E. (1995). “Self-serving evaluation of conflict behavior and escalation of the dispute,” Journal of Applied Social Psychology, 25, 2049–2066. De Vaus, D.A. (1991). Surveys in social research. London: UCL Press/Allen and Unwin. Dillman, D.A. (1978). Mail and telephone surveys: The total design method. New York: Wiley. Euwema, M.C., and Van de Vliert, E. (1990). “Gedrag en escalatie bij hiërarchische conflicten [Behavior and escalation in hierarchical conflicts],” Toegepaste Sociale Psychologie, 4:28–41. Flanagan, J.C. (1954). “The critical incident technique,” Psychological Bulletin 51:327–359.
132
AUKJE NAUTA AND ESTHER KLUWER
Hall, J. (1969). Conflict management survey. Conroe, Tex.: Teleometrics International. Janssen, O., and Van de Vliert, E. (1996). “Concern for other’s goals: Key to de-escalation of conflict,” International Journal of Conflict Management 7:99–120. Kenny, D.A. (1995). “The effect of nonindependence on significance testing in dyadic research,” Personal Relationships 2:67–75. Kenny, D.A. (1996). “Models of non-independence in dyadic research,” Journal of Social and Personal Relationships 13:279–294. Kenny, D.A., and Kashy, D.A. (1991). “Analyzing interdependence in dyads.” In B.M. Montgomery and S. Duck (Eds.), Studying interpersonal interactions. New York: The Guilford Press. 275–285. Kenny, D.A., and La Voie, L. (1985). “Separating individual and group effects,” Journal of Personality and Social Psychology 48:339–348. Kidder, L.H., and Judd, Ch.M. (1986). Research methods in social relations. New York: CBS College Publishing. Killman, R.H., and Thomas, K.W. (1977). “Developing a forced choice measure of conflict handling behavior,” Educational and Psychological Measurement 37:309–325. Kluwer, E.S. (1998). “Responses to gender inequality in the division of family work: The status quo effect,” Social Justice Research 11:337–357. Kluwer, E.S. (2000). Procedural and distributive justice in close relationships: The moderating role of gender. Paper presented at the 10th International Conference of the International Society for the Study of Personal Relationships, Brisbane, Australia, June–July. Kluwer, E.S., De Dreu, C.K.W., and Buunk, B.P. (1998). “Conflict in intimate versus nonintimate relationships: When gender role stereotyping overrides biased self-other judgment,” Journal of Social and Personal Relationships 15:637–650. Kluwer, E.S., Heesink, J.A.M., and Van de Vliert, E. (1996). “Marital conflict about the division of household labor and paid work,” Journal of Marriage and the Family, 58, 958–969. Kluwer, E.S. Heesink, J.A.M., and Van de Vliert, E. (1997). “The marital dynamics of conflict over the division of labor,” Journal of Marriage and the Family 59:635–653. Kluwer, E.S. Heesink, J.A.M., and Van de Vliert, E. (2000). “The division of labor in close relationships: An asymmetrical conflict issue,” Personal Relationships 7:263–282. Landy, F. (1978). “Conflict management survey.” In O.K. Buros (Ed.), Eight mental measurement yearbook. Vol. 2. Highland Park, N.J.: Gryphon. 1173–1174. Murray, S.L., and Holmes, J.G. (1997). “A leap of faith? Positive illusions in romantic relationships.” Personality and Social Psychology Bulletin 23: 586–604. Nauta, A. (2003). “Perceived influence in team decisions: The more, the better for conflict management and health.” Paper presented at the 2003 Conference of the European Association for Work and Organizational Psychology, Lisbon, Portugal, May. Nauta, A., and Sanders, K. (2000). “Interdepartmental negotiation behavior in manufacturing organizations,” International Journal of Conflict Management 11:135–161. Nauta, A., and Sanders, K. (2001). “Causes and consequences of perceived goal differences between departments within manufacturing organizations,” Journal of Occupational and Organizational Psychology 74:321–342. Nauta, A., and Van Sloten, G. (2003). “Successfully dealing with imbalance at work [Succesvol omgaan met onbalans in arbeidsrelaties].” Manuscript submitted for publication. Nauta, A., De Dreu, C.K.W., and Van der Vaart, T. (2002). Social value orientation, organizational goal concerns and interdepartmental problem-solving behavior. Journal of Organizational Behavior, 23, 199–213. Nauta, A., De Vries, J., and Wijngaard, J. (2001). “Power and biased perceptions of interdepartmental negotiation behavior,” Group Processes and Intergroup Relations 4:263–270.
THE USE OF QUESTIONNAIRES IN CONFLICT RESEARCH
133
Pruitt, D.G., Peirce, R.S., McGillicuddy, N.B., Welton, G.L., and Castrianno, L.M. (1993). “Long-term success in mediation.” Law and Human Behavior 17:313–330. Putnam, L.L., and Wilson, C.E. (1982). “Communicative strategies in organizational conflicts: Reliability and validity of a measurement scale.” In M. Burgoon (Ed.), Communication Yearbook. Vol. 6. Beverly Hills, CA: Sage. 629–652. Rahim, A., and Magner, N.R. (1995). “Confirmatory factor analysis of the styles of handling interpersonal conflict: First-order factor model and its invariance across groups,” Journal of Applied Psychology 80:122–132. Raudenbush, S.W., and Bryk, A.S. (1986). “A hierarchical model for studying school effects,” Sociology of Education 59:1–17. Rubin, J.Z., Pruitt, D.G., and Kim, S. (1994). Social conflict: Escalation, stalemate, and settlement. New York: McGraw-Hill. Schuman, H. and Scott, J. (1987). Problems in the use of survey questions to measure public opinion. Science, 236, 957–959. Shaughnessy, J.J., Zechmeister, E.B., and Zechmeister, J.S. (2000). Research methods in psychology. Boston: McGraw Hill. Shockley-Zalabak, P. (1988). “Assessing the Hall Conflict Management Survey,” Management Communication Quarterly 1:302–320. Snijders, T.A.B., and Bosker, R.J. (1999). Multi-level analysis. London: Sage. Tanur, J.M. (1992). Questions about questions: Inquiries into the cognitive bases of surveys. New York: Russell Sage Foundation. Thomas, K.W. (1992). “Conflict and negotiation processes in organizations.” In M.D. Dunnette and L.M. Hough (Eds.), Handbook of industrial and organizational psychology 2nd ed., Vol. 3. Palo Alto, CA: Consulting Psychologists Press. 651–717. Thomas, K.W., and Killman, R.H. (1974). The Thomas-Kilmann conflict mode instrument. Tuxedo, N.Y.: Xicom. Thomas, K.W., and Killman, R.H. (1978). “Comparison of five instruments measuring conflict behavior,” Psychological Reports 42:1139–1145. Van de Vliert, E. (1997). Complex interpersonal conflict behaviour. Theoretical frontiers. Hove: Psychology Press. Van de Vliert, E. “Conflict and conflict management.” In P.J.D. Drenth and H. Thierry (Eds.) Handbook of work and organizational psychology 2nd ed., Vol. 3. Personnel psychology. Hove, UK: Psychology Press. 351–376.
A Multilevel Approach to Investigating Cross-National Differences in Negotiation Processes XU HUANG and EVERT VAN DE VLIERT
A successful instance of international negotiation is largely dependent on the extent to which the negotiating parties are capable of understanding the negotiating patterns of their counterparts who come from another culture (Brett 2000). Analogously, the cross-national validity of a negotiation theory is largely dependent on the extent to which scholars are capable of understanding and incorporating differences in negotiation processes across oceans and borders. Those observations add a national level of analysis to the commonly used individual and dyadic level in negotiation and conflict research. By far the greatest number of studies on cross-national negotiations are concerned with how national context shapes the psychological states of the negotiators (Brett & Okumura 1998, Fu & Morris 2000; Gelfand & Christakopoulou 1999), the social conditions of the negotiations (Carnevale, Pruitt, and Britton 1979; Graham, Kim, Lin, and Robinson 1988), and the behavior of the negotiators (McCusker 1994). In the same vein, many researchers have observed that the impact of the psychological states of the negotiators (e.g., motives and biases of judgment), the social conditions of the negotiations (e.g., constituencies and negotiator relationships), and the behavior of the negotiators (e.g., tactics and communication) on the outcomes of negotiations systematically differ from country to country. They have proposed several theoretical models examining the moderating effects of national contextual variables on such negotiation processes (Gelfand & Dyer 2000; Morris & Fu 2001). However, only a few models of this kind have been empirically tested (e.g., Gelfand & Realo 1999; Tinsley & Pillutla 1998). A common feature of these studies is that the moderating effects of national contextual variables, such as cultural individualism, were tested at the level of individual workers. This was done by measuring cultural individualism at the individual level. However, the meaning of a construct measured at the individual level may differ from that of a construct represented by its aggregated scores at the country level (Van de Vijver & Poortinga 2002).
International Negotiation Series 2: P. Carnevale and C.K.W de Dreu (eds.) Methods of Negotiation Research, 135–148 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
136
XU HUANG AND EVERT VAN DE VLIERT
Consequently, it remains problematic to generalize the individual-level findings to an explanation of cross-national differences in negotiation processes. To tackle this problem, we propose a new approach to research on crossnational negotiations, the cross-level approach, which integrates country-level analysis and individual-level analysis. We demonstrate that, using multilevel modeling, a statistical technique designed for relating distinct levels of analysis in a holistic analytical model, researchers are able to more accurately and directly test the moderating effects of country-level variables on individuallevel relationships. As a result, we may develop a better understanding of why the psychological states of the negotiators, the social conditions of the negotiations, and the behavior of the negotiators impact differently on negotiation outcomes across countries. In the following sections, we first briefly review the major methodological problems of individual-level examinations of the moderating effects of national variables on the negotiation processes. Then, based on a series of crossnational research projects that used multilevel modeling, we delve into how recent methodological developments contribute to a more refined cross-level approach to investigating the moderating effects of national contextual variables. Finally, we discuss the valuable methodological contributions of this cross-level approach and how they may instigate new perspectives on theory building in international negotiation research.
The Nation-As-Moderator at the Individual Level Cross-cultural differences in negotiation processes abound. For instance, Graham (1983) found that the role of the negotiator was the most important predictor of negotiation outcomes in Japan, deceptive tactics were the most important predictors of outcomes in Brazil, and cooperative tactics were the most important predictors in the U.S. Although such facts are certainly interesting, scientifically they are of no great help as they do not offer theorybased explanation of such cross-national variations. In order to address such variations in negotiation processes on a theoretical basis, various attempts have been made to specify country-level conditions that change the links between negotiation phenomena. For instance, Shapiro and Rognes (1996) used national culture to understand why the perceived orientation to dominate of the opponents was positively linked with joint outcomes among Americans, but negatively linked with joint outcomes among Norwegians. Using Hofstede’s (1991) framework, the authors explained that Americans, having a more individualistic culture than the Norwegians, tend to
CROSS-NATIONAL DIFFERENCES IN NEGOTIATION PROCESSES
137
socialize their inhabitants to be more comfortable with the dominating behavior of their opponents. Although Shapiro and Rognes’s method does provide a crude theoretical explanation of cross-national differences in negotiation processes, a major weakness of this inference-based approach is that, in the absence of concrete statistical evidence, it is hard to rule out rival explanations (e.g., Lytle, Brett, Barsness, Tinsley, and Janssens 1995). Gelfand and Realo’s (1999) recent work was among the first to statistically test the moderating effects of national contextual variables on negotiation processes. The model they used can be referred to as the nation-as-moderator model. Their findings showed that the well-known link between accountability to constituents and negotiation outcomes was moderated by cultural individualism, which was measured at the individual level. More precisely, accountability was found to be negatively related to joint profit among individualists, but positively related to joint profit among collectivists. The authors reasoned that, among individualists, accountability activated competitive construals and behavior, and resulted in lower outcomes. By contrast, among collectivists, accountability activated cooperative construals and behavior, and resulted in higher outcomes. However, cultural individualism and collectivism measured and analyzed at individual level do not necessarily equate to individualism and collectivism at the country level (Van de Vijver & Poortinga 2002), with the effect that the implications of Gelfand and Realo’s (1999) study were in fact limited to the individual level. Even if a cultural dimension measured at the individual level is structurally equivalent to its aggregation at the country level, there is still a serious potential problem with estimating the contextual effect at the individual level; that is, individuals living in the same country are influenced by the same national context, which may be reflected in the within-country statistical interdependence of their responses. Yet, testing the moderating effects of cultural dimensions measured at the individual level will largely ignore such within-country between-subject interdependence, leading to an underestimation of standard errors and an overestimation of effect sizes (Bryk & Raudenbush 1992; Kreft & De Leeuw 1998; Snijders & Bosker 1999). To sum up, the above method of examining the nation-as-moderator model at the individual level has three major problems. First, cultural or other national dimensions measured at the individual level may have meanings that are different from those of their aggregations at the country level. Second, testing the moderating effects of national contextual variables measured and analyzed at the individual level ignores the statistical dependence of individuals who share the same culture, and leads to an underestimation of standard errors. Third, rival national moderating variables cannot be statistically
138
XU HUANG AND EVERT VAN DE VLIERT
controlled. Therefore, in the following section, based on a series of crossnational research projects, we discuss a way to tackle the above problems by employing multilevel modeling, a statistical technique that can be used to test the nation-as-moderator model across levels of analysis.
Cross-Level Nation-As-Moderator Model A series of cross-national studies on job satisfaction and motives for volunteer work have contributed to a more sophisticated analytical model that permits a systematic and quantitative examination of cross-national differences in the theoretical relationships at the individual level. The common thread of those studies is that the national context can be statistically treated as a moderator of the link between two variables at the individual level in a cross-level model. This cross-level nation-as-moderator model integrates two levels of analysis, country-level analysis and individual-level analysis. Statistically, the integration of the two levels of analysis can be achieved by using multilevel modeling. Multilevel Modeling Multilevel modeling or hierarchical linear modeling (HLM) is designed to analyze data with a nested structure, such as people nested within countries. It is particularly useful to combine two levels of analysis in a single multilevel model. Data analyses are usually performed using Mlwin or HLM/2, which are computer packages for multilevel modeling (Bryk, Raudenbush, and Congdon 1994; Goldstein et al. 1998). For example, Mlwin produces an estimate for each predictor variable along with the associated standard error. These estimates are comparable to the unstandardized regression coefficients in ordinary regression analysis, and their significance level can be tested using t-distribution tests. Moreover, Mlwin produces a statistic called the deviance, which indicates how well a given model fits the data. If two models are nested, the difference in the deviances of the two models has a chi-square distribution with degrees of freedom equal to the difference in the number of parameters estimated. A formal chi-square test can then be performed to examine whether the more general model fits significantly better than the simpler model. In addition, Mlwin or HLM/2 splits the total variance of the dependent variable into two residual variances – in our case, the residual variance at the individual level and the residual variance at the country level. Thus, independent variables from the individual level and country level can be entered into the model to “explain” the residual variances at both levels, respectively.
CROSS-NATIONAL DIFFERENCES IN NEGOTIATION PROCESSES
139
These residual variances will decrease if the independent variables from the two levels have a significant impact on the dependent variable. Mlwin can further split the country-level residual variance of the dependent variable into the residual variance of intercepts and the residual variance of slopes. The residual variance of intercepts (cf. means) refers to the variations in the level of the dependent variable across nations. The residual variance of slopes (cf. correlations) refers to the cross-national variations in the relationship between an independent variable and the dependent variable. The residual variance of slopes will also decrease if a cross-level interaction term has a significant impact on the dependent variable. Job Satisfaction Research In this section, we describe a recent large-scale cross-national research project that has led to useful insights into how the cross-level nation-as-moderator model can be applied in research. Moreover, we illustrate that some variants of the cross-level model can be developed further to deal with specific theoretical issues in cross-national research. By employing a cross-level nation-as-moderator model (Figure 1), we (Huang & Van de Vliert, 2004) conducted a study on how the relationship between job level and job satisfaction is contingent on a national contextual variable, that of cultural individualism. Individualism, one of Hofstede’s (1991) cultural dimensions, was measured at the country level, while job level and job satisfaction were measured at the individual level. We based our analyses on data from around 100,000 workers employed in 39 countries. The data were drawn from a survey conducted in a multinational company in 2000. All of the items were translated into the languages of the countries under investigation by professional translation agencies in those countries. The questionnaires were administered to employees through the management. Employees were told that the answers would be kept completely anonymous and that the management would not be able to identify the individual respondents. The response rate was about 75 percent. The sample contained 52 percent blue-collar workers, 48 percent white-collar workers, 62 percent males, and 38 percent females. The average age was 36 and the average length of tenure was 10 years. The multilevel analyses were conducted using an Mlwin computer package. The findings contradict some long-held beliefs, principally the myth of the universal positive association between intrinsically motivating jobs and job satisfaction (e.g., Locke 1976). First of all, we confirmed the expectation that job level (blue-collar workers versus white-collar workers) is positively related to job satisfaction in individualistic countries, but is not significantly associated with job satisfaction in
140
XU HUANG AND EVERT VAN DE VLIERT
collectivistic countries. We reasoned that more intrinsically motivating whitecollar work is more likely to engender more job satisfaction in individualistic countries because in those countries workers are socialized to place more emphasis on self-actualization needs and self-esteem needs. In collectivistic countries, however, more intrinsically motivating white-collar work does not necessarily produce more satisfaction than less intrinsically motivating bluecollar work due to the fact that, in collectivistic countries, self-actualization and self-esteem are not ranked as higher-order needs by workers (e.g., Stigler, Smith, and Mao 1985; Page & Cheng 1992; Chiu 1993; Diener & Diener 1995; Diener, Diener, and Diener 1995). It should be noted that the above analysis was conducted after the main effect of national wealth and its interactive effect with job level had been controlled for. This was done by entering national wealth (in terms of income per capita) and the interactive term of national wealth and job level in the very first step of the analysis. The effect of national wealth was controlled because national wealth correlates strongly with cultural individualism (Hofstede 1991). In addition to the simple cross-level two-way interaction model, we also explored the more complex cross-level three-way interactive effect of cultural individualism, job level, and opportunity to use one’s skills and abilities on job satisfaction. Presented in Figure 1 is the corresponding hypothetical path diagram, which depicts a national contextual variable (cultural individualism) and an individual-level organizational variable (opportunity to use skills and abilities) jointly influencing the link between job level and job satisfaction. From a somewhat different perspective, national culture was expected to be systematically associated with cross-national variations in the individual-level two-way interactive effect of job level and opportunity to use skills and abilities on job satisfaction. One of the interesting findings shown in Figure 2 indicates that, in collectivistic countries, blue-collar work versus white-collar work and job satisfaction were strongly negatively linked where there was little opportunity to use one’s skills and abilities in the job (simple slope: b = –.18, p < .01), but were moderately positively linked in case of many such opportunities (simple slope: b = .11, p < .05). We argued that a possible explanation is that white-collar workers in collectivistic countries may generally have a higher level of education, and thus be more easily influenced by the individualistic values of the West (cf. Inglehart 1997), to the extent that they miss the highly valued intrinsic rewards of opportunities to use one’s skills and abilities more than do blue-collar workers. Apparently, the crosslevel nation-as-moderator model seems very useful in quantitatively testing crossnational variations in even more complex individual-level relationships.
CROSS-NATIONAL DIFFERENCES IN NEGOTIATION PROCESSES
Country Level
141
Cultural Individualism
Individual Level
Job Level
Job Satisfaction
Opportunity to Use Skills and Abilities Figure 1. Country-Level Cultural Individualism and Individual-Level Opportunity to Use Skills and Abilities as Joint Moderators of the Individual-Level Link between Job Level and Job Satisfaction Much Opportunity to Use Skills and Abilities
Little Opportunity to Use Skills and Abilities 4
4
Job Satisfaction
Job Satisfaction
Individualism 3.5 Collectivism 3
3.5 Collectivism 3
Individualism 2.5 –1.5
2.5 –0.5 Blue-Collar
0.5
1.5
–1.5
White-Collar
–1
–0.5
Blue-Collar
0
0.5
1
1.5
White-Collar
Figure 2. Three-Way Interactive Effect of Cultural Individualism, Blue-Collar versus WhiteCollar Work, and Opportunity to Use Skills and Abilities on Job Satisfaction
Research on Motives for Doing Volunteer Work In another study, we (Van de Vliert, Huang, and Levine, 2004) investigated reasons for doing volunteer work. We included 33 countries whose residents’ reasons for doing unpaid voluntary work had been assessed through the 1990–1993 World Values Survey (World Values Study Group 1994). The surveys were carried out through face to face interviews by professional researchers, with
142
XU HUANG AND EVERT VAN DE VLIERT
a sampling universe consisting of all adult citizens, aged 18 and older (after a listwise deletion of missing values N = 13,584; M = 412 per nation; Mage = 40, SDage = 15.2; 47 percent female; 65 percent with paid jobs). There are two kinds of reasons for doing voluntary work. People may take up volunteer jobs for either self-serving or egoistic reasons, or for other-serving or altruistic reasons, or for both. We argued that the two types of motives might cooperate or compete with each other depending on national contextual variables. In particular, in rich nations, self-serving and other-serving motivations might well be positively linked because self-sacrificing is a way of ‘establishing one’s self-identity, confirming one’s notion of the sort of person one sees oneself to be, and expressing the values appropriate to this self-concept’ (Katz & Kahn 1978: 361). By contrast, in poor nations, volunteer workers might have to make a relatively serious choice between serving their own interests and acting out of altruistic or humanitarian concerns. In addition, as people have to consume more resources to cope with extremely cold or hot climates, we hypothesized that, in countries with more demanding cold or hot climates, national wealth produces competition rather than cooperation between self-serving and other-serving motivations. In countries with more temperate climates, however, the degree of national wealth was supposed to have less influence on the cooperative or competitive nature of the complex motivation to engage in volunteer work. Depicted in Figure 3 is the path diagram of the variant of the cross-level nation-as-moderator model that represents those expectations. A stepwise multilevel analysis was conducted to predict other-serving motivations for doing voluntary work. In step 1, country-level individualism was controlled for because this cultural dimension is positively associated with both colder climates and national wealth. In steps 2 and 3, the individual-level motivation to serve one’s own interests was entered (the degree of self-serving motivation in step 2, the cross-national variation in the relationship between self-serving motivation and other-serving motivation in step 3). Step 4 contained average climatic temperature, temperature-squared and gross national income per capita. The two-way interactions followed in step 5, and the threeway interactions in step 6. The analysis revealed that cultural individualism at the country level does not play a part (step 1), that self-serving motivation at the individual level is positively related to other-serving motivation (step 2), and that the link between self-serving motivation and other-serving motivation varies considerably from one nation to another (step 3). The most important finding, in step 6, represented in Figure 4, showed that a voluntary worker’s self-serving and other-serving motivations tend to be positively linked in high-income regions
CROSS-NATIONAL DIFFERENCES IN NEGOTIATION PROCESSES
Country Level
Individual Level
National Wealth
Self-Serving Motives
143
Climatic Temperature
Other Serving Motives
Figure 3. Country-Level National Wealth and Climatic Temperature as Joint Moderators of the Individual-Level Link between Self-Serving and Other-Serving Motives for Volunteer Work
with cold or hot climates (e.g., Scandinavia), tend to be unrelated in high – and low-income regions with temperate climates (e.g., Southern Europe and the southern part of South America) and in low-income regions with hot climates (e.g., West Africa around the Gulf of Guinea), and tend to be negatively linked in low-income regions with cold climates (e.g., Baltic States). We concluded that, in depressed countries where the financial resources are minimal and the climatic demands maximal, volunteering seems to be experienced as subordinating self-serving reasons to other-serving reasons for doing unpaid work. Some scholars think that such a multilevel approach is unnecessarily complicated, and does not add much value. As a case in point, one of the reviewers of our paper made the following comment: “I have no objection to Multilevel Modeling, but an alternative is to use the correlation between the two motives (after a Fisher transformation) as the dependent variable. A countrylevel ordinary regression can then be performed on this variable with income, temperature, and their interaction as predictors. This analysis is much easier to follow, and it is not clear to me in what way the current analysis based on Multilevel Modeling is superior to this country-level regression analysis.” This is not an isolated opinion. What is more, others have actually used the method proposed by this reviewer. Diener and Diener (1995), for example, used 31 within-country correlations to show that the link between self-esteem and life satisfaction varies substantially across nations. They reported that the correlation between self-esteem and life satisfaction was .60 for people from North America, but less than .01 for people from India. Across the 31 countries the size of the correlation coefficients correlated .53 with cultural individualism. Viewed from our perspective, Diener and Diener (1995) used multilevel analysis, but did so in a sequential rather than simultaneous fashion. By conducting two subsequent analyses, first at the individual level and then at the country level, they demonstrated that the cross-national variations in a
144
XU HUANG AND EVERT VAN DE VLIERT
Self-Serving Motivation
1.5
4.5 3.5 2.5 1.5 Rich Countries 0.5 –0.5 –1.5 Poor Countries –2.5 –3.5 –4.5 –1.5 –0.5 0.5 Self-Serving Motivation
Hot Climate Other-Serving Motivation
4.5 3.5 Poor Countries 2.5 1.5 0.5 –0.5 –1.5 –2.5 Rich Countries –3.5 –4.5 –1.5 – 0.5 0.5
Temperate Climate Other-Serving Motivation
Other-Serving Motivation
Cold Climate
1.5
4.5 3.5 2.5 1.5 0.5 –0.5 –1.5 –2.5 –3.5 –4.5 –1.5
Poor Countries
Rich Countries –0.5
0.5
1.5
Self-Serving Motivation
Figure 4. National Wealth and Thermoclimate Jointly Influence the Link between Self-Serving and Other-Serving Motivations
particular psychological relationship, the link between self-esteem and life satisfaction, is systematically related to cultural individualism. However, this two-step approach neglects the problem of differential reliability. Differential reliability occurs when the number of respondents varies considerably across the nations under investigation. Oishi, Diener, Lucas, and Suh (1999) pointed out that the number of respondents in Diener and Diener’s (1995) study ranged from 29 to 985. Consequently, the correlation coefficient computed in Cameroon with 29 respondents is less reliable than the correlation coefficient obtained in Canada with 985 respondents. Oishi et al. (1999) further argued that multilevel modeling resolves the problem of differential reliability caused by directly relating correlation coefficients drawn from separate countries to the national characteristics of these countries. In multilevel analysis, the differential reliability becomes less problematic because regression lines instead of correlation coefficients are used as dependent variables. In short, the studies on job satisfaction and motives for volunteer work discussed in the preceding paragraphs demonstrate that each of the variants of the cross-level model shown in Figures 1 and 3 can be used to tackle specific issues in cross-national research. More importantly, by linking national characteristics with the cross-national variations in individual-level relationships in a cross-level model, one can predict and explain the variations in these individual-level relationships across nations. Last, but not least, using multilevel modeling, researchers can control for a number of rival explanatory variables at the country level, and can even examine the joint effects of different country-level variables on individual-level relationships.
CROSS-NATIONAL DIFFERENCES IN NEGOTIATION PROCESSES
145
Implications for Research on Negotiations In this article, we challenged the methodological approaches used to investigate and explain cross-national differences in the links between the psychological states of the negotiators, the social conditions of the negotiations and the behavior of the negotiators, on the one hand, and outcomes of negotiations, on the other. We pointed out that there are three major problems in existing methods of testing the nation-as-moderator model at the individual level. Using the results of large-scale cross-national studies on job satisfaction and motives for volunteer work, we illustrated that multilevel modeling allows researchers to establish a more statistically rigorous cross-level nation-asmoderator model, which can help resolve or mitigate the three problems. First, as shown in Figure 5, researchers can directly use the country-level variables in a multilevel model of the negotiation processes; thus avoiding the problem of the cross-level construct equivalence of the cultural variables (Van de Vijver & Poortinga 2002). Second, multilevel modeling yields more accurate estimates of standard errors than do zero-order correlation analysis and ordinary regression analysis conducted at the individual level, because it takes into account variances at both the country level and the individual level. Lastly, several country-level variables, representing rival explanations, can be controlled and analyzed at the same time in a multilevel model. Cultural individualism is just one of the national contextual variables that may exert a significant influence on the negotiation processes. Other national contextual variables such as political rights, civil liberty, and dominant religion may also influence processes of negotiation (Bercovitch & Elgström 2001). Figure 4, for example, suggests the crude hypothesis that negotiators living in more demanding climates experience more interdependence between concern for their own goals and concern for the other party’s goals, and that this interdependence is of a more positive nature in wealthier countries. In other words, using multilevel analysis, researchers can test the joint impact of two national characteristics on individual-level negotiation processes. Additionally, in reality, individuals are subject to the influence of variables from different levels, such as the team level and organizational level. Multilevel modeling allows researchers to build a three-level model – individuals are nested within teams that are nested within nations – to examine how national contexts and team contexts jointly influence individual-level negotiation processes. Despite all of these virtues, the application of multilevel modeling to crossnational research has some limitations as well. To test a cross-level model by using multilevel modeling, a minimum of 25 countries is usually required (Snijders
146
XU HUANG AND EVERT VAN DE VLIERT
Country Level
Individual Level
National Contextual Variables e.g., Individualism Power Distance Political System
Negotiators’ Psychological States Social Conditions of Negotiation Negotiators’ Behavior
Negotiation Outcomes
Figure 5. Multilevel Model of Cross-National Differences in Negotiation Processes
& Bosker 1999). However, in practice, it is costly to collect data from such a large number of countries. Another limitation is that only rather simple individual-level relationships can be tested. As shown in the studies reported above (Huang & Van de Vliert, forthcoming; Van de Vliert et al., 2004), multilevel modeling only allows for tests of the moderating effects of national contextual variables on a single individual-level relationship or on an individual-level two-way interaction effect. As a rule, negotiation models proposed by various researchers are characterized by a rather complex structure (Gelfand & Dyer 2000; Morris & Fu 2001). It is difficult to test these negotiation models using multilevel analysis. Lastly, the research model proposed here can be used to study and explain cross-national variations in negotiating behavior and outcomes, but not the behavior and outcomes of people from different countries during negotiation. These limitations notwithstanding, the cross-level nation-as-moderator model may bring about new perspectives on theory building. The basic assumption of the cross-level model is that the impact of the negotiators’ psychological states, social conditions of negotiation, and behavior of the negotiators on negotiation outcomes may systematically vary across nations. In some nations, some strategies may have a positive effect on outcomes; in others, there may be no effect at all. There may even be a negative effect in certain nations. These cross-national variations may exist in a systematic manner. And this systematic variation may well be explained by some national contextual variables. Put in a different way, using multilevel modeling, we can predict the applicability of certain theoretical relationships across nations. The cross-level nation-as-moderator model permits researchers to generalize about the cross-national applicability of individual-level negotiation processes. We
CROSS-NATIONAL DIFFERENCES IN NEGOTIATION PROCESSES
147
call this higher level of generalization of theories, a meta-theoretical perspective. In a nutshell, the cross-level approach discussed here may not only offer an alternative and a more statistically rigorous way of testing crossnational variations in the effectiveness of negotiation practices, but may also offer an alternative theoretical perspective for researchers to map and explain cross-national variations in negotiating behavior.
References Brett, J.M. (2000). “Culture and Negotiation,” International Journal of Psychology 35(2): 97–104. Brett, J.M., and Okumura, T. (1998). “Inter – and intracultural negotiations: US and Japanese negotiators,” Academy of Management Journal 41: 495–510. Bryk, A.S. and Raudenbush, S.W. (1992). Hierarchical Linear Models: Applications and Data Analysis Methods. London/New Delhi: Sage. Bryk, A.S., Raudenbush, S.W., and Congdon, R.T., Jr. (1994). Hierarchical Linear Modeling with the HLM/2L and HLM/3L Programs. Chicago: Scientific Software International. Carnevale, P.J., Pruitt, D.G., and Britton, S.D. (1979). “Looking tough: The negotiator under constituent surveillance,” Personality and Social Psychology Bulletin 5: 118–121. Chiu, L.H. (1993). “Self-esteem in American and Chinese (Taiwanese) children,” Current Psychology: Research and Reviews 11: 309–313. Diener, E., and Diener, M. (1995). “Cross-cultural correlates of life satisfaction and selfesteem,” Journal of Personality and Social Psychology 68(4): 653–663. Diener, E., Diener, M., and Diener, C. (1995). “Factors predicting the subjective well-being of nations,” Journal of Personality and Social Psychology 69(5): 851–864. Fu, H.Y. and Morris, M.W. (2000). “Need for closure fosters adherence to cultural norms: Evidence from cross-cultural studies of conflict resolution choices,” Stanford GSB Research Paper, #1649. Gelfand, M.J., and Realo, A. (1999). “Individualism-collectivism and accountability in intergroup negotiations,” Journal of Applied Psychology 84(5): 721–736. Goldstein, H., Rasbash, J., Plewis, I., Draper, D., Browne, W., Yang, M., Woodhouse, G., and Healy, M. (1998) A user’s guide to Mlwin. University of London. Graham, J.L., Kim, D.K., Lin, C.Y., and Robinson, M. (1988). “Buyer-seller negotiations around the Pacific Rim: Differences in fundamental exchange processes,” Journal of Consumer Research 15: 48–54. Hofstede, G.H. (1991). Cultures and Organizations: Software of the Mind. London: McGraw-Hill. Huang, X., and Van de Vliert, E. (2004). “Job level and national culture as joint roots of job satisfaction,” Applied Psychology: An International Review, 53: 329–348. Inglehart, R. (1997). Modernization and Postmodernization: Cultural, Economic and Political Change in 43 societies, Princeton: Princeton University Press. Kreft, I., and De Leeuw, J. (1998). Introducing Multilevel Modeling. London, Thousand Oaks; New Delhi: Sage Publications. Locke, E.A. (1976). “The nature and causes of job satisfaction.” In M.D. Dunnette, (Ed.) Handbook of Industrial and Organizational Psychology. Chicago: Rand McNally College Publication.
148
XU HUANG AND EVERT VAN DE VLIERT
Lytle, A.L., Brett, J.M., Barsness, Z.I., Tinsley, C.H., and Janssens, M. (1995). “A paradigm for confirmatory cross-cultural research in organizational behavior,” Research in Organizational Behavior 17: 167–214. Maslow, A.H. (1954). Motivation and Personality. New York: Harper & Row. McCusker, C.R. (1994). “Individualism-collectivism and relationships in distributive negotiation: An experimental analysis,” Unpublished Dissertation. University of Illinois at UrbanaChampaign. Oishi, S., Diener, E.F., Lucas, R.E., and Suh, E.M. (1999). “Cross-cultural variations in predictors of life satisfaction: Perspectives from needs and values,” Personality and Social Psychology Bulletin 25(8): 980–990. Snijders, T.A.B., and Bosker, R.J. (1999) Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling. London: Sage. Stigler, J.W., Smith, S., and Mao, L. (1985). “The self-perception of competence by Chinese children, Child Development,” 56: 1259–1270. Van de Vijver, F.J.R. and Poortinga, Y.H. (2002). Structural equivalence in multilevel research, Journal of Cross-Cultural Psychology 33: 141–156. Van de Vliert, E., Huang, X., and Levine, R.V. (2004). “National wealth and thermal climate as predictors of motives for volunteer work,” Journal of Cross-Cultural Psychology, 35: 62–73. World Values Study Group (1994). World values survey, 1981–1984 and 1990–1993. Ann Arbor: Inter-university Consortium for Political and Social Research (ICSPR).
Methodological Challenges in the Study of Negotiator Affect BRUCE BARRY and INGRID SMITHEY FULMER
The role of affect is theoretically and empirically underdeveloped in the literature on conflict and negotiation. Although negotiation can be construed as “a natural setting for the study of affect” (Brief & Weiss 2002: 295), negotiation research with its emphasis on cognition “has ignored most emotion-relevant variables” (Bazerman, Curhan, Moore, and Valley 2000: 285). Beginning with a trickle of papers in the late 1980s, rising to the level of a modest stream by the mid to late 1990s, the subject of affect in negotiation has begun to attract substantial attention from negotiation researchers. Still, the overall volume of completed research in this area is sufficiently small that a new review chapter (Barry, Fulmer and Van Kleef, 2004) was able to descriptively review virtually every published study in a handful of pages. Affect is used here to refer to the constellation of responses that comprise both emotions and moods. Emotions are said to be more differentiated and of shorter duration, whereas moods are more enduring and pervasive, if generally of lower intensity (Forgas 1992; Park, Sims and Motowidlo 1986). Emotions and moods are interdependent: moods influence the likelihood that emotions will be triggered, while emotions are able to elicit particular moods (Davidson 1994). Affect captures ephemeral responses that are situationally driven (state affect) as well as stable tendencies to experience emotion or mood states that are dispositionally based (trait affect; e.g., Watson, Clark and Tellegen 1988). This article examines methodological issues that are posed by attention to affective processes in bargaining and negotiation. As a starting point, we must recognize that the role of affect in negotiation can be said to have both experiential and strategic components. The experience of affect speaks to the genuine emotional responses that negotiators encounter as they react to interaction at the bargaining table. The strategic component captures the affective stances and reactions that negotiators consciously bring to bear on their opponents as ploys designed to influence the other party’s perceptions and/or actions within the encounter (Barry 1999). Both perspectives raise challenging methodological issues.
International Negotiation Series 2: P. Carnevale and C.K.W de Dreu (eds.) Methods of Negotiation Research, 149–164 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
150
BRUCE BARRY AND INGRID SMITHEY FULMER
In the pages that follow, we first briefly review existing research on affect in negotiation with particular attention to methodological approaches taken in empirical studies. We then point to methodological issues raised in these studies and consider possible remedies. The latter discussion is organized around the distinction between affective experience and affect strategy in negotiation. Our objective is to augment the likelihood that researchers can capture the “real” emotions that individuals experience, express, mask, and strategically deploy within the negotiation encounter.
Existing Research on Emotion in Negotiation Following Barry et al. (2004), we can divide studies of affect in negotiation into three categories: (a) research examining the effects of pre-existing mood states or emotions on negotiation – these are studies where emotion is treated as a predictor of process or outcome; (b) investigations of emotion experienced during or following negotiation – these are studies where emotion is an experienced consequence of interaction with a negotiation counterpart; and (c) researching exploring the tactical use of emotion within the bargaining encounter and its strategic value for achieving negotiation aims. Affect as Predictor (Anterior Mood/Emotion) Research exploring the predictive role of affect in negotiation has been grounded methodologically in the practice of mood induction. A long line of experimental research has shown that moods have a priming effect on various cognitive and information processing behaviors related to altruism, flexibility, problem interpretation, information use, problem solving, and reward allocation (for reviews see Barry & Oliver 1996; Forgas 1998; Isen 1987). In studies of negotiator mood, participants in a laboratory or classroom setting are typically exposed to an experimental manipulation of mood, and then asked to role-play participation in a simulated (typically mixed-motive) negotiation exercise. Researchers have induced positive pre-negotiation moods by using humorous cartoons and small gifts (Carnevale & Isen 1986), and humorous videotapes (Kramer, Newton, and Pommerenke 1993), finding that positive mood is subsequently related to enhanced joint outcomes. Kramer et al. also found that positive moods distort negotiators’ perceptions of performance, increasing pre-negotiation confidence and post-negotiation self-ratings of performance.
METHODOLOGICAL CHALLENGES IN THE STUDY OF NEGOTIATOR AFFECT
151
Baron (1990) also attempted to manipulate positive mood, but with an ambient approach to induction; he exposed subjects to a pleasant odor. Negotiators exposed to pleasant odors had higher pre-negotiation aspirations and made more concessions than negotiators not exposed. Forgas (1998) took a broader approach to mood (and different approach to experimental induction), using false-feedback on a test of verbal ability to induce moods. In the positive mood condition, experimental participants were led to believe that they had achieved above-average performance on a difficult test. In the negative mood condition, participants believed that they had performed poorly on an easy test. A control group received no cues about test difficulty and no feedback on performance. Positive-mood negotiators were more likely to plan to be cooperative before the negotiation, used more cooperative negotiation strategies, achieved better outcomes, and were more inclined afterwards to honor the deal they reached. This study also explored the effect of mood on negotiators’ opponents, finding that negotiators in positive moods elicited more cooperative behavior and post-negotiation compliance with agreed-upon deals from their opponents. One study focused exclusively on the manipulation of negative affect: Allred, Mallozzi, Matsui, and Raia (1997) generated perceptions of the negotiation opponent’s culpability for actions as a way to induce opponent-directed anger. Following a simulated job contract negotiation, participants with these perceptions reported less regard for opponent’s interests, diminished accuracy of their judgments of the other party’s interests, lower joint gains, and diminished desire for future interaction with the other party. We are aware of one study that examined the effects of actual (rather than laboratory manipulated) affective states and traits on conflict resolution behaviors. Rhoades, Arnold, and Jay (2001) had employed individuals complete measures of mood and affective disposition, and document experiences of workplace conflict over three successive workdays. They found that positive mood and positive dispositional affect were associated with more concern for the other party, collaboration, and problem-solving during the resolution of organizational conflicts. Negative mood and negative affectivity were associated with higher concern for self and competitive behavior. Mood mediated the effects of disposition: affective disposition had no effect on motivation and behavior independent of mood. Affect as Consequence (Experienced Emotion) Here we are concerned with investigations of the post-negotiation or intranegotiation emotional reactions experienced by negotiators. As with studies
152
BRUCE BARRY AND INGRID SMITHEY FULMER
described above exploring moods as predictors, the prevailing empirical paradigm is the use of laboratory negotiation simulations. O’Connor and Arnold (2001) found that negotiators who failed to reach an agreement in a simulated two-party employment negotiation experienced more anger and frustration than negotiators who reached agreement. Self-efficacy moderated this association: more confident negotiators experienced less negative emotion following impasse. In a study of e-mail-based negotiation, Moore, Kurtzberg, Thompson, and Morris (1999) manipulated group affiliation and self-disclosure. Although this study was not oriented around emotion per se, findings indicated that a context favoring a personal relationship (negotiation with someone in one’s own group with self-disclosure of personal information) yielded expressions of positive affect within the interaction that increased rapport and decreased the likelihood of impasse. A study by Hegtvedt and Killian (1999) explored the connection between justice perceptions and post-negotiation emotion following a pay allocation task. Negotiators who saw the bargaining process as fair were more likely to experience positive emotion and less likely to express negative emotions such as agitation and resentment afterwards. Perceptions of the fairness of one’s own outcome increased satisfaction and decreased disappointment and resentment; outcome fairness was, however, also positively related to guilt. (Perceptions of the fairness of the other party’s outcomes had no effect on emotional reactions.) Pillutla and Murnighan (1996) used an ultimatum game to explore perceptions of anger and unfairness, and spiteful reactions in a correlational study. Findings indicated that perceptions of unfairness were substantially related to subsequent anger, particularly when individuals responding to a take-it-or-leave-it offer had complete information about the size of the resource being divided. As Pillutla and Murnighan put it, “Respondents expect fairness: When they perceive that an offer is unfair, many react angrily and spitefully reject beneficial offers” (1996: 220). A few studies have examined negotiator affect in non-simulation contexts. Scanzoni and Godwin (1990) investigated marital negotiations through retrospective recall of couple decision-making (a critical incident technique). After recalling and reconstructing incident details, husbands and wives each separately completed an “epilogue questionnaire” that included a measure of perceived affect toward the partner. Findings indicated that bonds between spouses that existed before negotiation predicted post-interaction affect. Also, past cooperative behavior by husbands was associated with wives’ report of affect, but curiously the converse was not found. In the communication literature, an exploratory study of crisis negotiation utilized a combined measure of language intensity and message valence as a marker for negotiator affect
METHODOLOGICAL CHALLENGES IN THE STUDY OF NEGOTIATOR AFFECT
153
(Rogan & Hammer 1995). Transcript analyses of professional negotiator and perpetrator speech patterns during three authentic crisis negotiations (suicide threat, domestic/hostage situation, and mental instability situation) suggested that affective responses often proceed through predictable phases, and that speech patterns may be related to successful versus unsuccessful crisis resolution. Affect as Tactic (Emotion Strategically Deployed) The strategic use of emotion as a tactical gambit in negotiation has been a recurrent theme in the theoretical and applied analysis of negotiation (see Lewicki, Barry, Saunders and Minton 2003), but has not been systematically addressed in the empirical negotiation literature. Studies of tactical communication behaviors by negotiators tend to emphasize cognitive, information-processing maneuvers, with attention to emotionally-oriented gambits as an isolated or peripheral matter. For example, Kimmel, Pruitt, Magenau, Konar-Goldband and Carnevale (1980) examined tactics of information exchange, heuristic trial and error, and distributive behavior. The use of threats and put downs, which could be construed as emotionally manipulative in some circumstances, was pooled with other actions (commitments; extraneous arguments) into a factor describing distributive behavior. Weingart, Thompson, Bazerman and Carroll (1990) examined the relationship between negotiated outcomes and tactics, including a category labeled “negative reaction,” which in turn included what the authors called “negative affect statements” by a bargainer. Subsequent studies by Weingart and her colleagues, however, addressed only cognitive tactics related to the exchange, evaluation, and substantiation of information and offers (e.g., Weingart, Bennett and Brett 1993; Weingart, Hyder and Prietula 1996). One study that indirectly considered the tactical role of emotion was Barry’s (1999) investigation of the relative perceived ethical appropriateness of emotionally deceptive negotiation tactics compared with other (cognitive) deceptive tactics such as bluffing or misrepresentation. Emotionally deceptive tactics include the false expression or suppression of either positive or negative emotion. Although results suggest that people are substantially more approving of and express more self-efficacy in their ability to deploy emotionally deceptive tactics as compared to other tactics of informational deception, methodological challenges (to be discussed at length later) have limited researchers’ ability to study these tactics in the laboratory or classroom.
154
BRUCE BARRY AND INGRID SMITHEY FULMER
Summary The most prevalent empirical technique for the study of affect in negotiation is the manipulation or measurement of affect as a precursor to interaction. Typically the form of affect that is induced or measured is a diffuse mood state (rather than a more ephemeral and intense emotional state). A few studies have assessed affect that arises within the negotiation encounter or as a consequence of a negotiated outcome. There has been almost no empirical attention to the use of emotion strategically as a negotiating gambit.
Methodological Issues and Remedies Theoretical treatments have offered up a rich variety of roles and mechanisms regarding affective processes at the bargaining table, few of which have been tackled empirically. Barry and Oliver (1996), for example, proposed that emotion states are relevant at multiple points within a negotiation encounter: (a) anticipatory emotion states resulting from ambient conditions, prior exchanges, and dispositional affect; (b) affect experienced within the negotiation that follows from offers, concessions and tactics; and (c) post-negotiation affect that results from outcomes and outcome-related attributions. Morris and Keltner (2000) proposed a social-functional model that identifies relational problems which trigger particular social emotions. These emotions, in turn, give rise to interaction behaviors within the negotiation encounter. In a related vein, Lawler and Yoon (1995) modeled relational development that emerges from the emotional consequences of repeated negotiations between the same parties. Thompson, Nadler, and Kim (1999) discussed processes of emotional transmission among negotiating parties, including emotional “contagion” (the transfer of emotion from one party to another) and emotional “tuning” (the construction of messages designed to control or regulate the other party’s emotional responses.) Allred (1999) proposed a model of retaliatory conflict describing how parties perceive each other’s harmful behavior, and either retaliate or defuse tensions, depending on the nature of cognitive appraisal and attribution. Taken together, the models just identified consider a broad and deep set of phenomena related to emotion in negotiation, yet few of the ideas embedded within them have been examined directly in empirical research. Arguably, the methods that have prevailed to this point are not well suited to testing these
METHODOLOGICAL CHALLENGES IN THE STUDY OF NEGOTIATOR AFFECT
155
kinds of models. In the following sections, we discuss methodological constraints and possible remedies. A Major Problem: Capturing the Experience of Affect Outside the Laboratory Negotiation researchers have been relatively successful in making the case that simulations (together with incentives like extra credit or token cash payments) generate sufficient mental engagement among study participants to justify the generalizability of laboratory findings conducted in the cognitive domain. But when it comes to the study of emotion in laboratory-type settings, “. . . the ecological and external validity of laboratory paradigms and measures can sometimes be uncertain” (Cacioppo & Gardner 1999: 193). Researchers restricted to the laboratory or classroom are limited in the range of meaningful outcomes and social complexity of negotiations that can be studied. Real-world negotiations may involve substantial financial risks, are often conducted with people with whom one shares a significant ongoing personal relationship, and sometimes are related to major life events (divorce, child custody, judicial proceedings, etc.), all of which heighten outcome significance and potentially result in intensified emotional responses. Furthermore, protracted real-world negotiations may introduce an element of time that could either serve to dampen emotions or, conversely, facilitate the intensification of emotions via the opportunity for rumination. And not least of all, contextualized negotiations often involve the expectations and emotions of multiple negotiators and constituencies simultaneously. Such scenarios are difficult to reproduce in the lab. It is in precisely these types of high-stakes, grueling negotiations, however, that one is presumably likely to observe the strongest manifestations of emotion, and also perhaps the greatest strategic use of emotion. An additional consideration is that lab studies and classroom simulations effectively impose a particular situation on the participant, who voluntarily agrees to take part. While offering the advantage of control, this approach eliminates the situational effects of choice and of coercion. Individuals in real-world settings often have some choice about the negotiations into which they will enter, and sometimes even about their negotiation counterparts. As a result, one might observe either more or less emotional response in the field than in a corresponding lab study because the negotiator has chosen to engage a negotiation where he or she would be more or less likely to become emotional, depending on his or her personal preferences, (e.g., Goates, Barry, and Friedman [2003] provide evidence that negotiators vary in the extent to which they “prefer” emotion-laden bargaining encounters). On the flip side, although most university human subjects committees frown on the use of coercion to
156
BRUCE BARRY AND INGRID SMITHEY FULMER
force participants to participate in laboratory negotiation studies, in the real world, individuals may sometimes feel compelled to negotiate when they would really rather not do so (e.g., spouse asks for a divorce), or to negotiate using particular tactics, including emotion tactics such as pretending to be happy about some turn of events when one is not (e.g., “emotional labor” [Hochschild 1983] in negotiations with important business clients). Such circumstances may result in differential emotional experiences and/or use of emotion tactics, and correspondingly different negotiation outcomes. Another Problem: The Disingenuous Expression or Suppression of Affect With respect to the strategic side of affect, one issue (among others) is whether the tactical use of emotion can be reliably detected and measured. Content coding of audiotapes, videotapes, or transcripts has been used to study negotiation tactics and process (e.g., Brett, Shapiro and Lytle 1998). While there is evidence that coders can reliably code facial emotional expression (Kring & Gordon 1998), it is less certain that coders can accurately detect whether emotion is being systematically suppressed, exaggerated or simulated as a tactical ploy. Another problem arises when we consider that the duration of genuine emotions and mood states varies considerably. Measuring the life span of individual emotions is difficult but probably necessary in order to study their effects on negotiation processes. As difficult as it is to measure the persistence of genuine emotion, establishing the temporal boundaries of strategically generated expressions of affect through observation is even more challenging. Remedies So what are researchers to do? It seems that a shift toward the study of context-rich negotiation is not only inevitable, but also crucial to the advancement of knowledge about the role of emotion in negotiation. The discussion that follows offers examples of several approaches that have been utilized successfully in other research domains, and which may prove adaptable to the study of affect in negotiation. The remedy that is probably most straightforward to imagine but rather more difficult to implement is to place a greater emphasis on the observational study of live (not simulated) negotiations. As with all field studies, gaining access to observe such negotiations is half the battle. Some targets for observational studies might include: organizations with departments or employees whose primary function involves negotiation (e.g., sales, purchasing, customer service, employee recruitment/hiring, labor relations), community mediation
METHODOLOGICAL CHALLENGES IN THE STUDY OF NEGOTIATOR AFFECT
157
centers, judicial settings (e.g., plea negotiation conferences, small claims court mediation), and other nonprofit or governmental agencies where disputes are frequently negotiated or mediated. For the observational approach to bear fruit, one must accept that a researcher can accurately detect and reliably classify the experience of emotion by interactants. As we mentioned above, there is evidence that observers can reliably assess visible emotional expression – but expression and experience are not the same thing. As studies of affect in retrospectively constructed episodes (Scanzoni & Godwin 1990) attest, recall techniques can elicit from interacting parties a sense of affective experience that occurred in the past. Clearly, however, the veracity of recalled emotion states is likely to be enhanced to the extent that the temporal distance between emotional experience and recall is diminished. A tool that has been employed to address this problem is experience sampling methodology (ESM) (also referred to as ecological momentary assessment or augmented diary methods). Study participants are asked to report on their current mental states (cognitive and/or emotional) at frequent intervals (randomly, episodically, or at particular time intervals prompted by signals from the researcher) over some time period (Bolger, Davis, and Rafaeli 2003; Kubey, Larson, and Csikszentmihalyi 1996). For example, Weiss, Nicholas, and Daus (1999) used an ESM approach to study the dynamic relationships among affect, expectancy theory-based job beliefs, and job satisfaction. They utilized written questionnaires (Current Mood Report, Larsen & Kasimatis 1990) and beepers to prompt responses, finding that average mood contributed to job satisfaction over and above job beliefs and dispositional tendencies toward happiness. Ilies and Judge (2002) explored the longitudinal interrelationships among mood (positive and negative affect), personality and job satisfaction. Four times a day over a 19-day period, participants completed a web-based mood adjective questionnaire; other measures of personality and job satisfaction were also collected at the same time (responses were prompted by email). They found that level and variability of mood predicted level and variability of job satisfaction, and that within-person, these varied together over time. Given issues with compliance when paper diaries are used (e.g., Feldman Barrett and Barrett 2001), electronic devices such as cellular telephones, personal digital assistants or palmtop computers are being used to collect questionnaire responses in ESM studies, particularly in health studies (e.g., Collins, Kashdan, and Golnish 2003; Kamarck, Shiffman, Smithline, Goodie, Paty, Gnys and Jong 1998). Bolger, Davis and Rafaeli (2003) and Feldman Barrett and Barrett (2001) discuss practical and technological considerations related to use of such electronic devices in ESM.
158
BRUCE BARRY AND INGRID SMITHEY FULMER
Limitations of ESM include intrusiveness and potential for self-selection bias (not just anyone would participate in such a study) (Alliger & Williams 1993). In addition, given the frequency with which participants are asked to respond in such studies, ESM approaches often use questionnaires or checklists that solicit responses to predefined discrete emotional states. While convenient, the use of questionnaires potentially constrains the range of emotions reported to those that fit neatly into discrete categories. Feldman Barrett found that people seem to vary in how they discriminate among their emotions (global pleasantness or unpleasantness of emotions versus distinct affective states), a situation that “present[s] a certain challenge to researchers who focus their research efforts on discrete emotions while using self-report inventories” (1998: 596). ESM approaches in conjunction with more open-ended emotion diaries (Oatley & Duncan 1992), although too time-consuming and intrusive for some negotiation settings, could provide the kind of richer information on the breadth and dimensionality of the emotional landscape that would be desirable for theory development. For example, Oatley and Duncan’s emotion diary not only asks people to categorize their mood, but also to describe the length of time that the emotion lasted, cause, and whether the feeling was a “mixed” emotion (where two emotions were occurring at the same time). Bolger, Davis, and Rafaeli note that “advances in voice recording and recognition as well as in linguistic analysis technology allow the inclusion of verbal reports into diary research” (2003: 599), which may enable the use of more openended questions. ESM utilizing such emotion diaries could prove especially useful in conjunction with corroborating data (facial observation and/or physiological data) as a means of studying tactics of emotional deception. Another approach to studying emotional responses in real-time involves the physiological monitoring of emotional arousal. Experimental studies in various fields have explored the emotional arousal responses of the autonomic and central nervous systems utilizing measures of physiological activity (including facial and vocal responses) (for reviews see Cacioppo, Berntson, Larsen, Poehlmann and Ito 2000; Hagemann, Waldstein and Thayer 2003; Lane and Nadel 2000; Russell, Bachorowski and Fernandez-Dols 2003). There has been relatively little use of these approaches outside the laboratory, however, primarily due to their invasiveness and lack of portability in many cases. A recent exception is found in a study by Lo and Repin (2002), who used portable equipment and biofeedback software to collect data on skin conductance, pulse, heart rate, EMG, respiration rate and body temperature from a pilot group of ten professional foreign exchange and derivatives traders. They simultaneously collected real-time data on market events to
METHODOLOGICAL CHALLENGES IN THE STUDY OF NEGOTIATOR AFFECT
159
which the traders were responding, and found, among other things, preliminary evidence suggestive of experience effects (i.e., experienced traders seemed to be less emotionally sensitive to certain market changes than less experienced traders). A recent health psychology study combined ambulatory physiological monitoring and electronic experience sampling methods to examine the effect of everyday stress and emotion on blood pressure (ABP) (Kamarck et al., 1998). One hundred twenty participants wore ABP monitors that inflated and took blood pressure readings every 45 minutes during the day for six days; when the cuff inflated, respondents also responded to questionnaires (Diary of Ambulatory Behavioral States) on palm-top computers. Although the effects of negative affect and emotional arousal on blood pressure were generally modest, researchers noted substantial individual differences in blood pressure responsiveness to emotion.1 Realistically, the collection of physiological data is probably not feasible in all field studies of emotion, but as the data collection process becomes technically easier to do, collecting at least some basic physiological data from negotiators could provide additional insights into the biological processes that accompany negotiators’ emotional responses. At the present time it is not clear that physiological data (including brain imaging) are precise enough to differentiate among different discrete emotions; they are better at distinguishing between broad positive and negative emotions (Cacioppo et al. 2000). However, these approaches may be useful in conjunction with other approaches to further corroborate whether emotional arousal was indeed experienced physiologically, which could prove especially valuable in detecting the use of suppressive emotion-based tactics. In situations where negotiation transcript data are available, techniques of natural language analysis may be useful. On the one hand, some researchers urge caution in using counts of emotion words to study emotion: From an evolutionary perspective, language did not emerge as a vehicle to express emotion. In natural speech we generally use intonation, facial expression, or other nonverbal cues to convey how we feel. Emotional tone is also expressed through metaphor and other means not directly related to emotion words. Taken together, it is our sense that emotion researchers should hesitate before embarking on studies that rely exclusively on the natural production of emotion words (Pennebaker, Mehl and Niederhoffer 2003: 571).
On the other hand, there is preliminary evidence (Rogan & Hammer 1995), that patterns of language intensity (use of metaphors, obscure words, profanity statements, qualifiers, etc.) and valence may reflect negotiator affective patterns and be related to crisis negotiation outcomes.
160
BRUCE BARRY AND INGRID SMITHEY FULMER
Shifting from the quantitative end of the research design spectrum to the qualitative end, we discover still more tools for exploring the role of emotion for the negotiator-in-context. A hallmark of ethnography and related forms of qualitative research is the ability to collect data unconstrained by prior expectations about the information that can be acquired and the variables that can be examined (Becker 1996/2001). Emotion is arguably more socially and culturally embedded, and more contextually influenced than cognitive processing and so may lend itself more to qualitative methods in order to obtain sufficiently rich or “thick” description (Geertz 1973) about how emotion manifests itself in vivo. Ethnographic methods such as interviews and participantobservation are particularly well suited to this task, with researchers acting both as observers of interaction and as interviewers regarding the emotional tenor and experience of interaction that has just occurred. Friedman’s (1994) study of labor negotiations is an example of the extensive application of ethnography to negotiation. Other examples of this approach used to study emotion-related phenomena (although not in negotiation contexts per se) include: the work of Thoits (1996), who utilized participant-observation to study how people manage the emotions of others in support groups; a study of emotional contrast strategies of bill collectors that employed interviews and participant-observation (Rafaeli & Sutton 1991); and an interview-based study of reciprocal emotion management by paralegals (Lively 2000). Open-ended interview methods could conceivably be applied to laboratory negotiations as well, with experimenters poised to debrief subjects about the affective components of the interaction just experienced, perhaps in the form of verbal protocol analysis that accompanies a replay of a tape of the interaction – although again, one is mindful that the range of emotional experience in the lab may be constrained by the artificiality of the task or its attendant incentives, risks, and consequences. Still, there are a strong basis to infer from the literature on the psychology of emotions that individuals can do quite well at reporting their own emotions during and following events that elicit emotional experience and expression (e.g., Kring & Gordon 1998). Given the strengths and limitations inherent in each of these approaches, it may make sense to consider the application of several methods in the same study. This could be accomplished by simultaneously combining several quantitatively oriented approaches such as self-report measures and/or physiological monitoring (e.g., Kring & Gordon 1998). Taking advantage of the full spectrum of research approaches, negotiation researchers might also consider utilizing “mixed-methods” approaches that combine, either sequentially or concurrently, both quantitative methods and qualitative approaches (see Creswell 2003, for a review of mixed methods research strategies).
METHODOLOGICAL CHALLENGES IN THE STUDY OF NEGOTIATOR AFFECT
161
Conclusion The position taken in this article can be distilled to six observations: (1) emotion is fundamental to the negotiation encounter; (2) research on negotiation has focused on cognitive processes and underemphasized the role of emotion; (3) emotion is relevant to the negotiation encounter as anterior, experiential, and strategic phenomena; (4) conceptualizations of how emotion fits into negotiation are far out in front of empirical investigation in terms of sophistication and complexity; (5) prevailing methods in the empirical literature on negotiation are not well suited to the investigation of emotional processes; and (6) creative new methods (in the sense of being novel to research on the psychology of negotiation) are needed to do empirical justice to the subject of affect in negotiation. In this article, we survey a variety of research approaches already used to study emotion in other fields that might also be adaptable to the study of affect in context-rich negotiation. We encourage negotiation scholars to consider the use of multiple methods and varied techniques such as those described here as we progress toward the testing of existing theories, and as a stepping stone toward future theory building regarding the role of emotion in negotiation.
Note 1. See Bolger, Davis and Rafaeli (2003), for more on physiological data collection in conjunction with diary studies.
References Alliger, G.M., and Williams, K.J. (1993). “Using signal-contingent experience sampling methodology to study work in the field: A discussion and illustration examining task perceptions and mood,” Personnel Psychology 46:525–549. Allred, K.G. (1999). “Anger and retaliation: Toward an understanding of impassioned conflict in organizations.” In R.J. Bies, R.J. Lewicki, and B.H. Sheppard, (Eds.) Research on negotiation in organizations, Vol. 7. Stamford, CT: JAI Press. Allred, K.G., Mallozzi, J.S., Matsui, F., and Raia, C.P. (1997). “The influence of anger and compassion on negotiation performance,” Organizational Behavior and Human Decision Processes 70:175–187. 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. Barry, B. (1999). “The tactical use of emotion in negotiation.” In R.J. Bies, R.J. Lewicki, and B.H. Sheppard (Eds.) Research on negotiation in organizations, Vol. 7. Stamford, CT: JAI Press.
162
BRUCE BARRY AND INGRID SMITHEY FULMER
Barry, B., Fulmer, I.S., and Van Kleef, G.A. (2004). “I laughed, I cried, I settled: The role of emotion in negotiation.” In M. Gelfand and J. Brett (Eds.) The handbook of negotiation and culture: Theoretical advances and cross-cultural perspectives. Palo Alto, CA: Stanford University Press. Barry, B., and Oliver, R.L. (1996). “Affect in dyadic negotiation: A model and propositions,” Organizational Behavior and Human Decision Processes 67:127–143. Bazerman, M.H., Curhan, J.R., Moore, D.A., and Valley, K. (2000). “Negotiation,” Annual Review of Psychology 51:279–314. Becker, H.S. (2001). “The epistemology of qualitative research.” In R. Emerson, (Ed.) Contemporary field research: Perspectives and formulations. Prospect Heights, IL: Waveland Press. (Reprinted from Jessor et al., [Eds.] [1996]. Ethnography and human development. Chicago: University of Chicago Press.) Bolger, N., Davis, A., and Rafaeli, E. (2003). “Diary methods: Capturing life as it is lived.” Annual Review of Psychology 54:579–616. Brett, J.M., Shapiro, D.L., and Lytle, A.L. (1998). “Breaking the bonds of reciprocity in negotiations,” Academy of Management Journal 41(4):410–424. Brief, A.P., and Weiss, H.M. (2002). “Organizational behavior: Affect in the workplace.” Annual Review of Psychology 53:279–307. Cacioppo, J.T., Berntson, G.G., Larsen, J.T., Poehlmann, K.M., and Ito, T.A. (2000). “The psychophysiology of emotion.” In M. Lewis, and J.J. Haviland-Jones, (Eds.) 2nd edition. Handbook of Emotions. New York: Guilford Press. Cacioppo, J.T., and Gardner, W.L. (1999). “Emotion.” Annual Review of Psychology 50: 191–214. Carnevale, P.J.D., and 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. Collins, R.L., Kashdan, T.B., and Gollnish, G. (2003). “The feasibility of using cellular phones to collect ecological momentary assessment data: application to alcohol consumption.” Experimental and Clinical Psychopharmacology 11, 1:73–78. Creswell, J.W. (2003). Research design: Qualitative, quantitative and mixed methods approaches. Thousand Oaks, CA: Sage. Davidson, R.J. (1994). “On emotion, mood, and related affective constructs.” In P. Ekman and R.J. Davidson (Eds.) The nature of emotion: Fundamental questions. New York: Oxford University Press. Feldman Barrett, L. (1998). “Discrete emotions or dimensions? The role of valence focus and arousal focus.” Cognition and Emotion 12, 4:579–599. Feldman Barrett, L., and Barrett, D.J. (2001). “An introduction to computerized experience sampling in psychology,” Social Science Computer Review 19(2):175–185. Forgas, J.P. (1992). “Affect in social judgments and decisions: A multiprocess model,” Advances in Experimental Social Psychology 25:227–275. 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. Friedman, R.A. (1994). Front stage, backstage: The dramatic structure of labor negotiations. Cambridge: MIT Press. Geertz, C. (1973). The interpretation of cultures: Selected essays. New York: Basic Books. Goates, N., Barry, B., and Friedman, R.A. (2003). Good karma: How individuals construct schemas of reputation in negotiation contexts. Paper for the annual meeting of the International Association for Conflict Management, Melbourne, Australia.
METHODOLOGICAL CHALLENGES IN THE STUDY OF NEGOTIATOR AFFECT
163
Hagemann, D., Waldstein, S.R., and Thayer, J.F. (2003). “Central and autonomic nervous system integration in emotion.” Brain and Cognition 52:79–87. Hegtved, K.A., and Killian, C. (1999). “Fairness and emotions: Reactions to the process and outcomes of negotiations,” Social Forces 78:269–303. Hochschild, A.R. (1983). The managed heart: Commercialization of human feeling. Berkeley: University of California Press. Ilies, R., and Judge, T.A. (2002). “Understanding the dynamic relationships among personality, mood, and job satisfaction: A field experience sampling study,” Organizational Behavior and Human Decision Processes, 89:1119–1139. Isen, A. (1987). “Positive affect, cognitive processes and social behaviour.” In L. Berkowitz (Ed.) Advances in experimental social psychology, Vol. 20. New York: Academic Press. Kamarck, T.W., Shiffman, S.M., Smithline, L., Goodie, J.L., Paty, J.A., Gnys, M., and Jong, J.Y. (1998). “Effects of task strain, social conflict, and emotional activation on ambulatory cardiovascular activity: Daily life consequences of recurring stress in a multiethnic adult sample,” Health Psychology 17:17–29. Kimmel, M.J., Pruitt, D.G., Magenau, J.M., Konar-Goldband, E., and Carnevale, P.J.D. (1980). “Effects of trust, aspiration, and gender on negotiation tactics,” Journal of Personality and Social Psychology 38: 9–22. Kramer, R.M., Newton, E., and Pommerenke, P.L. (1993). “Self-enhancement biases and negotiator judgment: Effects of self-esteem and mood,” Organizational Behavior and Human Decision Processes 56:110–133. Kring, A.M. and Gordon, A.H. (1998). “Sex differences in emotion: Expression, experience, and physiology,” Journal of Personality and Social Psychology 74:686–703. Kubey, R., Larson, R., and Csikszentmihalyi, M. (1996). “Experience sampling method applications to communication research questions,” Journal of Communication 46:99–120. Lane, R.D., and Nadel, L. (2000). Cognitive neuroscience of emotion. New York: Oxford University Press. Larsen, R.J., and Kasimatis, M. (1990). “Individual differences in entrainment of mood to the weekly calendar,” Journal of Personality and Social Psychology 58:164–171. Lawler, E.J., and Yoon, J. (1995). “Structural power and emotional processes in negotiation.” In R.M. Kramer and D.M. Messick (Eds.) Negotiation as a social process. Thousand Oaks, CA: Sage. Lewicki, R.J., Barry, B., Saunders, D.M., and Minton, J.W. (2003). Negotiation. 4th Edition. Boston: McGraw Hill/Irwin. Lively, K.J. (2000). “Reciprocal emotion management: Working together to maintain stratification in private law firms,” Work and Occupations 27(1):32–63. Lo, A.W., and Repin, D.V. (2002). “The psychophysiology of real-time financial risk processing,” Journal of Cognitive Neuroscience 14(3):323–339. Moore, D.A., Kurtzberg, T.R., Thompson, L.L., and 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., and Keltner, D. (2000). “How emotions work: The social functions of emotional expression in negotiations.” In B.M. Staw and R.I. Sutton (Eds.) Research in organizational behavior, Vol. 11. Amsterdam: JAI. Oatley, K., and Duncan, E. (1992). “Incidents of emotion in daily life.” In K.T. Strongman (Ed.) International Review of Studies on Emotion, Vol. 2. Chichester: John Wiley and Sons.
164
BRUCE BARRY AND INGRID SMITHEY FULMER
O’Connor, K.M., and Arnold, J.A. (2001). “Distributive spirals: Negotiation impasses and the moderating role of disputant self-efficacy,” Organizational Behavior and Human Decision Processes 84:148–176. Park, O.S., Sims, H.P., and Motowidlo, S.J. (1986). “Affect in organizations: How feelings and emotions influence managerial judgment.” In H.P. Sims and D.A. Gioia (Eds.) The thinking organization. San Francisco: Jossey-Bass. Pennebaker, J.W., Mehl, M.R., and Niederhoffer, K.G. (2003). “Psychological aspects of natural language use: Our words, our selves,” Annual Review of Psychology 54:547–577. Pillutla, M.M., and Murnighan, J.K. (1996). “Unfairness, anger, and spite: Emotional rejections of ultimatum offers.” Organizational Behavior and Human Decision Processes 68:208–224. Rafaeli, A., and Sutton, R.J. (1991). “Emotional influence strategies as means of social influence: Lessons from criminal interrogators and bill collectors,” Academy of Management Journal 34: 749–775. Rhoades, J.A., Arnold, J., and Jay, C. (2001). “The role of affective traits and affective states in disputants’ motivation and behavior during episodes of organizational conflict,” Journal of Organizational Behavior 22:329–345. Rogan, R.G., and Hammer, M.R. (1995). “Assessing message affect in crisis negotiations: An exploratory study,” Human Communication Research 21(4):553–574. Russell, J.A., Bachorowski, J., and Fernandez-Dols, J. (2003). “Facial and vocal expressions of emotion,” Annual Review of Psychology 54:329–49. Scanzoni, J., and Godwin, D.D. (1990). “Negotiation effectiveness and acceptable outcomes,” Social Psychology Quarterly 53:239–251. Thoits, P.A. (1996). “Managing the emotions of others,” Symbolic Interaction 19(2):85–109. Thompson, L.L., Nadler, J., and Kim, P.H. (1999). “Some like it hot: The case for the emotional negotiator.” In L.L. Thompson, J.M. Levine, and D.M. Messick, (Eds.) Shared cognition in organizations: The management of knowledge. Mahwah: Erlbaum. Watson, D., Clark, L.A., and 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. Weingart, L.R., Bennett, R.J., and Brett, J.M. (1993). “The impact of consideration of issues and motivational orientation on group negotiation processes and outcome,” Journal of Applied Psychology 78:504–517. Weingart, L.R., Hyder, E.B., and Prietula, M.J. (1996). “Knowledge matters: The effect of tactical descriptions on negotiation behavior and outcome,” Journal of Personality and Social Psychology 70:1205–1217. Weingart, L.R., Thompson, L.L., Bazerman, M.H., and Carroll, J.S. (1990). “Tactical behavior and negotiation outcomes,” International Journal of Conflict Management 1:7–31. Weiss, H.M., Nicholas, J.P. and Daus, C.S. (1999). “An examination of the joint effects of affective experiences and job beliefs on job satisfaction and variations an affective experiences over time,” Organizational Behavior and Human Decision Processes 78(1):1–24.
Comparative Case Studies I. WILLIAM ZARTMAN
Introduction Case studies are one of, if not the, most frequently used methods for conducting research on negotiation. They can vary from purely historical studies that seek to establish all the relevant facts of the encounter to analytical studies chosen to illustrate specific theoretical propositions. The most successful cases for the purposes of generating useful knowledge are those located somewhere in the middle of that spectrum, leaning toward the latter side of center. “Case” is used here to refer to the story of negotiations on a single conflict or problem, either as a single set of encounters or as a number of successive instances. Cases are the best way of combining empirical data (a redundant expression) with theory and concepts, but their use raises further, more interesting methodological questions. These questions concern the number of cases used by a research project and the type of data to be drawn from them, questions that are at the forefront of importance in the current wave of scholarship on the subject. Any research activity designed to create knowledge involves a question requiring answers or a category of events needing an explanation, a theory embodying those answers and explanations, and a method for gathering and using data. Each step in the research process poses its challenges. Given the need to bridge idiosyncrasies and to combine depth of Weberian understanding with the breadth of multiple instances, there is much to be said for using a comparative case method to answer questions and provide explanations about negotiations, focusing on the basic question of how outcomes are obtained. Case studies can be exploratory or confirmatory, providing inductive ideas for generalized explanations or deductive testing of logical constructs. Case studies can show causal links; they shed light on process and allow an exploration of the dynamic path from components to results, thus satisfying the needs of both analysts and practitioners. Comparative case studies lie at the crossroads of reality and theory; they present their evidence through the eyes of a knowledgeable specialist and they test it against the hypothetical constructs of a creative conceptualist. So the Janus-faced challenge to case study authors is dual and the standards of quality are high. International Negotiation Series 2: P. Carnevale and C.K.W de Dreu (eds.) Methods of Negotiation Research, 165–176 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
166
I. WILLIAM ZARTMAN
Theory and Data To begin with some extremely simple basics, knowledge can be considered to be ordered and generalized data. We live by generalizations, to get us away from suffocation in a world of infinite, unique data; leaves need to be aggregated onto branches and branches onto trees lined up in forests. Leaves are of course interesting in their own right, but unless we are innate historians, interested only in understanding correctly what happened in a single past event, data from the past are interesting only to the extent that they provide guidance in the present for the future. To be of such use, data must be aggregated into generalized knowledge as concepts and, if the concepts themselves are arranged in a dynamic relation to each other, as theories. Theories can either be concocted out of whole conceptual cloth, as a logical exercise, or can be built out of the record of data, generalized from idiosyncratic instances – exercises that can be termed deductive and inductive, respectively. Either way, theories face two tests of internal and external validity, respectively: a logical verification to see whether they hold together and make sense, and an empirical verification to see whether they summarize regularities in the data on the ground. Although it is the second test that concerns us here, it is worth a small paragraph to address the first. Building theory depends on prior identification of its components and purposes, taking us back to concepts. Phenomena or regularities in data need to be named, as concepts, and then explained, and explanation depends on the chosen terms of analysis, or concepts and categories of data to be applied to the explanation. “Theory” is a rather big and overused word; we are lucky if we can identify regularities, relations, effects, and generalizations and then – extremely important – the reasons behind them. These two elements – regularities and reasons – are the what and the why of theory-building, and the why is crucial in understanding rather than just observing the mechanisms of behavior. “What we are trying to explain” is a crucial question for analysis, and the terms of analysis chosen to pursue that explanation constitute the next crucial choice. There are many answers to that first question but an important one is “What is it in the process of negotiation that explains the outcome?” Thereafter, there are a myriad terms of analysis that can be used to pursue the answer to the question; none is trumps but a few leading ones emerge to face the test of logic as providing the most convincing reasoning. The exciting thing about current research on negotiation is that it has produced important conceptual development and cumulative knowledge. Old effects – since the process of negotiation itself is millennia old – have been identified and given names (things only “come into existence” when they
COMPARATIVE CASE STUDIES
167
have been named). I have been associated with such concepts as formulas and ripeness, but there are plenty of others – flexibility, prenegotiation, BATNA and security points, integration and distribution, toughness dilemma, value making and value taking, among many others – that have made the understanding of the negotiation process and the answering of the crucial question move ahead. But for all their logic, concepts and effects such as these (to dodge the big word “theory” for the moment) need to be tested against data to make sure that they work and are not just figments of creative imagination. Testing logic against real data has three purposes: to see if the logic is “real,” to provide an explanation of real events if it is, and to incite a search for alternative explanations – alternative logics – for exceptions. Where do these data come from? The best source of data is historical reality, from negotiations that have actually happened, been recorded and analyzed in case studies. Case data are authoritative; they record what happened, not what could or might have happened (although there is value in counterfactual analysis as well).1 There are many good historical case studies of negotiation that seek no particular conceptual guidance or verification but work merely to establish a good record and understanding of the event itself, finding an explanation of outcomes in the events (and often personalities) of the case itself. Fine examples of such historical studies are numerous, ranging from Nicolson (1946) to Preeg (1970).2 Often such accounts provide generalized bits of wisdom that emerge from the events and can be helpful in understanding and producing successes or failures in other (or the same) cases, beginning the inductive process of theory- and concept-building, as in the very studies just cited. The contrast is evident in the two oldest accounts of negotiations: the negotiations recorded in Genesis between God and Abraham over Sodom, which is rich in concepts but only implicitly, and the negotiations recorded in Thucydides (1960) between Sparta and Athens, where the purported lessons are made most explicitly. Other essentially historical studies refer to concepts already formulated, either induced from other cases or deduced from logical premises. But whether one is the writer of Genesis, Thucydides, Preeg, Nicolson, or any other author of quality about cases of negotiation, s/he has been steeped in the ambiance and context of the case, as a participant observer (as in the case of the first three names cited) or as a diligent researcher, and has developed a feel for the subject that makes a deep understanding analysis of the case possible. Whatever the level of conceptual sophistication, a case study writer never stops being a diligent historian too if s/he does a good job, and that feel for the case allows the writer to get behind the data to give it context and meaning and achieve a Weberian understanding of its dynamics (Skocpol and Somers 1980/1994).
168
I. WILLIAM ZARTMAN
If the theory fits the historical data (never the reverse), a presumptive explanation is provided; if not, an alternative explanation is needed. As philosophers of science tell us, the theory is never proven in the sense of being fully verified; it merely gains support, even though exceptions can always take place. Beyond that, as social scientists (should) know, humans have a happy capacity of free choice, the ability to do what they damn well please, included wildly stupid and gloriously creative things, and can never be caught inescapably in theories, mechanisms and regularities. But all that said, general regularities in events, contexts and behavior do occur, to be expressed in concepts and illustrated through data. The question remains, what cases, and what data to draw from them?
Choice of Cases The simplest answer to the choice of data sources is to pick the case in which the analyst is interested, and to seek guidance in an explanation of its outcome from the available concepts and theories. In single case studies, the choice is generally directed by considerations external to the concept; the purpose is to find explanations for the case, not tests of the concept. The case is the dependent variable or explanans, the thing being explained, whereas the concept is the independent variable or explanandum, the thing explaining. Such use of cases assumes either that the concepts are already well established or that the case can be used inductively to derive them. Two excellent examples are studies of multilateral negotiations, one by Bunn (1992) on arms control negotiations and the other by Winham (1986) on the GATT Tokyo Round of negotiations. While focusing on their subject per se, both found the concept of formulas useful in their analysis. Another insightful pair is Rubin’s (1981) collection of interpretations of the Kissinger shuttle negotiations in the Arab-Israeli conflict and Pruitt’s (1997) collection of analyses of the Oslo negotiations on the Palestine-Israeli conflict; in these cases, the concepts tested and applied varied among contributing authors. However, single case studies are of inherently limited utility in producing knowledge about negotiation as opposed to data on the unique case. Things that happened once, however engrossing as a story, have no way of telling us whether they represent regularities or exceptions; truth is stranger than fiction not because it is exceptional but because the story leaves us wondering whether it is really normal or indeed an exception to normality. The only way to test and reinforce concepts’ and theories’ claims to normal regularity rather than exceptionality is to look at a number of cases, not just one, and the more
COMPARATIVE CASE STUDIES
169
the better (knowing, as noted, that the generality can never be proved or expected to be universal). Zartman (1994) and Hampson (1995) are the first works to attempt a theoretical analysis of multilateral negotiations, each using cases somewhat differently; the first subjects two cases (the Single Europe Act and the GATT Uruguay Round) to a competition among six different theoretical approaches (game, decision, small group, leadership, coalition, and organization theories) to try to draw out a common analytical perspective, whereas the second examines nine cases from which to draw conceptual characteristics and insights.3 Comparative case study exhibits the advantages of in-depth analysis of reality while overcoming the weaknesses of focusing on one case alone. But how and what to compare? There is a large literature on case studies, starting with Mill (1843/1967) and going on to the most authoritative recent statement made by Alexander George (1982; George & McKeown 1985). With limited space and an argument to be made, cases are unlikely to be chosen at random, if indeed there were a notion of randomness that was operationalizable and applicable. Most likely, the analyst will begin by choosing a number of cases that are salient and relevant. Saliency involves importance in the general discourse about negotiation problems, including simple current events. Relevance concerns applicability to the conceptual issues involved. The more cases can be chosen to focus on variations relevant to the conceptual issues and hold other features constant, the more explanatory factors can be isolated and identified, a condition termed structured, focused comparisons. It would also be useful to bring in negative cases as a control, rather than including only positive cases, although comparing why it did not happen with why it happens significantly increases the difficulty of holding constant the elements to be analyzed. The simplest way to achieve comparison is to examine multiple instances in the same case. Instances of failure can be compared with instances of success in the same country in a comparative analysis that uses specific concepts to provide an explanation of outcomes, as was well and explicitly done in Touval (1982) examining nine attempts to mediate the Arab-Israeli conflict, and in Stedman (1991) comparing three failures and one success in mediating the Zimbabwean anti-colonial war. Interestingly but only coincidentally, both works focus on ripeness as a major explanatory variable, making important contributions to an understanding of the concept. Another comparative single-case analysis can be provided by examining the role of various components of a single set of negotiations, as is done in the first study of European integration as negotiation (Meerts & Cede 2004), which analyzes negotiations in each of the major institutions of the European Union. If multiple instances or segments within a single case are not available or chosen,
170
I. WILLIAM ZARTMAN
then similar cases can be the corpus of analysis, following the same rules of structured, focused comparison. As one moves away from multiple instances within a single case, it becomes more difficult to hold elements constant in order to focus on particular explanatory aspects of negotiation. The problem with multiple case studies is that the more the cases, the bigger the book and the further the account gets from the important details of reality. Some excellent multiple case studies guided by or testing concepts get rather voluminous, such as Crocker, Hampson and Aall (1999, 21 cases in 735 pages) on mediation or Stedman, Rothchild, and Cousens (2002, 9 cases in 728 pages) on post-agreement settlement, leaving in the dust behind them other studies such as Ali and Matthews (1999, 8 cases in 322 pages) on negotiating African civil wars and Zartman (1995, 11 cases in 353 pages) on the difficulties of negotiating centralist and regionalist civil conflict. Yet, the ability to make comparisons across a number of negotiations gives a rich harvest of lessons and insights. In addition to their supporting case material, such studies also contain varying amounts of conceptual or theoretical knowledge drawn and tested from the cases. Obviously, support is not measured by the number of pages, but more and longer case accounts can provide more data in order to test and apply the theory and concepts. It might be worthwhile examining the last-mentioned work (Zartman 1995) in greater detail, as an example of the process and advantages (and disadvantages) of a comparative set of case studies. Cases were chosen, not “scientifically,” but by the criteria of saliency and interest already mentioned. Most cases were significant and unresolved at the time the project started – Sri Lanka, Sudan, Eritrea, Lebanon, South Africa, Angola, Mozambique, Afghanistan, Colombia – plus two other cases, also unresolved but less well known – Euskadi and the Philippines. The challenge was to analyze the negotiations to date and then provide conceptually-derived prescriptions on ways to bring them to success. Interestingly, a number of cases – Eritrea, Lebanon, South Africa, Mozambique – came to a conclusion while the project was under way, while some others – Angola and Afghanistan – reached a successful conclusion of the current phase of the conflict, only to be followed by a new phase. The cases further divided into conflicts for control of the central government and conflicts for control of a region. A conceptual framework, built on contrasting notions of stalemate and on characteristics of internal wars – asymmetry, phases, agendas, escalation – and mediator tactics, was set up at the beginning and refined inductively in interaction with the case authors, providing the basis at the end for a review of factors in failure and success and two parts of a bottom line – weakness of both parties that hinders ripeness, and weakness of mediators that impedes patronage or relationship – to give
COMPARATIVE CASE STUDIES
171
a final account for failure. It was above all the repeated interaction between the inductive and deductive parts and people of the project that made the study so effective. On the other hand, it is worth recalling that nothing was “proven;” old and new propositions and concepts were supported and proposed, open for further testing. Multiple case accounts allow the analyst to develop a deeper understanding of the details and idiosyncrasies of the case, so that the fit between the generalizations and the data can be fully explored, explained, and understood. This analytical formula has been an integral part of the studies of the Processes of International negotiation (PIN) Program of the International Institute of Applied Systems Analysis (IIASA) in Austria. These works include comparative case-and-concept studies of negotiations on such issues as civilian and military use of nuclear material in order to test a number of propositions regarding the impact of the world’s most dangerous substance on the process of negotiation (Avenhaus, Kremenyuk & Sjöstedt 2002), economic issues in order to compare the strength of economic vs. negotiation explanations of outcomes (Sjöstedt and Kremenyuk 2000), symmetry and asymmetry in order to derive strategies and employ a new concept of power (Zartman & Rubin 2000), environmental issues in order to develop analytical and strategic concepts (Sjöstedt 1993), and a broad range of historic and contemporary encounters in order to answer some major conundrums about resolution vs. transformation and peace vs. justice (Zartman & Kremenyuk 2004). But events, unlike concepts, do not naturally come in boxes, with sharp sides and square corners, and calling an event one thing or another has to be done with extreme care and support. To evoke the concepts mentioned above, was there really a formula in the negotiations? What was it, specifically; and when did negotiations pass from diagnosis to formulation? Or was there really a mutually hurting stalemate (MHS)? How do we know and how did they (the participants) know, and how long did that perception last (even if they didn’t say so)? Debates about the existence and effects of the concepts and theories in reality can only be resolved by evidence from reality, and that can only be supplied by detailed case studies.
Beyond Cases and Back If eight cases are better than one and 21 cases better than eight, what about going to truly large sets containing many – tens and even hundreds of cases? How can we provide even larger collections of comparative data for more conclusive application and testing? Ostensibly, the answer seems to be found
172
I. WILLIAM ZARTMAN
in aggregate data, large collections of precisely identified statistical indicators applicable to all cases and associated with characteristics treated in relevant theories and concepts. If the analysis can enter into truly large data sets, firm generalizations and regularities can be established, both for knowledge and for appropriate policymaking. Usually, there is no unanimity of support for a theory or concept, but merely statistically significant correlations among a number of variables.4 A number of recent works have taken up the challenge of combining case data into large wads that can then be subjected to statistical significance tests. This research uses aggregate data on the largest number of cases possible either to test deductive propositions or generate inductive findings through correlation or factoring. Despite careful coding, it needs to group large numbers of diverse cases together into types, and is more interested in showing statistically significant correlations than in finding causality or in explaining the category of exceptional cases. Recent works by Walter (2002)5 and Fortna (2004) have turned to a particular aspect of negotiation of contemporary concern, the question of durability of negotiated agreements. Hampson (1999, five cases) and Stedman, Rothchild and Cousens (2002) previous sought explanations based on selected, structured, focused comparisons but the Walter and Fortna studies took on samples of 72 and 24, respectively. But in the process of doing so, they raise a number of significant methodological questions (for a fuller development, see Sambanis 2004 and Ragin 1997). First, in seeking to compress events into statistics, the data move far away from the subtleties of reality, as coders make sharp judgments on the nature or category of complex events. Many events do not lend themselves to binary (yes-no) or even plural statistics. To cite one example, in a recent preliminary study involving 60 instances of negotiation in 12 cases (Zartman 2004), I ran into frequent coding problems: How long does an MHS have to last to be coded an MHS (Walter 2002 at 56)?6 Which (power-sharing, powerdivision, elections) was the significant procedural provision when more than one occurred? When was a ceasefire a ceasefire? When was an instance of conflict autonomous rather than a continuation/revival of the previous instance? Concepts are clean, statistics are sharp, but events are messy. Yet they require procrustean sharpening to be subjected to statistical analysis. Such analysis can only handle data that are quantitatively, objectively measurable and explain only that for which it has data, and in the interest of precision it must make inflexibly quantitative definitions. Second, not only do the statistical data turn difficult evaluations into absolute indicators but they also hide the reasoning and details that support those choices. The basis of individual categorizations is always open to ques-
COMPARATIVE CASE STUDIES
173
tion when the evidence is not given, and an assurance of intercoder reliability is simply not adequate to relieve questions. Case study accounts may be compressed, but at least one can spot missing data and questionable judgments. Third, direct data are often not available, only indicators, sometimes termed proxies. Since the method can only handle comparable, quantifiable data, and so, because it has no “feel” for its subject, it has to rely on indicators or “proxies;” subjective elements must be objectified to become data. As a result, its indicators, such as per capita income or economic growth, are often far away from the effect they are proxying, such as proneness to the breakdown of order. Inequalities (in household incomes and in land ownership) are used as indicators of “grievance,” as if there were some universal threshold of envy or economic inequalities were the only and direct cause of protest and revolt (Collier and Hoeffler 2002; Collier et al. 2003: 66). Fourth, what cannot be measured or proxied is not analyzed. Since it is difficult to objectify intensity of feelings, such as nationalism or commitment, or degree of satisfaction with outcomes, these phenomena become ignored in analysis, even though they have been identified as crucial elements in negotiation. Fifth, as a result of the above, data become no longer data but are themselves events squeezed into generalizations. They become shorthand for the event and so enter into a tangling tautology: they become theoretical generalizations required to test theoretical generalizations. If household incomes are used to “proxy” grievance in order to test its role in causing conflict, they contain the unproven theory that income inequality directly causes or relates to grievance. But whether or how it does so or operates is outside the analysis. Of course, the research is not as simple-minded as that; the analyst looks for evidence or at least an indicator for the effect being evaluated, but the notation that appears on the correlation chart sets the tautology trap. Finally, there is remarkably little process in this analysis. Outcomes and conditions are noted but they are static road signs, neither roads nor driving skills. Getting there does not tell how, and so the dynamic of the process is lost. That loss is serious indeed, and marks a step backward in the analysis of negotiation. When the analysis was in the hands of the historians, attention tended to be focused on results, with much less on the way in which they were obtained. Process analysis in social science has been working to correct the aim, building on the historical analysis. It would be a step backward to focus simply on correlations between conditions and results.
174
I. WILLIAM ZARTMAN
Conclusion This is a rich list of problems; an equally full list can be made for case studies as well. Case studies exchange feel for precision and thrive on it; their strength is an understanding of the situations they analyze, even if it is hard to place numbered pieces of those situations into columns in a chart. Such studies are more interested in arguing and illustrating how perceptions, processes, communications, and grievances operate in known instances of negotiation than in correlating inputs and outcomes, and they spend little time on absent effects, non-instances, or control cases. Case students may be satisfied to understand one case well and produce some lessons for someone else to test on other cases, rather than finding correlations in universes of cases of varying importance. They are even happier when comparative case studies can be undertaken, either through successive negotiation attempts in the same conflict or through several negotiations of several problems or conflicts. Yet their data suffer from loose formalization, necessarily small numbers of cases, and deference to case idiosyncrasy. At the other extreme of one or a few cases lie studies involving many, many cases, summarized in aggregate data analyzed by statistical methods. While useful for establishing correlations, this method has its own problems: apples and oranges are often crammed into the same indicator, sensitive concepts are crudely operationalized, the variables used for analysis are often so distant from the phenomena named in the theory that it is hard to be sure the theory is being tested, and process dynamics are almost invariably lost. All this is not to say that such studies and their methodology are useless, as this critique may imply. It does indicate that enormous refinement is still awaited, that conceptual links and assumptions still need analysis, that subjective data still call for their place in the analysis, and that the statistical correlations can well be used to provide hypotheses that closer analysis can test. Yet the balance of advantages and weaknesses, inevitable in any method of analysis, places case studies in the midst of a search for breadth and depth, for data and theory. Much of the greatly expanded understanding of the negotiation process made available over the past four decades involves case studies – largely comparative case studies – used either to generate or to test conceptual and theoretical generalizations. Empirical soundness, including a feel for the subject, harnessed to a concern for usefulness through accurate generalizations and concepts, can be achieved – perhaps even best achieved – through comparative case studies.
COMPARATIVE CASE STUDIES
175
Notes 1. Too little work is done on historic possibilities at particular decision points, comparing the possible against the actual and analyzing why a particular decision was made. For some such case studies, see Tuchman (1984), Parker (1993), Jentleson (2000), and Zartman (2005). 2. I have tried to cite significant case studies to illustrate my analysis. There are many more of them than those cited here and I apologize to their authors for the omissions. 3. Ten years later, the 1994 book was used as the basis for further case studies in Crump and Zartman (2003) and Crump (2003), expanding the conceptual and empirical development. 4. For a serious effort to bridge this gap, see Sambanis 2004. 5. Walter (2002) combines both methods by examining two cases (Zimbabwe and Rwanda) in addition to the 72 sets of data. 6. Although Walter (2002: 56) says it does not matter.
References Ali, T.M. & Matthews, R.O., eds. (1999). Civil Wars in Africa. Toronto: McGill/Queens University Press. Avenhaus, R., Kremenyuk, V., & Sjöstedt, G., eds. (2002). Containing the Atom: Nuclear Negotiations for Safety and Security. Lanham, MD: Lexington Books. Bunn, G. (1992). Arms Control by Committee. Stanford, CA: Stanford University Press. Collier, John & Hoeffler, Anke (2002). “Greed and Grievance in Civil War.” Working Paper 2002–01. Oxford University, Centre for the Study of African Economies. Collier, John, et al. (2004). Breaking the Conflict Trap. Washington: World Bank. Crocker, C.A., Hampson, F.O., and Aall, P., eds. (1999). Herding Cats: Multiparty Mediation in a Complex World. Washington: United States Institute of Peace Press. Crump, L., editor (2003). Multiparty Negotiation and Complexity, special issue of International Negotiation, 8, 2. Crump, L., and Zartman, I.W. eds. (2003). Multilateral Negotiation and Complexity, special issue of International Negotiation 8, 1. Faure, G.O. & Rubin, J.Z., eds. (1993). Culture and Negotiation. Thousand Oaks, CA: Sage. Fortna, V.P. (2004). Peacetime: Ceasefire Agreements and the Durability of Peace. Princeton, NJ: Princeton University Press. Genesis 18: 20–32. George, A. (1982). “Case Studies for Theory Development,” paper presented to the Symposium on Information Processing in Organizations, Carnegie-Mellon University. George, A. & McKeown, T. (1985). “Case Studies and Theories of Organizational DecisionMaking,” Advances in Information Processing 2: 21–58. Hampson, F.O. (1999). Nurturing Peace. Washington: United States Institute of Peace Press. Jentleson, B., editor (2000). Opportunities Missed, Opportunities Seized. Lanham, MD: Rowman and Littlefield. Meerts, P. & Cede, Franz, eds. (2004). Negotiating European Union. New York: Macmillan. Mill, John Stuart, (1843/1967). A System of Logic: Ratiocinative and Inductive. Toronto: University of Toronto Press.
176
I. WILLIAM ZARTMAN
Nicolson, H. (1946). The Congress of Vienna. New York: Harcourt Brace. Parker, R. (1993). The Politics of Miscalculation in the Middle East. Bloomfield, IN: Indiana University Press. Preeg, E. (1970). Traders and Diplomats: An Analysis of the Kennedy Round of Negotiations under the GATT. Washington: Brookings Institution. Pruit, D.G., editor (1996). Lessons Learned from the Middle East Peace Process. Special issue of International Negotiation 2, 2. Ragin, Charles C. (1997). “Turning the Tables: How Case-Oriented Research Challenges Variable-Oriented Research,” Social Research XVI, 1: 27–42. Rubin, J.Z., editor (1981). Dynamics of Third Party Intervention. New York: Praeger. Sambanis, Nicholas (2004). “Using Case Studies to Expand Economic Models of Civil War.” Perspectives on Politics II, 2: 259–79. Skocpol, Theda and Somers, Margaret (1980/1994). “The Uses of Comparative History in Macrosocial Analysis.” Comparative Studies in Society and History XXII, 2: 174–97 (April), reprinted in Skocpol, Theda, editor, Social Revolutions in the Modern World. Cambridge, UK: Cambridge University Press, 1994. Sjöstedt, G., editor (1993). International Environmental Negotiations. Thousand Oaks, CA: Sage. Sjöstedt, G. & Kremenyuk, V., eds. (2000). International Economic Negotiations: Models vs. Reality. Cheltenham, UK: Elgar. Stedman, S.J. (1991). Peacemaking in Civil War. Boulder, CO: Lynne Rienner. Stedman, S.J., Rothchild, D., and Cousens, E., eds. (2002). Ending Civil Wars: The Implementation of Peace Agreements. Boulder, CO: Lynne Rienner. Touval, S. (1982). The Peace Brokers. Princeton, NJ: Princeton University Press. Thucydides (1960). The Peloponnesian Wars. New York: Oxford University Press. Tuchman, B. (1984). The March of Folly. New York: Knopf. Walter, B. (2002). Committing to Peace: The Successful Settlement of Civil Wars. Princeton, NJ: Princeton University Press. Winham, G. (1986). International Trade and the Tokyo Round Negotiation. Princeton, NJ: Princeton University Press. Zartman, I.W., editor (1995). Elusive Peace: Negotiating an End to Civil Wars. Washington: Brookings Institution. Zartman, I.W. (2004). “MEOs and Durable Settlements: A Theoretical and Empirical Evaluation of the Reasons for Durability of Peaceful Settlements in Civil Wars,” paper presented to the American Political Science Association, Chicago. Zartman, I.W. (2005). Cowardly Lions: Missed Opportunities to Prevent Deadly Conflict and State Collapse. Boulder, CO: Lynne Rienner. Zartman, I.W & Kremenyuk, V., eds. (2005). Peace vs. Justice: Negotiating Forward- and Backward-Looking Outcomes. Lanham, MD: Rowman & Littlefield. Zartman, I.W. & Rubin, J.Z., editors (2000). Power and Negotiation. Ann Arbor, MI: University of Michigan Press.
Discourse Analysis: Mucking Around with Negotiation Data LINDA L. PUTNAM*
Clearly, I am the type of scholar who loves to muck around in the data. Qualitative coding of communication processes allows me to get my fingers, hands, and arms deep into the negotiation data. This mucking in the thick of things is the key to discovering subtle nuances of not only what negotiators say but also what they do not say. It provides opportunities to extend the functions of messages to their forms and structures, their symbolic features, and the broad context of bargaining. The coding of communication processes in negotiation encompasses a wide array of qualitative and quantitative research methods. Qualitative methods fall into the categories of textual and discourse analyses while quantitative studies typically employ variations of content or interaction analyses (Weingart, Olekalns and Smith 2004). In both instances, scholars examine language, messages, talk, or written texts that emanate from a negotiation. Clearly, discourse units can be counted and converted to data amenable for quantitative analysis. However, this particular paper focuses on qualitative methods that are central to an in-depth examination of meaning and interpretations that arise from the negotiation process. In this way, language and symbols are keys to understanding how bargaining is constituted in a particular way, how it is maintained or changed over time, and how participants make sense or interpret this process (Putnam and Roloff 1992). Research in negotiation typically centers on three different types of discourse analysis: conversational analysis, pragmatics, and rhetorical analyses. To illustrate these methods and how they have been applied to negotiations, I provide an overview of these approaches to discourse analysis. Then, I distinguish them from content and interactional analyses, methods that typically employ quantitative coding. Finally, I focus on ways to conduct discourse analyses, some strengths and weaknesses of this approach, and some “insights” gleaned from personal experience in mucking around with the data. I draw from negotiation research that employs discourse analysis to highlight the strengths, difficulties, and pitfalls of this approach.
International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 177–192 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
178
LINDA L. PUTNAM
What Is Discourse Analysis and How Is It Used? Discourse analysis is an extensive arena of research that crosses linguistic, literary, and communication studies. This approach focuses not solely on the words or the texts of negotiation; rather it examines the micro-processes and nuances of speaking during negotiation. It is not just a set of techniques, rather it entails assumptions about the nature of language and reality, particularly the way a phenomenon such as negotiation is constructed through social interaction. Conversational analysts use both verbal and nonverbal paralanguage cues such as pauses, talk-overs, voice tones, and turn-taking, to analyze the way bargainers shift identities from the caucus to the table (Francis 1986), to reaffirm a differential status between union and management (O’Donnell 1990), to initiate new concessions through statements of reflection (Walker 1995), and to shape offers and counteroffers through cultural differences in turn taking and styles of speaking (Fant 1989, Grindsted 1989). Other studies explore how conversations are structured to accomplish plea bargaining (Maynard 1984, 1989) or to enact trade negotiations over the telephone (Firth 1994). These studies very precisely detailed their tracking of sequences and structures of talk, and they reveal interesting insights about how inconsistencies between verbal and nonverbal behaviors shape negotiation processes (Firth 1995). However, negotiation concepts rarely enter into the design or conclusions of these studies. Researchers seem more interested in uncovering patterns that distinguish ordinary conversations from negotiation talk. Another type of discourse analysis, pragmatics, focuses on the way that language signals action, for example, the act of giving a promise to enact a commitment, asking a question to make a request, and providing a justification to produce an account for one’s behaviors. These studies move beyond the micro-processes of talk to the meanings that words have in the interaction context. They examine the way language use, such as types of pronouns, verbs, and adjectives, make demands or distinguish between threats and warnings (Donohue and Diez 1985; Gibbons, Bradac and Bush 1992); increase or decrease relational distance between opponents (Donohue, Ramesh and Borchgrevink 1991; Donohue and Roberto 1994), and build relationships through expressing positive emotional effect and decreasing negative arousal (Donohue and Ramesh 1992; Rogan and Hammer 1995). These studies tap into the use of micro patterns of talk such as adjectives and pronouns, but they also root these processes in broader negotiation constructs, for example, integrative and distributive processes, relational messages between perpetrators and hostage negotiators, and successful and unsuccessful outcomes.
DISCOURSE ANALYSIS: MUCKING AROUND WITH NEGOTIATION DATA
179
A third type of discourse analysis draws from rhetoric or the study of persuasion and symbolic meaning (van Dijk 1997). Researchers who employ this approach center their analyses on broad-based language patterns, such as phrases that serve as evidence, claims, and reasons for arguments; metaphors and symbolic language that invoke meanings in context; and patterns of talk that construct narratives, such as references to plots, scenes, and story lines. In studies of argumentation, discourse surfaces as persuasive attacks and defensive reactions similar to bargaining strategies and tactics (Putnam and Jones 1992; Roloff, Tutzauer and Dailey 1989), as legitimating a bargainer’s claims (Schuetz 1978), as ways of redefining issues through policy deliberations (Putnam and Wilson 1989), and as reason-giving for proposed changes in the status quo (Putnam and Geist 1985; Putnam, Wilson and Turner 1990). Literary analyses adopt a symbolic view of communication by focusing on the ways words and phrases become short hand expressions for past discussions and shared experiences. For example, Hamilton, (1997, 2000) research on talk in pay negotiations between management and trade unions illustrates how phrases such as “stake in the organization” and “take out heads” become short-hand expressions that refer to a past sense of shared community and current practices of company downsizing. Thus, researchers who focus on figures of speech identify language patterns that have multiple and often contradictory meanings within the larger bargaining context. The final type of rhetorical analysis stems from narrative or dramaturgical texts. In narrative analyses, researchers focus on the way talk constructs stories, complete with villains and heroines, plot lines, time orientation, motives, and values. Hence, negotiation surfaces as a ritualized performance in which the parties develop narratives of their experiences within their bargaining cultures (Friedman 1994). Narrative analysis reveals how administrators and teachers build common ground through sharing stories of mutual enemies, and yet hold very different understandings for what the bargaining ritual means (Putnam, Van Hoeven and Bullis 1991). These areas of discourse analysis represent only a sample of the many ways that language forms a lens for understanding the complexities of negotiation. The most informative analyses treat negotiation as more than simply another type of conversation, but rather as a speech event in which the setting, participants, norms, and history of the process influence language choices (Cicourel 1988) and the language choices, in turn, shape bargaining activity. Discourse analysis, then, is a qualitative approach to examining negotiation texts. Researchers focus on the nuances of meaning within the broad context of negotiation interaction and aim to untangle how the bargaining evolves as
180
LINDA L. PUTNAM
it does. Rather than rely on a priori categories to analyze texts, discourse analysts let the themes or patterns grow out of the transcripts as part of the meanings that are co-developed during the negotiation. Researchers use language patterns to demonstrate how bargaining concepts evolve and take shape in the negotiation. For example, Putnam (2004) illustrates how symbolic uses of the terms “language” and “money” shift in teacher-administrator negotiations, moving from standing for sections of the contract itself to signaling a bargaining formula for packaging items, and finally to establishing shared control of the form and nature of the agreement (see Table 1). In this study, shifts in the meanings of these words function as units of analysis to decipher how negotiators come to a common understanding and how the bargaining changes over time. Thus, qualitative analysis fits interpretative social science that centers on how texts evoke multiple meanings. In this way it differs from quantitative analyses that rely on unitizing a code, determining the frequency of its use, and calculating reliability and validity of the codes. Instead, discourse analysts seek to ascertain how language makes certain negotiation practices possible, ones such as legitimating positions and identities, managing role conflicts with constituents, and reaching particular outcomes (Donohue and Roberto 1994; Putnam, Van Hoeven and Turner 1991; Rogan and Hammer 1995).
Table 1. Discourse Analysis of Transcripts of Teachers’ Negotiation: District #1 Discourse Units and Exemplars from Text
Themes and Patterns
Inferences from Themes and Patterns
Phase 1 “We need to figure out the language here.” “Let’s set forth some language on this.” “Let’s talk about money. What is the cost of this package and what are the sources of these funds”
Language as synecdoche functions as “the whole” to refer to all the policy issues in the contract. Money as synecdoche is a short hand word to refer to money items in the contract.
The use of the word, language, serves as an organizing device to cluster issues in the contract. The use of the terms, money, cost, and funds indicts overall cost of the contract.
Phase 2 “That’s our language on dues deduction.” “He’s added our language.” “We can give them some language.”
Both sides begin to own, claim, and exchange language through the use of pronouns such as “our” and “theirs.” Language and money
The terms, language and money, shift in meaning to become commodities to be traded as a concession strategy. Money refers to dollars in the teacher’s
DISCOURSE ANALYSIS: MUCKING AROUND WITH NEGOTIATION DATA
181
Table 1 (cont.)
Discourse Units and Exemplars from Text
Themes and Patterns
Inferences from Themes and Patterns
“Let’s give on some minor items and hold out on our money.”
become used with verbs to exchange, to give and to claim.
pockets versus costs to the Board.
Phase 3 Teachers: “The Board will not give both a good money settlement and language items” Board: “This package does not contain your standard throwaway language. It will take serious effort to get them to let go of this language. I assume that means money, not your autographed picture”
Money becomes the counterpart of language. The two possessions can be pitted against each other in a win-lose struggle. To keep policy items out of the contract, the Board can trade money to buy out language items.
The terms form a duality such that gains in one represents losses in the other. Negotiators manage this dichotomy of language versus money by separating the two arenas and shifting back and forth between them.
Overall Pattern The tacit norm for getting a settlement in District 1 was a dichotomy: trading language for money.
It becomes a bargaining formula that both enables and constrains the process and the two sides.
The language versus money duality gives the Board control over policy issues in the contract and the teachers control over salary issues.
Conducting Discourse Analyses of Negotiations Texts and Research Questions An array of textbooks exist on methods of conducting discourse analyses (Boje 2001; Phillips and Hardy 2002; Phillips and Jorgensen 2002; Titscher, Meyer, Wodak and Vetter 2000; Wodak and Meyer 2001; Wood and Kroger 2000). These texts review different types of analyses and provide guidance for using particular approaches. Although most texts advocate employing naturally-occurring texts such as transcripts of negotiations, they also recognize that interview transcripts, contracts, documents, policies, and laws are open to discourse analysis. Thus, the researcher’s ultimate conundrum lies in selecting
182
LINDA L. PUTNAM
the corpus of texts to use in a study. For example, researchers could focus on the interaction at the table or the talk in the caucus sessions (Donohue and Roberto 1994; O’Donnell 1990) or they could center on written proposals, newspaper clippings, interviews, memos, or some combination of the above that encompasses the context and culture of negotiation (Friedman 1992; Putnam, Van Hoeven and Bullis 1991). The choice among texts often hinges on the research questions that drive the particular project. Usually, a researcher’s interests and philosophical position guide the choice of texts and help a scholar work through the conundrum of how broad or narrow, how inclusive or exclusive, and how fine grained or contextual the study needs to be. A general guideline is that the researcher should select a manageable subset of texts, make reference to the broader discourses that impinge on these texts, and have sufficient data to justify claims that emerge from analyses (Wood and Kroger 2000). For example, to ascertain the way that trade negotiators rely on different types of strategies, transcripts of the interaction might suffice, but to determine how cultural differences and preferences for commodity trades guide these patterns, researchers might need interview data and copies of trade policies that impinge on the interactions (Grindsted 1989). Hence, beginning with an explicit research question provides a frame for the study and aids in making decisions about data collection and analysis. Another factor that influences choices about texts is the goal of the researcher; that is, how descriptive, normative, or critical a researcher wants to be. Often goals arise from the project itself, but researchers have their own implicit biases as to the ultimate aim of their projects. Descriptive studies focus on explaining and understanding how language use constitutes the negotiation while normative studies center on ways to make the bargaining process effective. Critical studies examine the practices that privilege or marginalize certain players and the ways that power and control are managed through the negotiation process (Hamilton 1997; Putnam 2004). In my experience, goals typically emerge from a combination of the phenomenon under study, the extant literature, and the puzzles that drive the project. In effect, most researchers do not enter a discourse study with a blank template. They are searching for patterns, insights, and new concepts for future work. In summary, goals and research questions intertwine to shape the choices that researchers make about the scope and depth of texts to use, the finegrained or broad-based nature of a study, and the type of discourse analysis to employ. In the absence of one “right” way to proceed, researchers use coherence as an overarching criterion, ask interesting questions, and make their decisions as transparent and reasonable as possible. Researchers often under-
DISCOURSE ANALYSIS: MUCKING AROUND WITH NEGOTIATION DATA
183
estimate the importance of “an interesting question.” Interesting questions are often determined by what a research project challenges rather than by what it confirms and what is surprising rather than what is commonplace (Weick 1979). For example, to discover that threat statements may actually promote integrative bargaining is an interesting question because it challenges the assumptive ground that threats are typically linked to conflict escalation. Discourse analysis, however, reveals that vague and ambiguously worded threats signal to the other party that a resistant point is near; thus, helping both parties become serious about making concessions (Putnam and Wilson 1989). In effect, discourse studies often challenge assumptions that we hold to be true to reveal the subtleties of language and how they constitute negotiations. Data Analysis, Interpretation, and Research Reports To conduct discourse analyses, researchers need to locate patterns of language use within the context of the full negotiation. These language units emanate from different types of discourse analyses as they fit into the broad context of a particular negotiation. For example, a conversational analyst might isolate moments when bargainers talk simultaneously or alter their typical patterns of turn taking. A rhetorical researcher might single out reasoning processes used to define and redefine issues (Putnam, Wilson and Turner 1990) or particular words that symbolize part-whole relationships in language use (Putnam, 2004, see Table 1). A systematic tracking of arguments over time might reveal how a bargaining issue, like “equity in insurance premiums,” transformed to become a concern about “the insurance carrier.” Thus, the content of a negotiation interfaces with the patterns of argument to show how bargainers creatively co-develop issues. Selecting the discourse units and patterns for analysis also emanates from puzzles that arise while observing the bargaining situation. A puzzle is a situation that occurs within the process, for instance, how did bargainers move from inequitable insurance premiums to a discussion of the carrier? How do discourse patterns help some hostage negotiators form collegial relationships while transcripts of other hostage negotiations reveal adversarial patterns between hostage taker and negotiator? Since discourse analysis is rooted in naturalistic events, ethnographic knowledge of the context, the participants, and the larger picture of negotiations is crucial to selecting language units, tracking them, and linking them back to the broader context. Thus, qualitative studies are reflexive in that discourse units arise from the talk itself at the same time the negotiation is being constructed. These units draw from and reflect back on interactions that influence the bargaining in
184
LINDA L. PUTNAM
particular ways. For instance, an analysis of words and phrases that signifies drawing closer together or distancing from each other (e.g., “we” versus “they,” “that” versus “this”) shows how hostage negotiators protect the perpetrator’s face, manage identity issues, and build trust with the hostage taker (Rogan and Hammer 1994). The use of pronouns serves as the discourse units and a comparison between the perpetrator’s use of these units and the hostage negotiator’s response reveals the patterns or themes linked to distancing or pulling together. The overall interpretations occur reflexively by seeing how these patterns evolve over time and how they interface with the context of the negotiation, for example, type of hostage situation, training for the hostage negotiator, length of the incident, and federal versus local level of the event. Some discourse analyses are amenable to tracking through the use of computer programs such as ATLAS and NVIVO. These programs consolidate different units across themes in a large data set and plot patterns pictorially. Rather than consolidating data through statistical analyses, these programs sort texts into patterns. However, they only help with sorting and organizing the data, not with making interpretations or testing claims (Phillips and Hardy 2002). Making interpretations from the themes and patterns that emerge in discourse studies requires sticking close to the data and then moving back and forth between the broader context and the process. For example, a researcher might discover that both sides in a labor-management negotiation make negative remarks about an accountant who handles finances for the organization. Tracking these comments throughout the duration of the negotiation and in the caucus meetings demonstrates that both parties view this person as a villain or as someone who makes bargaining difficult for all of them. Moreover, when they mention this person’s name at the table, it serves as a bonding experience that leads them to sequential concessions and agreements on difficult matters. A conclusion that could be drawn from these patterns is that the accountant serves as a common enemy to unite the opposing sides. This inference could then be verified through interview data with team members about the role this person plays in the negotiations. Inferences, then, are drawn from patterns of discourse, themes and content of talk, and are related back to the larger context in which the parties work. They are then tested through collecting additional data and comparing these findings with the extant literature. For example, Putnam (2004) draws inferences from shifts in the meanings of the words language and money in a teachers’ negotiation (see Table 1). This study reveals themes in the use of words that shift across negotiation phases. These shifts show how both sides treat language and money issues as commodities that become dualities to be traded in a win-lose
DISCOURSE ANALYSIS: MUCKING AROUND WITH NEGOTIATION DATA
185
fashion. Both parties were aware of this bargaining formula, lamented the way it constrained them, but felt it was essential for an effective settlement. Reflecting back on the context and the bargaining relationship revealed a duality of control in which board members used the formula to control the length of the contract and teachers pressured the board for extra dollars in raises and fringe benefits. Both sides wanted to preserve their ways of controlling the process while both felt highly constrained by their tacit norms and bargaining formula. Writing up a qualitative study is similar to, yet different from, quantitative research. Consistent with quantitative studies, researchers make arguments and provide evidence to support their claims. If they choose a social science format, they follow the standard outline for preparing articles. Even with qualitative discourse studies, a deductive model that lays out key themes and conclusions to the study is easier to follow than inductive studies that leave the interpretations to the end of the article (Wood and Kroger 2000). Unlike quantitative coding, discourse studies need exemplars from the text to illustrate and support claims about the patterns, themes, and conclusions of the study. Scholars evaluate the quality of a study based on how well the evidence supports the claims, the plausibility of the findings, and the insights gleaned from the analytic approach (Phillips and Hardy 2002). A study’s overall coherence and the way it “rings true” to the reader are also important criteria for evaluating this research. For example, Friedman’s (1995) observations that negotiations are both constrained and enabled by the networks that bargaining team members employ resonates both with the participants and with other researchers. These insights not only challenge the assumptive ground of dyadic negotiations, but they also parallel observations drawn from engaging in actual bargaining experiences.
Pros and Cons of Discourse Analysis Qualitative discourse analysis, as with quantitative coding, is very labor intensive; so why would a researcher do this work? Four main reasons spur investigators to employ discourse analyses of negotiations. First, this method provides a way for researchers to unpack the developmental and contextual aspects of negotiation. A text of a lengthy negotiation event can never be studied in its entirety. Patterns that evolve over time provide clues as to what is happening and what voices are present or missing. In my own research, discourse analysis of interviews with teachers and administrators revealed how past negotiations impinged on the current bargaining and how the current
186
LINDA L. PUTNAM
deliberations differed from past practices. Participants found rules and formulas to make the process work for them and in doing so became constrained by their own norms for social interaction. Discourse analyses exposed the rules and practices, exhibited how they functioned as constraints, and indicated why they made it difficult for both sides to reach optimal settlements. A second reason for using discourse analysis is that it provides a way to link macro political, legal, and organizational processes to the micro behaviors in negotiation. These connections stem from the way that discourse enacts the process while it refers back to the larger context in which the bargaining occurs. Thus, a third strength of discourse analysis is its reflexivity or using language to reflect back on the bargaining context, how a study was conducted, and how processes evolve over time (Holland 1999). Reflexivity also addresses how discourse reveals instrumental, relational, and identity levels of negotiation simultaneously. In this way discourse analyses demonstrate how bargaining about issues and positions also includes negotiating one’s identity and the relationship among the parties, including ties to communities and organizations. The fourth reason to use discourse analysis is to uncover new concepts and to extend the extant knowledge of negotiation in different directions. Qualitative discourse processes can lead to the discovery of new concepts, such as issue development, a concept that shows how reframing an issue alters the way it is defined within the negotiation (Putnam and Holmer 1992). This concept enriches the notion of a negotiation package by exhibiting how splintering, combining, and dropping issues do more than create a bargaining mix. Rather they lead to new understandings, ones that transform the way disputants label a conflict. Transformation of issues, in turn, opens up new avenues for developing agreements through introducing topics created within the negotiation. In effect, discourse analysis has the advantage of revealing subtleties in the negotiation, providing methods of linking micro-behaviors to macro-context, employing reflexivity, and discovering new concepts to enrich bargaining theory. Discourse analysis also has its drawbacks and may be difficult to implement in a number of settings. Researchers need texts, either transcripts of negotiation interactions, documents, informal exchanges, interviews, field notes, proposals, or some form of verbal/written material to analyze. Its difficulty as a method stems in part from being labor intensive and from requiring trial and error to select texts, decide on units, and link them to the larger negotiation and social context. Researchers have to be willing to muck around in the data; they have to follow a trail like a bloodhound or a detective to see where it leads. To illustrate, a graduate student and I decided to perform a discourse
DISCOURSE ANALYSIS: MUCKING AROUND WITH NEGOTIATION DATA
187
study on the way that negotiators positioned themselves in the bargaining talk. Relying on detailed methods of conversational analysis, we examined positioning patterns that revealed what we knew already about offensive and defensive roles in labor-management negotiation. The lesson that I learned from this experience was that discourse analysis is not just a method. As with any research project, if you do not have an interesting problem, puzzle, or dilemma, the study may lead nowhere. Thus, another shortcoming of this approach is designing the study to address an important problem or puzzle, one that arises from within the setting or from the extant negotiation literature. Discourse analysis is often criticized for lacking rigor. It is a relatively new and unproven method and not widely used in bargaining research. However, as with most qualitative studies, rigor is a matter of precision, clarity, and systematic procedures. One way to conduct discourse analysis is to track patterns through mapping language use in charts and spreadsheets. For example, in studies of turning points in negotiations, Druckman and his colleagues (1991, 2001) tracked abrupt shifts in events through mapping internal and external precipitants, subsequent departures, and the effects of these shifts on negotiated outcomes. Through a systematic tracing of different paths, the investigators examined links between these negotiation shifts and conflict escalation. These detailed mappings enabled the researchers to uncover turning points, to demonstrate how they evolved, and to distinguish between trade and security negotiations. One of the most difficult aspects of discourse analysis lies in drawing inferences from the data. Researchers need to avoid inferences that are too general or cannot be substantiated by detailed references to the text. For instance, in our study of teachers’ bargaining, state law required some items to be negotiated (e.g., salary, benefits, reduction in force) and others (e.g., academic freedom, teacher evaluation) were open to discussion but not negotiation between teachers and administrators. However, since we did not collect data at the state level, we could not make claims that this legal context influenced how issues were bargained in the two districts that we studied. We could speculate about the effects of the law on the size and scope of issues included in the contract, but we needed more information about how the law was interpreted throughout the state to make major claims about its effects. A final problem with discourse studies is keeping the research focused on language analysis and not on a case study of the bargaining. In discourse analysis, language forms the basis for the research and the context or sequence of events aids in interpreting these patterns. The discourse must be analyzed through a particular method or approach. Discourse studies that get lost in a ‘play by play’ description of the event have turned into case studies.
188
LINDA L. PUTNAM
Thus, the study must focus on the way that patterns of discourse construct, alter, and produce a negotiation. The bargaining events, the relationships among players, the negotiation policies, and the history and cultural factors are relevant only as they are produced and reproduced in the negotiation texts.
Insights from Mucking Around with the Data Researchers often develop implicit guidelines for conducting their trade. These implicit guidelines form “helpful hints” that are shared with students but are rarely mentioned in the method books. Four implicit guidelines come to mind for conducting discourse analysis of negotiations. • • • •
Let the text and context talk to you Work back and forth between the text and the concepts Look for inconsistencies, ironies, or unexpected occurrences Dispute your own interpretation and explanation
The first implicit guideline, let the text and context talk to you, refers to a way of selecting themes and discourse patterns. Researchers typically design laboratory studies with a clear notion of the concepts and research questions prior to their investigations. This approach is difficult for qualitative studies because the parameters that are salient in a situation guide the researcher to particular puzzles, questions, or issues. Letting your text talk to you means that you design the study through relying on the data itself, the nature of the situation, and the events that emerge. In Friedman’s (1992, 1994) investigation of the International Harvester negotiation, the idea of a dramaturgical lens came from observing negotiation not only from the “front stage” performance but also from the “back-stage” networking and interaction rituals that influenced front-stage scripts. Drama, as a metaphor of the process, arose through asking questions about traditional scripts and the way negotiators varied from these scripts to manage around their roles. The text, then, guides the researcher to particular language patterns and research questions. A second guideline is to work back and forth between the text and the concepts. Discourse analysts need to move from text to negotiation concepts to make broader claims about their studies. To illustrate, Martin Carcasson and I began our study of the Oslo negotiations with the goal of developing a network analysis of communication contacts at Oslo (Putnam and Carcasson 1997). As the study evolved, the text that we examined led to modes of communication and roles of agents and we shifted our focus to the concepts of legitimacy and the preservation of secrecy in the back-channel process. By
DISCOURSE ANALYSIS: MUCKING AROUND WITH NEGOTIATION DATA
189
working back and forth from the case to concepts in the literature, we broadened our focus to cover a complex communication system that merged well with literature on back-channel negotiations. In like manner, Cobb’s (1991) study of neutrality as a discursive practice in mediation, illustrates how tying particular language patterns to issues of legitimacy uncovered power moves that mediators endorsed and used to marginalize some disputants and privilege others. Similar to the first guideline, working back and forth between the text and the concepts keeps inferences about discourse patterns close to the observations of the data. A third guideline, look for inconsistencies and ironies in the text, focuses on the discovery of puzzles and how researchers can engage in puzzle solving. Inconsistencies are contradictions that appear in the text and call for the researcher to question the data. For example, illustrating how a particular pattern of argument in negotiation both conceals and reveals information simultaneously leads the researcher to probe different questions about information exchange (Putnam 1997). Friedman’s (1992) discovery that intergroup bargaining promotes “flexible stability” aids in unpacking how negotiation groups both perpetuate and counteract on-going struggles. These ironies lie at the roots of new discoveries and provide puzzles for future studies. A final guideline for conducting discourse analysis is to dispute your own interpretations. Qualitative data analysis involves discovering and assembling pieces of a puzzle and drawing inferences to form a broad image. By disputing interpretations of findings, the researcher engages in a process of eliminating explanations, particularly ones that are not plausible. Some explanations do not “ring true” for the situation; others are too disconnected from the data. Thus, discourse analysts need to make decisions based on coherence among analytic schemes, evidence drawn from the texts, and interpretations that resonate with the situation. In affect, several guidelines derived from mucking around with the data aid in selecting discourse patterns by letting the text talk to you, integrating patterns and negotiation concepts by working back and forth from text to theory, discovering new concepts through focusing on ironies and inconsistencies, and making appropriate claims through a process of elimination among plausible interpretations.
Conclusion Mucking around with the data, like making mud pies, is messy but fun. Just as children create imaginative and often provocative images from this activity, researchers can uncover and construct original concepts from mucking around
190
LINDA L. PUTNAM
with negotiation data. Becoming immersed in bargaining texts requires that researchers have systematic methods for analysis. Discourse analysis, as one of these methods, uses textual data to uncover patterns of language in use, to provide links between micro behaviors and macro context, to develop new constructs, and to unpack levels of meaning. Historically, negotiation literature is rich in knowledge about individual motives, aspirations, strategies, and outcomes. In contrast, it is only beginning to build a body of knowledge about process and the dynamics of social interaction. In constructing this knowledge base, researchers need a variety of tools, particularly ones that will allow them to explore multiple levels of meaning evident in mixed-motive interactions.
References Boje, David (2001). Narrative Methods for Organizational & Communication Research. London: Sage. Cicourel, Aaron (1988). “Text and Context: Cognitive, Linguistic, and Organizational Dimensions of International Negotiations.” Negotiation Journal, 4, 3: 257–266. Cobb, Sara and Rifkin, Janet (1991). “Practice and Paradox: Deconstructing Neutrality in Mediation.” Law & Social Inquiry, 16: 35–62. Donohue, William A. and Diez, Mary (1985). “Directive Use in Negotiation Interaction.” Communication Monographs, 52: 305–318. Donohue, William A. and Ramesh, Closepet (1992). “Negotiator-Opponent Relationships,” in Linda L. Putnam and Michael E. Roloff, editors, Communication and Negotiation. Newbury Park: Sage. Donohue, William A., Ramesh, Closepet and Borchgrevink, Carl (1991). “Crisis Bargaining: Tracking Relational Paradox in Hostage Negotiation.” International Journal of Conflict Management, 2, 4: 257–274. Donohue, William A. and Roberto Anthony (1994). “Relational Development as Negotiated in Hostage Negotiation.” Human Communication Research, 20, 2: 175–198. Druckman, Daniel (2001). “Turning Points in International Negotiations: A Comparative Analysis.” Journal of Conflict Resolution, 45: 519–544. Druckman, Daniel, Husbands, Jo L. and Johnston, Karin (1991). “Turning Points in the INF Negotiations.” Negotiation Journal, 7, 1: 55–67. Fant, Lars M. (1989). “Cultural Mismatch in Conversation: Spanish and Scandinavian Communicative Behavior in Negotiation Settings.” Hermes, Journal of Linguistics, 3: 247–265. Firth, Alan (1994). “‘Accounts’ in Negotiation Discourse: A Single-Case Analysis.” Journal of Pragmatics, 23: 199–226. Firth, Alan (1995). “Talking for a Change: Commodity Negotiating by Telephone,” in Alan Firth, editor, The Discourse of Negotiation. Oxford: Pergamon. Frances, David W. (1986). “Some Structures of Negotiation Talk.” Language in Society, 15: 53–79.
DISCOURSE ANALYSIS: MUCKING AROUND WITH NEGOTIATION DATA
191
Friedman, Raymond A. (1989). “Interaction Norms as Carriers of Organizational Culture: A Study of Labor Negotiations at International Harvester.” Journal of Contemporary Ethnography, 18: 3–29. Friedman, Raymond A. (1992). “The Culture of Mediation: Private Understandings in the Context of Public Conflict,” in Deborah M. Kolb and Jean M. Bartunek, editors, Hidden Conflict: Uncovering Behind-the-Scenes Disputes. Beverly Hills: Sage. Friedman, Raymond A. (1994). Front Stage, Backstage: The Dramatic Structure of Labor Negotiations. Cambridge: The MIT Press. Gibbons, Pamela, Bradac, James J. and Bush, Jon D. (1992). “The Role of Language in Negotiations: Threats and Promises,” in Linda L. Putnam and Michael E. Roloff, editors, Communication and Negotiation. Newbury Park: Sage. Grindsted, Annette (1989). “Distributive Communicative Behavior in Danish and Spanish Negotiation Interaction.” Hermes, Journal of Linguistics, 3: 267–279. Hamilton, Peter M. (1997). “Rhetorical Discourse of Local Pay.” Organization 4, 2: 229–254. Hamilton, Peter M. (2000). “Attaining Agreement: A Rhetorical Analysis of an NHS Negotiation.” The International Journal of Public Sector Management, 13: 285–300. Holland, R. (1999). “Reflexivity.” Human Relations, 52, 4: 463–485. Maynard, Douglas W. (1984). Inside Plea Bargaining: The Language of Negotiation. New York: Plenum. Maynard, Douglas W. (1989). “On the Ethnography and Analysis of Discourse in Institutional Settings.” Perspectives on Social Problems, 1: 127–146. O’Donnell, Katherine (1990). “Difference and Dominance: How Labor and Management Talk Conflict,” in A.D. Grimshaw, editor, Conflict Talk. Cambridge: Cambridge University Press. Phillips, Nelson and Hardy, Cynthia (2002). Discourse Analysis: Investigating Processes of Social Construction. Thousand Oaks, CA: Sage. Phillips, Louise and Jorgensen, Marianne W. (2002). Discourse Analysis as Theory and Method. London: Sage. Putnam, Linda L. (2004). “Dialectical Tensions and Rhetorical Tropes in Negotiations.” Organizational Studies, 25, 1: 35–53. Putnam, Linda L. (1997). “Productive Conflict: Negotiation as Implicit Coordination,” in Carsten De Dreu and Evert Van De Vliert, editors, Using Conflict in Organizations. London: Sage. Putnam, Linda L. and Carcasson, Martin (1997). “Communication and the Oslo Negotiation: Contacts, Patterns, and Modes.” International Negotiation, 2, 2: 251–278. Putnam, Linda L. and Geist, Patricia (1985). “Argument in Bargaining: An Analysis of the Reasoning Process.” The Southern Speech Communication Journal, 50: 225–245. Putnam, Linda L. and Holmer, Majia (1992). “Framing, Reframing, and Issue Development,” in Linda L. Putnam and Michael E. Roloff, editors, Communication and Negotiation. Newbury Park: Sage. Putnam, Linda L. and Jones, Tricia S. (1982). “Reciprocity in Negotiations: An Analysis of Bargaining Interaction.” Communication Monographs 49: 171–191. Putnam, Linda L. and Roloff, Michael. E. (1992). “Communication Perspectives on Negotiation,” in Linda L. Putnam and Michael E. Roloff, editors, Communication and Negotiation. Newbury Park: Sage. Putnam, Linda L., Van Hoeven, Shirley A. and Bullis, Connie A. (1991). “The Role of Rituals and Fantasy Themes in Teachers’ Bargaining. Western Journal of Speech Communication 55: 85–103.
192
LINDA L. PUTNAM
Putnam, Linda L. and Wilson, Steve R. (1989). “Argumentation and Bargaining Strategies as Discriminators of Integrative Outcomes,” in M. Afzalur Rahim, editor, Managing Conflict: An Interdisciplinary Approach. New York: Praeger. Putnam, Linda L., Wilson, Steve R. and Turner, Dudley B. (1990). “The Evolution of Policy Arguments in Teachers’ Bargaining.” Argumentation 4: 129–152. Rogan, Randall G. and Hammer, Mitch R. (1994). “Crisis Negotiations: A Preliminary Investigation of Facework in Naturalistic Conflict Discourse.” Journal of Applied Communication Research 22: 216–231. Rogan, Randall G. and Hammer, Mitch R. (1995). “Assessing Message Affect in Crisis Negotiations: An Exploratory Study.” Human Communication Research 21: 553–574. Roloff, Michael E., Tutzauer, Frank E. and Dailey, William O. (1989). “The Role of Argumentation in Distributive and Integrative Bargaining Contexts: Seeking Relative Advantage But at What Cost?” in M. Afzalur Rahim, editor, Managing Conflict: An Interdisciplinary Approach. New York: Praeger. Schuetz, Jan (1978). “Argumentative Competence and the Negotiation of Henry Kissinger.” Journal of the American Forensic Association 15: 1–16. Titscher, Stefan, Meyer, Michael, Wodak, Ruth and Vetter, Eva (2000). Methods of Text and Discourse Analysis. London: Sage. Van Dijk, Teun A. (1997). Discourse as Structure and Process: Volume 1. London: Sage. Walker, Esther (1995). “Making a Bid for Change: Formulations in Union/Management Negotiations,” in Alan Firth, editor, The Discourse of Negotiation. Oxford: Pergamon. Weick, Karl E. (1979). The Social Psychology of Organizing. Reading, MA: Addison-Wesley Publishing. Weingart, Laurie R., Olekalns, Mara and Smith, Philip (2004). “Quantitative Coding of Negotiation Behavior.” International Negotiation, 9: 441–455. Wodak, Ruth and Meyer, Michael (2001). Methods of Critical Discourse Analysis. London: Sage. Wood, Linda A. and Kroger, Rolf O. (2000). Doing Discourse Analysis: Methods for Studying Action in Talk and Text. Thousand Oaks, CA: Sage.
Field Experiments on Social Conflict DEAN G. PRUITT
Advancements in the study of social conflict, as in other sectors of science, depend on gathering clearly interpretable data. Field experiments are an important means to this goal. In field experiments, the investigator manipulates one or more variables in a naturally occurring setting and measures the resulting behavior or outcome. In “true,” as opposed to “quasi” field experiments, the investigator manipulates the variables by randomly assigning participants (individuals, hospitals, mediation cases, etc.) to the levels of these variables. True field experiments will be the topic of this article and will be referred to simply as “field experiments.” Good discussions of this methodology can be found in Boruch (1997), Cook and Campbell (1979), and Mosteller and Boruch (2002). An example of a field experiment involving social conflict is a study run by my students and me in a community mediation center (McGillicuddy, Welton and Pruitt 1987). We attempted to assess the impact of med-arb (mediation followed by binding arbitration if mediation fails to produce an agreement) on the process of mediation. Buffalo City Court cases were randomly assigned to three conditions: med-arb (same), in which the same person was mediator and potential arbitrator; med-arb (diff), in which the mediator and potential arbitrator were different people; and straight mediation, in which it was not clear what would happen if agreement was not reached. Two observers sat in the room and content-analyzed the mediations. The results showed that the disputants behaved more constructively under med-arb (same) than under straight mediation, making fewer hostile comments and invidious comparisons and proposing more novel alternatives for dealing with the issues.1 Other data suggest a possible explanation for these results, that the disputants feared binding arbitration and hence were especially intent on settling their dispute under med-arb. Field experiments may be contrasted with two other kinds of methods: laboratory experiments and observational field studies. Field experiments and laboratory experiments are alike in that the investigator manipulates at least one variable, but laboratory experiments take place in artificial settings constructed by the investigator rather than in naturally occurring settings. For International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 193–209 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
194
DEAN G. PRUITT
example, in an earlier study related to the one just described (Johnson and Pruitt 1972), undergraduates negotiated about wage increases and hospitalization benefits in a simulated labor-management task. Three conditions were imposed by instructions about what would happen if agreement were not reached in 25 minutes. They were told in the first condition that an arbitrator would enter and make a binding decision; in the second condition that an arbitrator would make a nonbinding decision; and in the third condition that the negotiation would temporarily stop and then resume. The participants made more concessions and were more likely to reach agreement under the first condition than the latter two. In observational field studies,2 the investigator measures all the variables instead of manipulating some of them. For example, Kochan and Jick (1978) conducted a study of archival records of mediated New York public employee disputes, comparing outcomes before and after a med-arb procedure came into law. The prior procedure was med-fact-finding, mediation followed by factfinding if mediation failed. This was an observational study rather than an experiment because the investigator did not assign the cases to the conditions. Rather assignment was done by the authorities who were following provisions in the law. The results of this study suggest that more cases were settled in mediation under the med-arb law than under the med-fact-finding law. This finding is consistent with the results of the two experiments described earlier, in that the anticipation of binding arbitration produced conciliatory behavior, which was presumably designed to escape the uncertainties of arbitration. All three methods have their particular strengths and weaknesses. To understand the arguments for and against field experiments, we should first examine the pros and cons of conducting experiments (in either the laboratory or the field) versus those of observational studies. Subsequently, we will consider the respective pros and cons for doing field and laboratory experiments.
Experiments vs. Observational Studies Advantages of Experiments over Observational Studies Creating novel conditions. One advantage of experimental research is that it allows researchers to study conditions that do not ordinarily occur. Thus observational studies are hopeless for testing the effect of a medicine that has not come onto the market or a mediation technique that has not yet been introduced. The investigator himself or herself must produce these conditions.
FIELD EXPERIMENTS ON SOCIAL CONFLICT
195
Assessing cause and effect. Studies involving manipulated variables make it easier to distinguish cause and effect. The reasoning is as follows. When a study of any kind shows a relationship (covariation) between two variables, X and Y, there are four possible explanations: (1) X (or an associated variable) influences (has a causal impact on) Y; (2) Y influences X; (3) some third common factor, Z, influences both X and Y, producing a “spurious” relationship between them; (4) the relationship between X and Y is due to chance. Statistical tests of significance allow us to rule out explanation 4 at an acceptable level of confidence. Beyond that, it is a matter of reasoning, weighing the plausibility of each of the other three explanations. If the study is an experiment and X is a manipulated variable, we can usually rule out explanations 2 and 3, arguing that the investigator is the source of variation in X and hence neither Y nor Z can have influenced that variable. This reasoning leaves explanation 1 as the only plausible account, that the relationship was produced by X (or an associated variable) influencing Y. Distinguishing cause and effect is often a problem in observational studies. Consider a hypothetical, but quite typical, variant of the Kochan and Jick study described earlier. States that mandate a med-arb procedure for public employee controversies are compared with states that mandate some other procedure, and it is found that more cases are settled in the former than the latter. Such data would be hard to interpret, because all of the first three explanations would be plausible. The nature of the procedure could plausibly be influencing the ease of reaching agreement (explanation 1). But the typical ease of reaching agreement in a state could also plausibly have influenced what procedure was adopted (explanation 2). Furthermore, many third common factors, such as the availability of funds to pay public employees, could plausibly be influencing both the nature of the procedure adopted and the ease of reaching agreement (explanation 3). The Kochan and Jick study is an exception among observational studies in that its before-after design allows us to draw fairly firm conclusions about cause and effect. Explanation 2, (that Y influenced X) can be ruled out by the order of events. The fact that more agreements were reached after than before the law was changed could not possibly have produced the change in law. Furthermore, explanation 3 seems implausible, since it is hard to imagine a third common factor that could have produced a change in the law and a simultaneous improvement in mediation outcomes. Hence, explanation 1 is by far the most plausible option, that the change in the law or some associated change (variable X) made agreement more likely (variable Y). Before-after studies of this kind, which examine the impact of an environmentally produced change, are sometimes called “natural experiments.” It is
196
DEAN G. PRUITT
often easier to sort out causation in such studies than in garden-variety observational studies. Still their name is a misnomer; they are not really experiments because the investigator does not control the independent variable. Reducing confounding. Another important advantage of experiments is that they make it easy to control for systematic error, that is for variables that are confounded with the one manipulated. When we manipulate or measure a variable, X, we are always inadvertently manipulating or measuring a number of other “confounded” variables (A, B, C, D, E, etc.) that covary with X. If an X-Y relationship is shown in our data, it could be because one or more of these confounded variables is influencing Y rather that because of a causal relationship between X and Y. But there is an advantage to experiments because variables can be manipulated with much more precision than they can be measured. Hence, the number of confounded variables is usually greatly reduced and it is easier to pinpoint the causal variable. This is where a problem arises in interpreting the results of the Kochan and Jick study. The cases involving med-fact-finding were all negotiated before 1 June 1974, whereas those involving med-arb were negotiated after that date. Hence, time period was completely confounded with which procedure was used, which means that other changes that occurred during this period could have produced the findings.3 Perhaps tougher cases were negotiated in the early 1970s than in the mid 1970s. Perhaps the labor-management climate improved over this period, or there was an improvement in the skills of the negotiators or mediators. If an experiment had been undertaken, these potential confounding factors would have been eliminated, facilitating interpretation of the data. Though people who design experiments try hard to avoid confounds, they are never completely successful. Thus the expectation of arbitration in our med-arb (same) condition necessarily carried with it a sense that the mediator was powerful, anxiety about the relationship with the mediator, belief that a definitive decision would be made soon, etc. These confounds were also present in the Kochan and Jick study as well as a host of others that were due to their non-experimental design. In short, it is usually easier to pinpoint the causal variable in an experiment than an observational study, but we can never be 100% certain in a single study.4 Preventing selection effects. A major confound in many observational studies results from self-selection, whereby participants exercise some control over the conditions they are in. Self-selection produces different kinds of participants in each condition, making it hard to interpret the data. Similar problems
FIELD EXPERIMENTS ON SOCIAL CONFLICT
197
are produced by agency-selection, whereby an organization (e.g., a mediation center) assigns different kinds of participants to each condition. Random assignment of participants to conditions – a defining feature of true experiments – is designed to eliminate these selection confounds.5 Selection effects were probably not a major factor in the Kochan and Jick study, since assignment to condition depended on whether a dispute went to mediation before or after the law changed rather than disputant or agency preference. But consider an observational study by McEwen and Maiman (1984), which compared adjudication with mediation in small claims courts in Maine. The most interesting finding from this study was that, “Defendants who went to mediation were considerably more likely to pay their debts than those whose cases were adjudicated” (20). Unfortunately, self-selection was built into part of the design, in that some courts gave disputants a choice between whether to work with a mediator or a judge; and agency-selection was built into another part of the design, in that a judge made this decision in other courts. It seems quite possible that these selection procedures channeled easier, less conflicted cases into mediation and tougher, more hostile cases into adjudication, which could account for these findings. In observational studies such as this, investigators often try to rule out selection problems logically or by statistical control after the data are gathered. Thus McEwen and Maiman present two arguments against the notion that their findings were due to selection effects. They note that perception of having freely chosen mediation was unrelated to whether the defendant subsequently paid his or her debts. This suggests that those who chose to enter mediation were not especially likely to reach a lasting agreement. They also note that in courts where the judge made the decision, assignment to condition usually hinged on whether a mediator was available to hear a case. This implies that type of case did not determine whether a case went to mediation. These are good arguments, but they are not completely convincing. The observations about how judges made their decisions are rather informal, and the disputant perception argument rests on the validity of the disputant reports about whether they were free to choose between the procedures. If it had been possible, a field experiment would have provided more solid support. Advantages of Observational Studies over Experiments It is straightforward that experiments are not useful for studying all phenomena. Some variables, such as the behavior of the heavenly bodies or of nation states, cannot be manipulated. If we want to perform experiments on these phenomena, we must employ simulations, which always run the risk of
198
DEAN G. PRUITT
differing in critical ways from the phenomena we wish to study. It is also not possible to manipulate participant characteristics, such as sex, age, and race. In addition, there are ethical objections to manipulating certain variables. Even if we are able to create serious marital quarrels in a study of their impact on the offspring, we should not try to do so. All of these variables must be observed rather than measured. Observational designs must also be used when one wants to look at the interrelations among a large number of variables, since experiments are limited to manipulating only a few variables at a time. Surveys, in which dozens or even hundreds of variables are measured, are a case in point. Hypotheses about cause and effect among survey variables may be explored by means of path analysis or causal modeling, but these methods seldom allow watertight conclusions about causation. In addition, in-depth, qualitative studies of one or a few cases can sometimes produce a rich theoretical tapestry concerning the interrelations among a number of variables (Pruitt, in press). In such studies, causal sequences can be followed in still other ways, for example, by asking people why they behaved as they did. However, such methods only yield speculative hypotheses, which should ultimately be tested by experimentation (Cook and Payne 2002). A third advantage of observational designs is that they allow us to estimate the strength of relationships among naturally occurring variables. Experiments are not so useful for this purpose, because good experimental design requires augmenting the strength and potency of the manipulated variables and reducing variation in the measured variables. This allows us to reach statistical significance if our variables are related to each other, but unnaturally magnifies the strength of those relationships.
Laboratory vs. Field Experiments Both laboratory and field settings are good places to observe new phenomena and generate new hypotheses – the laboratory because it isolates and magnifies relationships that might not be noticed in the field, and the field because phenomena occur there that are unlikely to be programmed into the laboratory until they have been noticed. Thus, a mix of laboratory and field research is bound to be more heuristic than exclusive reliance on either approach. Once new hypotheses have been developed, they should, if possible, be tested experimentally to avoid confounding and establish cause and effect. But should they be tested in the laboratory or the field?
FIELD EXPERIMENTS ON SOCIAL CONFLICT
199
Advantages of Doing Experiments in the Laboratory Rather than the Field Laboratory experiments are far more common than field experiments in research on social conflict for two primary reasons: greater control and a wider range of available manipulations and measures. After discussing these topics, we will look at some special problems associated with conducting field experiments in social service agencies. Control. Laboratory settings usually allow greater control over the elements of research than do field settings. This makes it easier to conduct good experiments. For example, randomization is usually more possible and more secure in laboratory settings. In many field settings, random assignment to conditions is not possible because society (e.g., mediation agencies) insists on having this prerogative. We were lucky in the med-arb study to secure the cooperation that allowed us to randomize. At the end of the study, however, the director of the mediation center told us that she would never again allow a randomization study because it was too disruptive for her staff. Greater control also makes it possible to create more precise manipulations in the laboratory, holding more variables constant between conditions and thus reducing the number of confounds and alternative interpretations of the results. Holding variables constant also reduces random error, making it easier to reach statistical significance and hence to discover subtle effects with smaller numbers of participants. Mortality – participants dropping out of the sample before the study is done – is also a larger problem in the less-controlled environment of field experiments, an issue we will return to toward the end of the article. Range of manipulations and measures. Another reason for conducting laboratory experiments is that they usually allow a wider range of manipulations and measures than do field experiments. Investigators are king in the laboratory and, within ethical limits, can do almost anything they want. More constraints exist in the field. Constraints are especially severe when field experimenters try to disguise the existence of a study for fear that participants will act differently because they know they are being observed. This produces three kinds of constraints: (1) Investigators must impose conditions that would plausibly be encountered in everyday life so as to avoid detection. (2) They are limited to producing the most innocuous of conditions that do not stress the participants, because they cannot solicit informed consent. (3) The use of questionnaires to explore process and to get at longer-term outcomes is quite limited.
200
DEAN G. PRUITT
This kind of study was done by Cialdini and his colleagues (Cialdini, Vincent, Lewis, Catalan, Wheeler and Darby 1975) to test the hypothesis that concessions are reciprocated in everyday negotiation. Participants, who were students walking alone on university walkways, were randomly assigned to one of three conditions: a concession condition, in which they were asked a large favor and, when they refused, asked a small favor; a no-concession condition, in which they were only asked the small favor; and an exposure condition, in which both favors were described and they were asked to make a choice. The favors all had to do with volunteering for charitable activity. The hypothesis was supported in that 50% agreed to do the small favor in the concession condition, but only 17% agreed in the no-concession condition and only 25% in the exposure condition.6 It was not possible to interview participants about why they made the decisions they did because this would have blown the investigator’s cover. Field experiments in social service agencies. Another type of field experiment assigns cases to different treatments in a social service agency. Participants know that they are participating in a study, but the researchers are limited in what they can manipulate, because they must stick to conditions that the agency can implement cheaply and can justify to clients and supporters as part of the agency’s mission. There is much more freedom of action in the laboratory. Our med-arb experiment (McGillicuddy et al., 1987) is an example of a study that was shaped by an agency’s program. Two of the conditions were part of the mediation center’s repertoire, and the third condition [med-arb (diff)] was a minor variation on what they usually did. Med-arb was only one of many procedures we might have studied, but most of the others would have bent the agency’s program too much out of shape. We were also constrained in our measurement, in that we could not tape record the mediation sessions. This meant that we could not check the accuracy of our content analysis or go back to the cases with new measures after the fact. Fortunately, we were able to administer extensive questionnaires to most of the mediators and disputants because they knew they were in a study and were willing to stay after the mediation session. Experiments in agencies are usually more difficult to do than laboratory experiments because they must be sold to busy agency officials, data gatherers must travel extensively to the agency headquarters, and operational problems are more likely to emerge.
FIELD EXPERIMENTS ON SOCIAL CONFLICT
201
Advantages of Doing Experiments in the Field Rather than the Laboratory If laboratory experiments provide more control and flexibility at lower cost, why should anyone do a field experiment? One answer to this question is that some conditions and variables can only be realistically produced in the field. Thus, one must move to the field in order to study whether arrest or counseling is more effective at stopping men from abusing their wives. Sherman and Berk (1984) did such an experiment and found only half as much recidivism over a six-month period (10% vs. 19%) in men who were sent to jail rather than counseled. Likewise, Emery and his colleagues (Emery, Matthews and Wyer 1991; Emery, Laumann-Billings, Waldron, Sbarra and Dillon 2001) could not have used laboratory methods to study whether mediation of divorce custody produced a different relationship between children and the nonresidential spouse than did litigation. It is also easier to study long-term effects in the field because laboratory effects are usually too weak to persist very long. The Emery study is a good example of the strength of field experiments for this purpose. It found that, “in comparison with families who litigated custody, nonresidential parents who mediated were more involved in multiple areas of their children’s lives, maintained more contact with their children, and had a greater influence in co-parenting 12 years after the resolution of their custody disputes” (Emery et al. 2001: 323). A third problem with laboratory experiments is that they are sometimes of questionable external validity. In other words, it is not clear that their findings can be generalized to real-life settings that involve the same variables. Laboratory settings differ from field settings in many ways, but most of these do not affect our capacity to generalize. Where we are in trouble is if our laboratory manipulations – our independent variables – produce atypical psychological or social processes that affect our dependent variables. Then the outcomes we get in the laboratory will not match those that occur in the real-life settings to which we hope to generalize our findings. Motivational and emotional impact. Disputants usually experience stronger passions – such as frustration, anger, and the desire for revenge – in the field than in the laboratory, because they are dealing with issues that are more important to them (Barry, Fulmer and Van Kleef in press). This means that effects that are mediated by such passions will usually be strengthened as we move from the laboratory to the field. It also means that effects that are defeated by strong passions – such as awareness of one’s surroundings and
202
DEAN G. PRUITT
concern about the impression one is making on others – will usually diminish as we move from the laboratory to the field. Our med-arb study (McGillicuddy et al., 1987) is an example of an effect that was strengthened by moving into the field. Before doing the field experiment, we ran some participants in a laboratory negotiation task involving the prices of three products in a simulated wholesale market. A mediator, who could become an arbitrator in the med-arb condition but not in the straight mediation condition, helped them with the negotiation.7 This laboratory experiment yielded only weak trends in the direction later taken by the results of our field experiment. In retrospect, it is quite certain that the issues faced in the laboratory were less motivating and emotion producing than those faced in the field. Hence, the participants in the med-arb condition were probably less anxious about the prospect of having the mediator take over their decision making, and hence less motivated to reach agreement. A case in which a robust laboratory finding was weakened to the point of disappearance can be seen in a pair of studies done by Druckman and Arai (private communication). In a laboratory experiment, they found that mediation was less likely to lead to agreement if the disputants sat across a table than if they faced each other across open air. But when they moved to a field setting and ran the same experiment in small claims court, this effect disappeared. This effect may have been partly drowned out by the greater error variance that inevitable occurs in the field. But it was probably also a victim of the strong passions that were quite evident in the field setting, where people were dealing with frustrating and ego-involving issues from their own lives. Heavy, emotion-laden involvement tends to block perception of background features of the environment. Hence, the conflict and the adversary probably stood out starkly while the rest of the scene – including the presence or absence of a table – probably receded into oblivion. If I am right about the source of the difference between Druckman and Arai’s two studies, it implies that their laboratory experiment was not a failure. The results of that study suggest that the absence of a table facilitates agreement in low key, non-emotional conflicts – a subclass of all conflicts. What is needed next is a study that crosses presence or absence of a table with high vs. low emotional involvement, to test the hypothesis that putting a table between the disputants makes a difference when passions are weak but not when they are strong. Other differences. Laboratory settings usually strip phenomena down to what are considered their essentials, eliminating many features that would be costly to implement or could contribute unwanted variance. For example, laboratory
FIELD EXPERIMENTS ON SOCIAL CONFLICT
203
research on negotiation usually employs tasks such as those used by Johnson and Pruitt (1972), where participants talk to each other in search of agreement on a few straightforward issues. Other elements that are often found in realworld bargaining – long-term working relationships, parallel secret negotiations, complicated constituent decision making, statements to the press, etc. – are not available. If any of these moderate the phenomena under study, the variables manipulated may work differently in the laboratory than the field, and the laboratory findings may lack external validity. Other possible sources of artificiality derive from the presence of an experimenter and the impact of such an authority figure on the undergraduates who are the usual participants in these studies (see Chapter 9 in Aronson, Ellsworth, Carlsmith and Gonzales 1990). Efforts to improve laboratory settings. Laboratory researchers have long been aware of the various threats to external validity just described and have often tried to remedy them. One approach to the problem of weak emotional impact is to recruit people from either side of a severe real-life conflict (such as that between Israel and the Palestinians) and have them discuss or react to information about that conflict (Robinson, Keltner, Ward and Ross 1995; Vallone Ross; Lepper 1985). Another approach, which also combats experimenter effects, is to employ confederates who enrage the participants in ways that seem genuine and not under the experimenter’s control. A whole industry of research on aggression has grown up around this paradigm (Berkowitz 1993; Geen 2001), and many of the findings in that tradition generalize nicely to the real world (Anderson and Bushman 1997). This approach has also been used for research on escalation (Mikolic, Parker and Pruitt 1997) and on conflict in subcultures of honor (Nisbett and Cohen 1996). Studies such as these still simplify reality, for example by only measuring the short-term impact of the manipulations, but they are a step in the direction of improving external validity.
Problems Specific to Field Experiments and Some Suggested Solutions Mortality Mortality is a common problem in field experiments because of the weaker control over participants. After being assigned to conditions, participants do not show up or fail to complete the study. This is a particular problem in experiments that take advantage of one of the strengths of field settings – the possibility of studying long-term effects.
204
DEAN G. PRUITT
Why worry about this issue? Mortality poses a threat to random assignment if participants drop out of one condition for different reasons than out of another – a phenomenon called “differential mortality.” For example, if the med-arb conditions in our study had seemed more threatening than straight mediation, braver people might have stayed in the former than the latter condition, which could have biased the results.8 Mortality also poses a threat to external validity if certain kinds of people drop out of all conditions, because the results may only apply to the type of people who stay in the study (Aronson et al., 1990). A major problem of differential mortality was narrowly averted in our medarb study. Officers of the court became alarmed when some of the cases assigned to straight mediation did not settle their disputes and came back to the court. (This could not happen in the med-arb conditions, because the arbitrator always imposed a settlement.) They complained to the mediation center employee who was doing the random assignment, and she changed her procedure without notifying us. As each new case came into her office, she was supposed to open the next envelope in a loose-leaf binder and read out the instructions contained in that envelope. Instead, she began to open several envelopes ahead of time and assign the easier cases to straight mediation so that they would not go back to the court. We detected this problem during one of our weekly inspections of the binder, because we found some folders already opened. To avoid differential mortality, we were forced to drop from our sample all of the cases that had gone through her office between the time at which she said she had started this practice and the time we detected it. If there is substantial mortality in an experiment, it is necessary to check whether significantly more participants have dropped out of one condition than another, a sure sign of differential mortality. Mortality was a big problem in our med-arb study; indeed so many cases were dropped or withdrew that we had to process 114 cases to get 12 cases in each of our three conditions. Fortunately, the dropout rate was not significantly greater in one condition than another. This was a comforting finding but it by no means ruled out differential mortality. If there is substantial mortality, it is also important to check whether different kinds of cases have dropped out of one condition than another, using whatever information one has on the dropouts. Differential mortality exists if the dropouts differ between conditions on any variable. It is a problem if that variable is correlated with the dependent variable in a direction that could produce one’s findings. An example where it was not a problem can be seen in the Emery et al. (2001) twelve-year follow-up of divorce mediation. They found evidence of differential mortality in that fathers from heavily conflictladen relationships were less likely to drop out of the mediation sample than
FIELD EXPERIMENTS ON SOCIAL CONFLICT
205
the litigation sample. However, this did not endanger their conclusions because conflict should produce less contact with the children whereas they found more contact after mediation than after litigation. Such post-hoc analyses are worth doing in studies that involve heavy mortality, but they are seldom definitive. Even if one’s findings are exonerated by all the variables one has measured, it is still possible that differential mortality is a problem on some other unmeasured variable. Hence, it is always better to conduct an experiment in which there is little or no mortality. Boruch (1997) provides much useful advice about how to accomplish this goal (see especially Chapter 6). Gaining Access and the Right to Randomize All experiments, whether run in the laboratory or the field, require permission from a human participants review board. But many field experiments require additional approval from the agency that supplies the cases. For example, in the med-arb study, we had to persuade the mediation center to allow us access to their cases and the right to randomize assignment of cases to conditions. Center personnel also had to vouch for our study with the court from which the cases were drawn; and, halfway through the study, they had to clamp down on the employee who was disturbing our randomization. To get agency cooperation – especially at the high level we obtained – requires a “courtship” process whereby one does favors for the agency and/or promises it something of value in exchange. Druckman and Arai were able to get agency cooperation by promising that the agency would learn something about its own procedures and make a contribution to science; but a larger quid pro quo seemed necessary in our case. Before we proposed the med-arb study, we evaluated one of the agency’s programs for its annual report, and we helped recruit mediators on our campus. Since the agency was making heavy use of med-arb, we were also able to sell our study as an evaluation of that procedure.9 Furthermore, we detected that sponsorship of our research made the agency look progressive in the eyes of its sponsors and the broader mediation profession. Hence, we made presentations of our research at the yearly conferences of the New York state mediation community, often asking the agency director to make preliminary remarks. Our approach to courtship was essentially guided by a principle recently articulated by Gueron (2002), “Remember that you want them more than they want you” (35). It is especially hard to get agencies to agree to random assignment because that procedure curtails some of an agency’s freedom of action with its clients and can provide an administrative burden to the agency. Because of this difficulty, Campbell and his associates (Campbell and Stanley 1963; Cook and
206
DEAN G. PRUITT
Campbell 1979) have developed a number of “quasi-experimental” research designs for evaluating procedures when randomization is not possible. All of these designs produce more confounds than do randomized designs, but some of them are quite good. There are even quasi-experimental designs that can be used in studies involving only a small number of cases or a single case (Hersen and Barlow 1976). Confidentiality Disputants often want to keep the details of their conflicts secret for fear of embarrassment, benefit to an adversary, or legal liability. Hence, field experiments on social conflict require particular attention to confidentiality. Concerns about confidentiality usually make it impossible to tape record negotiation and mediation sessions. However, it is sometimes possible to do on-site coding of negotiator or mediator behavior, using coding categories that reflect the process of what is happening but reveal little or nothing about the content of the sessions. This approach, which was taken in our med-arb study and the Druckman-Arai study, makes it difficult to do inter-coder reliability checks and impossible to go back to the data with new coding categories. In a later observational study (Pruitt, Peirce, McGillicuddy, Welton and Castrianno 1993; Zubek, Pruitt, McGillicuddy, Peirce and Syna 1992), we solved these problems by getting permission to take verbatim notes of what was said in mediation. We argued successfully that these notes were tantamount to hearsay and hence could not be admitted to a court of law. We also put case numbers instead of names on these notes and other records; and when we had assembled all the data for our cases, we destroyed the codebook that linked these numbers to the participants’ names.
Conclusions Writing this article has led me to conclude that field experiments should be employed more often then they are in the study of social conflict. They are more useful than observational studies for assessing the impact of novel conditions, establishing cause and effect, and reducing confounding. And they are more useful than laboratory experiments for examining long-term effects and discovering externally valid effects that involve strong passions and complex social interactions. However, field experiments clearly have their limitations. Some variables cannot be practically or ethically manipulated and require the use of observational methods. Observational methods are also more useful for looking at the
FIELD EXPERIMENTS ON SOCIAL CONFLICT
207
relationships among a large number of variables and for estimating the strengths of relationships between variables. Furthermore, laboratory experiments allow more control of conditions and greater flexibility in designing manipulations, and they are usually easier to implement than field experiments in social service agencies. What these points suggest is that all three of these methods have their value, and in many cases more than one method should be used for addressing a given problem. For example, the conciliatory effect of the expectation of arbitration was first worked out in a low-cost laboratory study by Johnson and Pruitt (1972) and then tested in real-world settings by Kochan and Jick (1978), who used an observational approach, and McGillicuddy et al. (1987), who ran a field experiment. As a result, we can be more confident about the external validity of this effect, and more certain about the mechanisms underlying it, than if a study had been done with only one of these methods.
Notes 1. Med-arb (diff) was intermediate between the other two conditions on these measures. 2. Sometimes called “correlational” field studies. 3. The reader will recall that in the just prior section we said that the most plausible interpretation of the Kochan and Jick finding is that the change in the law or some associated change made agreement more likely. The confounded variable, time period, is an example of an “associated change.” 4. To definitively pinpoint a causal variable, one must run several studies in each of which that variable is confounded with a different set of alternative variables. 5. Randomization also removes confounds with other features of the situation such as the time of day people enter the study, the room they are run in, the identity of the researcher, etc. 6. Nobody agreed to do the large favor. 7. The med-arb condition was quite similar to the med-arb (same) condition in the later field experiment. The straight mediation conditions were identical in the two experiments. 8. Differential mortality has an effect similar to self-selection, a topic discussed earlier. Both produce confounds by placing different kinds of people in the conditions of the study. 9. The agency director was very pleased with the way the results turned out.
References Aronson, E., Ellsworth, P.C., Carlsmith, J.M. and Gonzales, M.H. (1990). Methods of Research in Social Psychology, 2nd edition. New York: McGraw-Hill. Anderson, C.A. and Bushman, B.J. (1997). “External validity of ‘trivial’ experiments: The case of laboratory aggression,” Review of General Psychology, 1:19–41. Barry, B., Fulmer, I.S. and Van Kleef, G.A. (in press). “I laughed, I cried, I settled: The role
208
DEAN G. PRUITT
of emotion in negotiation,” in M.J. Gelfand and J.M. Brett, editors, Culture and Negotiation: Integrative Approaches to Theory and Research. Berkowitz, L. (1993). Aggression: Its Causes, Consequences, and Control. New York: McGraw-Hill. Boruch, R.F. (1997). Randomized Experiments for Planning and Evaluation. Thousand Oaks, CA: Sage. Campbell, D.T. and Stanley, J.C. (1963). Experimental and Quasi-Experimental Designs for Research. Chicago: Rand McNally. Cialdini, R.B., Vincent, J.E., Lewis, S.K., Catalan, J., Wheeler, D. and Darby, B.L. (1975). “Reciprocal concessions procedure for inducing compliance: The door-in-the-face technique.” Journal of Personality and Social Psychology, 31:206–213. Cook, T.D. and Campbell, D.T. (1979). Quasi-Experimentation: Design and Analysis Issues for Field Settings. Chicago: Rand McNally. Cook, T.D. and Payne, M.R. (2002) “Objecting to the objections to using random assignment in educational research,” in F. Mosteller and R. Boruch, editors, Evidence Matters: Randomized Trials in Education Research: 150–178. Washington, DC: Brookings Institution Press. Emery, R.E., Laumann-Billings, L., Waldron, M.C., Sbarra, D.A. and Dillon, P. (2001). “Child custody mediation and litigation: Custody, contact, and coparenting 12 years after initial dispute resolution.” Journal of Consulting and Clinical Psychology, 69:323–332. Emery, R.E., Matthews, S.G. and Wyer, M.M. (1991). “Child custody mediation and litigation: Further evidence on the differing views of mothers and fathers.” Journal of Consulting and Clinical Psychology, 59:410–418. Geen, R.G. (2001). Human Aggression, 2nd edition. Buckingham, England: Open University Press. Gueron, J.M. (2002). “The politics of random assignment: Implementing studies and affecting policy,” in F. Mosteller and R. Boruch, editors, Evidence Matters: Randomized Trials in Education Research: 15–49. Washington, DC: Brookings Institution Press. Hersen, M. and Barlow, D.H. (1976). Single Case Experimental Designs. New York: Pergamon. Johnson, D.F. and Pruitt, D.G. (1972). “Pre-intervention effects of mediation vs. arbitration.” Journal of Applied Psychology, 56:1–10. Kochan, T.A. and Jick, T. (1978). “The public sector mediation process: A theory and empirical examination.” Journal of Conflict Resolution, 22:209–240. McEwen, C.A. and Maiman, R.J. (1984). “Mediation in small claims court: Achieving compliance through consent.” Law and Society Review, 18:11–49. McGillicuddy, N.B., Welton, G.L. and Pruitt, D.G. (1987). “Third party intervention: A field experiment comparing three different models.” Journal of Personality and Social Psychology, 53:104–112. Mikolic, J.M., Parker, J.C. and Pruitt, D.G. (1997). “Escalation in response to persistent annoyance: Groups vs. individuals and gender effects,” Journal of Personality and Social Psychology, 72:151–163. Mosteller, F. and Boruch, R., editors (2002). Evidence Matters: Randomized Trials in Education Research. Washington, DC: Brookings Institution Press. Nisbett, R.E. and Cohen, D. (1996). Culture of honor: The psychology of violence in the South. Boulder, CO: Westview. Pruitt, D.G. (in press). “Escalation, readiness for negotiation, and third party functions” in I.W.
FIELD EXPERIMENTS ON SOCIAL CONFLICT
209
Zartman and G.O. Faure, editors, Escalation and Negotiation. Cambridge, England: Cambridge University Press. Pruitt, D.G., Peirce, R.S., McGillicuddy, N.B., Welton, G.L. and Castrianno, L.B. (1993). “Long-term success in mediation.” Law and Human Behavior, 17:313–330. Robinson, R.J., Keltner, D., Ward, A. and Ross, L. (1995). “Actual versus assumed differences in construal: ‘Naïve realism’ in intergroup perception and conflict,” Journal of Personality and Social Psychology, 68:404–414. Sherman, L.W. and Berk, R.A. (1984). “The specific deterrent effects of arrest for domestic assault.” American Sociological Review, 49:261–272. Vallone, R.P., Ross, L. and Lepper, M.R. (1985). “The hostile media phenomenon: Biased perception and perceptions of media bias in coverage of the Beirut massacre.” Journal of Personality and Social Psychology, 49:577–585. Zubek, J.M., Pruitt, D.G., McGillicuddy, N.B., Peirce, R.S. and Syna, H. (1992). “Short-term success in mediation: Its relationship to prior conditions and mediator and disputant behaviors.” Journal of Conflict Resolution, 36:546–572.
Laboratory Experiments on Negotiation and Social Conflict PETER J. CARNEVALE and CARSTEN K.W. DE DREU
This article addresses the laboratory experiment from the perspective of the experimental social psychologist interested in making statements about social conflict and negotiation. The main concern is the validity of these statements – whether the conclusions or inferences about the data are justified (Cook & Campbell 1979; see also Borsboom, Mellenbergh, & van Heerden 2004). Note that it is not that a particular method or study is valid, per se; rather, it is the statements made about the data – the propositions – that have validity or not (Brewer 2000). Such statements are best judged in light of a study’s purpose. One purpose might be to demonstrate an effect or relationship between variables. For example, Thompson and Hastie (1990, Study 2) wanted to demonstrate that when people face many issues of opposed interests, issues of common interest go unseen (this happened in 85% of the negotiations!). O’Connor and Carnevale (1997) wanted to demonstrate that such an issue of common interest would be exploited by people to gain advantage in an agreement (in one situation, such exploitation occurred in 53% of the negotiations!). Both studies were conducted by employing naive university students in conducting an artificial task in an artificial setting. One wonders if these patterns would occur in other settings, with other kinds of people, other periods of history, other cultures, or even with other levels of task incentive (see Hertwig & Ortmann 2001). Such issues of generality are always present in laboratory research or any research for that matter; indeed, any question about boundary conditions is a question about underlying process. I observe X here; over there, I do not observe X; since process Y does not operate over there, maybe that is why I do not observe X over there. In other words, I believe that process Y is critical for X to occur. Any situation without Y means no X. The nice thing about psychological processes is that they occur not only in the experimental psychologist’s laboratory, but they occur in the “real world” as well (Mook 1983).
International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 211–225 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
212
PETER J. CARNEVALE AND CARSTEN K.W. DE DREU
More often than not, laboratory experiments in social psychology (and related domains, like organization behavior), are driven by a desire to show causality or to reveal a process that explains a relationship between variables. The investigator experimentally controls (“manipulates”) one or more variables in a laboratory setting – an artificial environment that the participants agree to attend (e.g., a room in a building, or a website; see Cha 2005), and the researcher randomly assigns participants to conditions thus ensuring equality of conditions prior to any treatment. Often the manipulated variables (the independent variables) are situations or conditions that surround the negotiation (e.g., time pressure; accountability to a constituent) and the measures (the dependent variables) are outcomes (agreement reached; quality of agreement; symmetry of agreement) in a negotiation task. The connection between the situation and outcome variables is often assumed to be mediated by psychological states – cognitions, motivations, and emotions. Sometimes researchers assume that the connection between situation and outcome variables is mediated by what people say to one another, strategies and tactics, and these often are measured (see Weingart, Olekalns, & Smith 2005). Indeed, it is a good idea to have a measure of all variables in a psychological experiment (but there is some debate about the necessity; see Sigall & Mills 1998). One advantage of a laboratory experiment is that motivational, cognitive, perceptual, and decision processes can often be examined as they unfold, permitting the tracking of complex sequences of events that are hard to observe in other settings.
Background There are numerous excellent discussions of laboratory method, including Aronson, Ellsworth, Carlsmith, and Gonzales (1990) and Kerlinger (1986), the latter well known for the notion that one should maximize systematic variance (e.g., make treatment conditions as distinct as possible), minimize error variance (e.g., have accurate assessments of processes), and control extraneous systematic variance (e.g., have homogeneous conditions). The essential classic on design is Campbell and Stanley (1963). Abelson (1995) presented a treatment of statistics in experimental design that is destined to be a classic. There are, of course, numerous issues, for example, whether one should generate explanations of effects after seeing the data (suggested by Bem 2002), or not (suggested by Kerr 1998). Some argue that this issue becomes moot in programmatic research where follow-up studies replicate and extend past studies, which means that capitalizing on chance patterns when “hypo-
LABORATORY EXPERIMENTS ON NEGOTIATION AND SOCIAL CONFLICT
213
thesizing after the results are known” is not a problem. McGrath (1982) provides great wisdom on the many choices facing the researcher. Aronson and Carlsmith’s (1968) review of experimental method inspired many young social psychologists when they noted, in the last line of the paper, that the enterprise is fun. Many social psychology experiments involving negotiation and social conflict arise from observations gleaned in natural settings. Triplet’s (1897) foundational laboratory experiment designed a task investigating the impact of others’ presence on competition and performance. Having observed bicycle racing, Triplet modeled the task to simulate performance on a bicycle race track. Pruitt and Johnson’s (1970) study of the impact mediator’s suggestions have on negotiator concession-making built off Stevens (1963) real-world mediation observations.
Laboratory Tasks for Negotiation and Social Conflict A central element of any laboratory study of negotiation and social conflict is the task. It represents the reward structure, the incentives, the set of alternatives that people choose among, and the outcomes that are the possible results of these choices. In negotiation, the alternatives are represented as the topics being talked about, and usually are represented as issues that require a decision by two or more negotiators. The participants in these experiments are usually undergraduate volunteers. Several studies examine the policies of the laboratory, for example, Sieber and Saks (1989) conducted a survey of 326 psychology departments, finding that about three-fourths maintained a “participant pool” that had undergraduate students from introductory classes, and where participation was a class requirement. Often in a laboratory experiment, two or more people interact with each other, or a single participant may (unknowingly) deal with a confederate or a computer program. Communication is sometimes face-to-face, sometimes by means of note passing, and sometimes over a computer network. The tasks are often simplified versions of reality, yet retain key elements of the structure or processes of negotiation. Economics has influenced the social psychology of negotiation, as seen in the classic work of Siegel and Fouraker (1960; the former a psychologist and the latter an economist; see Smith 2001). Researchers often distinguish between negotiation tasks with a single dimension of value and those with multiple dimensions (see Froman & Cohen, 1970). The “Game of Nines” task used by Kelley, Beckman, and Fischer (1967) requires two people to divide
214
PETER J. CARNEVALE AND CARSTEN K.W. DE DREU
nine points (where points represent money), by holding up cards displaying numbers ranging from one to eight. If their sum on any trial is nine or less, then an agreement occurs; each person receives their card’s shown value (e.g., three for one person, six for the other). In the absence of agreement, each gets zero (or some pre-specified minimum). Clearly a competitive incentive exists (i.e., “I would rather have six than three; or five than four; or eight than one”), but there is a cooperative incentive as well: we should both prefer any agreement to no agreement. Thus, the task is mixed-motive as identified by Schelling (1960). Other single dimension of value tasks include the singleissue negotiation task (e.g., the price of a car in a buyer-seller negotiation), for example, a task developed by Liebert, Smith, Hill, and Keiffer (1968). Table 1 shows a variation on a single-issue negotiation task that simulates a union-management negotiation. Participants play the roles of management and a union negotiator considering a wage increase. Management’s issue chart is shown at the top of Table 1, and the union negotiator is given the issue chart shown at the bottom. Each chart lists five possible settlement points, represented by the labels. Beside each label is the value of settlement at that point (typically indexed to money). Each side views only their own issue chart, but all parties are typically instructed that they may say anything about the charts during the negotiation. Because parties are dividing a fixed sum of value (in this case, 240), the task is considered “fixed-sum.” The existence of a no-agreement alternative with, for example, a value of zero to both parties, alters the structure of the game, making it “variable sum” (since the outcome values are either 240 or 0). The labels used to describe negotiation tasks (“union/management” and “wages,” or “buyer/seller” and “price,” or “country A/country B” and “territory”) may have important cueing effects, although many researchers have assumed that the label is irrelevant (Rettinger & Hastie 2001). Burnham, McCabe, and Smith (2000) found that simply changing the text in the materials so that the other is labeled “partner” as opposed to “opponent” largely affects the context. Siegel and Fouraker (1960), Kelley (1966), and Pruitt and Lewis (1975) developed negotiation tasks with multiple dimensions of value and multiple issues, each offering several possible settlement points with each side possessing different preference orderings (see Pruitt 1981, for an early review of this task). Multiple dimensions tasks allow the development of integrative or win-win agreements of the sort called logrolling. Possible outcome values vary, meaning that some agreements represent better outcomes for both parties. An example is offered in Table 2. The participants again play roles of management and union negotiators, faced here with the task of reaching agreement on four issues. Management’s negotiator employs the issue chart
LABORATORY EXPERIMENTS ON NEGOTIATION AND SOCIAL CONFLICT
215
shown at the top of Table 2, and the union negotiator is assigned the issue chart displayed at bottom. These charts list five possible settlement points on each issue. Again, each side sees only their own chart, and is typically instructed that they can say anything to one another about it. In Table 2, the issues are integrative in the sense that they can be traded for one another. That is, each side can get what they want on their most valued issue and gives in on their least valued issue. In this task, many possible agreements exist and they vary in their value to the negotiators, both individually and collectively. The two issues in the center offer equal value for both parties and, in the parlance of negotiation research, are distributive issues, following the unfortunate terminology used by Walton and McKersie (1965) (unfortunate since the term “distributive” has other meanings in the expansive social psychology of justice literature). They noted that “distributive bargaining is the process by which each party attempts to maximize his/her own share in the context of fixed-sum payoffs. Integrative bargaining is the process by which the parties attempt to increase the size of joint gain without respect to the division of the payoffs” (8). Joint gain in the Table 2 task settles upon 70k on “salary” and 20% on “medical.” Such an outcome results in greater value – to each individual and the pair – than an agreement that splits these two issues down the middle ( joint gain of 800 versus 460 on these two issues). Table 3 presents a rough taxonomy of laboratory tasks; it indicates some areas where tasks simply do not exist (see Bornstein 2003, for taxonomy of intergroup tasks that also identify gaps). An organizing feature of the taxonomy is whether the task models single or multiple dimensions of value. The former include one-issue negotiation tasks (as shown in Table 1); the latter includes integrative bargaining tasks and their variations (as shown in Table 2, including contingent agreement games and the integrative ultimatum game). Another feature of the taxonomy concerns whether the task represents a game of agreement or one of coordination. A game of agreement allows the behavioral act of agreeing, reaching accord or developing consensus. Consider the ultimatum game, where a proposer offers an exchange and the responder either accepts or rejects it, and the game ends (Güth, Schmittberger, & Schwarze 1982). In this game, non-agreement is allowed, as determined by one or more participant saying no, withdrawing, or reaching a deadline that terminates negotiation; no agreement implies that the status quo is maintained. Any agreement usually represents a change in the status quo. Games of coordination model individual decisions to distribute outcomes, usually over time. These include, for example, the prisoner’s-dilemma and trust games. Coordination games typically generate simple response variables (“C” or “D”
216
PETER J. CARNEVALE AND CARSTEN K.W. DE DREU
Table 1. The Singe Issue Negotiation: Issue Charts for Management and Union Negotiators Management Issue Chart Annual Raise
(value)
15% 12% 9% 6% 3%
(00) (60) (120) (180) (240)
Annual Raise
(value)
15% 12% 9% 6% 3%
(240) (180) (120) (60) (00)
Union Issue Chart
Note: All agreements sum to a constant for the dyad, 240, making the task “fixed-sum.”
in a matrix choice; see Pruitt 1970). The notion of disagreement is not clearly represented, and coordination games typically do not model more than one dimension of value. In negotiation experiments, the main measurement is the outcome, which might be whether an agreement is reached or the quality of agreement, including its joint value, degree of efficiency, or its symmetry (the extent to which it favors one side or the other). A useful discussion of coding negotiation agreements can be found in Weingart, Hyder, and Prietula (1996). In more complex simulations, the primary measure may be an escalatory sequence of behaviors that reflect ever-increasing levels of conflict (see Mikolic, Parker, & Pruitt 1997).
Design: The Road to Strong Inference A laboratory experiment should be appreciated not only for its message, for what it says about theory, but for the techniques used to produce the message. This point became clear to this article’s first author while he was a graduate student working in Dean Pruitt’s laboratory in Buffalo, New York in the early
LABORATORY EXPERIMENTS ON NEGOTIATION AND SOCIAL CONFLICT
217
Table 2. Issue Charts for Management (Top) and Union (Bottom) Negotiators Management Issue Chart Salary 70.000 65.000 60.000 55.000 50.000
Vacation Days (00) (15) (30) (45) (60)
3 weeks (00) 2.5 weeks (30) 2 weeks (60) 1.5 weeks (90) 1 week (120)
Annual Raise 15% 12% 9% 6% 3%
(00) (60) (120) (180) (240)
Medical Coverage 100% (00) 80% (100) 60% (200) 40% (300) 20% (400)
Union Issue Chart Salary 70.000 65.000 60.000 55.000 50.000
Vacation Days (400) (300) (200) (100) (00)
3 weeks (120) 2.5 weeks (90) 2 weeks (60) 1.5 week (30) 1 week (00)
Annual Raise 15% 12% 9% 6% 3%
(240) (180) (120) (60) (00)
Medical Coverage 100% 80% 60% 40% 20%
(60) (45) (30) (15) (00)
Note: Agreement values for the dyad vary from 820 to 1160 making the task “variable-sum.” Source: De Dreu, Giebels, and Van de Vliert (1998).
Table 3. Taxonomy of Laboratory Tasks for the Study of Mixed-Motive Interaction
Single Dimension of Value
Multiple Dimensions of Value
COORDINATION GAMES
AGREEMENT GAMES
• • • • • • •
• 1-Issue Negotiation • Ultimatum Bargaining
Prisoner’s Dilemma Chicken N-Person Dilemma Locomotion Games Coalition Games Trust Game Etc.
• not done
• Integrative Bargaining • Integrative Ultimatum
218
PETER J. CARNEVALE AND CARSTEN K.W. DE DREU
1980s. One day, Pruitt walked into the project room with a sheet of paper in hand – and a gleam in his eye – and announced that he had something – and it was something special – about experimental design. The topic of the past months had been the concept of accountability and how it might play out in bilateral negotiation. We defined accountability in terms of social power and as an aspect of the relationship between negotiator and constituent. One idea posited that a mediator’s accountability to constituents might interfere with the development of a trusting relationship between the opposing negotiators and thereby interfere with problem solving and the development of integrative agreements. Pruitt developed an idea for testing that hypothesis, and the paper in his hand depicted a “2 × 2” experimental design, alongside some path notation. The design employed a moderator variable to help make inferences about mediating processes. The logic was straightforward: if the difference between high and low accountability was due to process X (in this case, a trusting relationship between negotiators), then controlling X should moderate the accountability effect. It soon became apparent that Pruitt’s design was elegant and structurally reflected Platt’s notion of strong inference (Platt 1964). To illustrate this type of design and others, what follows is a brief discussion of experimental design as it was developed, employed, and taught in the Pruitt laboratory. Path-diagram notation. Path-diagram notation is useful for presenting causal propositions schematically. The basic building blocks of propositions are variables and the links postulated between them. Variables are represented in path diagrams by letters or words in open boxes, and causal links are represented by arrows pointing in the direction of assumed causation. Simple path diagrams can be linked in chains. For example, if we know that A has a direct effect on B and B has in inverse effect on C, we can write: A
B +
C −
Other combinations of signs are also possible, + +, − +, − −, and chains consisting of four or more variables are possible. In such a chain, A is often called the independent variable, B the mediating variable, and C the dependent variable. If there are more than three variables in a chain, then there will be more than one mediating variable. Pruitt pointed out that such chains always permit theoretical derivations. Above, the relationship between A and B and the relationship between B and C imply a relationship between A and C, namely that increases in A lead to decreases in C. For example, if we believe that frustration (A) leads to anger (B) and that anger (B) reduces lik-
LABORATORY EXPERIMENTS ON NEGOTIATION AND SOCIAL CONFLICT
219
ing (C), then we might conclude that frustration reduces liking. The concepts in propositions are stated as variables rather than as conditions or states of nature. Thus the parameter A might be “time pressure,” a variable ranging from low to high (in other words, the A parameter would not be written as “high time pressure”). More complex forms of path diagrams express the joint effect of two variables on a third, for instance, additive and interactive effects. Several basic experimental designs help assess the processes underlying a research finding. These designs complement various statistical procedures (e.g., Baron & Kenny 1986): conceptual replication, precise manipulation, and the use of a moderator variable. Each of these fosters conceptual internal validity, concerning the impact of one variable on another and the quality of explanation. Conceptual internal validity provides a basis for generalizing findings and thus new research. It also fosters strong inference (Platt 1964), a way of building cumulative knowledge. Conceptual replication. We are on much stronger theoretical ground if we manipulate variables in more than one way. Often this requires multiple studies that each tests the variable differently. An example is negotiator positive mood, demonstrated by Carnevale and Isen (1986) to produce more cooperative negotiation and more integrative agreements; positive mood was implemented via a small gift and reading cartoons just prior to negotiation. Forgas (1998) also implemented positive mood and obtained a similar effect. Since his procedure differed from that of Carnevale and Isen – he used positive feedback on a “test of verbal abilities” just prior to negotiation – we can be more certain about the effect. The Forgas study was not just a replication in part (he also examined negative mood), but it was a conceptual replication for positive mood – that is, important for ruling out alternative explanations of the Carnevale and Isen results due to possible artifacts or confounds associated with the gift/cartoon manipulation of positive mood. In path diagram notation, the Carnevale and Isen study might be represented as “A” leads to “B” leads to “C,” where “A” is “receiving a gift,” “B” is “positive mood,” and “C” is “cooperation.” However, “B” could be composed of B1 and B2, where B1 is “sense of obligation having received a gift,” offering an alternative explanation to “positive mood.” The Forgas study rules that out as a possible explanation since it is hard to conceive of such obligation arising from receiving a positive test score. Of course, some unknown common process (apart from mood) could be important in interpreting the Carnevale and Isen mood procedure and the Forgas mood procedure, and this might be an interesting basis of a future study. Precise manipulation. Another approach is to try to refine the experimental manipulation. This is often done by including one or more control groups
220
PETER J. CARNEVALE AND CARSTEN K.W. DE DREU
in an experimental design, each holding constant a variable or set of variables. For example, a study of constituent surveillance of the negotiation process might require a control group that has surveillance per se, that is, surveillance by someone not associated with the negotiation (as opposed to surveillance by a constituent) (cf. Pruitt, Carnevale, Forcey, & Van Slyck 1986). Without such a control, we cannot know if the effect is driven by constituent surveillance or any surveillance. The experimental design that controls for the impact of a pre-test questionnaire reflects this notion of precise manipulation (see Solomon 1949). The use of a moderator variable. Here we use a factorial design that employs a moderator variable that is assumed to affect the relationship between the independent and dependent variables established in a prior study. It is essential that the moderator variable be selected on the basis of the hypothesized process. Pruitt told us that the most elegant type of moderator variable is one with two states: (a) under which the hypothesized intervening process is free to vary, and (b) under which it is not free to vary. The path notation for such a design is beyond the scope of this article, but the basic prediction is as follows: given we have a moderator variable called “Z” that has two levels, high and low, where a mediating process “B” is allowed to vary freely when “Z” is high, but is not free to vary under “low Z.” Variable “A” produces mediating process “B,” which in turn causes “C.” A factorial design that manipulates “A” and “Z” leads to the following prediction: we expect an effect of the independent variable “A” on the dependent variable “C” when variable “Z” is high (it is free to vary). But when “Z” is low (i.e., process “B” is removed) we predict no effect of the independent variable “A” on the dependent variable “C.” Neuroscience demonstrates this logic when a learning effect is hypothesized to occur in a specific brain region and the effect disappears when that area of the brain is removed. Carnevale and Conlon (1988) adopted a moderator variable design to assess the mediating processes associated with the impact of time pressure on third party intervention behavior. Past work had suggested that time pressure causes third parties to become more forceful in their intervention style (more likely to use sticks as well as carrots, as opposed to problem solving). Thus, Carnevale and Conlon hypothesized that time pressure would affect mediator behavior – enhancing the use of pressing tactics. But the question was: what is the mediating process? The strategic choice model of mediation suggests two possibilities: the mediator’s beliefs about the likelihood of agreement (the mediator’s “perceived common ground”) or the mediator’s concern for the parties’ aspirations. Carnevale and Conlon constructed moderator variables (“Z”) that controlled both possible mediating processes. The data were clear:
LABORATORY EXPERIMENTS ON NEGOTIATION AND SOCIAL CONFLICT
221
time pressure affected perceived common ground, not concern. When perceived common ground was free to vary, time pressure had an effect; but when it was not free to vary, there was no relationship between time pressure and the use of forceful mediation tactics. In the spirit of strong inference, future studies should pursue perceived common ground, and not concern, as the critical process that explains the impact of time pressure on mediator behavior. This, of course, does not mean that mediator concern is unimportant, but rather that it is likely less valuable in understanding the impact of time pressure on mediation.
Coda Like a great work of art, a laboratory experiment can be appreciated not only for its message, that is, what it says about theory, but for the technique used to produce the message. The latter is especially relevant to its truthfulness and its validity. Indeed, the latter is especially relevant to its beauty (see Overbye 2002). If we ask, “What is a good experiment?” the answer will often depend on the purposes of the experiment (see Brewer 2000; Dawes 1996). If the purpose is explanation, then the more the experiment controls for and rules out alternative explanations, the better. And if the experiment helps with strong inference, which some argue is the basis for cumulative knowledge in science, perhaps this is best, although perhaps unrealistic (see Hafner & Presswood 1965). Of course, the character of explanation, whether it is a neural mechanism, a higher order mental process, the structure of an incentive, a property of a group decision process, or a cultural process, will vary as a function of the tastes of the researcher, and is very much a matter of the zeitgeist. A key feature of laboratory experimentation is that it is programmatic. That is, experiments are conducted in a series, with each experiment in the series building on a prior experiment. This allows the researcher to narrow the set of theoretical possibilities that might explain an effect. The main benefit of conducting a series of experiments, over a single experiment, is that we learn far more. This is seen even when researchers with very different views agree to collaborate on a followup study to resolve a theoretical controversy (e.g., Latham, Erez, & Locke, 1988). This occurred also in a program of research on mediation of disputes (Carnevale, 1986, 1992; Murnighan, 1986; see Conlon, Carnevale, & Murnighan, 1994). This helps strong inference (Platt 1964). There are numerous other benefits; for example, they allow us to test theories in a safe place, and we can develop tests that are less expensive than in natural settings. In experiments, we can allow behavior that approximates
222
PETER J. CARNEVALE AND CARSTEN K.W. DE DREU
intense behavior in natural settings, but host it in a relatively safe place where it is less likely to produce harm. The wind tunnel in aviation research offers a great analogy: differently-shaped wings can be built and flown, but wind tunnel testing saves lives and cost. Of course, the wind tunnel needs to represent processes that occur in nature or at least in that part of nature where a plane with that wing will fly. Another benefit is that laboratory designs allow measurement and tracking of processes that would be impossible or at least extremely difficult to get at in natural settings, for example, measures of facial electromyograph in conditions of cooperation and competition obtained in the clever experiment by Lanzetta and Englis (1989) or fMRI data obtained by Sanfey, Rilling, Aronson, Nystrom, and Cohen (2003) in the ultimatum game. The main shortcoming of laboratory research is that it does not reveal the relative importance of different variables as they affect negotiation. It only tells what effects can occur (Mook 1983; and Locke 1986). Moreover, it is often difficult to generalize results from laboratory settings to natural settings, a problem shared with field research since natural settings differ from one another. There are two solutions: do parallel research in both laboratory and natural settings, and conduct research derived from theory. With theory-based research, the results may be generalized because the theoretical processes occur in the laboratory as well as in natural settings.
References Abelson, R.P. (1995). Statistics as principled argument. Hillsdale, NJ: Erlbaum. Aronson, E. & Carlsmith, J.M. (1968). “Experimentation in social psychology,” in G. Lindzey & E. Aronson (Editors), The handbook of social psychology (2nd ed.), Reading, MA: Addison-Wesley: 1–79. Aronson, E., Ellsworth, P.C., Carlsmith, J.M., & Gonzales, M.N. (1990). Methods of research in social psychology. 2nd Edition. Reading, MA: Addison-Wesley. 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. Bem, D. (2002). “Writing the empirical journal article.” in Darley, J.M., Zanna, M.P., & Roediger III, H.L. (Editors.), The complete academic: A career guide, Washington, DC: American Psychological Association: 171–201. Bornstein, G. (2003). “Intergroup conflict: Individual, group, and collective interests.” Personality and Social Psychology Review, 7: 129–145. Borsboom, D., Mellenbergh, G.J., & van Heerden, J. (2004). “The concept of validity.” Psychological Review, 111: 1061–1071. Brewer, M.B. (2000). in H.T. Reis & C.M. Judd (Editors) Handbook of research methods in social and personality psychology. Cambridge, UK: Cambridge University Press. Burnham, T., McCabe, K., & Smith, V.L. (2000). “Friend-or-foe intentionality priming in an extensive form trust game.” Journal of Economic Behavior & Organization, 43: 57–73.
LABORATORY EXPERIMENTS ON NEGOTIATION AND SOCIAL CONFLICT
223
Campbell, D.T., & Stanley, J.C. (1963). Experimental and quasi-experimental designs for research. Chicago: Rand McNally. Carnevale, P.J. (1986). “An unnecessary neologism in two systems of mediation.” Negotiation Journal, 2: 357–361. [reply to J.K. Murnighan (1986), “The structure of mediation and intravention: Comments on Carnevale’s strategic choice model.” Negotiation Journal, 2: 351–356.] Carnevale, P.J. (1992). “The usefulness of mediation theory.” Negotiation Journal, 8: 387–390. [reply to E. van de Vliert (1992), “Questions about the strategic choice theory of mediation.” Negotiation Journal, 8: 379–386.] Carnevale, P.J., & Conlon, D. (1988). “Time pressure and strategic choice in mediation.” Organizational Behavior and Human Decision Processes, 42: 111–133. 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. Cha (2005). “All that glitters is not gold: Examining the perils and obstacles in of collecting data on the internet.” International Negotiation. Conlon, D.E., Carnevale, P.J., & Murnighan, K. (1994). “Intravention: Third-party intervention with clout.’ Organizational Behavior and Human Decision Processes, 57: 387–410. Cook, T.D., & Campbell, D.T. (1979). Quasi-experimentation: Design and analysis issues for field settings. Chicago: Rand McNally. Dawes, R.M. (1996) “The purpose of experiments: Ecological validity versus comparing hypotheses.” Behavioral and Brain Sciences, 19: 20. Froman, L.A., & Cohen, M.D. (1970). “Compromise and logroll: Comparing the efficiency of two bargaining processes.” Behavioral Science, 30: 180–183. 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. Güth, W., Schmittberger, R. & Schwarze, B. (1982). “An experimental analysis of ultimatum bargaining.” Journal of Economic Behavior and Organization 3: 367–388. Hafner, E.M., & Presswood, S. (1965). “Strong inference and weak interactions.” Science, 149: 503–510. Hertwig, H., & Ortmann, A (2001). “Experimental practices in economics: A methodological challenge for psychologists?” Behavioral and Brain Sciences, 24: 383–451. Kelley, H.H. (1966). “A classroom study of the dilemmas in interpersonal negotiations. In K. Archibald.” in Strategic interaction and conflict: Original papers and discussion. Berkeley, CA: Institute of International Studies. Kelley, H.H., Beckman, L.L., & Fischer, C.S. (1967). “Negotiating the division of reward under incomplete information.” Journal of Experimental Social Psychology, 3: 361–98. Kerlinger, F.N. (1986). Foundations of behavioral research. New York: Holt, Rinehart and Winston. Kerr, N. (1998). “Harking: hypothesizing after the results are known.” Personality and Social Psychology Review, 2: 196–217. Latham, G.P., Erez, M., & Locke, E.A. (1988). “Resolving scientific disputes by the joint design of crucial experiments by the antagonists: Application to the Erez-Latham dispute regarding participation in goal setting.” Journal of Applied Psychology, 73, 4: 753–772. Lanzetta, J.T., & Englis, B.G. (1989). “Expectations of cooperation and competition and their effects on observers’ vicarious emotional responses.” Journal of Personality and Social Psychology, 56: 543–554. Liebert, R.M., Smith, W.P., Hill, J.H., & Keiffer, M. (1968). The effects of information
224
PETER J. CARNEVALE AND CARSTEN K.W. DE DREU
and magnitude of initial offer on interpersonal negotiation. Journal of Experimental Social Psychology, 4: 431–41. Locke, E. (1986) (editor). Generalizing from laboratory to field settings: Research findings from industrial-organizational psychology, organizational behavior, and human resource management. Lexington. McGrath, J. (1982). “Dilemmatics: The study of research choices and dilemmas,” in J.E. McGrath, J.M., R.K., editor, Judgment Calls in Research. Sage, Beverly Hills, CA: 69–102. Mikolic, J.M., Parker, J.C., & Pruitt, D.G. (1997). “Escalation in response to persistent annoyance: Groups versus individuals and gender effects.” Journal of Personality and Social Psychology, 72: 151–163. Mook, D.G. (1983). “In defense of external invalidity.” American Psychologist, 38: 379–387. Murnighan, J.K. (1986). “The structure of mediation and intravention: Comments on Carnevale’s strategic choice model.” Negotiation Journal, 4: 351–356. Overbye, D. (2002, March 26). “The most seductive equation in science: Beauty equals truth.” The New York Times, Science Desk, Late Edition – Final, Section F: 5. O’Connor, K.M., & Carnevale, P.J. (1997). “A nasty but effective negotiation strategy: Misrepresentation of a common-value issue.” Personality and Social Psychology Bulletin, 23: 504–515. Platt, J. (1964). “Strong inference.” Science, 146: 347–353. Pruitt, D.G. (1970). “Motivational processes in the decomposed prisoner’s dilemma game.” Journal of Personality and Social Psychology, 14: 227–38. Pruitt, D.G. (1981). Negotiation behavior. New York: Academic Press. Pruitt, D.G., Carnevale, P.J., Forcey, B., & Van Slyck, M. (1986). “Gender effects in negotiation: Constituent surveillance and contentious behavior.” Journal of Experimental Social Psychology, 22: 264–275. Pruitt, D.G., & Johnson, D.F. (1970). “Mediation as an aid to face-saving in negotiation.” Journal of Personality and Social Psychology, 14: 239–46. Pruitt, D.G., & Lewis, S.A. (1975). “Development of integrative solutions in bilateral negotiation.” Journal of Personality and Social Psychology, 31: 621–633. Rettinger, D.A., & Hastie, R. (2001). “Content effects on decision making.” Organizational Behavior and Human Decision Processes, 85: 336–359. Sanfey, A.G., Rilling, J.K., Aronson, J.A., Nystrom, L.E., & Cohen, J.D. (2003). “The neural basis of economic decision-making in the ultimatum game.” Science, 300: 1755–1758. Schelling, T.C. (1960). The strategy of conflict. Boston, MA: Harvard University Press. Sieber, J.E. & Saks, M.J. (1989). “A census of subject pool characteristics and policies.” American Psychologist, 44: 1053–61. Sigall, H., & Mills, J. (1998). “Measures of independent variables and mediators are useful in social psychology experiments: But are they necessary?” Personality and Social Psychology Review, 2: 218–226. Siegel, A.E. & Fouraker, L.E. (1960). Bargaining and group decision making. New York: McGraw-Hill. Smith, V.L. (2001). “From old issues to new directions in experimental psychology and economics.” Behavioral and Brain Sciences, 24: 428–429. Solomon, R.L. (1949). “An extension of control group design.” Psychological Bulletin, 46: 137–150. Stevens, C.M. (1963). Strategy and collective bargaining negotiation. New York: McGraw-Hill.
LABORATORY EXPERIMENTS ON NEGOTIATION AND SOCIAL CONFLICT
225
Thompson, L.L. and Hastie, R. (1990). “Social perception in negotiation.” Organizational Behavior and Human Decision Processes, 47: 98–123. Triplett, N. (1897). “The dynamogenic factors in pacemaking and competition.” American Journal of Psychology, 9: 507–533. Walton, R.E., & McKersie, R. (1965). A behavioral theory of labor negotiations: An analysis of a social interaction system. New York: McGraw-Hill. Weingart, L.R., Hyder, E., & Prietula, M.J. (1996). “Knowledge matters: The effect of tactical descriptions on negotiation behavior and outcome.” Journal of Personality and Social Psychology, 70: 1205–1217. Weingart, L.R., Olekalns, M., & Smith, P.L. (2005). “Quantitative coding of negotiation behavior.” International Negotiation.
Managing Conflict in the Literature: Meta-analysis as a Research Method ALICE F. STUHLMACHER and TREENA L. GILLESPIE
Meta-analytic work populates a visible chunk of the social science literature. A review of PsycINFO for 2003–2004 revealed over 450 articles with “metaanalysis” included in the title. This presence in academic publications illustrates the accessibility of meta-analysis to researchers and its acceptance within the scientific community as a viable research methodology. Within the realm of negotiation and social conflict, use of this technique has emerged (e.g., de Dreu & Weingart 2003; de Dreu, Weingart, & Kwon 2000; Druckman 1994; Stuhlmacher & Citera, in press; Stuhlmacher, Gillespie, & Champagne 1998; Stuhlmacher, & Walters 1999; Walters, Stuhlmacher, & Meyer 1998; Zetik & Stuhlmacher 2002), but its use remains relatively limited. This article offers a glimpse into this popular research methodology to assist readers in being informed consumers of meta-analyses. Through our coverage of meta-analysis, we hope to foster an understanding of this technique and an interest in its application to the conflict and negotiation field. As users of this method, we include examples from our own experience with meta-analysis, including its strengths, challenges, and predictions for how meta-analysis will continue to evolve. For those readers who may want to venture deeper into the topic, this article offers many excellent technical references. The heart of meta-analysis is its use of a standardized metric (an effect size) to be combined across studies to summarize confidence in findings. Providing an approach for quantitatively cumulating research study results, metaanalysis gives us a clearer picture of our past research efforts (Glass, McGaw, & Smith 1981). A good meta-analysis can evaluate where we have been and also identify where opportunities exist for further exploration in a field of study. Thus, the picture painted by meta-analysis is believed to be subject to less distortion than historical approaches to synthesizing research findings (Cooper 1990a; Cooper & Rosenthal 1980; Glass, et al. 1981; Rosenthal 1991).
International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 227–238 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
228
ALICE F. STUHLMACHER AND TREENA L. GILLESPIE
However, those drawn to meta-analysis for its objectivity soon realize that the approach encompasses more than a straightforward recipe for integrating research. While it has a standard methodology, meta-analysis takes on a life of its own, defined by the creativity, planning, and decision-making of the researchers who use it (Hedges 1990).
Development of Meta-analyses Historically, researchers synthesized work in a given area by comparing the outcomes of studies in terms of significance levels and counting the number of studies with findings in the expected direction versus those in an opposite direction. This focus on significance, or “vote-counting,” provided a heuristic for understanding a body of research, but did not account for such differences between studies as quality or sample size. Plus, a focus on significance levels failed to provide information on how much impact or change was common across variables. Prior to the development of meta-analysis, researchers recognized that words were not enough to convey the wealth of knowledge of past studies; they began searching for a method for pooling the data (e.g., Light & Smith 1971). With research studies continuing to pile up in a given area, there was a continuing need to better understand contradictory results. Meta-analysis emerged from this need, in response to a long-standing debate over the effectiveness of psychotherapy (Smith & Glass 1977). While others proposed similar concepts as early as 1904 (see Rothstein, McDaniel, & Borenstein 2002), Glass (1976) coined the term “meta-analysis” in publishing his methodology for estimating the effect size, or measure of change, across studies (see Glass et al. 1981 for a review). He and his colleagues articulated a new way for scientists to investigate past research in an area; measures from studies could be standardized and the effect size calculated across studies (Glass et al. 1981). Thus, meta-analysis was used to assimilate the vast amount of research on psychotherapy outcomes and, ultimately, demonstrated its benefit in terms of a measurable effect (Smith & Glass 1977). Although not all researchers were convinced initially of its value, meta-analysis remained in the methodological spotlight, possibly due in part to the continuing work of other researchers attempting the same type of research synthesis (e.g., Rosenthal 1984). In addition, further work illustrated the accuracy of meta-analysis over more traditional approaches in synthesizing literature (e.g., Cooper & Rosenthal 1980). Additional meta-analytic developments came from work by Rosenthal (1984), who offered a means for combining probability ( p) values across studies for an overall assessment of statistical significance. His methodology
MANAGING CONFLICT IN THE LITERATURE
229
was complementary to the effect size estimates (d ) posited by Glass et al. (1981), providing a more detailed picture of meta-analysis results across situations and research areas. These publications lay the foundation for what has been termed “modern” meta-analysis methods (Huffcutt 2002). Modern metaanalyses address not only effect size (d) and probability ( p) estimates, but include methodological guidance on issues such as assessing moderators and weighting individual studies (see Hedges & Olkin 1985; and Hunter & Schmidt 1990 for reviews).
The Meta-analytic Research Process Similar to other types of research approaches, meta-analysis is geared towards generalization (Glass et al., 1981); we are inferring the realities of the research domain from a sample of studies (Huffcutt, 2002). Rather than examining individuals as data points, meta-analysis examines studies as the individual data points. The steps of meta-analysis follow the framework typical for scientific inquiry: formulating the problem, collecting and evaluating data, analyzing the data, and interpreting the results (Hedges, Shymansky, & Woodworth 1989). Formulating the Problem. One of the initial hurdles in a meta-analysis is ascertaining what it is that we are studying and ultimately inferring from our sample of studies. Formulating the problem is more difficult than it sounds. This step initiates the “apples and oranges” criticism of meta-analysis (e.g., Glass et al. 1981). That is, if the problem is identified with relatively broad constructs, then the studies that are later chosen and aggregated may be very different from each other. Ultimately, combining the studies would result in aggregating “apples and oranges,” with the result being of limited value to the research area. For example, Stuhlmacher et al. (1998) considered the definition of time pressure in negotiation and what researchers might consider “time pressure.” Based on this, we included not only studies manipulating a time limit for negotiating but such factors as costs for each offer, fixed number of offers, and third party time pressure. These variables may seem like a bunch of “apples and oranges” but, conceptually, each limited the amount of time for negotiation. Similarly, Zetik and Stuhlmacher (2002) decided that “aspirations” in negotiation and “goals” in negotiation are similar constructs, although these originate in different research traditions. The “apples and oranges” criticism remains a prevailing issue for researchers and readers (Glass et al. 1981; Lipsey & Wilson 2001). Perhaps it is the interpretation of relevance, but “what is combined” remains an issue among supporters and critics of the meta-analytic methodology.
230
ALICE F. STUHLMACHER AND TREENA L. GILLESPIE
Practical issues may also influence formulating the problem. For example, at some point, the researcher will want a sense of how many studies might be included. If there are a limited number of studies, it might make sense to broaden the definition of the independent or dependent variables. Or, it might be feasible to paint a broader picture of the construct by including multiple dependent variables (e.g., effects on both cooperation and impasses; Stuhlmacher et al. 1998). Invariantly, beginning researchers seek guidance or directions for a “magic” number of studies. Rather than reiterating the concerns of an ideal quantity of studies to sample (e.g., Glass, et al. 1981; Rosenthal 1984), Hedges (1990) pointed out that researchers should examine the representativeness of the participants and the treatments in the studies relative to the area of interest. Thus, again, investigators are asked to use judgment in determining the selection of the participants, treatments, and studies to the research domain. Even with clearly defined constructs, investigator judgment calls are necessary. Hedges et al. (1989) warned, “Treatments will not have been implemented identically in all studies and different studies will not have measured the outcome constructs in exactly the same way. Thus, you will have to judge whether each operation is a legitimate representation of the corresponding construct” (5). Negotiation and conflict research is marked by several research traditions across multiple scholarly disciplines, thus creating some potential misclassifications. For example, Walters et al. (1998) questioned if research on prisoner dilemma games should be included with explicit negotiation. Ultimately, because the research literature has built on matrix games and explicit negotiations, both of these operationalizations were combined. Studies were coded for whether the task was a matrix game or an explicit negotiation so that each type of task could be compared. In Walters et al. (1998), these two tasks produced different results regarding gender differences. The ability to run comparisons across treatments is an unmistakable advantage of synthesizing the literature, and an example of how meta-analyses can clarify past research efforts. Similarly, because conflict and negotiation are vast research literatures, a common misperception is that large numbers of studies exist that can be combined on a particular topic. We have been extremely surprised over the sometimes small number of studies that actually address a particular question and report relevant data. The most startling finding was a search that resulted in only 21 explicit negotiation studies (yielding 53 effect sizes) directly comparing the profits men and women earned in explicit negotiation (Stuhlmacher & Walters 1999). Despite several hundred articles being screened, we were puzzled that such a simple relationship was not reported in a way that allowed
MANAGING CONFLICT IN THE LITERATURE
231
effect size calculations. In this case, reviewers were equally puzzled, and asked for validation of the search. It was certainly possible that gender comparisons may have been reported in a manuscript but were not retrieved by database keywords or other searches. To locate more studies, a manual search of frequent journal outlets of negotiation research was undertaken. When missing studies did not materialize, attention was turned to explain why so few studies were found and included. The smaller than expected sample size in the meta-analysis could be due to a combination of factors. Undoubtedly, some studies slipped past our search, particularly unpublished research. Ultimately, although many researchers could have run the analyses for gender, gender information was not incorporated in final reports. In many cases, gender-profit was not central to the reports, and thus probably omitted. Further, interest in reporting gender differences in negotiation decreased after reviews (e.g., Rubin & Brown 1975) found limited gender differences. It was also possible that researchers found gender differences that were mixed or non-significant (possible given the sample size to detect a small effect) and publication pressures resulted in omitting gender comparisons from the report. While it is unsatisfying to find a small number of useable studies, clarifying the amount and type of research on a topic is yet another important way that meta-analyses can summarize past efforts and direct future research. Collecting and Evaluating Data. The next stage, collecting data, involves many of the same issues that arise in other research designs, such as choosing participant characteristics. One of the main concerns with meta-analysis has been the suitability of the studies selected for inclusion in a meta-analysis and the implications for the inferences made from the results (Hedges 1990). Studies relevant to the research area may be scattered throughout academic libraries, presented at local conferences, or crammed into the dark recesses of a laboratory file cabinet. Some of these studies are easily accessible (e.g., journal articles, books), while others require more diligent investigation. Regardless, the comprehensive meta-analyst will search for pertinent work by using a variety of tactics such as contacting prominent researchers in the field, poring over conference programs, and posting requests on Internet sites. Once assembled, the studies generally undergo scrutiny by more than one researcher. At this juncture, certain characteristics of the research report are coded. Coding may require simple distinctions that are easy to code reliably (e.g., coding as a published vs. unpublished study) while some distinctions require more training and judgment in coding (e.g., relative power between parties, opponent strategy). Lipsey and Wilson (2001) clarify that variables are coded for at least one of three reasons. Some variables relate to substantive
232
ALICE F. STUHLMACHER AND TREENA L. GILLESPIE
issues, or those variables that relate to the topic of study and may be hypothesized to impact the results. A second group of variables is coded that relates to the methods and procedures (e.g., task, participant population, use of confederates). These variables can examine if differences in procedures influence the meta-analytic results. Finally, source descriptors are coded for describing the sample of reports and their context (e.g., published/unpublished, date of study). Each type of coding plays a unique and important purpose in capturing the studies and to assist in interpreting the findings. Analyses. After the data are collected, they often are analyzed using techniques associated with a methodological tradition. Again, this is a judgment call. Interestingly, the approach may be highly influenced by the academic discipline of the meta-analyst. In an analysis of meta-analytic trends (Mohr, Zickar, & Russell 2000), Schmidt and Hunter’s methods were employed in 85% of the meta-analyses by industrial/organizational psychologists, Glass’s methods were frequently used by clinical (50%) and educational psychologists (65.7%), while social psychologists were more likely to apply the Hedges method (38.1%) or the Rosenthal method (23.8%). The meta-analyses that have been published to date in negotiation have used the Hedges method (i.e., Hedges & Olkin 1985). Our decision to use Hedges’ procedures was driven by a desire to compare our results on gender differences with seminal gender meta-analyses by Alice Eagly and colleagues (e.g., Eagly & Johnson 1990; Eagly, Karau, & Makhijani 1995). Also, we found that negotiation measures were not clearly amenable to the assumptions and corrections for the study artifacts that Schmidt and Hunter propose. Across traditions, assumptions and formulas differ somewhat; ultimately, the choice of methods may influence the reception of the meta-analytic research. It is clear though that current developments require frequent re-evaluation and consideration of any particular approach. Also related to the analyses is the research design and choice of variables. The interdependent nature of negotiation creates issues regarding the level of analysis to address (Zetik & Stuhlmacher 2002). Many variables of interest in negotiation can be considered across multiple levels. For example, when do we want to look at joint or dyad level outcomes and when are we really interested in individual level variables? This creates yet more judgment calls for the meta-analyst. Meta-analysts face other judgment calls when data in the selected studies are missing or incomplete. Similar to sampling concerns, nonrandomly missing data impacts the validity of the inferences made from metaanalytic results. The researcher then must determine how to deal with missing data (Pigott 1994). In addition to the more traditional concerns of missing
MANAGING CONFLICT IN THE LITERATURE
233
data, negotiation research faces a dilemma of negotiations that may not come to an agreement or ones that end early. For example, in the case of an impasse, should the settlement be represented as a missing data point, a zero profit, a best alternative to a negotiated agreement (BATNA) or some sort of average? How should process issues be considered when the negotiation is not resolved? It becomes more complex when the original researcher omits how impasses were handled or how group statistics were calculated. Current researchers can certainly assist future research by including complete methodological information and relevant descriptive statistics (particularly means, standard deviations, sample sizes, and intercorrelations) across conditions and variables. The quality of the meta-analysis results rests on the quality of the data included in the review. In short, investigators often find themselves making judgment calls, some arising very early in the process and others emerging after investigators have logged extensive hours on the project. These junctures, while seemingly objective, allow subjectivity to creep into the meta-analysis methodology, influencing the project with the investigators’ biases. However, judgment calls are a part of the process and can be managed professionally. One way of approaching judgment calls is to construct a plan in advance for all the steps of the project, detailing how judgments will be made, and reporting deviations from the plan in the final report (Hedges 1990). For example, a decision can be to run the analysis in more than one way to incorporate a possible judgment call. As mentioned earlier, Walters et al. (1999) ran results together and separately for matrix games and explicit negotiations. Zetik and Stuhlmacher (2002) compared negotiation performance across goal conditions in two ways (optimal vs. suboptimal goals, and goal vs. no goal) to tease out if different operational definitions across studies made a difference in interpretation. In this way, more information is brought to evaluating the analyses. Interpretation and Reporting. After analyses are completed, the information is interpreted and written up into a final report. The final report ideally includes the judgments that were made throughout the course of the metaanalysis; readers are thus alerted to judgment calls that limit the generalizability of the results or the inferences that can be made. For example, specific information is needed on how the studies were selected, the descriptions of the subjects in the studies, as well as a depiction of the treatments (Hedges 1990). Judgment continues as meta-analysts try to make sense of the data. Given identical results, researchers may interpret it differently (Cooper 1990b), which influences the resulting report and potentially future research. As Cooper (1990b) explained, “Thus, while a great deal of agreement might
234
ALICE F. STUHLMACHER AND TREENA L. GILLESPIE
be reached on the observation that an eight-ounce glass contains four ounces of water, there can still be much disagreement about whether the glass is half empty or half full” (87).
Challenges of Meta-analysis Anticipating and resolving the judgment calls within a meta-analysis are central challenges of the meta-analysis process. There are resources devoted to the judgment call issue in greater depth (e.g., Wanous, Sullivan, & Malinak 1989) and ideally these judgments should be addressed at the beginning of a meta-analytic project. Anticipated in advance, the meta-analysts can agree on the best process for navigating these issues. Glass et al. (1981) remarked that many critics may be concerned about study quality, studies that include nonindependent data, and the problem of selection bias in the studies that are more easily accessible to researchers. However, they countered these arguments by showing the insignificance of the effects of these cited methodological weaknesses in actual data analysis. A further challenge is that meta-analyses require a substantial amount of expertise and effort (Lipsey & Wilson 2001). The uninitiated may not appreciate the amount of time, effort, and expertise involved in a completing a high-quality meta-analysis. Initially, the researcher may encounter obstacles convincing others about the value and difficulty of such projects. As more scientists become familiar with the technique, it will hopefully become less necessary for the researcher to educate promotion or tenure committees, grant agencies, and colleagues that meta-analyses involve significant time, effort, and resources but can yield a substantial contribution to the field. Ultimately, we find that high-quality research requires time and a meta-analysis is no different. These challenges are less likely as meta-analyses are published which clarify the status quo set by primary studies. The effects between variables can appear far weaker or even inverted when a comprehensive meta-analysis is conducted. One recent example is a meta-analysis by De Dreu and Weingart (2003) where the general belief that task-conflict can be beneficial to team performance was not supported. In fact, task-conflict appeared to be as dysfunctional as relationship conflict. Future research can clearly benefit from the research review that a meta-analysis can provide to a topic.
MANAGING CONFLICT IN THE LITERATURE
235
Future of Meta-analysis The future for meta-analysis is bright, and advancing with increasing sophistication. One way this can be seen is through an increase in software developed to assist the meta-analyst (e.g., Arthur, Bennett, & Huffcutt 2001; see also Rothstein et al. 2002 for summary of software). It also appears clear that one substantial advantage of the meta-analytic perspective is the opportunity for researchers to handle large amounts of information (Lipsey & Wilson 2001). As a result of planned coding systems, a meta-analysis can deal with large numbers of variables across many studies without the concern of studies being forgotten or receiving less attention from the investigator. If enough literature exists, predictions from competing models can be tested (e.g., De Dreu et al. 2000). In addition, moderators, mediators, or other relationships between variables can be investigated that may have been less obvious in the separate studies or if reviewed using more conventional methods of research synthesis (Lipsey & Wilson 2001). Meta-analyses have been criticized as emphasizing correlational rather than causal relationships. However, this criticism hinges on the type of research that has been conducted. Like other forms of modeling, causality assumptions rely on the design of the research. Thus, in many ways, the meta-analyst is constrained by the data that exists. One recent approach that expands the use of existing data involves using meta-analyses to test mediators through building causal models (Becker 1992; Shadish 1996). For example, a correlation matrix can be derived from several meta-analyses and can be used as the input for path-analysis. This then has the potential to test not just moderators but mediating pathways and competing models (e.g., Grant, Compas, Stuhlmacher, Thurm, McMahon, & Halpert 2003).
Conflict Resolution in the Literature In summary, meta-analysis offers a unique perspective for evaluating the progress in a research domain. Meta-analysis sifts through past research in a disciplined way, providing structure to the search for answers (Lipsey & Wilson 2001). It is our hope that this paper illustrates how meta-analysis can set clear ground rules, assumptions, and goals; involve ignored voices; while organizing, re-conceptualizing, and synthesizing disparate perspectives. More simply, a good meta-analysis should be an example of how to resolve, or at least manage, conflict. If it is true that “the field of conflict resolution has become, ironically, very competitive” (Coleman 2001: 597), then this area is
236
ALICE F. STUHLMACHER AND TREENA L. GILLESPIE
fertile ground for meta-analytic work. Competition between ideas and traditions has always existed in negotiation research. These pressures may have been intensified of late as negotiation has become more central to the curriculum in many areas of study, as more researchers enter the field, and as pressures to publish research has increased. As studies continue to emerge and results, however disparate from each other, continue to be reported, the role of meta-analysis becomes greater in helping make sense of it all by exposing and examining contradictory results. It is our belief that meta-analyses afford opportunities for constructive dialogue and advances in understanding social conflict.
References Arthur, Winfred. Jr., Bennett, Winston, Jr., & Huffcutt, Allen (2001). Conducting meta-analyses using SAS. Mahwah, N.J.: Lawrence Erlbaum Associates. Becker, Betsy Jane (1992). “Models of science achievement: Forces affecting male and female performance in school science,” in Thomas D. Cook, Harris Cooper, David S. Cordray, Heidi Hartmann, Larry V. Hedges, Richard J. Light, Thomas A. Louis, & Frederick Mostellar, editors, Meta-analysis for explanation: A casebook. New York: Russell Sage Foundation. Coleman, Peter T. (2001). “Concluding overview,” in Morton Deutch & Peter T. Coleman, editors, The Handbook of Conflict Resolution: Theory and Practice. San Francisco: Jossey-Bass. Cooper, Harris M. (1990a). “Moving beyond meta-analysis,” in Kenneth W. Wachter & Miron L. Straf, editors, The future of meta-analysis. New York: Russell Sage Foundation. Cooper, Harris M. (1990b). “On the social psychology of using research reviews,” in Kenneth W. Wachter & Miron L. Straf, editors, The future of meta-analysis. New York: Russell Sage Foundation. Cooper, Harris M., & Rosenthal, Robert (1980). “Statistical versus traditional procedures for summarizing research findings.” Psychological Bulletin, 87: 442–449. De Dreu, Carsten K., Weingart, Laurie R., & Kwon, Seungwoo (2000). “Influence of social motives on integrative negotiation: A meta-analytic review and test of two theories.” Journal of Personality and Social Psychology, 78: 889–905. De Dreu, Carsten K., & Weingart, Laurie, R. (2003). “Task versus relationship conflict, team performance, and team member satisfaction: A meta-analysis.” Journal of Applied Psychology, 88, 4: 741–749. Druckman, Daniel. (1994). “Determinants of compromising behavior in negotiation: A metaanalysis.” Journal of Conflict Resolution, 38: 507–556. Eagly, Alice H., & Johnson, Blair T. (1990). Gender and leadership style: A meta-analysis.” Psychological Bulletin, 108: 233–256. Eagly, Alice H., Karau, Steven J., & Makhijani, Mona (1995). “Gender and the effectiveness of leaders: A meta-analysis.” Psychological Bulletin, 117: 125–145. Glass, Gene V. (1976). “Primary, secondary and meta-analysis of research.” Educational Researcher, 5: 3–8.
MANAGING CONFLICT IN THE LITERATURE
237
Glass, Gene V., McGaw, Barry, & Smith, Mary L. (1981). Meta-analysis in social research. Beverly Hills, CA: Sage. Grant, Kathyrn E., Compas, Bruce E., Stuhlmacher, Alice F., Thurm, Audrey E., McMahon, Susan D., & Halpert, Jane A. (2003). “Stressors and child/adolescent psychopathology: Moving from markers to mechanisms of risk.” Psychological Bulletin, 129: 447–466. Hedges, Larry V. (1990). “Directions for future methodology,” in Kenneth W. Wachter & Miron L. Straf, editors, The future of meta-analysis. New York: Russell Sage Foundation. Hedges, Larry V., & Olkin, Ingram (1985). Statistical methods for meta-analysis. Orlando, FL: Academic Press. Hedges, Larry V., Shymansky, James A., & Woodworth, George (1989). Modern methods of meta-analysis. Washington, DC: National Science Teachers Association. Huffcutt, Allen I. (2002). “Research perspectives on meta-analysis.” in Steven G. Rogelberg, editor, Handbook of research methods in industrial and organizational psychology. Malden, MA: Blackwell. Hunter, John E., & Schmidt, Frank L. (1990). Methods of meta-analysis: Correcting error and bias in research findings. Newbury Park, CA: Sage. Light, Richard J., & Smith, Paul V. (1971). “Accumulating evidence: Procedures for resolving contradictions among different research studies.” Harvard Educational Review, 41: 429–471. Lipsey, Mark W., & Wilson, David B. (2001). Practical meta-analysis: Vol. 49. Applied social research methods series. Thousand Oaks, CA: Sage. Mohr, David C., Zickar, M.J., & Russell, S.S. (2000, April). An analysis of trends in metaanalytic research. Poster presented at the Annual Meeting of the Society of Industrial/ Organizational Psychology. New Orleans, LA. Pigott, Therese D. (1994). “Methods for handling missing data in research synthesis,” in Harris Cooper & Larry V. Hedges, editors, The Handbook of Research Synthesis. New York: Russell Sage Foundation. Rosenthal, Robert (1984). Meta-analytic procedures for social research. Beverly Hills, CA: Sage. Rosenthal, Robert (1991). Meta-analytic procedures for social research: Vol. 6. Social research methods series. Newbury Park, CA: Sage. Rothstein, Hannah R., McDaniel, Michael A., & Borenstein, Michael (2002). “Meta-analysis: A review of quantitative cumulation methods,” in Fritz Drasgow & Neal Schmitt, editors, Measuring and analyzing behavior in organizations. San Francisco: Jossey-Bass. Rubin, Jeffrey Z., & Brown, B.R. (1975). The social psychology of bargaining and negotiation. New York: Academic Press. Shadish, William R. (1996). “Meta-analyses and the exploration of causal mediating processes: A primer of examples, methods, and issues.” Psychological Methods, 1: 47–65. Smith, Mary L., & Glass, Gene V. (1977). “Meta-analysis of psychotherapy outcome studies.” American Psychologist, 32, 9: 752–760. Stuhlmacher, Alice F., & Citera, Maryalice. (in press). Negotiator hostility and profit in virtual negotiations: A meta-analysis. Journal of Business and Psychology. Stuhlmacher Alice F., Gillespie, Treena L., & Champagne, Matthew V. (1998). “The impact of time pressure in negotiation: A meta-analysis.” The International Journal of Conflict Management, 9: 97–116. Stuhlmacher, Alice F., & Walters, Amy E. (1999). “Gender differences in negotiation outcome: A meta-analysis.” Personnel Psychology, 52: 653–677.
238
ALICE F. STUHLMACHER AND TREENA L. GILLESPIE
Walters, Amy E., Stuhlmacher, Alice F., & Meyer, Lia L. (1998). “Gender and negotiator competitiveness: A meta-analysis.” Organizational Behavior and Human Decision Processes, 76: 1–29. Wanous, John P., Sullivan, Sherry E., & Malinak, Joyce (1989). “The role of judgment calls in meta-analysis.” Journal of Applied Psychology, 74, 2: 259–264. Zetik, Deborah C., & Stuhlmacher, Alice F. (2002). “Goal setting and negotiation performance: A meta-analysis.” Group Processes and Interpersonal Relations, 5: 35–52.
When, Where and How: The Use of Multidimensional Scaling Methods in the Study of Negotiation and Social Conflict ROBIN L. PINKLEY, MICHELE J. GELFAND and LILI DUAN
Multidimensional scaling (MDS) like other multivariate procedures is a datareduction technique that allows us to discover how and why variables are related. As such, the purpose of MDS is to uncover the spatial representation or “hidden structure” that underlies and defines behavioral data – such as negotiator or disputant perceptions and preferences (Kruskal & Wish 1978). A fundamental aspect of human behavior is the tendency to make judgments about the degree of similarity and difference among the myriad stimuli with which we are faced (Green & Carmone 1970). For example, scholars reading this journal are certain to have beliefs about how this journal and others are related to one another; although they may not fully recognize the criteria they are using to make such judgments. Multidimensional scaling techniques allow us to uncover the perceived attributes or “dimensions” that account for correlations among these judgments and label the criteria used for making them. The ultimate goal of MDS techniques is to produce a geometric map that illustrates the underlying structure of complex psychological phenomena. The distance between the stimuli in a spatial map represents judgments regarding how similar or dissimilar each stimuli is to others. The smaller the distance between two stimuli, the greater their proximity and thus, the greater the similarity between them. By producing a map of the evaluated stimuli, MDS techniques are able to illuminate the “hidden structure” and thus, the cognitive framework or underling dimensions that distinguish one class or category of stimuli from another. MDS has been applied in a variety of disciplines, including psychology (e.g., Johnston 1995), economics (e.g., Black 1991), sociology (e.g., Beardsworth & Keil 1992), political science (e.g., Lieske 1993), anthropology (e.g., Bernard 1994), and organizational behavior (e.g., Robinson & Bennett 1995; Jehn 1994). Indeed, MDS has been used to understand a wide range of phenomena – ranging from perceptions of nations (Wish, Deutsch, & Biener 1971), to perceptions of visual patterns (Hirschberg, Jones, & Haggerty 1978). Like factor analysis, MDS may be applied to any matrix of data, as long as International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 239–255 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
240
ROBIN L. PINKLEY, MICHELE J. GELFAND AND LILI DUAN
the elements of the matrix provide information about the relation among the objects, events, or behaviors (which we will refer to as stimuli) that comprise the rows and columns of the data matrix (Young & Hamer 1987). Unlike factor analysis, however, MDS does not require metric data, allowing for the use of both metric (interval or ratio) and nonmetric (ordinal) data. Given that data reflecting attitudes and cognitions are nonmetric, MDS is ideally suited to the study of conflict and negotiation. Although MDS has wide-ranging theoretical and applied appeal, it has been highly underutilized in the conflict and negotiation literature, which has tended to rely on factor analysis to understand hidden data structures. In this paper, we seek to illustrate the promise that MDS has in the study of conflict and negotiation. We begin with a discussion of how MDS can be differentiated from other multivariate techniques, such as factor analysis, illustrating its distinct advantages to conflict scholars. Next we provide a brief overview of multidimensional scaling techniques – highlighting the various methods available for collecting proximity data and the MDS computer analysis programs that can be used to analyze them. We further review the nature of the results and the ways in which they are interpreted. We conclude with some examples of the types of questions that have been addressed using MDS in the conflict and negotiation literature, highlighting the promise this technique holds for future research.
Multidimensional Scaling (MDS) versus Factor Analysis (FA) Conflict researchers have typically used factor analysis to understand the structure of data. Given that MDS techniques provide a number of advantages over the use of factor analysis and other multivariate techniques, it is important to gain an understanding of the relationship between the two. First, while each uses a very different set of statistics, the principle behind each method is quite similar. Factor analysis and multidimensional scaling are both based on the premise that when a bunch of variables (or in the case of MDS, a number of stimuli) are correlated with each other, they have something in common (Bernard 2002). In the case of factor analysis, that “something” is referred to as a factor, while in multidimensional scaling, it is referred to as a data cluster or category. Regardless of whether we are referring to factors, data clusters or categories, they are all “supervariables” or latent variants that subsume a number of variables (or stimuli) into fewer, broader variable classes. Thus, both factor analysis and multidimensional scaling are used to identify the structure or
THE USE OF MULTIDIMENSIONAL SCALING METHODS
241
interrelationships among these supervariables. For example, factor analysis is used to illuminate factors that capture the interrelationship among variables. Similarly, MDS uses dimensions to illustrate the structure among variable clusters and categories. This is beneficial, because when “we can discover (or, more correctly, intuit) these underlying supervariables, we can explain the variance in the dependent variable of interest, with a small number of independent variables” (Bernard 2000: 635). Despite the principle that underlies both statistical techniques, there are some notable differences between them. In the case of FA, factors represent the underlying relationships of a set of attributes with respect to a sample of individuals. As such, one subject’s attribution is not sufficient for the application of factor analysis. In contrast, MDS uses individuals as the unit of analysis. This means that one respondent’s evaluation of stimuli is sufficient (although rarely used) for the use of an MDS analysis, allowing researchers to obtain a solution for each individual. Thus, MDS focuses on how an individual perceives the objects, rather than on the objects themselves. Another important difference between FA and MDS is the nature of the responses that can be obtained from participants. There are generally two approaches to obtaining participants’ assessment of stimuli: attribute-free and attribute-based approaches (Hair, Anderson, Tatham, & Black 1998). As discussed below, both approaches may be employed with MDS, whereas only attribute-based approaches can be used with Factor Analysis. Attribute-free data are based on participants’ direct assessment of the similarity (or dissimilarity) between stimuli. With this approach, the investigator does not provide any criteria on which these judgments are to be made. For example, Gelfand Triandis, & Chan (1996) asked respondents to judge the degree to which 15 different concepts were each similar to one another, for a total of 105 paired comparisons. Likewise, Pinkley (1990) asked respondents to sort conflict stimuli into categories based on their similarity. In both studies, participants were free to judge the similarity between stimuli based on their own criteria, enabling their own mental models of the stimuli to surface. In this respect, because the attribute-free approach asks respondents to provide similarity judgments without the researchers’ criteria being provided, it is less likely to be contaminated by the preconceptions or hypotheses of the researcher. By contrast, attribute-based approaches ask participants to assess stimuli on a pre-defined set of attributes. For example, Hensen, Sarma, & Collins (1999) asked participants to rate their preferences on items reflecting Holland’s six occupational themes. These preferences were later transformed into Euclidean Distances for input into an MDS analysis. Importantly, MDS studies can use
242
ROBIN L. PINKLEY, MICHELE J. GELFAND AND LILI DUAN
either attribute-free or attribute-based approaches, whereas factor analysis can only be applied to the attribute-based approaches. That is, FA requires subjects to rate stimuli on some list of attributes provided by the researcher. Accordingly, it retains a higher risk of contamination by the researcher’s own criteria. A second advantage of using MDS techniques is that they provide a mechanism for detecting, quantitatively categorizing and labeling people’s perceptions and preferences, even when the criteria used to make such judgments are implicit or cognitively unavailable to respondents (Pinkley 1990). While people can readily compare and evaluate stimuli, they are often less able to conceptualize their perceptions and judgments in terms of specific categories or identify the dimensions that underlie them.
Different Types of MDS MDS techniques can be used with both homogeneous and heterogeneous samples. When individual differences are not of interest or assumed to be nonexistent (i.e., the respondent population of interest is assumed to be homogeneous), a traditional two-way matrix (sometimes referred to as in-group scaling) should be used. This would be the case, for example, if an experimenter wanted to determine the dimensions that underlie and account for scholar perceptions (with an n = 50 scholars) of 40 scholarly journals (to extend our earlier example), but was unconcerned with how individual differences or scholar characteristics might affect such judgments. In this case, the design would be a 40 × 40 design with a Cartesian product of < stimuli × stimuli > (i.e., a 40 × 40 matrix), with every journal compared to every other journal. When large or heterogeneous sample sizes are used however, it is dangerous to assume that all respondents will share the same point of view and thus, by extension that it is safe to assume homogeneity of similarity judgments across people (Green & Carmone 1970). As a consequence, a three-way MDS, or individual differences scaling procedure is often used in the place of the more traditional two-way procedure, because it allows researchers to examine the pattern of each respondents (or subgroup of respondents) perception of the stimuli. When this approach is used, homogeneity can be uncovered through analysis, instead of mandated by assumption, leading to the aggregation of subject judgments. The most widely used three-way MDS procedure is the INDSCAL model developed by Douglas, Green, and Schaffer (1986). This method develops
THE USE OF MULTIDIMENSIONAL SCALING METHODS
243
both a common or “group” space (which is very similar to the solution obtained using a two-way MDS method) and a set of respondent weights allowing the experimenter to examine how each point of view is related to the others (i.e., the dimensions that each demographic type has in common). Differences in dimension saliency (i.e., points of view) can then be related to individual differences and respondent/situation characteristics to test for hypothesized relationships. As a consequence, research investigates whether respondents who have high weights on a particular dimension (i.e., the dimension accounts for much of the variance in their judgments) are different from those with low weights on that dimension. Returning to our example, this technique would allow researchers to discover that scholars affiliated with research institutions weight a dimension labeled “rigor” more heavily than those affiliated with teaching colleges, and that those at teaching colleges weight a dimension labeled “corporate application” more heavily than those at research institutions. In this case, the design would be a “40 journals × 40 journals × 50 respondents” design with a Cartesian product of < stimuli × stimuli × respondents > (i.e., a 40 × 40 × 50 matrix), with each journal compared to every other journal. A three-way MDS is also an appropriate technique for comparing the perceptions of one or more groups to another. For example a three-way MDS could be used to determine if the perceptions of Japanese scholars varied from those of Latin American scholars and US scholars. In this case, the matrix would resemble 40 journals × 40 journals × 3 cultures. The INDSCAL model can also be used to evaluate the goodness of fit for each respondent (or demographic type) stimulus configuration (i.e., the amount of variance in respondent judgments explained by the multidimensional solution). An alternative method for aligning individual differences with points of view is to determine which respondent characteristics predict the pattern of dimension weight scores. For example, Jones and Young (1972) use discriminate function analysis to distinguish groups in terms of their different dimension patterns. In addition to INDSCAL, numerous other computing programs have been developed for multidimensional scaling and other related tasks such as cluster analysis. These programs include, but are not limited to ALSCAL (Takane, Young & Deleeuw 1976), M-D-SCAL (Kruskal 1968), TORSCA (Young & Torgerson 1968), and PROXSCAL (Young & Hamer 1987). The ALSCAL multidimensional scaling program is included in the SPSS 10.0 Base package and is also available in the SAS ALSCAL procedure. ALSCAL performs metric or nonmetric MDS and has individual differences scaling options and thus can compare the differences of several individual or group matrices.
244
ROBIN L. PINKLEY, MICHELE J. GELFAND AND LILI DUAN
PROXSCAL, another multidimensional scaling program, is also available with SPSS 10.0 Categories package. PROXSCAL offers several improvements upon ALSCAL, including algorithmic strategies that better ensure convergence, a wide range of data transformations, and a number of different options for fitting models to the data (see Busing, Commandeur, & Heiser 1997, for further discussion).
Data Collection Methods Regardless of the type of stimuli presented to respondents (e.g., objects, people, behavior, events) the input for MDS techniques is the similarity data, referred to as relational or proximity data. MDS handles all kinds of proximity data matrices including metric or nonmetric; matrices with or without missing proximity data; rectangular (i.e., two-way) or square matrices; and unequally replicated matrices (Young & Hamer 1987). Since most of the MDS computer analysis programs possess “missing data” features, missing cells pose no problem as long as the absolute number of proximity data entries is large relative to the number of dimensions necessary to account for the relationships among them. A number of methods are available for converting stimuli into proximity data. The most common method is to obtain direct, pair wise comparisons, by having respondents judge the degree to which each stimulus is similar to every other stimulus on a Likert-type scale (see Gelfand, Triandis, & Chan 1996 for an example). An alternative method is to randomly select a subset of the stimuli (say for example 10 out of our stimulus set of 40 journals) to be designated as “target” stimuli, against which all other stimuli must be compared (Pinkley, Brittain, Neale & Northcraft 1995). When this method is used, respondents are presented with one of the target stimuli (for example, one out of the 10 randomly selected target stimuli) and then asked to rank order the remaining stimuli (the remaining 39 out of 40 using our example) in terms of similarity in ascending or descending order (see Pinkley 1990 or Pinkley, Neale, Brittain, & Northcraft 1995 for examples). A third method, called the subjective clustering method, (Green & Carmone 1970) requires respondents to sort stimuli into groups so that those in the same group are more similar to each other than those in other groups (see Johnson 1995; Gelfand, Nishii, Holcombe, Dyer, Ohbuchi, & Fukuno 2001 for examples). Correlations among variables or any other indication of the interrelationship among stimuli are also acceptable for input.
THE USE OF MULTIDIMENSIONAL SCALING METHODS
245
Interpreting the MDS Configuration and Labeling the Dimensions There are two issues that scholars face when interpreting MDS configurations: 1) determining the number of dimensions that best represents the actual proximities between the data, and 2) labeling each dimension or determining the interpretability of each dimension. Each issue is discussed in turn below. Determining Dimensionality. Three criteria are used to determine the optimum number of dimensions needed to describe the stimulus space. While reliance on these methods varies, experimenters typically use all three criteria for determining dimensionality. The first criterion used for determining dimensionality, is Kruskal’s (1964) STRESS index or goodness of fit, which indicates how the distances displayed in the configuration reflect the actual proximities in the similarities data. Technically speaking, it is the square root of a normalized residual sum of squares, which exhibits the amount of variance that remains unaccounted for by the MDS model. Although measurements of STRESS vary from analysis program to analysis program (such as M-D-SCAL, TORSCA, KYST, or ALSCAL), the meaning of STRESS is always the same: Small STRESS indicates good fit, with good fit nearing zero and poor fit nearing one. As the number of dimensions increase, STRESS becomes closer to zero. It should be noted that a number of factors affect stress values. For example, when the number of stimuli (I) and the number of dimensions (R) of stress are similar, the STRESS index can be distorted, resulting in undue influence of the interpretation R. As a result, a good rule of thumb is to use at least four times as many stimulus items as the number of dimensions likely to underlie the stimulus space (i.e., I > 4R). A second criterion for evaluating the number of dimensions necessary and sufficient to adequately represent the stimulus space, is the RSQ index or squared multiple correlation between the proximities in the similarities data and the distances plotted by the MDS model. The RSQ index describes how much of the variance in the proximity data is accounted for by the MDS model. As with any squared correlation, a one indicates a perfect fit and a zero indicates no fit at all. To determine the appropriate number of dimensions, researchers typically plot the first two criteria (i.e., STRESS and RSQ) against the number of dimensions to discover an elbow or bend, which is designated by a sudden rise in RSQ and fall in STRESS. The number of dimensions that correspond with the elbow represent a particularly good fit of the MDS model to the proximity data (Young & Hamer 1987). The elbow test is usually accompanied by an assessment of how well the raw data fits the MDS model by
246
ROBIN L. PINKLEY, MICHELE J. GELFAND AND LILI DUAN
examining the amount of variance accounted for by the MDS procedure. If no elbow is found, the appropriate number of dimensions cannot be selected on that basis. A third criterion for selecting dimensionality, is the number of dimensions most easily interpreted (using the interpretation procedures discussed below), with the goal being to “select the space with the fewest dimensions and the richest interpretation” (Young & Hamer 1987: 205). Interpretability of Dimensions. After selecting dimensionality, the dimensions must be interpreted and labeled. One common method for labeling dimensions is to visually inspect the spatial maps produced by the MDS analysis to look for patterns in the attributes of stimuli clustered around one end of a dimensions continuum to those at the other end. This “massaging” of the data, can lead to interesting insights. As a complement to this subjective procedure, researchers often employ more rigorous, objective techniques to aid in the interpretation of the multidimensional space (configuration). One common technique is to have the participants who made the proximity judgments rate the stimuli on a number of unidimensional attributes and then use multiple regression to regress the unidimensional attributes onto the coordinate values in the multidimensional space. In order for an attribute to be useful in interpreting the space, it must have a: 1) significant multiple correlation and F-value, indicating that the configuration “explains” the attribute well, and 2) significant Beta weight (normalized regression coefficient) on a dimension, indicating that the attribute corresponds well with the multidimensional space (Kruskal & Wish 1978). The task of assigning a label to a particular dimension is simplest when each label loads on only one dimension (Pinkley, et al. 1995). An alternative method is to ask the participants making the proximity judgments to specify the criteria they use for making these judgments. If this is done, a second set of subjects can be given the original set of stimulus objects and the criteria list provided by the proximity-rating participants and asked to rate the degree to which each criteria describes each stimulus object on a Likert-type scale.
MDS and Research: Examples and Prospects for Future Research in Conflict and Negotiation Several examples may further illustrate the inherent benefits of using MDS techniques in the study of conflict and negotiation. Although we provide only a couple examples here, a handful of scholar’s have used MDS techniques to address such issues as negotiatior perceptions of conflict situations (Pinkley
THE USE OF MULTIDIMENSIONAL SCALING METHODS
247
1990), mediation tactics (Carnevale & Pegnetter 1985; McLaughlin, Carnevale, and Lim 1991), managerial third-party dispute intervention strategies (Pinkley et al. 1995), cross-cultural studies (Gelfand et al. 2001), and international conflicts (Druckman 1997; Druckman, Martin, Nanand, & Yagcioglu 1999). A survey of these studies will demonstrate the varied techniques for collecting proximity data, as well as, determining situations and labeling the dimensionality. Example 1 Pinkley, Brittain, Neale, & Northcraft (1995) used MDS to conduct an inductive analysis of managerial third-party dispute intervention strategies. The objective of this study was to identify the dimensions that distinguish one class or category of intervention strategies from others. In addition, the authors examined the relationship between strategy selection and the nature of the conflict (i.e., managerial dispute intervention goals, dispute intensity, time pressure, dispute importance, managerial power, and the relative power of the disputants). To fulfill this objective, the authors used a five-step method to collect and analyze the data. In step one, alumni from four universities filled out a survey in which they provided a description of the last time they intervened in a corporate conflict, as well as, specifics regarding the nature of the conflict. Of the 142 obtained descriptions, 40 were randomly chosen as step 2 stimulus materials. In step two, ten of the remaining 40 conflict descriptions were randomly selected as target descriptions. One hundred participants, were randomly separated into ten groups, each of which was assigned target description, such that ten participants were given the first target description, ten the second target description and so on. Each group was asked to rank-order the remaining 39 descriptions, in terms of how similar they were to their assigned target description. Participants were also asked to specify the criteria they used for making their similarity judgments. A three-way MDS analysis using SAS’s alternative least squares scaling (ALSCAL, see Takane, Young, & DeLeeuw 1977) implementation of the individual differences scaling model (Carroll & Chang 1970). A new set of allowed experimenters to test the hypothesized relationship between managerial strategy selection (as defined by dimensionality) and the nature of conflict. Two criteria were used to determine the optimum number of dimensions: 1) Kruskal’s (1964) STRESS index and 2) an elbow test accompanied by an examination of the amount of variance accounted for by each dimensional solution. Both procedures suggested that a six-dimensional solution did not significantly improve on the five-dimensional solution, which accounted for 95% of the variance.
248
ROBIN L. PINKLEY, MICHELE J. GELFAND AND LILI DUAN
In step three, one-hundred and forty potential labels were generated from two sources: 1) the labels suggested by the criteria used by step two participants to rank-order the intervention strategy descriptions, and 2) the thirdparty intervention categories used by past scholars such as Thibaut and Walker (1975) and Carnevale (1986). A third group of participants rated the degree to which each of the 40 intervention strategy descriptions reflected each of the 140 potential labels on a 9-point Likert-type scale. Two criteria were used to label the dimensions: 1) multiple-correlations and F-tests revealed that 22 of the 140 potential labels related to the dimensions at the p-value level of .01 or better and 2) multiple regression produced direction cosines (beta weights) to relate each potential label to each dimension. Ten labels most closely related to each of the five dimensions and were thus, used to label the configuration. In step four, the authors used confirmatory analysis to verify the appropriateness of each of the ten labels by asking five trained raters (unaware of the step one – three results) to rate the step one intervention strategy descriptions in terms of each label. Cronbach’s alpha was used to assess interrater reliability (found to be quite high at .85). The F values and beta weights found that all ten of the selected labels loaded onto the five-dimensional solution with reliability ranging from .77 to .97. As a result, the five dimensions were labeled: 1) Attention to the stated versus underlying problem, 2) Disputant commitment forced versus encouraged, 3) Manager decision control versus disputant decision control 4) Manager approaches conflict versus avoids conflict, and 5) Dispute handled publicly versus privately. Finally, step five used multiple regression to relate the five-dimensional solution to the intervenor goals and perceptions of conflict specified by the step one participant surveys. This step allowed the authors to evaluate when and under what circumstances various intervention strategies are used by managers. Example 2 Gelfand, Nishii, Holcombe, Dyer, Ohbuchi, and Fukuno, M. (2001) used MDS to examine the dimensions that are used to construe conflicts across cultures. The purpose of the study was to discern if there are universal (or etic) dimensions of conflict construal and if there are culture-specific (emic) construals that are consistent with prevailing cultural values and practices. This study involved five steps. In step one, students from the U.S. and Japan were asked to write a description of a conflict that they had experienced in the recent past. Consistent with Pinkley (1990), they were told they could describe any incident they chose, regardless of its nature, the type of relationship, or the
THE USE OF MULTIDIMENSIONAL SCALING METHODS
249
degree of severity of the dispute. Participants were asked to describe the conflict situation in terms of the following two questions: 1) Briefly, what is the conflict really about? and 2) What is at the heart of the conflict? In both countries, instructions were given in the native language, English or Japanese. All materials were first translated into Japanese, and then back-translated by another translator into English to check for discrepancies. In step two, 28 conflicts in the U.S. and 28 conflicts in Japan were randomly selected for the MDS portion of the study. Selected episodes had to be 1) brief (2–3 sentences); 2) clear and unambiguous as to the exact nature of the conflict; and 3) relevant in both cultural contexts. Japanese conflict episodes were translated into English (for U.S. participants), and U.S. conflict episodes were translated into Japanese (for Japanese participants) and were then backtranslated by different translators. In step three, a new set of respondents from the U.S. (N = 94) and Japan (N = 130) were given a set of 28 index cards, each of which contained a description of a conflict situation. Participants in both countries were randomly assigned to sort either the U.S. conflict episodes or the Japanese conflict episodes, and were unaware of the source of the conflict episodes (i.e., Japan or U.S.). Participants were asked to sort the conflict cards into as many piles as they desired, based on their perceived similarity. This design resulted in four MDS spaces: 1) American cognitive representations of U.S. conflicts; 2) Japanese cognitive representations of U.S. conflicts; 3) American cognitive representations of Japanese conflicts; and 4) Japanese cognitive representations of Japanese conflicts. This design enabled Gelfand et al. (2001) to use stimuli that were derived naturally in each culture (in the spirit of an emic approach) yet also allow for cross-cultural comparisons of cognitive construals of identical conflict episodes (in the spirit of an etic approach). It also enabled the identification of strong universals of conflict construal (i.e., dimensions of construal that are found regardless of the source of conflict and the cultural background of the participants). In step four, participants rated the conflicts on a number of unidimensional items to assist in the labeling of the dimensions. These items were derived from previous studies of conflict construal conducted in the U.S. (i.e., Pinkley 1990) as well as from literature on conflict in Japan and the U.S. Due to time restrictions and the cognitive load of the ratings (28 conflicts × 21 ratings would require 588 ratings), participants were randomly assigned to rate either the first 14 conflicts or the last 14 conflicts on the 21 unidimensional items. In step five, a conflict episode by conflict episode (28 × 28) diagonal matrix of dissimilarities was created for each set of the U.S. and Japanese participants, resulting in four upper triangular matrices. KYST 2-A statistical
250
ROBIN L. PINKLEY, MICHELE J. GELFAND AND LILI DUAN
program was used to analyze the matrices. An elbow test of Kruskals’ measure of stress suggested that stress values did not decrease substantially from the three to four dimensional solutions, yet did decrease substantially from the two to three dimensional solutions for all of the MDS spaces. The three dimensional solution was also chosen because it allowed for the most comprehensible interpretations for all of the MDS solutions. In step six, the dimensions were labeled based on a) an examination of the conflicts in the MDS spaces as well as b) multiple regression analyses which examined how well the location of each conflict on these unidimensional items was predicted by its location in the multidimensional space. Items with a significant multiple correlation and significant Beta weights indicate that the configuration “explains” the item well. Items that load on multiple dimensions, however, are not as useful for labeling the dimensions. The results demonstrated that Japanese and American participants construed both U.S. and Japanese conflicts through a Compromise versus Win frame (Pinkley 1990), providing evidence of a universal dimension of conflict construal. The results also illustrated that Japanese perceived both sets of conflicts to be more Compromise focused, as compared to Americans. In addition, there were unique dimensions of conflict construal among Americans and Japanese (e.g., Infringements to Self and Giri Violations, respectively), suggesting that identical conflict episodes can be perceived differently across cultures. Example 3 Druckman, Martin, Nan, & Yagcioglu (1999) used MDS to examine the structure of actual cases of international negotiation. The objective of the study was to test whether Iklé’s (1964) typology of international negotiation could account for similarities among the actual negotiations. This typology distinguished among five objectives in international negotiations, including extension, normalization, redistribution, innovation, and side effects, which were construed as distinct types of negotiations with particular processes and outcomes. Although the taxonomy had been widely discussed in the literature, Druckman et al. (1999) set forth to directly test the validity of Iklé’s notions using MDS. This study involved six steps. In step one, 30 cases were randomly selected from the Pew Case Studies in International Affairs (approximately 17% of the cases available). Cases that were sampled differed along numerous characteristics, including region and type of issue (e.g., economic cases, security issues, environmental, hostage negotiations). All cases were approximately ten to fifteen pages and of a common format, consisting of background information, a discussion of the unfolding of the negotiation, and an analysis of the
THE USE OF MULTIDIMENSIONAL SCALING METHODS
251
processes and the outcomes of the negotiation. After selecting the cases, the authors categorized the conflicts in terms of Iklé’s five objectives. Final decisions were reached by a consensus between at least two people (the first author and another student familiar with Iklé’s theory). The authors found that the cases that were chosen were representative of the five objectives identified by Iklé’s theory (eight innovation cases, five normalization cases, ten redistribution cases, five extension cases, and two side effect cases). In step two, the 30 cases were divided randomly into approximately four equal sets and who were assigned to one of four coders were unfamiliar with the taxonomy being examined. The cases were coded for sixteen categories, including characteristics of the parties (e.g., number, power, length and type of relationship), number and type of issues, the process of negotiation (e.g., length and types of exchanges), negotiation outcomes (e.g., type of agreement) and conditions surrounding the negotiation (e.g., time pressure and media coverage). A smaller sample of cases was subject to inter-rater reliability. There were high levels of agreement (generally over 90%) and disagreements were resolved through a refinement of the definitions of variables. In step three, correlations were computed among the 30 cases across the 16 coded variables. These correlations were used as an indication of similarity among the cases. Subsequently, a 30 × 30 matrix of correlations was subject to MDS analysis. Based on stress values as well as interpretability, the authors chose a two-dimension solution. In step four, the authors examined whether the negotiation cases clustered according to Iklé’s taxonomy. First, they visually inspected the clustering of the cases and found initial support for the notion that negotiations cluster according to their focus on innovation, redistribution, extension, side effects, and normalization. They also found a new negotiation category, labeled “multilateral regimes,” which they noted had not existed at the time of the publication of Ilké’s book. Next, they performed a Kruskal-Wallis ANOVA to test for significant differences among the clusters, and also performed a K-means cluster analysis. This analysis revealed three clusters (multilateral and normalization cases), innovation cases, and redistribution cases. In step five, the authors checked to see whether their initial categorization of the cases could be verified. In order to do so, they performed discriminant analyses that examined whether the sixteen coded variables could distinguish among Iklé’s categories. These analyses illustrated that a high percentage of cases were classified accurately. Finally, in step six, the authors created profiles of the clusters of types of negotiations along the sixteen variables, providing a parsimonious understanding of the processes that characterize different types of international
252
ROBIN L. PINKLEY, MICHELE J. GELFAND AND LILI DUAN
negotiations. This analysis showed, for example, that normalization negotiations generally consisted of a highly visible negotiation process, and frequent breakdowns that led to impasses or compromise outcomes. Example 4 McLaughlin, Carnevale, and Lim (1991) used MDS to identify the dimensions underlying professional mediators’ categorizations of mediation tactics. The authors had four phases to achieve their research purpose: (a) data collecting, (b) multidimensional scaling analyses, (c) regression analyses, and (d) cluster analyses. In step 1, surveys were mailed to 230 mediators sampled randomly from a list of members of the Society of Professionals Involved in Dispute Resolution (SPIDR). The mediators were asked to sort 36 stimulus tactics, taken from Carnevale and Pegnetter’s (1985), into as many mutually exclusive categories as they wanted. After the sorting task, the mediators rated each tactic on five bipolar scales: friendly-unfriendly, assertive-passive, controllinguncontrolling, use frequently-use infrequently, and effective-ineffective. All materials were mailed back to the researchers. In the next step, researchers created a tactics × tactics (36 × 36) diagonal matrix of similarities. The similarity of each pair of tactics was presented by the number of times across mediators that both tactics were included in the same category. The tactics × tactics diagonal matrix was used in MDS analyses and clustering analyses. A two-way nonmetric MDS program, KYST2A (Kruskal, Young, & Seery 1977) scaled the similarities data. Kruskal’s stress values were obtained for the one- through six-dimensional solution and the elbow criterion was used to determine the number of dimensions. Because stress values dropped greatly from the two- to three-dimensional solution and decreased minimally beyond the three-dimensional solution, the threedimensional solution appeared to be closest to the true structure of mediation tactics. In the third step, unidimensional scales were regressed onto the MDS configuration to label the dimensions. Those unidimensional scales, with (a) relatively large squared multiple correlations and (b) a large weight on one of the three dimensions but not on the other two, proved most useful for interpreting the dimensions. Based on these analyses, researchers labeled the three dimensions substantive versus reflexive, affective versus cognitive, and forcing versus facilitating. In the last step, the tactics similarities matrix was clustered hierarchically with the CLUSTER procedure in SPSS. The results of cluster analyses were compared with MDS results. In addition, the authors found that the cluster
THE USE OF MULTIDIMENSIONAL SCALING METHODS
253
and MDS results were compatible with the results of factor analyses of mediation tactics.
Conclusion Multidimensional scaling is a powerful method to uncover the hidden structures that underlie peoples’ judgments about themselves and their environments. In the domain of conflict and negotiation, MDS enables scholars to understand phenomena at multiple levels of analysis – from the individual, to the group, to the international level. It enables researchers to inductively study many issues in conflict and negotiation, such as modeling disputant perceptions of conflicts, characterizing the types of negotiation and mediation tactics that are used to resolve social conflict, and organizing the features or characteristics of conflicts and negotiations, among other issues. It also holds much promise to illuminate how individual differences – such as personality, education, and gender, as well as organizational or cultural differences – affect conflict and disputing. At the same time, compared to its cousin, factor analysis, MDS has been highly underutilized in the field. In this article, we hope we have begun to show the unique benefits that this technique can bring to the science and practice of conflict and negotiation.
References Beardsworth, A., & Keil, T. (1992). “The vegetarian option: Varieties, conversions, motives and careers.” Sociological Review 40: 253–293. Bernard, H.R. (2000). Social Research Methods: Qualitative and Quantitative Approaches. Thousand Oaks, California: Sage Publications, Inc. Black, J.K. (1991). Development in theory and practice: Bridging the gap. Boulder, Colorodo: Westview. Busing, F.M.T.A., Commandeur, J.J.F., & Heiser, W.J. (1997). PROXSCAL: A multidimensional scaling program for individual differences scaling with constraints,” in W. Bandilla & F. Faulbaum, editors, Softstat ’97: Advances in Statistical Software, Stuttgart: Lucius & Lucias, 6: 67–74. Carnevale, P.J.D., & Pegnetter, R. (1985). “The selection of mediator tactics in public sector disputes: A contingency analysis.” Journal of Social Issues 41, 2: 65–81. Carroll, J.D., & Chang, J.J. (1970). A quasi-nonmetric version of INDSCAL, a procedure for individual differences in multidimensional scaling. Paper presented at meetings of the Psychometric Society, Stanford, CA. Druckman, D. (1997). “Dimensions of international negotiations: Structures, processes, and outcomes.” Group Decision & Negotiation 6, 5: 395–420. Druckman, D., Martin, J., Nanand, S.A., & Yagcioglu, D. (1999). “Dimensions of international negotiation: A test of Iklé’s typology.” Group Decision & Negotiation 8: 89–108.
254
ROBIN L. PINKLEY, MICHELE J. GELFAND AND LILI DUAN
Gelfand, M., Nishii, L., Holcombe, K., Dyer, N., Ohbuchi, K., & Fukuno, M. (2001). “Cultural Influences on Cognitive Representations of Conflict: Interpretations of Conflict Episodes in the U.S. and Japan.” Journal of Applied Psychology 86, 6: 1059–1074. Gelfand, M., Triandis, H., & Chan, D. (1996). “Individualism versus collectivism or versus authoritarianism?” European Journal of Social Psychology 26, 3: 397–410. Green, P.E., & Carmone, F.J. (1970). Multidimensional Scaling and Related Techniques in Marketing Analysis. Allyn & Bacon, Inc., Boston, Massachusetts. Hair, J.F., Anderson, R.E., Tatham, R.L., & Black, W.C. (1998). Multivariate Data Analysis, fifth edition. Prentice-Hall, Inc., Upper Saddle River, New Jersey. Hansen, J.C., Sarma, Z.M., & Collins, R.C. (1999). “An evaluation of Holland’s model of vocational interests for Chicana(o) and Latina(o) college students.” Measurement and Evaluation in Counseling & Development 32: 2–14. Hirschberg, N., Jones, L., & Haggerty, M. (1978). “What’s in a face: Individual differences in face perception?” Journal of Research in Personality 12, 4: 488–499. Jehn, K.A. (1994). “A multimethod examination of the benefits and detriments of conflict.” Administrative Science Quarterly 40: 256–282. Johnson, C.S. (1995). “The Rokeach value survey: Underlying structure and multidimensional scaling.” Journal of Psychology 129: 583–597. Jones, L. & Young, F. (1972). “Structure of a social environment: Longitudinal individual differences scaling of an intact group.” Journal of Personality & Social Psychology 24, 1: 108–121. Kruskal, J.B. (1964). “Nonmetric multidimensional scaling: A numerical method.” Psychometrika 29: 115–129. Kruskal, J.B. (1968). How to use M-D-SCAL, a program to do multidimensional scaling and multidimensional unfolding, (Version 4 and 4M of M-D-SCAL, all in FORTRAN IV), Bell Telephone Laboratories, Murray Hill, New Jersey, March 1968 (mimeographed). Kruskal, J.B., and Wish, M. (1978). Multidimensional scaling. Sage University Papers Series. Quantitative Applications in the Social Sciences, no. 07–011. Beverly Hills, California: Sage. Kruskal, J.B., Young, F.W., & Seery, J.B. (1977). How to use KYST2A: A very flexible program to do multidimensional scaling and unfolding. Murray Hill, NJ: Bell Telephone Laboratories. Liske, J. (1993). “Regional subcultures of the United States.” Journal of Politics 55: 888–913. McLaughlin, M., Carnevale, P., & Lim, R. (1991). “Professional mediators’ judgments of mediation tactics: Multidimensional scaling and cluster analyses.” Journal of Applied Psychology 76, 3: 465–472. Pinkley, R.L. (1990). “Dimensions of conflict frame: Disputant interpretations of conflict.” Journal of Applied Psychology 75, 2: 117–126. Pinkley, R.L., Brittain, J., Neale, M.A. & Northcraft, G.B. (1995). “Managerial third party dispute intervention: The impact of situational factors on strategic choice.” Journal of Applied Psychology 80: 363–402. Robinson, S.L. & Bennett, R.J. (1995). “A typology of deviant workplace behaviors: a multidimensional scaling study.” Academy of Management Journal 38, 2: 555–572. Takane, Y., Young, F.W. & De Leeuw, J. (1977). “Nonmetric individual differences in multidimensional scaling: An alternating least squares method with optimal scaling features.” Psychometrika 42: 7–67. Thibaut, J., & Walker, L. (1975). Procedural justice: A psychological analysis. Hillsdale, New Jersey: Erlbaum. Wish, M., Deutsch, M., & Biener, L. (1970). “Differences in conceptual structures of nations: an exploratory study.” Journal of Personality & Social Psychology 16, 3: 361–373.
THE USE OF MULTIDIMENSIONAL SCALING METHODS
255
Young, F.W. and Hamer, R.M. (1987). Multidimensional Scaling: History, Theory and Application. Hillsday, New Jersey: Lawrence Erlbaum Associates. Young, F.W. and Torgerson, W.S. (1968). A FORTRAN IV program for nonmetric multidimensional scaling on the IBM 360/75 computer (obtained from Dr. Young at the University of North Carolina).
Markov Chain Models of Communication Processes in Negotiation PHILIP L. SMITH, MARA OLEKALNS and LAURIE R. WEINGART
In trying to predict the outcomes of two-party negotiations, researchers have focused mainly on how negotiators’ ability to identify mutually beneficial outcomes is affected by antecedent conditions. However, an understanding of the impact of antecedent conditions does not tell us how they exert their influence. To understand this, we turn to analyses of negotiation processes, which provide the link between antecedent conditions and outcomes. Such analyses enable us to assess not only how such variables shape the negotiation process, but also how that process promotes or inhibits the attainment of high quality outcomes. A process analysis also deepens our understanding of the emergent properties of negotiation. Because each negotiator reacts to the other party on a momentto-moment basis, we need to understand how strategies and tactics are sequenced. In our research, we have used Markov chain analysis to try to capture this dynamic aspect of negotiation. Not only does this kind of analysis allow us to capture patterns of action and reaction within a negotiating dyad, it also allows us to build more complex models that link these patterns to both antecedent conditions (e.g., power, social value orientation) and consequences (type of outcome). The kinds of questions we might seek to answer with a Markov chain analysis include: Is there a propensity for integrative tactics to be reciprocated? Is the reciprocation of integrative tactics related to the quality of the negotiated outcome? Does the relationship between outcomes and the reciprocation of integrative tactics hold, irrespective of the social value orientation of the dyad, or do dyads of different kinds achieve high-quality outcomes in different ways? These questions, although differing in complexity, all involve a question about the communication processes within the negotiating dyad, namely, how do the tactics used by the members of a dyad during the course of a negotiation depend on one another? To answer these questions using a Markov chain analysis involves several steps. First, we need to determine the number of strategies or tactics that will be included in the analysis (see Weingart, Olekalns & Smith 2004). Then we International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 257–272 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
258
PHILIP L. SMITH, MARA OLEKALNS AND LAURIE R. WEINGART
construct a contingency table that represents the dependencies among strategies in sequences of a particular length. Subsequently, we must determine the length of the strategy sequences that best captures the communication process in the empirical data. Finally, we must assess which strategy sequences contribute to the overall model fit as a way of identifying important relationships. In the following sections, we introduce the technical details of undertaking an analysis of this kind.
Describing Strategy Sequences Markov chain analysis assumes that the set of tactics used in a negotiation can be classified into a finite number of discrete tactics, or states, and that the sequencing of tactics, which is represented as a transition between tactics, can be described by a simple probability model. To carry out a Markov chain analysis, we think of the tactic used by a particular member of a dyad at speaking turn n as a random variable, denoted Xn. By this we mean that at any given point in time, negotiators may choose between one of several tactics, each with some probability. Which tactic they choose denotes the state of the dyad at that time. The appendix shows an excerpt from a transcript in which we have coded each strategy as either integrative or distributive. The sequential dependencies among these 12 strategies can be represented schematically as follows: integrative1(Party A) → distributive2(Party B) → distributive3(Party A) . . . . . . → integrative10(Party B) → distributive11(Party A) → distributive12(Party B) Put more formally, we can say that Xn is a dichotomous variable that takes the values “integrative” or “distributive,” each with some probability. We can therefore think of a negotiation consisting of N speaking turns as a sequence of state transitions of the form: X1(a) → X2(b) → X3(a) → Xn-1(a) → XN(b)
(1)
In this representation, the random variables are subscripted with the speaking turn and are superscripted a or b to show that they represent communication between members of the dyad (Parties a and b) in alternation. The arrows indicate a hypothesized casual, and hence statistical, dependency, between consecutive values of Xn. Within this framework, we may ask: What is the probability that a particular tactic is used at a given speaking turn, given the entire history of the process up until that point? In terms of our example, this is like asking: What
MARKOV CHAIN MODELS OF COMMUNICATION PROCESSES IN NEGOTIATION
259
is the likelihood that a given negotiator will choose an integrative tactic, given the entire preceding sequence of integrative and distributive tactics used by the two negotiators? Expressed in terms of random variables, the question becomes: What is P[Xn | Xn-1 . . . X2, X1, the conditional probability that Xn takes a given value, given the sequence of random variables, X1, X2, . . . Xn-1? The Markov assumption states that the probability of Xn depends only on a fixed number of the set of preceding X values. Specifically, the process is called a first-order Markov chain if P[Xn | Xn-1 . . . X2, X1] = P[Xn | Xn-1],
(2)
that is, if the probability of Xn depends only on the value of Xn-1, the state at the immediately preceding time. In terms of the negotiation process, this asserts that the tactic used by a negotiator at a given time depends only on the tactic used by the other party at the immediately preceding time step. The process is called a second-order Markov chain if P[Xn | Xn-1 . . . X2, X1] = P[Xn | Xn-1, Xn-2],
(3)
This equation asserts that Xn depends jointly on Xn-1 and Xn-2, the states at the two preceding time steps. In terms of the negotiation process, it asserts that the tactic used by a negotiator at a given time depends both on the tactic previously used by the other party and on the negotiator’s own previous tactic. In our work we have found that second-order chains are usually sufficient to describe the structure of empirically-observed negotiations (Olekalns & Smith, 1999, 2000, 2001, 2003; Weingart et al., 1999). In Markov chain models, the statistical dependency between consecutive states is expressed as a transition matrix. As we will discuss, in our analyses we determine the properties of the transition matrix by evaluating the statistical dependencies among frequencies in a contingency table. The levels of the variables that form the margins of the table represent the coded strategies. The number of dimensions in the table reflects the lengths of the sequences we wish to analyze. In the simplest analysis, we might choose to code behavior as either integrative or distributive and to focus on a first-order Markov chain (sequences of length 2). A transition matrix for such first-order, two-state chain might have the form: Time n 0 1 Time n-1
[ ]
0 .6 .4 . 1 .3 .7
(4)
This matrix summarizes the probability of finding the dyad in a particular state, given the preceding state or states. If we identify the states 0 and 1 with
260
PHILIP L. SMITH, MARA OLEKALNS AND LAURIE R. WEINGART
integrative and distributive tactics, respectively, this matrix asserts that the probability of reciprocating an integrative tactic at time n-1 (the row state) with an integrative tactic at time n (the column state) is .6. The probability of reciprocating a distributive tactic is .7. This transition matrix is equivalent to the pair of probability statements: P[X_Xn = 0|Xn-1 = 0] = .6 P[X_Xn = 0|Xn-1 = 1] = .3.
(5)
Because the probabilities in each row of the transition matrix must sum to 1.0, these two equations suffice to determine the transition matrix uniquely. The requirement that the rows sum to 1.0 expresses the fact that, at each time step, the chain must be in exactly one of a finite set of states.
Loglinear Analysis of Markov Chain Models As discussed by Agresti (1990) and Bishop, Fienberg and Holland (1975), Markov chain models may be analyzed using the same loglinear modeling techniques that are used to analyze multi-way contingency tables. This approach is an attractive one, because of the widespread availability of loglinear modeling software. Unlike most psychological applications of loglinear models, however, the unit of analysis in Markov chain models is the tactic, or the speech act, rather than the individual. In loglinear models, statistical dependencies among variables are represented as interactions. An association between two variables is a two-way interaction; an association among three variables is a three-way interaction, and so on. In loglinear analyses of Markov chains, the order of the chain (i.e., whether it is first-order or second-order, etc.) may be assessed by determining the highest order interaction needed to describe the dependencies in the sequence of coded tactics. To do this, the sequence of tactics must be put into contingency table form. The variables that form the margins of the contingency table are the tactics at consecutive time steps. The number of dimensions in the table depends on the hypothesized order of the underlying chain. In general, a Markov chain of order m has dependencies among the tactics at m + 1 time steps. The number of dimensions of the table must therefore equal the hypothesized order of the chain plus one. Thus a first-order chain requires a two-dimensional table; a second-order chain requires a three-dimensional table, and so on. Because the order of the chain is an empirical question, it is necessary to test for the nonsignificance of interaction terms that represent chains of higher order. Thus to conclude that a given set of data
MARKOV CHAIN MODELS OF COMMUNICATION PROCESSES IN NEGOTIATION
261
comes from a second-order chain, it is necessary to test for the nonsignificance of the interaction terms associated with a third-order chain. As thirdorder chains have dependencies among tactics at four consecutive time steps these tests, which involve four-way interactions, must be performed on a fourdimensional table. To test that a chain is at least second order, we form a contingency table which counts the number of different tactic sequences of length three. For the sequence in (1), the contingency table is formed from consecutive threeelement sequences of the form: X1(a) → X2(b) → X3(a), → X2(b) → X3(a) → X4(b), → X3(a) → X4(b) → X5(a), etc. For a dichotomous classification of states like that used by Weingart et al. (1995) there are eight different kinds of three-element sequences (see Table 1); for the seven-state classification used by Olekalns and Smith (2003) there are 343. In the case of the excerpt given in the Appendix, if we again let 0 denote integrative tactics and 1 denote distributive tactics, the first sequence, X1(a) → X2(b) → X3(a) would be coded as 011; the second would be coded as 111, and so on. The excerpt as a whole yields a total of 10 three-element sequences: 011, 111, 110, 101, 010, 100, 000, 000, 001, 011. The contingency table formed from this excerpt would contain one pure distributive sequence (111), two pure integrative sequences (000), and a number of different mixed sequences. It may be shown that the probabilities of consecutive (m + 1) – element sequences in a Markov chain of order m are conditionally independent of one another, given the value of their common segment of length m (Bartlett, 1951; Hoel, 1954; see also the appendix of Olekalns & Smith, 1997). This means Table 1. Sequence Frequencies for Simulated Markov Chains First-Order Data X1 0 0 0 0 1 1 1 1
X2 0 0 1 1 0 0 1 1
X3 0 1 0 1 0 1 0 1
Second-Order Data N 454 300 206 285 300 191 284 478
X1 0 0 0 0 1 1 1 1
X2 0 0 1 1 0 0 1 1
X3 0 1 0 1 0 1 0 1
N 332 176 143 332 177 288 322 738
Note: The variables X 1, X 2, X 3 code the first, second, and third element of each threeelement sequence. The variable N is the total number of three-element sequences of each type. Both simulations were based on 2500 observations.
262
PHILIP L. SMITH, MARA OLEKALNS AND LAURIE R. WEINGART
that a test that a chain is (at least) second order can be carried out by testing the appropriate three-way interaction terms in a contingency table in which all two-way interaction terms are present. To demonstrate how we fit first- and second-order Markov chains, we generated two data sets (shown in Table 1) using Monte Carlo simulations of two chains, each representing different underlying processes. The first data set (left side of Table 1) was based on the assumption that the underlying communication structure can be represented as a first-order chain. This chain was generated using the probability model of Equation 5 (i.e., the transition matrix (4)). The second data set (right side of Table 1) was based on the assumption that the underlying communication structure can be represented as a secondorder chain. This chain was generated using the probability model below: P[Xn P[Xn P[Xn P[Xn
= = = =
0|Xn-1 0|Xn-1 0|Xn-1 0|Xn-1
= = = =
0, 0, 1, 1,
Xn-2 Xn-2 Xn-2 Xn-2
= = = =
0] 1] 0] 1]
= = = =
.7 .5 .4 .2
(6)
Note that the first-order transition probabilities for this chain, averaged over values of Xn-2, are .6 and .3, which are the same as those for the first-order chain. However, the statistical structure of the two chains is very different. Table 2 shows the result of fitting Markov chain models to these simulated data. These models are specified using the notation of Fienberg (1980; see also Agresti, 1990). In Fienberg’s notation, a loglinear model is specified as a set of bracketed model terms. This notation provides a succinct way of indicating which variables in a model are independent of one another and which variables interact. Variables that interact are grouped together inside a set of brackets; variables that are independent of one another are placed in separate brackets. Thus, for example, the models [X1][X2][X3] and [X1, X2][X2, X3] are two possible representations of the structure of a three-dimensional contingency table with margins X1, X2, and X3. The first model states that the three variables that form the margins of the table are independent of one another. It may be thought of as a generalization of the usual chi-square test for independence in two-way tables. Psychologically, it asserts that there are no statistical dependencies in the tactics used in consecutive speaking turns in the negotiation. The second model, [X1, X2][X2, X3] is a first-order Markov chain. It states that the only statistical dependencies in the table are the one-step transitions between consecutive speaking turns: Xn-1 → Xn. Both of these models can be tested within a three-way table because neither of them is saturated. That is, the table possesses sufficient residual degrees of freedom to allow failures of
MARKOV CHAIN MODELS OF COMMUNICATION PROCESSES IN NEGOTIATION
263
Table 2. Markov Models For Simulated Data Model
G2
df
DG2
p
Ddf
p
First-order data, 3-way table (independence and first-order models) [X1][ X2][ X3] [X1, X2][X2, X3]
233.77
4
0
2.84
2
.24
–
–
–
230.93
2
0
–
–
–
–
–
5
0
–
–
–
Second-order data, 3-way table (first-order model) [X1, X2][X2, X3]
50.03
2
0
Second-order data, 4-way table (first- and second-order models) [X1, X2][X2, X3][X3, X4]
92.63
8
0
[X1, X2, X3] [X2, X3, X4]
1.41
3
.70
– 91.22
First-order data with two subpopulations, 3-way table (first-order models with subpopulation dependencies) [M] [X1, X2][X2, X3] [M, X1] [M, X2][M, X3] [X1, X2][X2, X3] [M, X1] [M, X2] [M, X3] [X1, X2][X2, X3] [M, X1, X2] [M, X2, X3]
426.74
9
0
11.64
6
.07
415.10
3
0
0.91
4
.92
10.73
2
.005
Note: G2 is the likelihood-ratio chi-square statistic for each model. It tests the null hypothesis that the data in the table are well described by the indicated model. DG2 is the sequential likelihood ratio test statistic. It tests whether the indicated model provides a significantly better description of the data than the model that precedes it.
fit to be detected. In the case of the first-order Markov model, the table has two residual degrees of freedom, one associated with the two-way interaction X1.X3 and one associated with the three-way interaction X1.X2.X3. Failure of a first-order model to fit implies that the chain is at least of second order. A direct test of the hypothesis that the chain is second order can be carried out in a four-dimensional table. To show this we need to show simultaneously that the pair of three-way interactions, X1.X2.X3 and X2.X3.X4 that represent the second-order effects in such a table are significant, and the four-way interaction X1.X2.X3.X4 that represents the third-order effect is not. Loglinear models like those described above may be fitted in a generalized linear modeling framework using a logarithmic link function and a Poisson
264
PHILIP L. SMITH, MARA OLEKALNS AND LAURIE R. WEINGART
distribution error. Generalized linear models generalize the classic linear (i.e., regression) model by permitting the error distribution to be non-normal and by specifying a relationship between the linear predictor and error distribution in a more general form than that of classical regression models (see McCullagh & Nelder, 1989; Venables & Ripley, 2002). The first part of Table 2 shows a fit of independence and first-order Markov models to the first-order data from Table 1. As can be seen, the data are best fitted by a first-order order. The values of G2 demonstrate that the independence model, [X1][X2][X3], fails to capture the structure of the data, whereas the first-order Markov model, [X1, X2][X2, X3], captures it accurately. Table 2 also shows conditional likelihood fit statistics DG2. This statistic tests the improvement in fit (i.e., the change in G2 obtained by progressing from a simple to a more complex model. In this case, it tests the improvement produced by moving from an independence model to a model with first-order sequential dependencies. The value of DG2 for the first-order Markov model in the top part of the table shows that it provides a significantly better description of the data than does the independence model. The second part of Table 2 shows a fit of a first-order Markov model to the second-order data from Table 1. The failure of a first-order model to fit these data implies that the underlying chain is at least of second order. This is tested directly in the third part of Table 2 which reports fits of the secondorder chain reclassified into a four-dimensional contingency table (i.e., into sequences of length four). The first-order Markov model, which in this table is [X1, X2][X2, X3] [X3, X4], again fails to fit. In contrast, the second-order model [X1, X2, X3] [X2, X3, X4] accurately describes these data, in agreement with the known statistical properties of the chain. This is demonstrated both by the nonsignificance of the residual G2 statistic and the notable improvement in fit over the first-order model, as indicated by the value of DG2.
Markov Models with Subpopulations An important use of Markov chain models is to compare how dyads in different subpopulations sequence their tactics during the course of a negotiation. These subpopulations may be defined prospectively, at the beginning of the negotiation, by the attributes of the negotiators or by the experimental conditions to which they are assigned. Structural power (high/low), motivational orientation (proself/prosocial), and tactical knowledge (present/absent) are examples of classification variables of this kind (Olekalns & Smith, 1999, 2001, 2003; Weingart et al., 1999). Alternatively, they may be defined retro-
MARKOV CHAIN MODELS OF COMMUNICATION PROCESSES IN NEGOTIATION
265
Table 3. First-Order Data with Two Subpopulations M
X1
X2
X3
N
0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1
0 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1
0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
302 148 83 141 149 76 142 207 83 81 27 134 81 81 135 626
Note: The variables X 1, X 2, X 3 code the first, second, and third element of each threeelement sequence. The variable M codes the subpopulations. The variable N is the total number of three-element sequences of each type. The simulated chain for each population was based on 1250 observations.
spectively, at the end of the negotiation by the kind of outcome that negotiators achieve. For example, Olekalns and Smith (2000, 2003) have considered models in which outcomes from simulated negotiations are classified as distributive, suboptimal, optimal, or impasse, according to where the joint outcome for the dyad fell in relation to the Pareto-optimum boundary for the task. In experimental settings, the prospective and retrospective variables are often the independent and dependent variables of the design and, consequently, are viewed as qualitatively different from one another. However, in a loglinear analysis framework, they are treated in an identical way. Specifically, hypotheses about any of the possible relationships among antecedent conditions, communication behavior, or outcomes are tested via interaction terms in a loglinear model. Table 3 shows data from two simulated, first-order, Markov chains from subpopulations with different transition matrices. The transition matrix for the first subpopulation M = 0 was 0
1
[ ]
0 .7 .3 . 1 .4 .6
(7)
266
PHILIP L. SMITH, MARA OLEKALNS AND LAURIE R. WEINGART
The transition matrix for the second subpopulation M = 1 was 0
1
[ ]
0 .5 .5 . 1 .2 .8
(8)
To give this example content, we can think of the two subpopulations as groups of negotiators with prosocial M = 0 and proself M = 1 motivational orientations, respectively. Note that the average of the transition matrices (7) and (8) is the transition matrix (4). This latter matrix may be thought of as the population transition matrix that would be obtained if differences among subpopulations were neglected. The transition matrices (7) and (8) suggest that, in comparison to the matrix (4), prosocial negotiators are more likely to reciprocate integrative tactics (.7) and less likely to reciprocate distributive tactics (.6), whereas proself negotiators tend to do the converse (.5 and .8, respectively). The bottom part of Table 2 shows the fits of three first-order Markov models with subpopulation effects to these data. The simplest of these is the model [M][X1, X2][X2, X3]. This is a model in which the transition probabilities are independent of the motivational orientation of the negotiators. The inclusion of the term M ensures the frequencies predicted by the model at each level of the motivational orientation variable are equal to those in the data. The test of this model term may thus be interpreted as a test of the hypothesis that negotiations involving prosocial and proself negotiators are, on average, of equal length. In general, all models involving subpopulations will include a term that represents the subpopulation main effect to equate the observed and predicted sample sizes for each subpopulation. The remaining models in Table 2 are subpopulation models in which tactic frequencies and sequence frequencies depend on motivational orientation. In the model [M, X1][M, X2][M, X3] [X1, X2][X2, X3] only tactic frequencies depend on motivational orientation. In the model [M, X1, X2][M, X2, X3] both tactic and sequence frequencies depend on motivational orientation. Table 2 shows that the addition of both kinds of dependency produces an improvement in fit, in agreement with the known properties of the data. Like many of the models that occur when modeling interaction sequences, the last two models in Table 2 have redundant model terms, which reflect the symmetrical properties of the underlying contingency table. Thus, for example, the model terms [M, X1], [M, X2] and [M, X3] all represent the same effect, namely, subpopulation dependencies in tactic frequencies, which are reflected in each of three margins of the contingency table. Similar redundancies exist elsewhere in Table 2 in tests of the order of the underlying
MARKOV CHAIN MODELS OF COMMUNICATION PROCESSES IN NEGOTIATION
267
Markov chains. When testing such effects in empirical data, rather than testing overall DG2 values for significance, as we have done in Table 2, it is preferable to base tests of improvement in fit on only one of a set of redundant terms to avoid inflating the Type I error rate. Further discussions of these issues may be found in Olekalns and Smith (1997, 1999, 2000) and Wasserman and Iacobucci (1986).
The Analysis of Residuals Once significant subpopulation effects have been identified in data, the next step is to try to characterize them qualitatively. The most direct way to do this is through the analysis of residuals. There are a number of different kinds of residuals that have been considered in the categorical data analysis literature, but for our purposes the most relevant and easily interpretable is the standardized residual. The standardized residual, ei, for any cell i of the contingency table is defined as ei
ˆ = ni − mi , mˆi
ˆ i is the frequency prewhere ni is the observed frequency of the cell and m dicted by the model. The square of each standardized residual is distributed approximately as a chi-square variable with one degree of freedom. This allows the residuals themselves to be interpreted roughly as z-scores (i.e., the square root of a single degree of freedom chi-square). As a rule of thumb, residuals of the order of ±2.0 can be viewed as making “large” or “significant” contributions to the fit of a model. However, an analysis of residuals should take into account the entire pattern of residuals, not just the large ones. Indeed, a strategy that is often useful, especially with large tables, is to ignore the magnitude of the residuals and to just look at their signs. This allows a researcher to discern systematic patterns of overprediction and underprediction in the model that would not be apparent from an exclusive focus on large residuals. When dealing with sequences of nested models like those considered here, one often has the problem of trying to understand why the addition of particular model terms, like subpopulation dependencies in sequential effects, significantly improves the fit. Under these circumstances, a useful strategy is to analyze the residuals from a model that does not include the term or terms in question. The large residuals from this model identify those cells of the table that make large contributions to the DG2 statistic when the term or terms in question are added to the model.
PHILIP L. SMITH, MARA OLEKALNS AND LAURIE R. WEINGART
268
Table 4 shows observed and predicted frequencies, together with standardized residuals, for the fit of the model [M][X1, X2][X2, X3] to the subpopulation data of Table 3. The observed and predicted values in this table were calculated from the M.X1 . X2 marginal table, by summing over levels of X3. The choice of this marginal table was made arbitrarily, because the same transition structure is also represented in the M.X2 . X3 table. Because the simulations for the two subpopulations were based on identical numbers of observations, the predicted frequencies for the subpopulations in this case are the same, although this will not be true in general. The pattern of residuals in Table 4 clearly reflects the differences between the transition matrices (7) and (8) that were used to generate the data. In particular, the prosocial group (M = 0) shows a large positive residual in the 00 cell and a large negative residual in the 11 cell. The proself group (M = 1) exhibits the opposite pattern. One would conclude from this pattern that prosocial dyads are more likely to reciprocate integrative tactics and less likely to reciprocate distributive tactics than are proself dyads, in agreement with the known structure of the data. Further examples of the analysis of residuals, based on more complex state representations than those considered here, may be found in the articles of Olekalns and Smith (1999, 2000, 2002).
Sample Size In practice, the complexity of the models that can be investigated using the methods described here is limited by the size of the sample. Negotiation experiments typically yield a corpus of a few thousand classifiable responses,
Table 4. Residuals for [M][X1, X2][X2, X3] M
X1
X2
ni
ˆi m
ei
0 0 0 0 1 1 1 1
0 0 1 1 0 0 1 1
0 1 0 1 0 1 0 1
450 224 225 3349 164 161 162 761
307.0 192.5 193.5 555.0 307.0 192.5 193.5 555.0
8.16 2.27 2.26 –8.74 –8.16 –2.27 –2.26 8.74
Note: Model [M][X1, X2][X2, X3] is a first-order model in which the transition matrix is the ˆ i and ei are the observed frequencies, predicted same for both subpopulations. The variables ni, m frequencies, and residuals for each cell of the M.X1.X2 table, summing over levels of X3.
MARKOV CHAIN MODELS OF COMMUNICATION PROCESSES IN NEGOTIATION
269
but with models of even moderate complexity, this may result in sparse contingency tables, in which the number of observations is small relative to the number of cells. Goodness-of-fit statistics like G 2 require samples of a sufficient size to allow them to be tested for significance against a chi-square distribution. A typical recommendation is that not too many of the cells (e.g., not more than 20%) should have an expected frequency of less than 5. If the majority of cells have expected frequencies of between 0.5 and 4, G2 tends to provide too liberal a test, in which reported p values underestimate true ones (Agresti, 1990; pp. 246–250). This may result in overfitting bias, in which models that are more complex than necessary are selected as providing the best description of the data. If the majority of expected frequencies are less than 0.5, G2 tends to provide too conservative a test, and an underfitting bias can result. For this reason, it is usually preferable to base inferences on the conditional likelihood ratio tests, DG2, rather than on the fit of the model to the table as a whole. This is because the adequacy of the chi-square approximation to the overall G2 statistic depends on the sparseness of the full table, whereas that of the conditional likelihood ratio tests depends on the sparseness of the marginal tables on which they are computed. Because marginal tables are obtained by collapsing the full table over one or more dimensions, they are usually much less sparse than the original table and consequently, provide a better basis for calculating fit statistics.
Further Reading Theoretical accounts of Markov chain models can be found in many books on probability theory, two of which include Norris (1998) and Ross (2000). Two standard texts in the categorical data analysis literature that treat Markov chain models are Agresti (1990) and Bishop et al. (1975). Section 7.4 of Bishop et al. is especially relevant to the problem of modeling negotiation tasks like those considered here. The key mathematical results in the statistical analysis of Markov chains include Anderson and Goodman (1957), Bartlett (1951), Billingsley (1961) and Hoel (1954). An account of Markov chain models from a communication theory perspective may be found in Hewes (1979). Applications of Markov methods in a communication setting are described by Hawes and Foley (1973), Thomas (1985), Thomas, Roger and Bull (1983), and Ting-Toomey (1973). An early application to dyadic negotiations is England (1973).
270
PHILIP L. SMITH, MARA OLEKALNS AND LAURIE R. WEINGART
Acknowledgements Preparation of this article was supported by Australian Research Council Discovery Grant DP0342967 and National Science Foundation grant SBR96–1671.
Appendix: Sample Transcript integrative
distributive distributive distributive
integrative
distributive integrative integrative integrative
integrative
distributive
There are four aspects that actually go together. There is the number of years of contract and the vacation, the removal can be separated out from the package. So if, if we keep the package at six thousand for you and we have 60% removal costs, that would be additional. No I’d need more than 60% I was thinking 80% actually. What if we kept it at 60% and upped it to $8000 for the package? I just want to get clear in my head about the vacation. Like I said I’m not prepared to relocate to Canberra unless I get more than the two weeks. Well it depends how many years you’re with us. The shorter the time that you were with us, I would expect the less time you would get in vacation. If you were going to be with us for six years say you’d get three weeks holiday. I would be prepared to go to Canberra for six years only for six weeks vacation. So if we were to do say three weeks vacation, how many years contract would you agree on? Perhaps if we get back to the package. I’d be prepared to go to Canberra for 6 years, if that’s important to you. . . . Well that’s related into vacation, and I know vacation’s important to you because of family commitments. So we could even look at four years and three weeks vacation each year for that time and include the airfares in an $8000 package with 60% removal costs on top of that. I think it’s a good starting point. It obviously needs some work. My contract can be longer than four years; I think that’s negotiable. Canberra if it’s really important to you. The thing with Canberra is that we’ve got three positions there. How open are you to challenge in new areas? Because we recognize that your strategic skills are applicable across any broad range and you have demonstrated very high levels of achieve-
MARKOV CHAIN MODELS OF COMMUNICATION PROCESSES IN NEGOTIATION
271
ment in the strategic areas. We’re not looking at the strategic abilities in HR, we’re looking at your strategic abilities as a separate set of skills and you’ve already demonstrated those abilities in HR to a very high degree, which is why you are sitting here having this conversation with me. We’d like to take those skills and apply them to a new area.
References Agresti, Alan (1990). Categorical data analysis. New York: Wiley. Anderson, Theodore W., & Goodman, Leo A. (1957). “Statistical inference about Markov chains.” Annals of Mathematical Statistics 28: 89–109. Bartlett, Maurice S. (1951). “The frequency goodness of fit test for Markov chains.” Proceedings of the Cambridge Philosophical Society 47: 86–95. Billingsley, Patrick (1961). “Statistical methods in Markov chains.” Annals of Mathematical Statistics 32: 12–40. Bishop, Yvonne M., Fienberg, Stephen E., & Holland, Paul W. (1975). Discrete multivariate analysis: Theory and practice. Cambridge, MA: MIT Press. Burnham, Kenneth P., & Anderson, David R. (1998). Model selection and inference. New York: Springer. De Dreu, Carsten K.W., & Van Lange, Paul A.M. (1995). “The impact of social value orientations on negotiator cognition and behavior.” Personality and Social Psychology Bulletin 21: 1178–1188. De Dreu, Carsten K.W., Weingart, Laurie. R., & Kwon, Soon (2000). “Influence of social motives on integrative negotiations: A meta-analytic review and test of two theories.” Journal of Personality and Social Psychology 78: 889–905. England, John L. (1973). “Mathematical models of two-party negotiations.” Behavioral Science 18: 189–197. Fienberg, Stephen E. (1980). The analysis of cross-classified, categorical data (2nd ed.) Cambridge, MA: The MIT Press. Hawes, Leonard C., & Foley, Joseph R. (1973). “A Markov analysis of interview communication.” Speech Monographs 40: 210–219. Hewes, Dean E. (1979). “The sequential analysis of social interaction.” The Quarterly Journal of Speech 65: 56–73. Hoel, Paul G. (1954). “A test for Markov chains.” Biometrika 41: 430–433. McCullagh, Peter, & Nelder, John A. (1989). Generalized linear models (2nd ed.) London, UK: Chapman and Hall. Neale, Margaret A., & Bazerman, Max H. (1991). Cognition and rationality in negotiation. New York: Free Press. Norris, James R. (1998). Markov chains. New York: Cambridge University Press. O’Connor, Kathleen (1997). “Motives and cognition in negotiation: A theoretical integration and an empirical test.” International Journal of Conflict Management 8: 114–131. Olekalns, Mara, & Smith, Philip L. (2003). “Testing a three-way relationship among negotiators’ motivational orientations, strategy choices, and outcomes.” Journal of Experimental Social Psychology 39: 107–117.
272
PHILIP L. SMITH, MARA OLEKALNS AND LAURIE R. WEINGART
Olekalns, Mara, & Smith, Philip L. (2003). “Social motives in negotiation: The relationship between dyad composition, strategy sequences and outcomes.” International Journal of Conflict Management 14: 233–254. Olekalns, Mara, & Smith, Philip L. (2000). “Negotiating optimal outcomes: The role of strategic sequences in competitive negotiations.” Human Communication Research 24: 528–560. Olekalns, Mara, & Smith, Philip L. (1999). “Social value orientations and strategy choices in competitive negotiations.” Personality and Social Psychology Bulletin 25: 657–668. Olekalns, Mara, & Smith, Philip L. (1997). “Understanding optimal outcomes: The role of strategic sequences in competitive negotiations.” Working Paper No. 110, Department of Management & Industrial Relations, University of Melbourne. Olekalns, Mara, Smith, Philip L. & Kibby, Rachel (1996). “Social value orientations, negotiator strategies and outcomes.” European Journal of Social Psychology, 26: 299–313. Ross, Sheldon M. (2000). Introduction to probability models (7th ed.). San Diego, CA: Academic Press. Thomas, Andrew (1985). “Conversational routines: A Markov chain analysis.” Language and Communication 5: 287–296. Thomas, Andrew, Roger, Derek, & Bull, Peter (1983). “A sequential analysis of informal dyadic conversation using Markov chains.” British Journal of Social Psychology 22: 177–188. Ting-Toomey, Stella (1983). “An analysis of verbal communication patterns in high and low marital adjustment groups.” Human Communication Research 9: 306–319. Wasserman, Stanley, & Iacobucci, Dawn (1986). “Statistical analysis of discrete relational data.” British Journal of Mathematical and Statistical Psychology 39: 41–64. Wasserman, Staney, & Weaver. Sheila O. (1985). “Statistical analysis of binary relational data: Parameter estimation.” Journal of Mathematical Psychology 29: 406–427. Weingart, Laurie R., Olekalns, Mara & Smith, Philip L. (2004) “Quantitative coding of negotiation behavior.” International Negotiation 9 (3). Weingart, Laurie R., Prietula, Michael J., Hyder, Elaine, & Genovese, Christopher (1999). “Knowledge and the sequential processes of negotiation: A Markov chain analysis of response-in-kind.” Journal of Experimental Social Psychology 35: 366–393. Venables, William N., & Ripley, Brian D. (2002). Modern applied statistics with S-Plus. (4th ed.). New York: Springer.
All that Glitters is Not Gold: Examining the Perils and Obstacles in Collecting Data on the Internet CHA YEOW SIAH
With the end of the semester looming ahead, I received notification that the subject pool in the department has run dry again. Without other economical avenues to recruit participants, I am resigned to the fact that my research has to be shelved for the time being, until the new semester begins and until the subject pool is open for use . . . Many researchers in small and mid-sized psychology programs around the world who do not have large research funding and who lament the difficulty of obtaining data are familiar with the above scenario. Unfortunately, nothing is more disruptive and frustrating to an enthusiastic researcher when his progress is constantly hindered by such data collection difficulty. There is intense competition for participants when the subject pool is open for use to a large group of undergraduates, graduates and faculty members. In my department, for instance, about 70 projects are vying for a limited pool of about 1000 introductory psychology students in each academic year. Given that each introductory student contributes five hours of his/her time for research participation, this averages to about 70 subject hours per project – a number that is hardly sufficient for a 2 × 2 between-subjects experimental design! Fortunately, there is now a real possibility that such woeful tales of insufficient data and slow research progress will largely disappear as a result of the internet revolution. The internet offers almost instantaneous access to a world of participants. With a continual increase in connectivity and simultaneous decrease in connection cost due to advent in technology and economy of scale, the number of users for the World Wide Web is growing at a rapid pace. From a widely cited figure of about 30 million people world-wide connected to the internet in 1995 (GVU’s 7th WWW User Survey 1995), this estimation has witnessed a steady increase. For instance, Couper (2000) cited a study by the Strategies Group that reported the number of Americans connected to the internet at the end of 1998 to be around 84 million and this figure further increased to 106.1 million by December 1999. This large pool of internet users International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 273–288 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
274
CHA YEOW SIAH
makes the medium practically a goldmine for those involved in the data collection endeavor, whether their aim is for entertainment (e.g., polling websites such as misterpoll.com or news website such as CNN Quickvote), commercial (e.g., marketing research companies or the on-line marketing research departments found in many companies), or serious academic research purposes. In addition to the virtually unlimited pool of potential participants on the internet, the obvious benefits of using internet as the new medium for research are many. At the top of the list, many lauded the ease in identifying special population groups (e.g., homosexuals, identical twins, person with disability or special illnesses) through the help of newsgroups or listservs (automated mailing lists). This provides researchers the capability to conduct research that they would otherwise find too costly, time-consuming, and hence impossible, to carry out a decade ago. Once the survey or experiment is posted on a website, the data collection process runs itself like an automaton, without the need to employ assistants to administer repeated data collection sessions that stretch over weeks or months. There is also no need to cover for the potentially large cost of papers, envelopes, postage, and for copying materials. Transcription errors are minimized, if not entirely eliminated, when one can program for the transfer of data directly to a spreadsheet or statistical programs for analysis. Most attractive of all, imagine your data bank increases continuously 24 hours a day even while you sleep. It is as exhilarating an image that a researcher could dream of in terms of data collection! For instance, Pettit (1999) collected over 800 responses in 21 days for a Computer Anxiety Scale she puts on the Web. Similarly, Epstein and Klinkenberg (2002) collected 1116 completed surveys in their first month of the data collection period. Nancarrow, Pallister and Brace (2001) succinctly summarized the above obvious advantages as the speed, ease and cost appeals of conducting an internet-based study. Considering the appeals of internet-based study, the temptation for most researchers would be to jump in the fray immediately. If the Web is akin to a goldmine of easily available data, why wait? After all, an on-line study is merely an electronic implementation of what used to be a paper-and-pencil task – a conversion that could be easily accomplished by someone literate in programming language. Or is this not the case? What are the difficulties of implementing an internet-based study? What are the hidden threats to validity that may render data collected from such a medium unusable? What are the related issues (e.g., ethics) that ought to be tackled with the introduction of this new medium of research? Are there any ramifications for the current deluge of requests for research volunteers on the internet? The major objective of this article is to answer the above questions, by pointing out a few of the major problems encountered by researchers con-
ALL THAT GLITTERS IS NOT GOLD
275
ducting internet-based study. We use the term “internet-based study” to refer to research that uses the internet as the medium for data collection, and does not include all kinds of computer-based research (e.g., computerized adaptive testing or simply computerized versions of psychological instruments). Moreover, rapidly advancing technology has allowed technology-savvy researchers to develop a range of inventive internet-enabled computer-based research techniques such as behavioral observations in virtual reality, and naturalistic observations using gadgets such as webcams and smartcards (Stanton and Rogelberg 2002). However, given that my aim is not to explore new data collection techniques but to discuss the quality of data collected using the internet as the medium, I restrict my focus to conventional experimental and survey research otherwise conducted in face-to-face, paper-and-pencil format but which are now transferred to an on-line version by capitalizing on the popularity of the internet medium. The purpose of this article is to raise awareness of some of the problems encountered during data collection and the limitations in drawing inferences for readers new to internet-based research. The appeal of recruiting participants quickly, easily and cheaply ought to be weighed against the stumbling blocks one may encounter when making causal inferences from an experiment or when providing population estimates from a survey. In sum, while the internet is akin to a goldmine of data for researchers, this article hopes to deliver the caution that all that glitters is not gold, and it is important for researchers to carefully consider the costs and benefits before jumping on the bandwagon to conduct research on the internet.
Potential Pitfalls in Internet-based Studies The two most commonly employed research methods on the internet are surveys, followed by experiments. The discussion of problems related to internet-based study will be made relative to these two research methods. One reason why surveys tend to be the more commonly employed methodology on the internet is due to the relative ease in adapting paper-and-pencil forms of a survey instrument to the on-line format. On the other hand, experiments conducted on the internet typically require additional interactivity (e.g., writing scripts for random assignment of participants to various conditions) and hence additional demands on programming (refer to Birnbaum (2000) for a discussion of the computer techniques used for internet-based study). However, internet-based research has witnessed a steady increase in recent years, as technological barriers are gradually lowered through the increased availability of more user-friendly Web programming language and a concerted
276
CHA YEOW SIAH
cooperative effort by researchers to share resources. Birnbaum, for instance, has generously offered his programs, SurveyWiz and FactorWiz (http://psych. fullerton.edu/mbirnbaum/programs), for free to researchers unfamiliar with HTML to create simple survey or factorial studies on a Web page. A number of centralized Web laboratory studies have also been set up over the years to host internet-based studies, including the Web Experimental Psychology Lab at the University of Zurich (http://www.psychologie.unizh.ch/genpsy/ Ulf/Lab/WebExpPsyLab.html) and the PsychExps site at the University of Mississippi (http://psychexps.olemiss.edu/). Readers are referred to Reips (2001) and Birnbaum (2004a) for a listing of existing Web laboratories in the world. There are four major areas of concern when conducting internet-based research, namely, 1) sampling error and generalizability; 2) subject fraud; 3) measurement errors resulting from extraneous factors, and 4) the ethics of conducting research on the internet. Sampling Error and Generalizability The main argument for conducting internet-based studies is that we net a more heterogeneous group of participants (Murray and Fisher 2002). While it is true that data collection on the internet yields an equally, if not more, diverse group of participants than is found in the subject pool of introductory psychology college students (Gosling, Vazire, Srivastava and John 2004), this may not necessarily translate into any real gains in terms of generalizability for both surveys and experimental studies conducted on the internet. There are two reasons why this is so for survey studies. First, a more heterogeneous group does not necessarily mean a more representative group of participants that allows accurate generalization to be made from results. Despite the diversity, certain groups may have been disproportionately excluded from the sampling population available in the internet. When the exclusion of some demographic groups (e.g., African Americans) is systematically related to the variable being measured (e.g., opinion on the pervasiveness of racial discrimination), sampling bias will produce a distorted picture of the population. In fact, the general demographic pattern of participants in the internet-based study that emerged from various studies is one in which participants are predominantly white, young, well-educated males with at least a college degree, who live in metropolitan areas, and who belong to the middle to upper socioeconomic status (see reviews by Couper 2000; Epstein and Klinkenberg 2002; Gonzalez 2002). However, there are certainly grounds for optimism as the over-representation of certain demographic groups is fast disappearing because of the rapid
ALL THAT GLITTERS IS NOT GOLD
277
growth of internet-access across all segments of the society. The sample obtained by Gosling et al. (2004), for instance, compares favorably in terms of gender and race composition to those obtained by Epstein and Klinkenberg (2002) two years earlier. Second, unlike telephone or mail surveys where we can construct an adequate frame population for sampling based on phone or mail directories, there are still tremendous difficulties at the current stage to obtain an adequate frame population for internet-based study from which to make statistical inferences. Technological innovation, despite its far-reaching impact, will take time to become a constant in all household. Given that the internet medium has not achieved the same level of penetration as the telephone or the television, we are therefore still missing a considerable proportion of people in the target population (to which one wants to make inference) from the frame population, creating what is known as coverage error. This prevents researchers from carrying out probability-based sampling (Couper 2000; Truell, Bartlett and Alexander 2002), which in turn disallows any precise population estimates to be made due to the unknown standard error of measurement. Therefore, though the internet provides an efficient medium for a large amount of data to be collected quickly and easily, the price to pay for such conveniences may currently still be too high for survey researchers when they are unable to make population estimates from the data. This leads researchers such as Gonzalez (2002) to exercise extra care when interpreting results from internet-based surveys. For experimental research on the internet, the advantage of yielding a heterogeneous sample seems persuasive considering that the most common criticism on psychological research is its over-reliance on college student samples (McNemar 1942; Gosling et al. 2004). Thus, having a heterogeneous sample is a commonly cited reason in favor of Web experimentation (Montgomery and Ritchie 2002; Reips 2000), with the assumption that increased heterogeneity leads to increased external validity. However, one should refrain from exaggerating the importance of external validity in experimental research. To the extent that experimental research should rightfully be more concerned with theory-testing and not with establishing population estimates, obtaining a representative sample is secondary to the aim of ensuring internal validity because the former is not necessary for causal hypothesis testing (Berkowitz and Donnerstein 1982; Cook and Campbell 1979). In fact, the heterogeneous nature of a Web sample opens itself to the criticism that it invariably increases error variance (Greenberg 1987), and thus constitutes a weakness instead. In response, Birnbaum (2004b) argued that the large increase in sample size from an internet-based study is probably sufficient to counter the added error
278
CHA YEOW SIAH
variance due to diverse demographics. This remains predicated on the fact that the medium continues to attract participants easily, an unlikely scenario as internet users are increasingly inundated with requests for participation and the idea of participating in a study to help research is no longer a novelty. More practically, he suggested that given a sufficiently large sample size, data could also be partitioned into different strata based on the demographic variables and analyzed separately. Nonetheless, it is worthwhile to note that the argument of conducting internet-based experiments for the benefit of enhancing generalizability misses the target in experimental research, given that this quality adds little to the rigor of experimental studies. Subject Fraud While the internet empowers researchers with the ability to collect data from all over the world, it is also a double-edged sword that handicaps researchers in aspects that are otherwise easily controllable in face-to-face administration of survey or experiment. A prime example would be subject fraud, which is likely one of the most intractable problems in internet-based studies. Subject fraud refers to participants lying on critical demographic information (e.g., sex, race, age, etc.) in a survey or making multiple submissions that compromise the data integrity. The consequence of lying on demographic information in a survey is that it causes erroneous classification that results in inaccurate generalization. Though the problem also exists in conventional research setting (e.g., a respondent may still lie about his age in a telephone survey), the danger is considerably greater in internet-based study because of the anonymous and faceless setting. Similarly, multiple submissions, one of the most commonly reported problems in the literature, create havoc during data analysis because it violates the independent observation assumption of statistical techniques. Multiple submissions are more likely to concern researchers who employ personality inventories or intelligence tests on the internet. This is because participants may be tempted to try again for curiosity’s sake, just to see how changing one response would affect their overall score. Birnbaum (2004a) suggested a list of solutions for multiple submissions though none of them is foolproof. These solutions can be classified generally into proactive identification measures such as checking of Internet Protocol (IP) addresses, and preventive ones which mainly focus on removing the incentive to participate more than once. Popular measures such as checking for repeated IP addresses have limitations because the repetition could be due to different participants using the
ALL THAT GLITTERS IS NOT GOLD
279
same machine in a computer laboratory to participate in the study. In addition, the use of dynamic IP addresses, where internet service provider assigns the same IP to another user once it is freed up adds complexity to the detection task. Other strategies such as setting a cookie in the participant’s browser in order to identify whether the machine has ever visited the website runs into ethical concerns pertaining to intrusion of privacy. Alternatively, programming the server to check for certain identifiers (such as participants’ ID or password allocated prior to the start of the experiment) and to reject such submissions when they are flagged requires the allocation of password that raises participants’ concern that their responses are no longer confidential. Such proactive measures to detect multiple submissions are typically accompanied by preventive measures that attempt to discourage respondents from participating again through the removal of the incentive to do so. For instance, when there are rewards for participation, participants could be told in the instruction that repeated attempts could disqualify the individual from winning the reward. Or interested participants could be linked to an alternative site to try the same task again when they click a button to identify their attempts as repeated ones. Of course, many preventive measures rely on the cooperation of the participants for them to be effective. Participants’ cooperation could be enhanced by informing them that their action of multiple submissions would contaminate the data, rendering the study a futile exercise and ultimately wasting their own effort and time in volunteering. A key to such preventive measures is to make them transparent so that they do not appear laborious and thus act as a deterrent for cooperation. In the absence of a foolproof strategy to deter multiple submission, it is reassuring to hear from several experienced researchers that multiple submission is generally an infrequent problem based on their experience with the data (Birnbaum 2004a, 2004b; Musch and Reips 2000). Measurement Errors Resulting from Extraneous Factors Measurement errors are detrimental to research because it introduces noise into the data by contributing to error variance. In conventional face-to-face administration, this could originate from the experimenter (demand characteristics, experimenter expectation), the respondent (incomprehension, missing responses, response set) or the instrument (poor wording or unclear instructions, unreliable measuring instruments). The nature of internet-based study (anonymity, social distance from experimenter, program-controlled procedures) eliminates some of the above common sources of error variance while introducing new ones. Thus, it necessitates a re-examination of the various threats to reliability and validity.
280
CHA YEOW SIAH
Thus far, the picture that emerged from a comparison of results obtained from internet-based versus laboratory studies has been very encouraging. For instance, Krantz and Dalal (2000) concluded that the two methods yield very similar results based on his review of nine studies that compared the two versions. Likewise, Gosling et al. (2004) reached the same conclusion when they compared the score reliabilities and discriminant validities of scales derived from their Web sample with those obtained by John and Srivastava (1999). There are indeed good reasons to believe that mounting a study on the Web enhances the reliability and validity of its results. In surveys, missing responses and response errors are likely to become a thing of the past when researchers capitalize on the power of programming to prevent respondents from proceeding further until they provide a response. Programming to highlight an improbable response that lies outside the reasonable range of a value could also help to inform participants of their mistakes. In experiments conducted on the internet, demand characteristics are minimized because there is no longer direct contact between the experimenter and the participants (Piper 1998). This also prevents subtle experimenter bias from occurring because it eliminates the possibility that participants may respond to certain attributes (sex, age, race, physical appearances, and mannerism) of the experimenters (Hewson, Laurent and Vogel 1996). Reips (2001) and Murray & Fisher (2002) also argued that the increased social distance from the experimenter reduces social desirability concerns and allows respondents to be more truthful in their responses. On the other hand, conducting a study on the internet introduces new concerns. For instance, Montgomery and Ritchie (2002) found during the testing of their experiment program that as many as one-fifth of their respondents either failed to read the directions on the screen or simply could not figure out what they were supposed to do next. Whereas an experimenter present in the laboratory could easily explain the task to make sure that any confusions are cleared up, this is no longer possible for studies conducted on-line. A participant who is unsure what he is supposed to do next will submit random responses that are difficult for researchers to detect. This makes pilot testing and performing additional checks before launching the study extra important for researchers who wish to conduct internet-based study. In addition, the hardware and software employed in internet-based study also introduces extraneous influences on the data. For instance, the use of differently configured computer platforms, browsers and equipments (e.g., different sized computer monitors) lead to compatibility issues and varying presentation (Buchanan and Smith 1999) that cannot be controlled by researchers. Besides, the different connection speed among the participants may also lead
ALL THAT GLITTERS IS NOT GOLD
281
to unnecessary frustration and thus contributes to random error (though this problem could be technically resolved or minimized, see Montgomery and Ritchie (2002)) and attrition. Couper (2000) reviewed past research and found that the design of the survey instrument such as the layout of questions can affect responses. This means that extraneous influences such as those mentioned are likely to contribute significantly to the random error component. Do the different research media (face-to-face versus Web-based) produce any psychological differences in participants that would affect responding? To the extent that such psychological influences may affect the reliability of the data collected from different media, or even lead to different predictions of behaviors, it is indeed an issue worthy of researchers’ attention. There is certainly evidence attesting to the possibility that some psychological variables are at work when participants respond to survey on the internet. For instance, Epstein and Kinenberg (2002) replicated an earlier paperand-pencil survey on HIV risk behavior by soliciting a homosexual sample on the internet, and noted much to their surprise that there was “an amazing amount of self-disclosures” (41) among their Web sample who answered the survey on the internet. Similarly, Gosling et al. (2004) also cited a study by Turner, Rogers, Lindberg, Pleck and Stonenstein (101) who revealed an increase in reporting of stigmatized behaviors among adolescent respondents under likewise circumstances, though it is worthy for us to note that Turner et al.’s study was a computerized survey that was not conducted on the internet. In fact, we can already gain substantial knowledge from extant literature that investigated equivalence between the computerized instrument and the traditional paper-and-pencil format. In particular, Richman, Kiesler, Weisband and Drasgow (1999) concluded, based on a meta-analysis of 61 studies, that computerized instruments indeed lead to less social desirability distortion when conducted in an anonymous setting where respondents feel more comfortable and less wary of evaluation. More importantly, Richman et al. noted that it is not the computer instrument per se, but moderating factors, such as whether participants are tested alone or in the presence of others, which has an effect on social desirability distortion. Obviously, anonymity on the internet helps to induce more candid responses in surveys, especially those which ask sensitive questions. This is welcoming news for researchers who study conflicts or investigate stigmatized or deviant behaviors that are traditionally plagued with the problem of untruthful reporting. But to ensure that this quality continues to exist, researchers should take care not to damage this veil of anonymity, for instance, when they are devising ways to curb the problem of multiple submissions.
282
CHA YEOW SIAH
On the other hand, for researchers conducting social psychological experiments on the internet, it is important to consider the anonymity factor, and to understand how it influences social behaviors. Bargh and McKenna (2004) pointed to the anonymity factor as a critical difference of the internet medium from previously available communication media and settings that can possibly affect the process and outcome of social interactions. According to Social Identity theory, whether an individual adheres or defies normative behavior in an anonymous setting very much depends on whether there exists strong group-level social identity in the group that he or she interacts in (Reicher, Spears and Postmes 1995). The prediction made by Social Identity theory is interesting and can be surprising for researchers unfamiliar to the theory. Thus, it is important for any prediction of social behavior made based on a Web sample to take into account the depersonalizing influence of an anonymous setting and how this is moderated by group-level identity. An understanding of such effects would be especially important in any research that is concerned with the interdependency between groups and individuals (e.g., social dilemmas research). Apart from anonymity, another factor that operates as a critical factor in internet-based research is the variable of trust. For instance, Bargh and McKenna (2004) gave the example of how “e-negotiation” could go awry due to greater levels of distrust when parties interact on-line (e.g., over e-mail).1 Needless to say, trust is a crucial element for studies that investigate social conflicts and their resolution. Substantively, trust (or the lack of ) in an online setting opens up new and intriguing avenues of research for social conflicts researchers. Methodologically, trust could also affect studies conducted on the internet via its influence on the treatment effect. For instance, in a recent investigation where I hoped to find out whether participants given a computerized social dilemmas task would behave differently if the task was carried out in the lab versus over the internet, I discovered an interesting difference owing to the setting in which the two groups of student participants completed their tasks. Specifically, the task was deliberately programmed to lead student participants to believe that they were connected to a group of seven other participants in a social dilemma task when in fact, they were given false feedbacks. Despite the fact that the computerized task was completely identical except for the setting (i.e., participating in the lab PC under the supervision of an experimenter versus participating on the internet by going to a computer lab on their own), 36% (n = 10) of the student participants who participated in the study on the internet were suspicious of the manipulation compared to 6% (n = 08) of the student participants who took part in the laboratory when they were queried immediately after the
ALL THAT GLITTERS IS NOT GOLD
283
experiment. The results imply that participants may become more skeptical of experimental procedures as a result of not seeing other real participants or experimenters around them, thereby weakening treatments designed by the experimenters. The possible attenuation of effect size means that internet-based researchers should be even more attentive to the issue of design and power. Finally, one other problem that plagues researchers of internet-based study is the lack of control over the environmental factors in which participants take part in the study (Buchanan and Smith 1999; Murray and Fisher 2002; O’Neil and Penrod 2001). It is no longer possible for researchers to provide a minimally distractive environment for participants when the study is carried out on-line. For instance, experimenters have little knowledge and control over the mental state of the participants (e.g., whether the participant is alert, intoxicated or overly tired), the time and day at which the data is collected, the level of background noises (e.g., music may be blaring in the background), and the kind of activities that participants may be carrying out concurrently during their participation (e.g., eating or watching television programs). It is a serious problem, particularly for experiments, because the inability to control such environmental factors introduces substantial error variances that affect the results in an unknown and significant manner. Can the problem be solved currently? Unfortunately, it seems that the answer is no. Can it be solved in the future? The answer is a hopeful yes when video data in addition to text data could be transmitted to provide more information to the experimenter. But until the technology is widely adopted, researchers conducting internet-based study may have to live with larger random error that has the possibility of masking the true treatment effects. Thus there are potential advantages and disadvantages when conducting an internet-based study. The possible cost of a weaker treatment effect and a lack of control in an experiment and the possible benefits of increasing selfdisclosure and decreasing social desirability have different implications for researchers conducting experiments or survey on the internet. In sum, there are strong reasons to believe that subtle differences between electronic and paper-and-pencil implementation of surveys and experiments produce dynamics and different experiences within individual participants that would interact in an unintended manner with the investigated variables. However, such issues have not been given sufficient attention to date. More empirical research in this area is necessary for researchers to probe in-depth on the list of factors mentioned.
284
CHA YEOW SIAH
Ethics of Conducting Research on the Internet Due to the recency of internet-based research, it is not surprising that the ethical arena for such methodology resembles the Wild West with no clearly set rules and guidelines for researchers. The latest revision of the American Psychological Association (APA) Ethics Code (2002), for instance, provides little help in guiding researchers who wish to make sure what they do falls within the boundary of the Ethics Code. Thus, the approval of research conduct for internet-based studies is largely left to the discretion of individual Institutional Review Board (IRB) which in some cases, may lead to different adoption of standards. For instance, while the IRB from which Epstein & Klinkenberg (2002) sought approval eventually allowed them to set cookies in participants’ browsers to avoid subject fraud, the same measure was not approved for reasons of intrusion to privacy by the Human Subjects Research Review Committee to which Montgomery & Ritchie (2002) belongs. The issue is an important one because there are many gray areas involved in the conduct of internet-based research, due to the fact that the medium represents largely uncharted ground for researchers. Many of these gray areas are found in domains such as subject recruitment, obtaining informed consent, protecting the well-being and privacy of participants, and ensuring data integrity. For instance, what are considered legitimate ways for researchers to recruit participants from internet or specific newsgroups? What are the dos and don’ts that researchers should take note of in order to protect participants’ anonymity and privacy? These various concerns need to be addressed in order for internet-based research to continue to flourish. To this end, experiences shared by various researchers (e.g., Buchanan and Smith 1999; Montgomery and Ritchie 2002; Pettit 1999) and related discussions on ethical issues (e.g., Smith and Leigh 1997) have served as useful guides in steering future researchers in the right direction. In particular, the set of 30 guidelines provided by Michalak and Szabo (1998) filled a much needed gap caused by the lack of code of conduct. In contrast to traditional research methodology, there is an even greater need to continually update the ethical codes for internet-based research because of the rapid advance of technology.
Conclusion This article discusses a few major problems which may be encountered by prospective researchers who are keen to take advantage of the ease in collecting data via the internet. While some of the problems present considerable
ALL THAT GLITTERS IS NOT GOLD
285
challenges to the viability of conducting research on the internet, it is important to take note that there is no lack of internet-based studies that have produced encouraging results. For instance, McGraw, Tew, and Williams (2000) showed that it is possible to carry out cognitive experiments involving timed displays to collect reaction time data with detectable effects of interest. Buchanan and Smith (1999) obtained better model fit and higher reliability for scores obtained on a personality inventory from an internet sample than from a conventional paper-and-pencil version. From a comparison of an internet-based survey and a mail survey, Truell, Bartlett, and Alexander (2002) found that not only were the response rates comparable, the response speed of internet-based survey was also about seven days faster than the mail survey, and it was more thoroughly completed than the traditional counterpart. In a similar vein, Pettit (2002) identified no difference in response set effects between an on-line version and a paper-and-pencil version of a personality inventory. In fact, the data obtained from on-line version was superior in terms of lower response errors due to illegible responses. Finally, Gosling et al. (2004) presented a spirited defense of Web-based studies and compared a set of traditional samples with a large internet sample to show that the latter was equally diverse with respect to a list of demographic variables, and that the data were also at least on par in quality, on a range of criteria, compared to those obtained by traditional paper-ad-pencil methods. As with all other research methodologies, there are benefits and costs associated with conducting research on the internet. Ultimately, researchers need to make a judgment call based on the variables studied and the design employed to decide on a case-by-case basis whether the benefits outweigh the cost of conducting research on the internet. Finally, it is undeniable that internet-based studies will continue to proliferate. The list of problems identified in this article should not be viewed as insurmountable, but rather, as new opportunities for innovation and refinement. More important, it is an additional tool in an arsenal of research methods that provide researchers with another means to triangulate their variables of interest, adding rigor to their research. Viewed in this light, the importance of attending to the ethical issues that arise from this rapidly expanding mode of research cannot be understated. The reason is simple. The promise of internet-based research depends very much on the perceived credibility of such a methodology. Without clear guidelines to prevent abuse from careless researchers, we may well face two dire consequences at both ends. First, the deep skepticism over the usefulness and quality of data collected through this medium would continue to mount. Second, prospective participants are driven away
286
CHA YEOW SIAH
from volunteering for studies conducted on the internet. When this happens, we may find ourselves staring into empty goldmines similar to the ones found at the end of the Gold Rush Era.
Note 1. For more on email- and internet-based negotiation approaches, see International Negotiation 9(1) and 9(2), thematic issues that focus on the impact of technology on negotiation processes.
References American Psychological Association (2002). “Ethical principles of psychologists and code of conduct.” Retrieved 19 November 2002 from http://www.apa.org/ethics/code2002.html. Bargh, John A. and McKenna, Katelyn Y.A. (2004). “The internet and social life.” Annual Review of Psychology, 55: 573–590. Berkowitz, Leonard and Donnerstein, Edward (1982). “External validity is more than skin deep: Some answers to criticisms of laboratory experiments.” American Psychologist, 37, 3: 245–257. Birnbaum, Michael H., editor (2000). “Psychological experiments on the internet.” San Diego, CA: Academic Press. Birnbaum, Michael H. (2004a). “Human Research and Data Collection via the Internet.” Annual Review of Psychology, 55: 803–832. Birnbaum, Michael H. (2004b). “Methodological and Ethical Issues in Conducting Social Psychology Research via the Internet,” in Carol Sansone, Carolyn C. Morf, and Abigail T. Panter, editors, The Sage Handbook of Methods in Social Psychology. London: Sage. Buchanan, Tom and Smith, John L. (1999). “Using the internet for psychological research: Personality testing on the World Wide Web.” British Journal of Psychology, 90: 125–144. Cook, Thomas C. and Campbell, Donald T. (1979). “Quasi-experimentation: Design and analysis issues for field settings.” Chicago: Rand McNally. Couper, Mick P. (2000). “A Review of Issues and Approaches,” in David de Vaus, editor, Social Surveys 2: 149–180. London: Sage Publications. Epstein, Joel and Klinkenberg, W. Dean (2002). “Collecting Data via the Internet: The Development and Deployment of a Web-Based Survey.” Journal of Technology in Human Services, 19, 2/3: 33–47. Gonzalez, J.E. (2002). “Present Day Use of the Internet for Survey-Based Research.” Journal of Technology in Human Services, 19, 2/3: 19–31. Gosling, Samuel D., Vazire, Simine, Srivastava, Sanjay and John, Oliver P. (2004). “Should We Trust Web-Based Studies? A Comparative Analysis of Six Preconceptions About Internet Questionnaires.” American Psychologist, 59, 2: 93–104. Graphic Visualization and Visibility Centre’s 7th WWW User Survey (1997). Retrieved 12 November 2002, from http://www.cc.gatech.edu/gvu/user_surveys. Greenberg, Jerald (1987). “The college sophomore as guinea pig: Setting the record straight.” Academy of Management Review, 12, 1: 157–159.
ALL THAT GLITTERS IS NOT GOLD
287
Hewson, Claire M., Laurent, Dianna and Vogel, Carl M. (1996). “Proper methodologies for psychological and sociological studies conducted via the internet.” Behavior Research Methods, Instruments & Computers, 28, 2: 186–191. John, Oliver. P. and Srivastava, Sanjay (1999). “The Big Five Trait taxonomy: History, measurement, and theoretical perspectives,” in Lawrence A. Pervin and Oliver P. John, editors, Handbook of personality: Theory and research. New York: Guilford Press. Krantz, John H. and Dalal, Reeshad (2000). “Validity of Web-based psychological research,” in Michael H. Birnbaum, editor, Psychological experiments on the internet. San Diego, CA: Academic Press. McGraw, Kenneth O., Tew, Mark D. and Williams, John E. (2000). “The Integrity of Webdelivered Experiments: Can You Trust the Data?” Psychological Science, 11: 502–506. McNemar, Q. (1942). Opinion-attitude methodology. Psychological Bulletin, 43: 289–374. Michalak, Erin E. and Szabo, Attila (1998). “Guidelines for Internet Research: An Update.” European Psychologist, 3, 1: 70–75. Montgomery, Paul and Ritchie, David (2002). “Kermitt: Conducting an Experiment on the Web.” Journal of Technology in Human Services, 19, 2/3: 135–149. Murray, Danielle M. and Fisher. Jeffrey D. (2002). “The Internet: A Virtually Untapped Tool for Research.” Journal of Technology in Human Services, 19, 2/3: 5–18. Musch, Jochen and Reips, Ulf-Dietrich. (2000). “A brief history of Web experimenting,” in Michael H. Birnbaum, editor, Psychological experiments on the internet. San Diego, CA: Academic Press. Nancarrow, Clive, Pallister, John and Bruce, Ian (2001). “A new research medium, new research populations and seven deadly sins for internet researchers.” Quantitative Market Research: An International Journal, 4: 136–149. O’Neil, Kevin M. and Penrod, Steven D. (2001). “Methodological variables in Web-based research that may affect results: Sample type, monetary incentives, and personal information.” Behavior Research Methods, Instruments & Computers, 33, 2: 226–233. Pettit, Frances A. (1999). “Exploring the use of the World Wide Web as a psychology data collection tool.” Computers in Human Behavior, 15, 1: 67–71. Pettit, Frances A. (2002). “A Comparison of World-Wide Web and paper-and-pencil personality questionnaires.” Behavior Research Methods, Instruments & Computers, 34, 1: 50–54. Piper, Allison I. (1998). “Conducting Social Science Laboratory Experiments on the World Wide Web.” Library & Information Science Research, 20, 1: 5–21. Reicher, Stephen D., Spears, Russell and Postmes, Tom (1995). “A social identity model of deindividuation phenomena,” in Wolfgang Strobe and Miles Hewstone, editors, European Review of Social Psychology, 6: 161–198. Chichester, UK: Wiley. Reips, Ulf-Dietrich. (2000). “The Web experiment method: Advantages, disadvantages, and solutions,” in Michael H. Birnbaum, editor, Psychological experiments on the internet. San Diego, CA: Academic Press. Reips, Ulf-Dietrich. (2001). “The Web experimental psychology lab: Five years of data collection on the internet.” Behavior Research Methods, Instruments & Computers, 33: 201–211. Richman, Wendy L., Kiesler Sara, Weisband Suzanne, and Drasgow Fritz (1999). “A Meta-Analytic Study of Social Desirability Distortion in Computer-Administered Questionnaires, Traditional Questionnaires, and Interviews.” Journal of Applied Psychology, 84, 5: 754–775. Smith, Michael A. and Leigh, Brant (1997). “Virtual subjects: Using the internet as an alternative source of subjects and research environment.” Behavior Research Methods, Instruments, & Computers, 29, 4: 496–505.
288
CHA YEOW SIAH
Stanton, Jeffrey M. and Rogelberg, Steven G. (2002). “Beyond Online Surveys: Internet Research Opportunities for Industrial-Organizational Psychology,” in Steven G. Rogelberg, editor, Handbook of research methods in industrial and organizational psychology. Malden, MA: Blackwell Publishers. Truell, Allen D., Bartlett, James E., II and Alexander, Melody W. (2002). “Response rate, speed, and completeness: A comparison of internet-based and mail surveys.” Behavior Research Methods, Instruments & Computers, 34: 46–49. Turner, Charles F., Ku Leighton, Rogers, Susan M., Lindberg, Laura D., Pleck, Joseph H. and Sonenstein, Freya L. (1998). “Adolescent Sexual Behavior, Drug Use, and Violence: Increased Reporting with Computer Technology.” Science, New Series, 280, 5365: 867–873.
The Method of Experimental Economics RACHEL CROSON
Introduction The 2002 Nobel Prize in Economics was awarded to Daniel Kahneman (an experimental psychologist) and Vernon Smith (an experimental economist). This award acknowledges an important trend in economics – the growth of experiments as a valid, accepted methodology and the influence of psychological research in that growth.1 There are many similarities between experimental economics and psychological research. Researchers in both fields are concerned with similar substantive areas; bargaining in economics and negotiation in psychology, public goods provision in economics and social dilemmas in psychology, just to name a few. The two fields also share many methodological practices. Both fields use convenient populations (like undergraduate students) as participants in their experiments. Both fields elicit decisions or introspections from their participants and use those to learn about the world. Both fields are concerned with careful experimental design, avoiding demand effects, and appropriate statistical analyses. However, the objectives of psychology and economics experiments are often different. Economics experiments are designed to address economic theories; psychology experiments are designed to address psychological theories. This distinction may seem obvious at first, but it has important and often unforeseen implications for methodological differences in the two fields. These methodological differences are the subject of this article. I will discuss five areas where experimental economists and experimental psychologists differ. They are ordered from most to least important for economists, although readers may disagree with my ordering. This is by no means the first article on experimental economics methodology. However, most previous articles have been aimed at traditional (nonexperimental) economists to argue that the experimental methodology is a relevant and valid one (e.g. Binmore 1987, 1999; Plott 1982, 1991a, 1991b, 1994; Roth 1986, 1991, 1994; Smith 1976, 1989, 1994). Friedman and Sunder
International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 289–306 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
290
RACHEL CROSON
(1994) have written the definitive text on experimental economics methodology aimed at economists who want to begin running or consuming experiments, and I highly recommend it for psychologists who want to see how economists think about experimental methodology. A very few papers explicitly compare economic and psychological methodology, Cox and Isaac (1986) present experimental economics methodology for an interdisciplinary audience. Hertwig and Ortmann (2001) provide a well-documented comparison of the methodologies. Holt (1995) presents his perceptions of methodology in experimental psychology for economists. Croson (forthcoming) compares the two methodologies for law and economics scholars. As many have noted, one main difference between the fields of economics and psychology is the existence of a unified core theory – expected utility theory.2 Experiments in economics are explicitly designed to address these economic theories. These experiments have an important place in the dialectic of the scientific method. First, the theory is constructed or described and predictions are generated from it deductively. Then, an experiment is run to test these predictions. Experiments designed to address economic theories need to have a high degree of internal validity. This requires the construction of a lab situation that exactly captures the theory’s assumptions. If the experiment is not internally valid, the data it produces is not relevant to the theory’s predictions. For instance, if the theory a researcher is trying to test assumes that a game is infinitely repeated, the experiment must implement such a game.3 Similarly one cannot test one-shot theories using repeated interaction among the same players.4 Many (possibly all) of the ways in which experimental economics methodology differs from that of experimental psychology, discussed below, stem from this objective of internal validity. Economic theories predict how people will act in the presence of real, salient and usually financial rewards; thus contingent incentives are critical for economic experiments. The theories are abstract, intending to apply to many different situations and individuals; thus there is little or no context in economic experiments and the subject pools used are also not of primary concern. The theory assumes that actors understand and believe the relationship between their actions and their payoffs; deception would endanger this belief and is almost nonexistent. Experimental details are chosen to influence participant’s perceptions of one or another of these areas. The final topic, data analysis is a bit different – variations in this methodology are based on the reference groups to whom researchers are appealing. For each topic, I will introduce the methodological issue, describe the experimental economics practice and its rationale and mention psychological
THE METHOD OF EXPERIMENTAL ECONOMICS
291
practice and rationale (much less thoroughly). My hope is that this article will lead not only to a deeper understanding of each field’s choice of methodology, but also to some practical advice to psychologists on how to have their work read and accepted by economists. I begin with perhaps the largest gulf between economic and psychological experiments: incentives.
Incentives Economic theories describe and predict decisions individuals will make in the presence of payoffs. Typically, theories specify the payoffs from taking one action or another. For example, individuals contributing to a public good (cooperating in a social dilemma) incur some private cost from doing so which outweighs the private benefit they receive from the public good. Since earnings are higher when individuals defect than when they contribute, theory predicts they will defect. It is critical for theory testing that the participants actually face the payoffs assumed by the theory. The fact that individuals cooperate in social dilemmas when there are no payoff consequences from their actions is simply not informative. Economic theory makes no predictions of what individuals will say they would do, only what they will actually do when faced with a given decision and the resulting payoffs. For example when studying ultimatum bargaining it is important to have participants actually earn the amounts that their decisions imply (e.g. Croson 1996, Forsythe et al. 1994). This has led to the practice of induced valuation in experimental economics (Smith 1976), where participants’ compensation is not simply positive, but importantly is responsive to the choices they make in a way that is consistent with the theory being tested. This practice replaces the flat-fee payments more common in psychology experiments, where individuals are paid some amount for their participation (perhaps zero, or perhaps earning extra credit in their course) which is not contingent on the decisions they make. For example, when two psychologists first identified the disjunction effect (nonconsequential reasoning), their papers included data from surveys and hypothetical questions (Shafir and Tversky 1992; Tversky and Shafir 1992). Participants were asked about gambles they might take after having won or lost a previous gamble or vacations they might purchase after having passed or failed their exams. When an economist wanted to further explore the boundaries of the effect, participants played games (prisoner dilemmas, public goods, and games of iterated dominance) and were paid based on their decisions and the decisions of their counterparts (Croson 1999, Croson 2000).
292
RACHEL CROSON
This is not to say that all psychologists use unpaid participants – many run experiments with incentives.5 But all economics experiments involve salient payments to participants. This practice is considered critical to the validity of the experiment and the objective of testing the theory being addressed. Typically, payments in economics experiments are made in cash directly after the experiment. There are a few reasons for the use of cash. First, everyone values it, in contrast with extra-credit points or other grade-related rewards which may be valued only by students who are grade-conscious and/or whose grade may be affected by the outcome. Second, cash is one good that is non-satiable – more is always better. That said, some notable experiments use forms of payment other than cash. All, however, use payments that are contingent on individual’s decisions and argue (successfully) that the payments involved are consistent with the theories being tested. For example, in Boyce et al. (1992) the authors want to elicit willingness-to-pay and willingness-to-accept values for specific environmental damages. They use dollars for the WTP and WTA measures, but for the environmental damage they bring a baby tree into the lab that the experimenter will chop down if the damage must be incurred. This tree-killing action captures nicely the theory’s assumptions of incurring environmental damage and is an excellent example of non-financial rewards that are nonetheless contingent on the participant’s actions. A second issue in the area of incentives is the amount participants receive. A number of papers have argued that participants must be paid enough to compensate them for their time (thus, the average payoff should translate into an hourly wage roughly equal to the salaries of on-campus jobs), and for the thinking costs they incur during the experiment (Smith and Walker 1993). One natural question is the extent to which paying participants contingent on their actions affects outcomes. There are now a number of meta-analyses on this question. Camerer and Hogarth (1999) review 74 studies and conclude that financial incentives have a large effect on judgment and decision tasks, but the effect is smaller in games and markets. Hertwig and Ortmann (1998) review 10 studies from the Journal of Behavioral Decision Making with and without incentives. They find that incentives decrease framing effects, bring auction bids closer to optimality and eliminate preference reversals. They also identify contexts when payment had no effect, specifically confidence judgment and information acquisitions. For some tasks contingent payment does not seem to affect the mean outcomes, while for others it does. There is general agreement, however, that contingent payment reduces the variance of responses one receives. Smith and Walker (1993) survey 31 experimental eco-
THE METHOD OF EXPERIMENTAL ECONOMICS
293
nomics studies varying incentives and conclude that “in virtually all cases, rewards reduce the variance of the data around the predicted outcome” (245). Paying contingent (and sufficient) payments can get expensive. There are a number of “tricks” that economists play in order to stretch limited funding. First, if participants in the experiment are playing the same game repeatedly, researchers sometimes choose one round at random for real payment. Of course, participants are told in advance that one round will be randomly chosen (see the section on deception, below), but not which one. This payment design has the advantage of avoiding wealth effects in which participants who are accumulating money throughout the session begin making riskier and riskier decisions. Similarly, for experiments with large numbers of participants, researchers sometimes choose one participant (or one dyad or one group) at random for real payment. Again, participants are told in advance that one will be chosen at random. While both these techniques have the potential to reduce expenditures, experimental economists using these designs typically make the expected earnings of any given participant equal to the wage rate. This is not to say that economists never use any data collected from a survey or hypothetical responses. Many experiments in economics use a postexperimental questionnaire where participants are asked to introspect as to why they made the decisions they made. But economists are very cautious when reporting and interpreting this data, and always ask these questions after they have collected the real-money decisions of interest. The perception is that responses to questions like this are “cheap talk” and may have no relation to what is actually in the minds of experimental participants. Personally, I consider this data like any other, taking into account the motivations of the subjects including honest reporting, impression management and other factors in interpreting these results. This data is often particularly useful in going beyond documenting an outcome and digging deeper into the motivations causing it. My recommendation to psychologists who want economists to use their work and to cite their results is simple – pay your participants. And pay them not a flat fee, but an amount contingent on their decisions. This is particularly important if the experiment’s results are inconsistent with economic theory. It is far too easy (and common) for economists to simply disregard experimental results inconsistent with their theories because the participants weren’t appropriately incentivized.
294
RACHEL CROSON
Context The second domain in which economic and psychology experiments differ is in their use of context in experiments. Economics experiments are primarily context-free (or context-neutral), even those that purport to be about something in the real world (e.g. Croson and Mnookin 1997). Participants are not told they are providing a public good (or, even more specifically, cleaning up the water in their city) as in many psychology experiments. Instead they are asked to allocate tokens between two accounts that offer varying payoffs (e.g. Croson 1996b). There are three main reasons why economists use little context. First, the theory being tested is supposed to apply generally – it should predict behavior in any context that involves the appropriate payoffs, so the experiments to test the theory should not rely on a particular context. Second, context often adds variance to the data. For example, if some participants think that going to court is a good thing and others think it is a bad thing, then describing the experimental decision as ‘going to court’ as opposed to ‘choosing option A’ could add noise (Croson and Johnston 2000). This additional noise might not change the average or aggregate decision, but it can impact the variance of those decisions, reducing the likelihood of detecting statistically significant differences between treatments of the experiment. Finally, and most importantly, context can add systematic bias or demand effects. For example, if participants in aggregate think there should be fewer court cases or want to be seen as kind, gentle types by their professor, then describing the decision in terms of going to court might reduce everyone’s likelihood of choosing that option. This would change the responses in a systematic way based on the context. Such systematic changes in the data will significantly change the conclusions reached, so economists try to avoid context in their experiments. There are some costs of avoiding context (see below), but in economics experiments they are not high, and context is generally considered a nuisance variable rather than a variable of interest. However, there are arguments in favor of context as well. As Loewenstein (1999) points out, even abstract instructions contain context, albeit unfamiliar. Additionally, experiments with context have more external validity, cueing subjects to behavior that we might more often observe in the real world. For example, the well-known result that Wason’s-task errors significantly decrease when context is added, suggests that the importance of cognitive errors may have been overestimated. For psychologists who want their work to be accepted by economists, the use of context is not as serious a methodological deviation as a lack of incentives. However, papers aimed at economists
THE METHOD OF EXPERIMENTAL ECONOMICS
295
need to argue that the results are not being driven by the particular context chosen or, even better, demonstrate the same effect in multiple (sufficiently different) contexts.
Subject Pools Just as economic theories are intended to apply generally to varying contexts, they are designed to apply generally to varying subject pools. While some theories focus on individual differences (e.g. risk-preferences), the primary goal in economics is to develop and test simple theories that explain behavior by many people in many contexts, even if only imperfectly, rather than developing more complicated theories that explain behavior by a subset of people in a smaller class of contexts more accurately. Thus there is limited concern by experimental economists about demographics. Many early economics experiments did not even collect demographic data of the participants and very little analysis was done examining individual heterogeneity. More recent research in experimental economics has begun to focus on individual differences, especially gender differences. Eckel and Grossman (forthcoming b) and Croson and Gueezy (2005) provide reviews of this literature. Other experimental economics research has started to investigate cultural differences. For example, Buchan, Croson and Dawes (forthcoming) use experimental economics methodology to look at cross-cultural direct and indirect trust. That said, economists are concerned with other dimensions of their subject pool. First, economists typically recruit volunteers as participants in their experiments, only rarely using students from the researchers’ own course (or other courses in the department). The latter is a common practice in psychology, where students in introductory psychology courses participate in experiments as part of their educational experience and/or for extra credit in their courses. Experimental economists are particularly concerned with avoiding demand effects present when students who have learned about the theories they are trying to test in class participate in experiments testing them. Some recent research suggests that there are systematic differences between decisions made by these “true” volunteers and by “pseudo” volunteers who are students in a class (Eckel and Grossman, forthcoming a). Second is the ongoing (and unresolved) debate about whether economists are different than other professionals. Papers have demonstrated that economic students (undergraduate and PhD) are more likely to free-ride in social dilemmas (Marwell and Ames 1981; Frank, Gilovich and Regan 1993, 1996) and to offer (and demand) less in ultimatum games (Carter and Irons 1991).
296
RACHEL CROSON
However, these results are by no means unchallenged. Yezer, Goldfarb and Poppen (1996) present evidence that economics students are more cooperative than other students in a lost-letter experiment (contrasting with the Frank et al. papers which were hypothetical survey results). Similarly, Laband and Biel (1999) show that professional economists cheat less on their association dues than professional sociologists and political scientists. My take on this debate is twofold. First, the jury is still out on whether economists are different than non-economists. Second, given this unresolved debate, the experimental economics practice of using volunteers from all over the university as participants in their experiments rather than students from their own (economics) classes is a sensible and conservative one. A final issue in the subject-pool debate concerns the use of students as opposed to professionals (or “real people” as critics of experiments sometimes say). In terms of economics experiments that test theories, this is not a problematic issue – the economic theory is supposed to be general and to apply to anyone facing a decision like the one described in the theory, not simply people who are above 30, for example. However, there are some experiments, particularly those that are aimed at testbedding policies, where the use of professionals as participants makes sense. For example, Cummings Holt and Laury (2002) use farmers to test competing designs for mechanisms to allocate water rights in Georgia. Dyer, Kagel and Levin (1989) and Dyer and Kagel (1996) test theories of auctions with contractors who submit competitive bids for a living. Once the researcher moves from students to professionals, the incentives used in the experiment become both more important and more difficult. An undergraduate student can be induced to think hard about a problem if the difference between making the right decision and the wrong one is around twenty dollars. A professional whose income (and opportunity cost of time) is higher may require significantly more money to participate. More troublesome is that high-earning professions may not be motivated by money, at least not on the scale most experimentalists can pay. Nonetheless, a growing number of economics experiments have attempted to demonstrate effects among professionals as well as among student participants. For psychologists who want their work to be accepted by economists, the use of Psych I and similar subject pools is somewhat problematic. If incentives are being offered, recruiting participants from across the university is a relatively painless way to avoid selection biases that may result from using only students in psychology courses, and demand effects from using one’s own students as participants in experiments.
THE METHOD OF EXPERIMENTAL ECONOMICS
297
Deception One methodological difference between experimental psychologists and economists which has recently received attention is the use of deception in experiments. A series of papers examines (and criticizes) standard practice in both fields, including Ortmann and Hertwig (1997, 1998), Bonetti (1998a, 1988b), McDaniel and Starmer (1998) and Hey (1998). One of the strictest rules in experimental economics is that the researcher may not deceive their participants. This prohibition on deception includes deception about the purpose of the experiment, the payoffs the participants will earn, or the characterization of the participants’ counterparts. As described above, the validity of an economic experiment rests on the link between behavior and payoffs (incentives). If that link is weakened, the experiment becomes an inferior test of the economic theory it is designed to address. Similarly, the reasoning goes, if participants are deceived about that link, the validity of their decisions is called into doubt. A second reason deception is disfavored in economics has to do with the public-goods nature of trust in the experimenter. If participants are routinely deceived in experiments, for instance, by being told they will take home their earnings in the game, and then actually receiving a $5 flat fee for their participation, they will begin to distrust the experimenter’s statements. This lack of trust could lead the participants to change their behavior in future experiments. In contrast, psychology experiments often deceive participants about the purpose of the experiment, the payoffs that will be earned and the existence (or nonexistence) of counterparts. Sometimes this deception is necessary for the experiment. For example, deception about the purpose of the experiment can aid in honest elicitation and overcome presentation effects; participants who know an experiment is about racial discrimination may act contrary to the way they normally would. In addition, deception is often used to examine situations which would not occur naturally, for example, how individuals respond to low ultimatum offers. That said, many of these benefits arising from deception can be enjoyed by simple omission (not informing the participants of the subject of the experiment, or doing so only very generally) rather than by commission (explicitly lying to the participants). After deceptive experiences, experimental psychologists often debrief their participants, revealing the deception and asking them not to tell other potential participants about their knowledge. Although this is recommended and often required by human subjects committees, to economists this practice
298
RACHEL CROSON
only enhances the likelihood of recruiting participants who enter the experiment expecting deception and further weakens the link between actions and earnings. The norm against deception is quite strong in experimental economics – an economic journal will not publish an experiment in which deception was used, while in psychology journals, deception is commonplace and accepted. The closest the experimental economics field has come to allowing deception is surprising participants with additional decisions when they initially believed the experiment had ended (e.g. Andreoni 1988; Croson 1996; Boles, Croson and Murnighan 2000; Croson, Boles and Murnighan (2003)). Note that this does not involve deception by commission; participants are never told anything that is not true. Instead, it involves deception by omission; participants are not told everything about the experiment when they begin. For psychologists who want their work to be read and accepted by economists, deception will prove to be a difficult barrier to overcome. If it’s simply a matter of saving money, economists react very skeptically to experiments that use deception. If the deception is necessary for some other reason, an explanation and justification will be needed in order to have the data taken seriously.
Experimental Procedures These last two topics differ from the previous four. Here, I describe a collection of practices in the implementation of economics experiments, rather than globally-accepted wisdom about experimental design. Many of these procedures are commonly used both psychology and economics. Experimental projects begin with an application to one’s human subject committee. Experimental economists tend to have relatively little trouble with human subjects committees, as many of the usual “flags” are not present in their designs. The experiments do not involve deception, there are rarely any consequences other than financial, and participants earn money for their participation. Little or no demographic data is collected, and participants are paid privately (see below), reducing the impact of social comparisons on earnings. Thus many projects receive expedited review and quick acceptance. The most troubling constraint for experimental economists imposed by human subject committees is guaranteeing positive earnings. Economists often want to explore decision-making in the face of losses rather than gains, but human subject committees (and other concerns) prevent us from taking money from our participants. There are a few alternative responses which have been
THE METHOD OF EXPERIMENTAL ECONOMICS
299
successful at various institutions. The easiest is to pay a large “show-up-fee” and then have losses from the experiment deduced from it. While it is not entirely clear that this induces a “loss frame” in participants, it is usually acceptable to human subject committees. One human subject committee has allowed researchers to take money from participants provided they agree to the risks they were to take. Another allows participants to “work off ” any debts they incur by photocopying articles at an appropriate wage rate. These are, however, the exception rather than the rule, and often experimental economists are called to construct creative experimental designs to both induce the appropriate incentives to test the theory and ensure the participants will all make money. A second set of concerns involves interactions with the participants. Economics experiments tend to be run with groups of participants rather than one at a time. And since there are no confederates, when the participants arrive in the lab there is often a period of waiting for their counterparts to arrive. Providing newspapers, magazines and an internet connection helps participants pass the time quietly, and without external discussion. During the experiment, instructions are read aloud while participants follow along silently. This is useful for three reasons. First, it ensures that participants have been exposed to the instructions, and have not simply skipped them and started the experiment. Second, it creates common knowledge, one of the theoretical conditions necessary to test many economic theories. This means not simply that everyone knows the game they will play, but that everyone knows that everyone knows the game, and that everyone knows that everyone knows that everyone knows . . . Reading the instructions out loud also serves a third purpose – it reassures the participants that everyone in the room has indeed been given the same instructions and reduces suspicion of deception. Instructions often include examples to help participants keep track of financial transactions. However, examples have the opportunity to induce demand effects – participants use the example as a signal of what the experimenter would like them to do. There are a few solutions to this. First, one can construct examples that are far outside the possible range of behavior. For example, when deciding how much to allocate to a public good out of 20 tokens, the examples can involve allocating 5000 tokens in some manner. Second, one can use variables like X and Y to substitute for numbers in the examples. This works well with mathematically sophisticated participants, but is often confusing for others. My personal preference is to use a quiz rather than examples. As in examples, it is important to make quizzes as unbiased as possible, either by using
300
RACHEL CROSON
different scales in the quiz and in the experiment, by making the quiz abstract rather than concrete, or by allowing participants to fill in their own numbers for their decisions. Then we ask participants to calculate their earnings (and sometimes, their counterpart’s earnings). Experimental monitors circulate throughout the room and check the answers, correcting and explaining for anyone who completed it incorrectly. A second issue that arises during experiments is how to deal with participant questions. Especially while reading instructions aloud, participants will often have questions about the procedures. These questions you would like to have asked and answered publicly. However, other participants ask leading or contaminating questions (e.g. “why don’t we just all cooperate so we can earn more money?”). These are attempts to influence the actions of others disguised as questions, and these the experimenter doesn’t want asked and answered publicly. One tactic I find useful is to ask those with questions to raise their hand. The experimenter can then go to the participant and hear the question privately. If the question is the type that should be publicly addressed, the experimenter can repeat the question and answer it publicly. If not, the question can be answered privately. A final issue that often arises during experiments involves randomization. While it is now technologically very easy to use computers to generate random numbers, participants are often skeptical about the unbiasedness of this method. Thus economics experiments typically use actual randomizing devices (e.g. dice as in Croson and Johnston 2000) to implement random outcomes. Once the experiment is over, participants receive their earnings (in cash) and are asked to sign a receipt before they leave. In many economics experiments it is important that these earnings be paid privately. Theories often assume that individuals value money absolutely, without social influence or comparative preferences. If we want to test the implications of such a theory, it is important to implement an experimental design where these other considerations are not present. Of course, one can also test the assumptions of such a theory, at which point knowing the earnings of others is an important design considerations. When payments are private, participants often want to know how well they did relative to others. I try hard to avoid answering that question, reminding them that earnings are private information. A final end-of-experiment consideration involves debriefing. As with our first topic (human subject committees) extensive debriefing is not typically needed when experiments involve no deception. Economists are concerned about communicating the nature of the experiment to participants when others will be coming to the lab later in the day or the week to participate
THE METHOD OF EXPERIMENTAL ECONOMICS
301
in another session or treatment. I usually ask participants to leave their names on a sign-up sheet if they are interested in a summary of the experimental objectives and results. I can then send them a writeup once all the data for the experiment is collected. This sheet also provides a vehicle to collect the contact information of participants who would like to participate in further experiments. The procedures described in this section are by no means critical for psychologists (or even economists) to use in their implementations. No economist will object to a psychological experiment on the basis of the debriefing procedure used, for example. Hopefully, however, this discussion will provide some advice and insight into the nitty-gritty of running economics experiments and perhaps even some ideas for psychologists to use.
Data Analysis A final methodological dimension along which the fields differ is the data analyses they employ. In my publications in both economics and psychology/management journals, I have found referees and editors surprisingly parochial about their favored statistical methods. Both sides use nonparametric statistics, although economists favor Wilcoxon tests (Mann-Whitney U tests) after an influential paper demonstrating its power in ultimatum and dictator game data (Forsythe et al. 1994).6 In contrast psychologists tend to use chi-squared and other tests that are appropriate for discrete data. Once the decision is made to move to parametric analysis, economists use regressions and psychologists ANOVAs. I have observed both these choices even when not appropriate – economists regularly use regressions even when the independent variables are all discrete and psychologists regularly discretize their continuous independent variables in order to be able to use ANOVAs. These two techniques are, of course, quite related to each other and I have never experienced a situation where they yield qualitatively different responses. My interpretation of the source of these differences has to do with the reference disciplines to which the various experimentalists are trying to speak. For example, empirical economists use regressions; thus experimental economists use regression techniques to convince them that experimental results are valid. A second methodological difference in analysis (and experimental design) is the concept of interactions. Economists are often uncomfortable with interaction effects (especially three-way or four-way interactions) while for psychologists interactions are the bread-and-butter of publication. My
302
RACHEL CROSON
interpretation of the source of these differences involves the different objectives of the groups. In psychology, a result’s cause can be understood only when one can make the result disappear. Thus interaction effects – treatments where the result is present and other treatments when it is absent (“now you see it, now you don’t”) illuminate the underlying cause of the result. In economics, the impulse is less to explore the cause of a given result than to explore its implications. Therefore the focus on interaction effects is reduced. A final methodological difference concerns the ex-ante and ex-post hypotheses. This distinction is central in economics experiments. In part this stems from the existence of a core (deductive) theory which experiments test and from which ex-ante hypotheses can be easily drawn. In contrast, Kerr (1998) describes a survey in which 156 researchers in social psychology, clinical psychology and sociology were asked how frequently they had personally observed some form of hypothesizing after the results are known (HARKing) in their professional life. Positive responses ranged from 32% to 48%. Respondents also advised using empirical inspiration for hypothesis generation 55% of the time, a similar rate as they advised the traditional ex-ante hypothesis testing methodology. Although no similar data exists among experimental economists, my interpretation is that this behavior occurs much less frequently, in part because of the existence of the unified body of theory. A hypothesis described as exante which contradicted this core theory would be extremely suspicions. Second, an important contribution of experiments is seen in the profession to be highlighting the shortcomings and omissions of this core theory. Thus disproving the “hypothesis” is a perfectly acceptable and often sought-after outcome. Kerr (1998) presents an outstanding discussion of HARKing and its costs and benefits for individuals and the profession.
Conclusion There are many similarities between experimental economics and psychological research. This article, however, was designed to illuminate some of the differences. I have discussed five methodological areas where experimental economists and experimental psychologists differ and have speculated as to the cause of those differences in each area. As I hope I have conveyed in this article, there are no right and wrong answers. Each researcher needs to make their own methodological decisions based on the objectives of their experiment, the methods currently used in their field, and the audience they wish to address. My hope is that this article has described the methodology used in experimental economics to a non-economist audience and offered some insight into which of these
THE METHOD OF EXPERIMENTAL ECONOMICS
303
considerations are viewed as non-negotiable (e.g. induced valuation) and which others may be relaxed (e.g. context). I have also offered some advice to psychologists who are interested in having their work read, cited and believed by economists. That said, one might imagine the parallel article to this one; aimed at experimental economists and specifying what they need to do in order to have psychologists take their work seriously. My topic headings for that article would include context (adding more), interaction effects and mediation/moderation analyses as methods to eliminate alternative explanations for demonstrated phenomena. I believe the questions that psychologists and economists are asking in their research have the potential to inform each other’s fields deeply and profoundly. Understanding and accepting methodological differences is an important first step toward generating surplus-creating gains from trade, and I hope this article will help in taking that first step.
Notes 1. This article will be discussing the methodology of experimental economics and providing comparisons with that of experimental psychology. I mean the latter term to apply quite broadly and include research in management, dispute resolution and other related fields that use experimental data and experimental psychological methods. 2. While many (perhaps almost all) economists acknowledge that the theory does not predict and explain economic outcomes in all settings, the general consensus is that the theory does exceptionally well in predicting and explaining in a large variety of settings. Furthermore, there are costs and benefits of making the theory more complex. The benefits are clear, by adding extra parameters you can increase the predictive ability of the theory (e.g. by adding a parameter for the status-quo point around which the utility [value] function is asymmetric you can capture different risk preferences in gains and losses, or by adding others’ payoffs into one’s own utility function you can capture social factors like altruism, envy and inequality-aversion). However, there are also costs to adding extra parameters. The parameters need to be estimated in any given situation, and when this is not possible the expanded theory makes no predictions. So when an expanded theory makes predictions they are more often correct, but there are fewer cases where any predictions are possible. Furthermore, the expanded theory is more complicated to work with and may not provide as many insights or unexpected predictions. Psychologists and economists have made different choices in this tradeoff between simplicity and descriptive ability. 3. The way to implement an infinitely repeated game in the lab is to derive the theory’s predictions under a discount rate of d, then implement the discount rate by setting the probability of the game ending in any given period equal to (1-d). 4. The way to implement a one-shot game is through a strangers or ‘zipper’ design where each participant meets each other participant at most once during the experiment. (Kamecke (1997), Andreoni (1988), Croson (1996b), Andreoni and Croson (forthcoming)).
304
RACHEL CROSON
5. Shafir and Tversky (1992) did include some studies on the prisoner’s dilemma game in which participants were paid for their earnings. Morris, Sim and Girotto (1998) ran similar experiments with a flat fee. 6. This paper also demonstrated significant differences between behavior in paid and unpaid ultimatum and dictator games.
References Andreoni, James (1988). “Why Free Ride? Strategies and Learning in Public Goods Experiments.” Journal of Public Economics, 37: 291–304. Andreoni, James and Croson, Rachel (forthcoming). “Partners versus Strangers: Random Rematching in Public Goods Experiments,” in Plott, C. and Smith V., editors, Handbook of Experimental Results. New York: Elsevier. Binmore, Kenneth (1987). “Experimental Economics.” European Economic Review, 31, 1/2: 257–264. Binmore, Kenneth (1999). “Why Experiment in Economics?” Economic Journal, 109, 453: F16–24. Boles, Terry; Croson, Rachel; Murnighan, Keith (2000). “Deception, Retribution, and the Negotiation Process.” Organizational Behavior and Human Decision Processes, 83: 235–259. Bonetti, Shane (1998a). “Experimental Economics and Deception.” Journal of Economic Psychology, 19, 3: 377–395. Bonetti, Shane (1998b). “Experimental Economics and Deception: Response.” Journal of Economic Psychology, 19, 3: 411–414. Boyce, Rebecca; Brown, Thomas, McClelland, Gary, Peterson, George and Schulze, William (1992). “An Experimental Examination of Intrinsic Values as a Source of the WTA-WTP Disparity.” American Economic Review, 82, 5: 1366–1373. Buchan, Nancy; Croson, Rachel; Dawes, Robyn (2002). “Swift Neighbors and Persistent Strangers: A Cross-Cultural Investigation of Trust and Reciprocity in Social Exchange.” American Journal of Sociology, 108: 168–206. Camerer, Colin; Hogarth, Robin (1999). “The effects of financial incentives in experiments: A review and capital-labor production theory.” Journal of Risk and Uncertainty 19, 1–3: 7–42. Carter, John; Irons, Michael (1991). “Are Economists Different, and If So, Why?” Journal of Economic Perspectives, 5: 171–177. Cox, James; Isaac, Mark R. (1986). “Experimental Economics and Experimental Psychology: Ever the Twain Shall Meet?” in Economic Psychology: Intersections in Theory and Application, MacFadyen and MacFadyen, editors, North-Holland, Amsterdam: 647–669. Croson, Rachel (1996a). “Information in Ultimatum Games: An Experimental Study.” Journal of Economic Behavior and Organization, 30: 197–212. Croson, Rachel (1996b). “Partners and Strangers Revisited.” Economics Letters, 53: 25–32. Croson, Rachel (1999). “The Disjunction Effect and Reason-Based Choice in Games.” Organizational Behavior and Human Decision Processes, 80: 118–133. Croson, Rachel (2000). “Thinking like a Game Theorist: Factors Affecting the Frequency of Equilibrium Play.” Journal of Economic Behavior and Organization, 41: 299–314. Croson, Rachel (2002). “Why (and How To) Experiment: Methodologies from Experimental Economics.” University of Illinois Law Review, 2002: 921–945. Croson, Rachel; Boles, Terry; Murnighan Keith (2003). “Cheap Talk in Bargaining Experiments: Lying and Threats in Ultimatum Games.” Journal of Economic Behavior and Organization, 51: 143–159.
THE METHOD OF EXPERIMENTAL ECONOMICS
305
Croson, Rachel; Gneezy, Uri (2005). “Gender Differences in Preferences.” Working Paper, Wharton School University of Pennsylvania. Croson, Rachel; Johnston, Jason (2000). “Experimental Results on Bargaining under Alternative Property Rights Regimes.” Journal of Law, Economics and Organization, 16: 50–73. Croson, Rachel; Mnookin, Robert. (1997). “Does Disputing through Agents Enhance Cooperation? Experimental Evidence.” Journal of Legal Studies, 26: 331–345. Cummings, Ronald; Holt, Charles; Laury, Susan (2002). “The Georgia Irrigation Reduction Auction.” Working Paper, Georgia State University, Department of Economics. Dyer, Douglas; Kagel, John; Levin, Dan (1989). “Comparison of Naive and Experienced Bidders in Common Value Offer Auctions: A Laboratory Analysis.” Economic Journal, 99, 394: 108–115. Dyer, Douglas; Kagel, John (1996). “Bidding in common value auctions: How the commercial construction industry corrects for the winner’s curse.” Management Science, 42, 10: 1463–1476. Eckel, Catherine; Grossman, Phillip (forthcoming a). “Volunteers and Pseudo-Volunteers: The Effect of Recruitment Method on Subjects’ Behavior in Experiments,” Experimental Economics. Eckel, Catherine; Grossman, Phillip (forthcoming b). “Differences in the Economic Decisions of Men and Women: Experimental Evidence,” In Plott, C. and Smith, V. Handbook of Experimental Results. New York: Elsevier. Forsythe, Robert; Horowitz, Joel; Savin, N.E.; Sefton, Martin (1994). “Fairness in Simple Bargaining Experiments.” Games and Economic Behavior, 6: 347–369. Frank, Robert; Gilovich, Thomas; Regan, Dennis (1993). “Does studying economics inhibit cooperation?” Journal of Economic Perspectives 7: 159–171. Frank, Robert; Gilovich, Thomas; Regan, Dennis (1996). “Do economists make bad citizens?” Journal of Economic Perspectives, 10: 187–192. Friedman, Daniel; Sunder Shyam (1994). Experimental Methods: A Primer for Economists. New York: Cambridge University Press. Gneezy, U., and A. Rustichini (2000). “Pay Enough or Don’t Pay At All.” Quarterly Journal of Economics, 791–810. Hertwig, Ralph; Andreas Ortmann (2001). “Experimental Practices in Economics: A Methodological Challenge for Psychologists?” Behavior and Brain Sciences, 24, 383–451. Hey, John (1998). “Experimental Economics and Deception: A Comment.” Journal of Economic Psychology, 19, 3: 397–401. Holt, Charles (1995). “Psychology and Economics: A Discussion of Methodology.” University of Virginia, Discussion Paper. Kamecke, Ulrich (1997). “Rotations: Matching Schemes That Efficiently Preserve the Best Reply Structure of a One Shot Game: Note.” International Journal of Game Theory, 26, 3: 409–417. Laband, David; Biel, Richard (1999). “Are Economists More Selfish than Other ‘Social’ Scientists?” Public Choice, 100, 1–2: 85–101. Loewenstein, George (1990). “Experimental Economics from the Vantage-Point of Behavioral Economics.” Economic Journal, 109, F25-F34. Marwell, Gerald; Ames, Ruth (1981). “Economists Free Ride, Does Anyone Else?: Experiments on the Provision of Public Goods IV.” Journal of Public Economics, 15: 295–310. McDaniel, Tanga; Starmer, Chris (1998). “Experimental Economics and Deception: A Comment.” Journal of Economic Psychology, 19, 3: 403–409. Morris, Michael; Sim, Damien; V. Girotto (1998). “Distinguishing Sources of Cooperation in
306
RACHEL CROSON
the One-Round Prisoner’s Dilemma: Evidence for Cooperation Decisions Based on the Illusion of Control.” Journal of Experimental Social Psychology, 34, 494–512. Ortmann, Andreas; Hertwig, Ralph (1997). “Is Deception Acceptable?” American Psychologist, 52, 7: 746–747. Ortmann, Andreas; Hertwig, Ralph (1998). “The Question Remains: Is Deception Acceptable?” American Psychologist, 53, 7: 806–807. Ortmann, Andreas; Hertwig, Ralph (2002). “The Costs of Deception: Evidence from Psychology.” Experimental Economics, 5: 111–131. Plott, Charles (1982). “Industrial Organization Theory and Experimental Economics.” Journal of Economic Literature, 20: 1485–1527. Plott, Charles (1991a). “Will Economics Become an Experimental Science?” Southern Economic Journal, 57, 4: 901–919. Plott, Charles (1991b). “Economics in 2090: The Views of an Experimentalist.” Economic Journal, 101, 404: 88–93. Plott, Charles (1994). “Market Architectures, Institutional Landscapes and Testbed Experiments.” Economic Theory, 4: 3–10. Roth, Alvin (1986). “Laboratory Experimentation in Economics.” Economics & Philosophy, 2, 2: 245–273. Roth, Alvin (1988). “Laboratory Experimentation in Economics: A Methodological Overview.” The Economic Journal, 98, 393: 974–1031. Roth, Alvin (1991). “Game Theory as a Part of Empirical Economics.” Economic Journal, 101, 404: 107–114. Roth, Alvin (1994). “Let’s Keep the Con out of Experimental Econ: A Methodological Note.” Empirical Economics, 19, 2: 279–289. Shafir, Eldar and Tversky, Amos (1992). “Thinking through uncertainty: Nonconsequential reasoning and choice.” Cognitive Psychology, 24, 4: 449–474. Smith, Vernon (1976). “Experimental Economics: Induced Value Theory.” American Economic Review, 66, 2: 274–279. Smith, Vernon (1989). “Theory, Experiment and Economics.” Journal of Economic Perspectives, 3, 1: 151–169. Smith, Vernon (1994). “Economics in the Laboratory.” Journal of Economic Perspectives, 8, 1: 113–131. Smith, Vernon and Walker, James (1993). “Monetary Rewards and Decision Cost in Experimental Economics.” Economic Inquiry, 31, 2: 245–261. Tversky, Amos and Shafir, Eldar (1992). “The Disjunction Effect in Choice under Uncertainty.” Psychological Science, 3: 305–309. Yezer, Tony; Goldfarb, Robert; and Poppen, Paul (1996). “Does Studying Economics Discourage Cooperation?: Watch What We Do, Not What We Say or How We Play.” Journal of Economic Perspectives, 10, 1: 171–186.
Legal Research on Negotiation REBECCA HOLLANDER-BLUMOFF
Lawyers and conflict are a classic combination. Lawyers are engaged in the resolution of conflict on a constant basis, whether they are negotiating settlements during litigation or crafting deals during corporate transactions. The study of legal conflict resolution is a natural area for legal scholarship to address. This article offers a sketch of the methodology used by legal scholars studying negotiation, describing the range of approaches used and suggesting some of the benefits of and obstacles to empirical work on legal negotiation. The first section provides a brief description of legal scholarship, including the role of empirical research. Knowledge of this broader framework helps in understanding the unique challenges to legal research on negotiation. The next section offers an overview of the treatment of negotiation by legal academics. The subsequent discussion highlights obstacles and challenges facing legal scholars, and others, who want to conduct empirical studies of legal negotiation. The conclusion discusses potential areas for future empirical exploration in the legal negotiation context.
The Nature of Legal Scholarship When law professors begin a traditional legal research project, they typically start – and finish – by reviewing legal opinions written by judges, statutes passed by Congress, and articles written by other law professors. Compiling the materials to conduct such research is a relatively easy task to accomplish using an online research engine like Lexis or Westlaw or using student research assistants who are well-versed in online and library document retrieval. The end goal of this kind of research is a well-written and thoughtfully argued law review article that offers the author’s theoretical vision of a legal issue to readers. The article may survey a vast collection of judicial opinions, statutory codes, and administrative regulations, or it may just focus on a handful of United States Supreme Court decisions, but it marshals legal authority to make its central argument. International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 307–322 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
308
REBECCA HOLLANDER-BLUMOFF
No empirical data are required – or even desired – most of the time.1 Empirical evidence lacks the force of law – so practitioners and legal scholars alike do not highlight it in the same way that they rely on controlling legal authority like statutes and cases. And even when “empirical” data is used, it is often merely presented to establish the factual existence of a situation – say, falling crime rates (Ehrenreich Brooks 2003: 2308) or the growing use of stock options (Dallas 2003: 791) – before the author moves on to discuss how the law does, or should, address the problem. These “prototypical” law review articles are sometimes thought of as a part of an ongoing dialogue about a theoretical framework or topic. For example, law and economics has been a popular theoretical framework for law professors in recent years, and articles discussing a problem from a law and economics perspective abound. Thus one might examine agreements to arbitrate employment disputes from a law and economics angle, showing why rational actors would only choose those agreements if they were economically beneficial and why the case law, therefore, is or is not correctly decided (Bodie 2004). That is not to say that the author’s own experience and thoughts are presented, transparently, as one person’s opinion and perspective. Rather, the classic law review article more often appears to be theoretical, or almost philosophical, in its normative approach to problems – an author presents (in the third person, almost always) a picture of the landscape of a certain topic, what others have said and thought about it, and finally the “way” that things should be seen. The author typically draws mostly on legal authorities in support of his or her propositions. Although it represents a small fraction of legal scholarship, there is, too, a more recent wave of empirically oriented articles. Typically, these have used research from other disciplines, like psychology and sociology, to buttress the “traditional” type of article discussed above. Findings from social science research are woven into, “sprinkled” onto, or even used as the basis for a more traditional law review article. Empirical findings may support the author’s premise, giving the author’s argument heft and making it more convincing. In this category, for example, Birke and Fox (1999) use data gathered by psychologists to describe potential obstacles to the rational resolution of legal disputes. Birke and Fox discuss how established psychological principles like anchoring and positive illusions might affect lawyers as they approach litigation. In a similar vein, Korobkin and Guthrie (2004) describe how common decision-making heuristics and biases may impact legal negotiation. Social science conclusions may form the basis for the thesis of an article, or may just inform a small part of a broader analysis.
LEGAL RESEARCH ON NEGOTIATION
309
The use of social science conclusions is not restricted to one area of legal scholarship. Areas as diverse as negotiation, litigation, general attorney advocacy, legislative rule-making, and judicial rule-making have been examined through the lens of empirical data. For example, Gary Blasi uses principles from social psychology to animate his discussion of attorney advocacy in light of human prejudice, citing the work of social psychologist John Bargh to discuss the priming effects of stereotypes on behavior (Blasi 2002). Blasi suggests that attorneys should take into account such effects when crafting their courtroom strategies. In recent years, a number of law review articles have begun to highlight primary data collected for the purpose of the article. Such primary data takes the form of empirical (both quantitative and qualitative) studies of things like settlement agreements, legal opinions, monetary outcomes, particular disputes, or the behavior of people engaged in negotiating. A number of legal scholars, including Ayres (1991, 1995), Kim (1997), Korobkin (1998a), Guthrie (Guthrie 1999) (Korobkin & Guthrie 1994, 1997), and Rachlinski (1996), have collected original data and used it as the centerpiece of a law review article in recent years. These writers offer useful illustrations of the potential for legal scholarship that includes primary empirical research, especially in the negotiation arena. For example, Korobkin (1998a) argues that the status quo bias extends to default contract terms. His argument is supported not just by a thorough discussion of previous social science research about the status quo bias but also by a tailored experiment of his own design that specifically addresses parties’ preferences for default contract terms. His research suggests that contracting parties view default contract terms as part of the status quo and therefore prefer those terms. In a subsequent article, Korobkin (1998b) uses his data to further develop a theory of contract negotiations called the “inertia theory.” The burgeoning field of “behavioral law and economics,” (Jolls, Sunstein & Thaler 1998) which adds “real” human behavior into a rational economic actor model, has helped such work gain popularity, prominence, and acceptance, but empirical work in the social science vein still accounts for a tiny minority of all articles published. Legal academia’s institutional biases in favor of normative theoretical analysis over empirical research pose a challenge to the study of legal negotiation. Very little case law or statutes concern legal negotiation, and thus normative approaches to negotiation that are wholly theoretical do not have the weight of any traditional legal authority to support them. Yet clearly, with so many legal disputes coming to a non-adjudicated resolution, negotiation is a
310
REBECCA HOLLANDER-BLUMOFF
critical component of the current legal system, and legal scholars have increasingly turned their attention to the process of dispute resolution outside the courts. In studying how disputes are resolved through negotiation, then, legal academics face a core tension between the study of an important and pervasive phenomenon in their field and a dearth of traditionally favored authorities. Because legal negotiation is largely “extra-legal,” that is, governed by few hard-and-fast legal rules and occurring for the most part between individuals in private settings, it makes sense that the legal study of negotiation would rely in some part on empirical work. The traditionally marginal role of empirical work in law scholarship, as described above, thus poses particular challenges to legal academics who want to study negotiation. An overview of the legal scholarship on negotiation – with an eye towards how legal academics are approaching this challenge – follows below.
Law Review Articles about Legal Negotiation The study of legal negotiation has been a growth industry over the past two decades. During the past ten years there have been 211 articles published in law reviews and legal journals with the word “negotiation” in their titles; in the ten years before that, there were 70.2 The Negotiation Journal, the Journal on Dispute Resolution (University of Missouri), and the Ohio State Journal on Dispute Resolution were founded in the 1980s; I participated in the founding of the Harvard Negotiation Law Review in 1995.3 Law review articles on negotiation take many forms. Typically, articles have at least some descriptive elements (e.g., how legal negotiation does work) and some prescriptive elements (e.g., how legal negotiation should work – that is, how practitioners should negotiate or how scholars ought to conceptualize the legal negotiation process). The descriptive element may be the end goal of the work or it might merely provide a backdrop. Descriptive, however, need not mean empirical. In line with classic legal scholarship, one might examine legal negotiation from a particular theoretical framework. For example, Donald Gifford considers the implications of a client-centered model of lawyer behavior on legal negotiation (Gifford 1987), offering a normative vision of how attorneys should act towards clients during the negotiation process. Others have examined elements of negotiation from a law and economics perspective (Wachtler & Cohen 1988). Some develop a coherent theory of negotiation to help guide practitioners and scholars in thinking about the dynamics of negotiation. For
LEGAL RESEARCH ON NEGOTIATION
311
example, Korobkin (2000) has developed what he calls a “positive theory” of negotiation, suggesting that all negotiation behavior can be understood in light of a dichotomy between the definition of a bargaining zone (“zone definition”) and the allocation of bargaining surplus (“surplus allocation”). Some legal academics examine legal negotiation through a conceptual lens: for example, in recent years a number of legal scholars have explored the role of the apology in negotiation. Articles about apology have ranged from an examination of the strategic use of apology to the potential impact on litigation of an apology (Cohen 1999; Latif 2001; Bartels 2000/2001). Similarly, many law review articles on negotiation explore a unique context – say, negotiation in a labor setting (Dau-Schmidt 1992), or negotiation in the context of a hostile corporate takeover (Subramanian 2003; Oesterle 1986), or negotiation over the sale of a car (Ayres 1991). Indeed, some commentators suggest that any discussion of legal negotiation must be context-specific for it to be meaningful (Neumann & Krieger 2003). Often these articles examine the implications of legal rules or endowments on negotiation in an individual context. Robert Mnookin and Lewis Kornhauser’s (1979) seminal article on the effect of legal endowments on negotiation in the divorce setting has spawned a veritable cottage industry of such articles. To name just a few diverse examples, authors have looked at how negotiation is affected by legal entitlements surrounding administrative rule-making (Rossi 2001), the death penalty (Hoffman & Kahn 2001), international trade (Busch & Reinhardt 2000), and internet transactions (Katsh, Rifkin & Gaitenby 2000). Some legal academics approach the study of negotiation with an expressly prescriptive orientation. Some offer broad suggestions for approaching negotiation: Roger Fisher is renowned for his writing about solutions to common negotiation pitfalls (Fisher, Ury & Patton 1991; Fisher & Jackson 1993), and helped to draw legal academics’ (and many others’) attention to the study of negotiation with his seminal book of negotiation theory and advice “Getting to Yes.” More recently, in “Beyond Winning,” legal negotiation scholars Robert Mnookin, Scott Peppet and Andrew Tulumello (2000) identified three crucial tensions in negotiation – between creating and disitributing value, between empathy and assertiveness, and between principals and agents – and offered sophisticated advice for managing them. Others suggest Robert Adler and Elliot Silverstein (2000) took a prescriptive tack in their survey of the problems posed by power differentials in negotiation, offering potential solutions for approaching negotiations in which there is a power imbalance. The works discussed above are typically grounded in the author’s own experience and thoughts, as well as a review of relevant literature in the field. There are also, however, more expressly empirical articles about legal negotiation.
312
REBECCA HOLLANDER-BLUMOFF
Empirical work about negotiation typically takes the form of surveys to legal practitioners or experimental or quasi-experimental studies using law students or other subjects.4 For example, a recent article by Andrea Kupfer Schneider (2002) highlights her empirical survey work on the effectiveness of different negotiation styles, concluding that hard bargaining is perceived as less effective than problem-solving negotiation tactics. Similarly, Janice Nadler (2004) recently conducted empirical research on e-mail negotiations, concluding that negotiators who engaged in small talk were more likely to reach agreement rather than impasse, and were less angry and more respectful of their adversary in negotiation, than those who did not engage in such small talk. Linda Babcock and Greg Pogarsky (2001, 1999), too, have explored negotiation empirically in their studies of the effects of damage cap rules on settlement negotiation, finding that damage caps act as anchors for negotiated outcomes. As noted earlier, a number of other legal scholars also employ empirical research in their work on legal negotiation, and the field of empirical research into legal negotiation continues to grow. The work surveyed above describes the state of largely mainstream law review and law journal literature on negotiation; for reasons described both above in both the preceding and the following sections, some legal academics who research legal negotiation may choose to publish their work in more specialized journals (Dickinson 2004), non-legal journals (Babcock and Loewenstein 1997), or even non-academic publications. Still others who study legal negotiation empirically may include psychologists, sociologists, or business school academics whose works are published outside of law reviews and legal journals and are thus less accessible to the legal community.
Obstacles to Researching Legal Negotiation Law differs from psychology and other disciplines that study conflict, like sociology or anthropology, because law is not expressly about the study of behavior. Law happens irrespective of whether anyone wants to do research into the practice of law; that is, people sue each other, negotiate contracts and settlements, and generally speaking engage in legal process all the time, regardless of whether anyone bothers to study any of it. While human behavior, too, happens whether or not psychologists bother to observe and study it, psychology as a discipline is expressly about the study of human behavior. The field of law, in contrast, encompasses both legal practice itself (by practitioners) and the study of the legal rules and legal practice (by legal academics). This makes it especially ironic that legal research on negotiation often
LEGAL RESEARCH ON NEGOTIATION
313
does not delve into the substance of real behavior. I attribute the dearth of such research to two different sources: first, obstacles within the legal academy, which I call institutional obstacles, and second, obstacles stemming from studying the community of legal negotiators, which I term methodological obstacles. Institutional obstacles A number of obstacles make empirical work in legal academia difficult. First, many legal academics simply lack training in empirical research. Law schools do not offer courses in research methods or data analysis; people who choose law as a career, too, do not tend to have undergraduate training in these areas. Therefore, many law professors have not received systematic training in quantitative analysis of any sort. Because legal academics are not trained in a tradition of empirical research, they are less likely to approach the problems in their field with an eye towards empirical analysis. Even when empirical research does seem to be called for, law professors rarely have the physical and human resources that may be necessary to conduct such research. Sample size, rules of inference, experimental method, and data analysis, not to mention the legwork, time, and space necessary to actually conduct research involving real people – all present challenges that law school does not train its graduates to meet. Additionally, empiricism is not always highly valued within the legal academy. Epstein and King have argued convincingly that the legal academy is not currently suited to produce sound empirical work – and that the bulk of the empirical work that it does produce is not, in fact, sound (Epstein & King 2002). Epstein and King noted that every single article that they reviewed in their own survey of empirical legal work in the decade from 1990 to 2000 violated at least one statistical, methodological, or inferential rule. Epstein and King identified problems with legal empirical work stemming from issues including, but not limited to, validity, reliability, replicability, measurement, and selection bias. But the problem with empirical work is not just on the production end. The primary outlet for work produced by legal scholars is law reviews, and as has been duly noted (Epstein & King 2002), these are publications in which students select the articles. As discussed above, these students receive no training in law school that enables them to understand or evaluate empirical work. And as long as empirical work is perceived as outside the norm of legal scholarship, student editors are choosing “risky” pieces when they select an article that is based on empirical work over a traditional law review article.5
314
REBECCA HOLLANDER-BLUMOFF
Additionally, the way that negotiation may be taught as a subject in law schools can sometimes send a message to students that rigorous intellectual study of negotiation is unwarranted. Many law schools, for example, hire individuals who are primarily practicing attorneys to teach negotiation classes. These practitioners may not be familiar with academic literature on negotiation and may rely largely on anecdotes to convey their negotiation lessons to students. Students who receive the message from their law schools that negotiation is a “soft” subject in which sophisticated intellectual analysis is not required may be less likely to select negotiation articles to publish in their law reviews. There lingers, too, a basic mistrust of empirical data by the audience for such articles – typically, other legal academics. Legal scholars believe they are students of social behavior – but they are not scientists. (Moe 2002: 373). They are not, as a rule, well-versed in statistical analysis or the interpretation of data. Ironically, an untested theoretical idea can be more convincing to a legal reader than the same idea supported by an empirical study. The study can be picked apart, its methodology criticized, its finding limited and narrowed until it only applies to the particular setting in which it occurred – while the theoretical idea need only hang together in the intellectual ether.6 Lawyers are, in fact, trained to use factual differences between cases to distinguish one case from another in order to develop the strongest argument for the side they represent. Unsurprisingly, then, legal scholars use the same process of analysis to dissect empirical work – and are often left, after such a process, with distinguishable details rather than generalizable principles. Methodological obstacles The structure of legal practice poses unique challenges for the researcher interested in field work. Practitioners are extremely busy and getting them to agree to participate in a study, by allowing someone to interview, survey, or observe them, can be difficult. Nonetheless, many lawyers are delighted to tell an interested listener all about their experiences – that is to regale an audience with what lawyers call “war stories.” These can be great fodder for analysis – but it takes lots of time and effort to deconstruct interviews into useful explication for an audience. In the “war story” vein, I interviewed eleven defense attorneys and prosecutors in 1996 for a project on negotiation in the plea bargaining context (Hollander-Blumoff 1997). The goal of my project was to examine some of the classic “Getting to Yes” negotiation principles to see whether they were relevant and applicable to the setting of plea bargaining. Most of the lawyers
LEGAL RESEARCH ON NEGOTIATION
315
that I interviewed allowed me to audiotape my interviews. My end result was hundreds of pages of transcriptions and notes that were rich with potential insights – but difficult to analyze in a systematic rather than an anecdotal way. This was ethnographic-type data that was not susceptible to statistical analysis or any rules of inference. The lawyers I spoke with were all very willing (and eager) to tell me about their experiences. Asking the lawyers the same questions gave me a good basis for comparison and a sense for which negotiation principles were often used and which were less often used. But there were, of course, drawbacks to this interview process that highlight some of the methodological obstacles to research about legal negotiation. There are several methodological issues that are unique to the study of legal practice, which I address in the first section below. There are several other methodological concerns that are familiar to anyone who has taken a basic research methods class, and they are outlined below as well. These more basic methodological issues are of concern to any evaluator – but are, perhaps, more troubling to legal academics because they are not typically trained to assess these factors. To better describe some of the methodological challenges of researching legal negotiation, I offer some illustrations from the research project described above. I begin with what I think is the biggest challenge and the one most specific to the legal context: the problem of confidentiality and privilege. 1. Confidentiality/Privilege The attorney-client privilege protects the relationship between lawyers and clients. It is likely to be very difficult for researchers to gain access to the critical interactions between principal and agent without waiving that privilege. Lawyers are by nature a conservative and risk-averse lot, and there is not much reason for most practicing attorneys to let anyone else get close to their client. If anything goes awry, could they be opening themselves up to malpractice liability? Could a client claim that the presence of a researcher had a negative impact on the representation provided by the attorney? And could a researcher be made to testify about the lawyer-client interactions, either in the context of a malpractice suit, or, perhaps worse, in the context of the original legal matter? Not only might a researcher’s presence at a client meeting be construed as a waiver of the attorney-client privilege, but sharing confidential information outside of the relationship could be detrimental to the relationship itself. Indeed, some attorneys are actually bound in what they can discuss by their office policies. For example, I interviewed lawyers working in the United States Attorney’s Office; this office has strict rules on who can talk to
316
REBECCA HOLLANDER-BLUMOFF
whom, and what the substance of the conversation can cover. In my own research, defense attorneys, bound only by their own clients and rules, rather than institutional guidelines, were far more voluble and far less circumspect in their storytelling. Given the restrictions on what attorneys may or will reveal about their clients and their clients’ problems, one might reasonably worry about the quality and completeness of any data collected. In order to uphold their professional responsibility, attorneys might have omitted important details of stories that could have compromised client confidentiality but that might have been important to an understanding of the problems they were describing. Lawyers simply cannot always give as much contextual information about a negotiation as would be helpful because of confidentiality issues. 2. Selection bias Lawyers who agree to discuss their cases with a researcher might practice law very differently than lawyers who are not inclined to participate in the research process. There was likely some selection bias among the lawyers who agreed to see me, for example – all were eager to help with my research and interested in academic endeavors. Lawyers who are not interested in academic study of and reflection on their behavior might somehow be different kinds of people with different styles and experiences of negotiating. My sample, too, was by no means random or unbiased. I was given several names by people that I already knew, and then used introductions by my original sources and others to network with other lawyers in search of individuals who would be willing to talk with me. 3. Internal Validity Likewise, the lawyers that I spoke with most certainly brought their own biases to their narratives and might have been less likely to tell me stories that portrayed their negotiation skills in a negative light. This effect might be exacerbated by the fact that, although the attorneys with whom I spoke remained anonymous in the final article, they were not anonymous to me as a researcher. As an attorney myself, I am part of the same greater community in which they work and this may have created incentives for how they presented themselves that could have skewed my results. As noted above, I networked with other attorneys to find my interviewees, and although I assured them of confidentiality, they may have been concerned about information getting back to their peers.
LEGAL RESEARCH ON NEGOTIATION
317
4. External Validity In my research described above, I sought to determine whether or not criminal plea bargaining could aptly be considered “negotiation” per se by exploring both the extant literature about plea bargaining and the real experiences of practitioners engaged in the process. I examined how well both what I read about the plea bargaining process and my empirical findings about plea bargaining fit within literature about the process of negotiation. My empirical research was unquestionably anecdotal. I did not have a sufficient sample size to draw any scientific conclusions about, say, how often prosecutors or defense attorneys considered their BATNA (best alternative to a negotiated agreement) or how relationships affected plea bargaining outcomes or processes. I intended my research to be anecdotal and illustrative, rather than statistically significant and dispositive in any scientific manner, so a critique of my work might fairly say that my results were not generalizable. Even a large scale study, though, might be susceptible to concerns about generalizability. For instance, Kupfer Schneider’s (2002) exploration of the effectiveness of negotiation style relied on a questionnaire sent to lawyers in Chicago and Milwaukee; her response rate overall was 29 percent, but almost 40 percent of the Milwaukee recipients responded, while only 18 percent of the Chicago recipients did. Proportionally, more women completed the survey. Perhaps geographical or gender issues affected her findings. Similarly, she asked attorneys to comment on a self-selected matter or transaction – there could have been confounding variables related to the underlying nature of the case selected that affected the results. As Korobkin (2001: 327) notes, “The great shortcoming of empirical observation is that it is difficult to generalize the findings of an empirical study to novel situations.” This concern is one that is shared by the legal academy. 5. Challenges to experimental or quasi-experimental studies Undeniably, field research in legal negotiation cannot offer any possibility for controlled experiments. Legal ethics (not to mention logistics) make assignment to a particular experimental condition impossible. Even in the context of analyzing non-experimental data, there is great potential for confounding variables – differences between individual clients, individual fact patterns, and legal jurisdictions – just to name a few factors. More controlled research into legal negotiation is possible, however. Currently, I am conducting research in which 200 simulated legal negotiations, conducted by first-year students at New York University law school, are videotaped. Additionally, all participants in the negotiations fill out a detailed survey immediately after completing the negotiation. The research centers on
318
REBECCA HOLLANDER-BLUMOFF
students’ behavior during the negotiations, their perceptions of fair procedures during the negotiation, and their desire to accept or reject the agreements afterwards. This research presents a tremendous opportunity because all 200 student dyads are participating in the same simulation, which means that we are able to measure outcomes in a way that is never possible in the real world. We can also examine student behavior and strategic choices with less worry over confounding variables because the fact pattern is a constant. As with many simulations, though, this project has limitations. These are students rather than experienced practitioners, so the applicability of the students’ behaviors to the “real” world of legal negotiation may be limited. Also, students may behave differently in simulations than they might in a “real life” experience. The “facts” of the situation, too, may produce unrealistic results. Students are assigned different “clients” who act the part, and this too may skew results. The research also presents more straightforward challenges: coding the behavior of individuals in a forty-five minute negotiation over fairly complex issues is laborious and time-intensive. In response to research like this, legal scholars tend to raise a real concern: are we learning more about what students think about how “real-live” lawyers behave than about how such lawyers would actually behave? Do years of experience in the law change the way that lawyers act so that the behavior of students reflects only that – the behavior of students? In my experience, most law students do approach the exercise with a remarkable degree of attention, care, effort, and professionalism, even if they do not approach the exercise in the same manner that a seasoned lawyer would. But the question still remains: how generalizable are the results? One might, perhaps reasonably, conclude that the results of such research tell us more about how individuals generally behave in negotiation, rather than how lawyers, in particular, do.
Areas for Potential Future Exploration Past research on negotiation in the law has stressed issues related to negotiating style; the effects of legal endowments; and cognitive biases, such as anchoring effects and reactive devaluation. My own research strikes out in a different direction. One line of research applies the principles of procedural justice to negotiation. Procedural justice literature has established that people place great importance on the procedures by which authority figures make decisions (Tyler & Lind 1992). When a third-party authority is involved, parties have been shown to be more satisfied with and accepting of outcomes when they are arrived at through the use of fair processes. Indeed, individu-
LEGAL RESEARCH ON NEGOTIATION
319
als may value features of the process above the actual outcome. The results of my work show that there is a robust procedural justice effect even when there is no third-party authority decision-maker. Perhaps even more interestingly, my research challenges the assumption that negotiators seeking to maximize gain ought to act adversarially in negotiation. The data show that those who are perceived as fair, and who perceive themselves as acting fairly, do not get any better or worse monetary outcomes than those perceived as less fair. There is, in fact, no correlation between the degree of perceived fairness and the outcome. But the degree of perceived fairness has powerful effects on how likely individuals are to accept, or to want to reject, the outcome. This data has the potential for powerful normative implications about appropriate and effective attorney behavior in negotiation. An intriguing concern about this research is how it relates to the attorney as an agent. Do the procedural justice effects transmit themselves to clients through the way in which lawyers “pitch” a certain outcome? That is, if an attorney feels a process is fair and is thus more ready to accept the resulting outcome, does that translate into a perception on the part of the client that the process is good or that the outcome is good? Does process between attorneys matter even when clients are not present? In essence, do clients care about process or not? This remains a largely unexplored area that is ripe for future study. Another line of my research explores how lawyers use “the law” in legal negotiations – that is, what differentiates the legal negotiation setting from any other type of negotiation? How does the use of legal endowments work practically during a negotiation? What effect does arguing the law, or not arguing the law, have on negotiation process and outcome? I have developed a coding scheme to analyze the data described above with respect to different uses of legal argument in negotiations and its effect on process, numerical outcome, and other outcome variables. Findings from such a study could be useful to practitioners and to academics alike, as we think prescriptively about what effective legal negotiation and able advocacy look like. It remains the great challenge for legal academics interested in negotiation research: how to extend social science methodology into a remarkably conservative, risk-averse field, so that we are able to amass more research on real attorneys and enable our empirical findings to become more generalizable. Empiricism has made inroads towards overcoming the institutional obstacles. As momentum favoring empirical work builds in the legal academy, scholars interested in legal negotiation can hope – and push – for similar receptivity by practitioners themselves that will help to erode some of the methodological barriers.
320
REBECCA HOLLANDER-BLUMOFF
Notes 1. I leave aside for this article the debate over what “empirical” means. First, a wide gulf separates qualitative from quantitative empiricism in the law; and then, there is hearty disagreement over what may be considered qualitative empirical data. For the purpose of this discussion, I define empirical work as that involving a systematic collection and analysis of data using social science methodology. 2. Lexis/Nexis search conducted April 30, 2004. 3. A number of articles about legal negotiation appear in journals that are not law reviews per se. For example, publications including Law and Social Inquiry, Law and Human Behavior, and the Journal of Legal Studies have included important work on legal negotiation. These publications, however, tend to be outside the “mainstream” of the most prestigious law reviews, and many legal academics are not familiar with the work of these journals; indeed, some do not even have access in any easy way to journals that are not within Lexis or Westlaw’s law review database. 4. Russell Korobkin (2002: 1038) identified four potential sources of empirical data on contract law: judicial opinions, studies of actual contracting practices of contracting parties, experimental studies of contracting behavior, and studies of contracting parties’ opinions. These are useful categories to consider in thinking about sources of data for studying legal negotiation. His first category is not relevant to the negotiation context, as very few opinions address the negotiation process. His second and fourth sources roughly correspond to survey research that asks real attorneys what they do as well as what they think about what they do, while the third source roughly corresponds with experimental work involving simulations. 5. There are several peer-reviewed law journals in specialized areas of the law; many do not emphasize empirical work, and in any event these journals generally have a smaller circulation and audience than more mainstream law reviews. 6. Korobkin raises a related point when he notes that “a single empirical study might be less fulfilling to the author than an article laying out a theory that seamlessly and completely resolves a legal problem on its own, with no questions left open” (Korobkin 2002: 1055).
References Adler, Robert S. & Silverstein, Elliot M. (2000). “When David meets Goliath: dealing with power differentials in negotiations.” Harvard Negotiation Law Review, 5: 1–112. Ayres, Ian (1991). “Fair driving: gender and race discrimination in retail car negotiations.” Harvard Law Review, 104: 817–872. Ayres, Ian (1995). “Further evidence of discrimination in new car negotiations and estimates of its cause.” Michigan Law Review, 94: 109–147. Babcock, Linda & Loewenstein, George (1997). “Explaining bargaining impasse: the role of self-serving biases.” Journal of Economic Perspectives, 11: 109–26. Babcock, Linda & Pogarsky, Greg (1999). “Damage caps and settlement: a behavioral approach.” Journal of Legal Studies, 28: 341–370. Bartels, William K. (2000). “Recent developments: the stormy sea of apologies: California Evidence Code Section 1160 provides a safe harbor for apologies made after accidents.” Western State University Law Review, 28: 141–157.
LEGAL RESEARCH ON NEGOTIATION
321
Birke, Richard & Fox, Craig R. (1999). “Psychological principles in negotiating civil settlements.” Harvard Negotiation Law Review, 4: 1–57. Blasi, Gary (2002). “Advocacy against the stereotype: lessons from cognitive social psychology.” UCLA Law Review, 49: 1241–1281. Bodie, Matthew T. (2004). “Questions about the efficiency of employment arbitration agreements.” Georgia Law Review, 39: 1–82. Busch, Marc L. & Reinhardt, Eric (2000). “Bargaining in the shadow of the law: early settlement in GATT/WTO Disputes.” Fordham International Law Journal, 24: 158–172. Cohen, Jonathan R. (1999). “Advising clients to apologize.” Southern California Law Review, 72: 1009–1069. Dallas, Lynne L. (2003). “The multiple roles of corporate boards of directors.” San Diego Law Review, 40: 781–820. Dau-Schmidt, Kenneth G. (1992). “A bargaining analysis of American labor law and the search for bargaining equity and industrial peace.” Michigan Law Review, 91: 419–514. Dickinson, David L. (2004). “A comparison of conventional, final-offer, and ‘combined’ arbitration for dispute resolution.” Industrial and Labor Relations Review, 57: 288–301. Ehrenreich Brooks, Rosa (2003). “The new imperialism: violence, norms, and the ‘rule of law.’” Michigan Law Review, 101: 2275–2340, 2308. Epstein, Lee & King, Gary (2002). “Empirical research and the goals of legal scholarship: The rules of inference.” University of Chicago Law Review, 69: 1–133. Fisher, Roger & Jackson, William (1993). “Teaching the skills of settlement.” Southern Methodist University Law Review, 46: 1985–1993. Fisher, Roger, Ury, William, & Patton, Bruce (1991). Getting to yes. New York, NY: Penguin Books. Gifford, Donald G. (1987). “The synthesis of legal counseling and negotiation models: preserving client-centered advocacy in the negotiation context.” UCLA Law Review, 34: 811–862. Guthrie, Chris (1999). “Better settle than sorry: The regret aversion theory of litigation behavior.” University of Illinois Law Review, 1999: 43–90. Hoffmann, Joseph L. & Kahn, Marcy L. (2001). “Plea bargaining in the shadow of death.” Fordham Law Review, 69: 2313–2391. Hollander-Blumoff, Rebecca (1997). “Getting to ‘guilty’: Plea bargaining as negotiation.” Harvard Negotiation Law Review, 2: 115–148. Jolls, Christine, Sunstein, Cass R., & Thaler, Richard (1998). “A behavioral approach to law and economics.” Stanford Law Review, 50: 1471–1550. Katsh, Ethan, Rifkin, Janet & Gaitenby, Alan (2000). “E-Commerce, e-disputes, and e-dispute resolution: In the shadow of ‘Ebay law.’” Ohio State Journal on Dispute Resolution, 15: 705–734. Kim, Pauline T. (1997). “Bargaining with imperfect information: a study of worker perceptions of legal protection in an at-will world.” Cornell Law Review 83: 105–156. Korobkin, Russell (1998a). “Inertia and preference in contract negotiation: The psychological power of default rules and form terms.” Vanderbilt Law Review 51: 1583–1651. Korobkin, Russell (1998b). “The Status quo bias and contract default rules.” Cornell Law Review 83: 608–687. Korobkin, Russell (2000). “A Positive Theory of Legal Negotiation.” Georgetown Law Journal, 88: 1789–1831. Korobkin, Russell (2001). “A multi-disciplinary approach to legal scholarship: economics, behavioral economics, and evolutionary psychology.” Jurimetrics Journal, 41: 319–336, 327. Korobkin, Russell (2002). “Empirical scholarship in contract law: possibilities and pitfalls.” University of Illinois Law Review, 2002: 1033–1066.
322
REBECCA HOLLANDER-BLUMOFF
Korobkin, Russell & Guthrie, Chris (1994). “Psychological barriers to litigation settlement: an experimental approach.” Michigan Law Review, 93: 107–192. Korobkin, Russell & Guthrie, Chris (1997). “Psychology, economics, and settlement: a new look at the role of the lawyer.” Texas Law Review, 76: 77–141. Korobkin, Russell & Guthrie, Chris (2004). “Heuristics and biases at the bargaining table.” Marquette Law Review, 87: 795–808. Latif, Elizabeth (2001). “Apologetic justice: evaluating apologies tailored toward legal solutions.” Boston University Law Review, 81: 289–320. Mnookin, Robert H. & Kornhauser, Lewis (1979). “Bargaining in the shadow of the law: the case of divorce.” Yale Law Journal 88: 950–997. Mnookin, Robert H., Peppet, Scott R., & Tulumello, Andrew (2000). Beyond Winning. Cambridge, MA: Belknap Press. Moe, Terry M. (2002). “Cynicism and political theory.” Cornell Law Review, 87: 362–383. Nadler, Janice (2004). Rapport in legal negotiation: how small talk can facilitate e-mail deal making.” Harvard Negotiation Law Review, 9: 223–251. Neumann, Richard & Krieger, Stefan (2003). “Empirical inquiry twenty-five years after the lawyering process.” Clinical Law Review, 10: 349–397. Oesterle, Dale A. (1986). “The negotiation model of tender offer defenses and the Delaware Supreme Court.” Cornell Law Review, 72: 117–157. Pogarsky, Greg & Babcock, Linda (2001). “Damage caps, motivated anchoring, and bargaining impasse.” Journal of Legal Studies, 30: 143–159. Rachlinski, Jeffrey J. (1996). “Gains, losses, and the psychology of litigation.” Southern California Law Review, 70: 113–184. Rossi, Jim (2001). “Bargaining in the shadow of administrative procedure: the public interest in rulemaking settlement.” Duke Law Journal, 51: 1015–1058. Schneider, Andrea Kupfer (2002). “Shattering negotiation myths: empirical evidence on the effectiveness of negotiation style.” Harvard Negotiation Law Review 7: 143–233. Subramanian, Guhan (2003). “Bargaining in the shadow of takeover defenses.” Yale Law Journal, 113: 621–686. Tyler, Tom R., & Lind, E. Allan (1992). “A relational model of authority in groups,” in Mark Zanna, editor, Advances in Experimental Social Psychology, 25: 115–191, New York: Academic Press. Wachter, Michael L. & Cohen, George M. (1988). “The law and economics of collective bargaining: an introduction and application to the problems of subcontracting, partial closure, and relocation.” University of Pennsylvania Law Review, 136: 1349–1417.
Methodologies for Studying Personality Processes in Interpersonal Conflict LAURI A. JENSEN-CAMPBELL and WILLIAM G. GRAZIANO
Interpersonal conflict is an inevitable part of life. In 1956, Muzafer Sherif noted “conflict . . . has no simple cause, nor is mankind yet in sight of a cure” (54). The inevitability of conflict stems from three apparently-panhuman psychological tendencies: 1) People differ in their attitudes, beliefs, knowledge, and life experiences; 2) Such differences induce people to be egocentric, and often to have difficulty perceiving the perspectives of others; and 3) People are generally motivated to protect and promote their own self-interests. Of course, people can be induced temporarily to take the perspectives of others, and to suppress briefly their own self-interests, but these brief golden moments seem to require special interventions, and when they do occur, to punctuate far longer, less enlightened periods. These three basic tendencies serve as catalysts for conflicts whenever people interact. For this perspective, conflict is an emergent property of relationships that appears during interaction between two or more persons. It is not primarily a characteristic of individual persons. There is, in fact, a growing consensus that mutual opposition or incompatibility is central to defining interpersonal conflict (Laursen & Collins 1994; Shantz & Hartup 1992). This mutual opposition, like many relationship interactions, is typically dyadic in nature, but may involve groups of persons. Interaction patterns present in certain relationships (e.g., parent/child) may be more likely to produce repeated conflicts than patterns in others (Furman & McQuaid 1992; Patterson 1982). Similarly, certain relationships may evoke different patterns of conflict resolution choices than do others (Jensen-Campbell & Graziano 2000). The emphasis on contexts and relationships is not a new concept. As early as the 1920s, Kurt Lewin was concerned with the impact on the situation on social behavior. Today, no doubt exists that situational factors contribute to all forms of social behavior, but the dominant view in social psychological theory in the 1970s and 1980s evolved toward an extreme situationist approach to human behavior at the expense to what an individual may bring to the situation (e.g., personality) (Cantor & Kihlstrom; Jones 1985; Mischel 1973).
International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 323–339 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
324
LAURI A. JENSEN-CAMPBELL AND WILLIAM G. GRAZIANO
Even recent conflict theories have focused mainly on the surrounding contexts involved in conflict (e.g., Shantz & Hartup 1992). Situational context provides only part of the conflict story. Given that interpersonal conflict is defined as two or more individuals opposing one another, personality differences should also contribute significantly to the dyadic relation of individuals. Here is one major source of individual differences in attitudes, beliefs, and life experiences discussed in the previous definition of conflict. It is too simplistic and atheoretical, however, to treat personality as a non-interacting set of individual differences. Instead, personality may be more usefully defined as structured system of individual differences organized to assist an individual person and his/her adaptation to the environment (Graziano 2003). For each individual, personality is a process summarizing the continuing negotiated compromises among competing motives and tendencies. One way to conceptualize these personality processes is in terms of levels of analysis. For example, McAdams (1995) theorized that individual differences in personality might be described at three different levels. Two of these levels may be particularly important for understanding personality and its influence within conflict situations. His first level consists of comparative dimensions of personality like the Big Five personality traits that are not conditional or dependent on the situation. In other words, Level I traits involve what a person “has” and brings to the conflict situation. The second level involves context-dependent strategies, plans, and concerns that enable individuals to resolve conflicts. In other words, Level II speaks to what a person “does” within a given conflict situation. These levels of analysis are not necessarily derivative of one another (McAdams 1995). Nonetheless, we suggest that there will be at least a loose linkage, between Level I traits and the organized systems of personal concerns within a conflict at Level II. When two people are involved in conflict, there is interdependence among effects due to the person, effects due to the partner, and effects due to the relationship and situation (Kenny & LaVoie 1984; Lewin 1935). Thus, understanding the influence of personality on social conflict must look past simple main effects models that focus on either situations or personality, per se. Here we start our paper by discussing various research methods that can be used to assess personality’s contribution to conflict behavior. We then focus on statistical advances that recognize the complex nature of interactions between individual, their partners, and the situation when studying conflict. The methods reported here are not intended as an exhaustive list for studying personality’s contribution to conflict, but will provide the reader with a starting point.
METHODOLOGIES FOR STUDYING PERSONALITY PROCESSES
325
Personality and Conflict Behavior We will also focus on agreeableness, a dimension in the Big Five factor model of personality, and its relation to interpersonal conflict to exemplify the utility of various methodologies. Obviously, there are many personality dimensions that are also worthy of study when discussing personality’s contribution to conflict behavior (De Dreu & Carnevale 2003; Terhune 1970). For example, conscientiousness is seen as an important factor in group living and concerns primarily matching behavior to performance standards (Digman & Inouye 1986; Graziano & Ward 1992). Indeed, meta-analytic results have revealed a positive relation between conscientiousness and performance and this finding was replicated in other cultures (e.g., Barrick & Mount 1991; Barrick, Mount & Judge 2001; Barrick, Stewart, Neubert & Mount, 1998; Salgado 1997). Although many personality characteristics may be related to negotiation and conflict behavior (e.g., Barry and Friedman, 1998), this article focuses on methodologies for studying personality processes in interpersonal conflict. Thus, providing an exhaustive list of personality characteristics that are related to interpersonal conflict processes remains beyond the scope of this article.
Hypothetical Situations/Role Playing A classic laboratory technique for studying conflict involves the hypothetical situations/role playing. Aronson and Carlsmith (1968) describe this methodology as an “ ‘as-if’ experiment in which the subject is asked to behave as if he were a particular person in a particular situation” (26). These methods can help elucidate the link between personality traits and the potential strategies, plans and concerns he/she may bring into a conflict situation. For example, Sternberg and Soriano (1984) used this method to uncover cross-situational consistencies in the perception and resolution of hypothetical conflicts. We also used this methodology to assess the link between the Big Five personality dimension of agreeableness and conflict resolution choices in adolescents and college students (Jensen-Campbell, Graziano, & Hair 1996 (see Study 1); Jensen-Campbell & Graziano 2001). In one study, a computerized version of Goldberg’s (1992) trait markers was used to measure personality (for details on the computer assessment methodology, see Graziano, Jensen-Campbell, & Finch 1997; Graziano, Jensen-Campbell, Steele, & Hair 1998). We then created stimulus stories representing conflict situations. Stories were divided equally among relationship types (e.g., mother, sibling) (see Jensen-Campbell
326
LAURI A. JENSEN-CAMPBELL AND WILLIAM G. GRAZIANO
et al. 2001 for actual vignettes). The adolescents’ task was to tell us how good or bad each choice was in solving conflicts in the different situations/ relationships. As predicted, the personality dimension of agreeableness uniquely predicted the endorsement of both constructive and destructive conflict resolution tactics. In terms of strengths, vignettes are direct, and focus participants’ attention on the variables of interest. In addition, the vignette method allows the researcher to control the conflict situation presented and to avoid some ethical problems associated with deception and manipulation of participants’ perceptions of conflict situations. There are also several potential weaknesses with using these role-playing methodologies (Graziano 1987; Greenberg & Folger 1988: 39–60). The primary argument against role-playing methodologies, however, is that social desirability processes may bias responding toward the ideal response rather than what participants would actually do in the situation. In his literature review, Laursen (1993) found that this data collection technique biased empirical results in the conflict literature. Specifically, he found that adolescents responded to hypothetical conflicts with ideal resolution choices (i.e., negotiation). In actual conflicts, however, power assertion and disengagement were used more frequently than negotiation. As a general rule, responses to hypothetical conflicts may be biased toward responses perceived to be desirable, but social desirability processes themselves are complex, elusive phenomena (Graziano & Tobin 2002). Furthermore, some studies report convergence. Using a multi-method format that collected data for both hypothetical vignettes and daily diary records across a two-week period, JensenCampbell and Graziano (2001) found convergence between hypothetical conflicts and diary records of actual daily interpersonal conflicts. Even so, our outcomes suggest that reactions to vignettes may have an underlying structure different from accounts of in-vivo conflict behaviors (Graziano 1987: 284– 287). In other words, the hypothetical vignette method relies heavily on simplistic, singular, deterministic views of personality on conflict behavior and fails to attend to contextual factors (e.g., conflict partner) that probably influence the behavior. (For a related discussion of personality and social desirability artifacts in conflict research, see Graziano & Tobin 2002).
Self-Report Methodologies Self-report questionnaires typically ask persons to report information about their conflicts during a specific period. Such measures may be useful as descriptions of individuals’ global perceptions of their conflict. For example, Jensen-Campbell and her colleagues (2003) asked adolescents to complete
METHODOLOGIES FOR STUDYING PERSONALITY PROCESSES
327
personality measures and the Styles of Conflict Resolution (SCR) questionnaire. The personality dimension of agreeableness again predicted global perceptions of appropriate conflict behavior. Potential threats to the validity of self-report summary measures in assessing personality’s contribution to conflict include temporality concerns, the control of third variables, social desirability artifacts, the use of retrospective reports, and shared method variance. Experimental studies are ultimately preferable for assessing causality, but correlated self-report studies can contribute to more comprehensive explanations. First, to establish that personality is systematically related to effects in social conflict, the researcher must show temporality. A study that collects all measures contemporaneously does not provide for an adequate test of causality. Prospective longitudinal studies provide clearer evidence of causality by establishing temporality (Shadish, Cook & Campbell 2001). For example, JensenCampbell and her colleagues (2002) assessed the associations between personality and being a victim of aggression. Because they controlled for fall levels of victimization when assessing the link between fall personality and spring victimization, their associations could be interpreted as changes in victimization from fall to spring (see also Hodges et al. 1995; Schwartz, Dodge, & Coie 1993). Correlational data in longitudinal designs does not yield unequivocal cause-and-effect conclusions, but it does foster more confidence in the predictive effects of personality on conflict behavior than does examining data in a cross-sectional design (see Biesanz, West, & Kwok 2003). Another problem in assessing the unique predictive validity of personality using self-report measures is the need to control for plausible “third variables” that may be causing the effect. For the relation between personality and conflict behavior not to be spurious, the relation must exist after explicitly collecting data and controlling for other variables that may affect the result. For example, the link between agreeableness and conflict resolution tactics might be an artifact of a correlation between agreeableness and the third variable of social desirability (SD) motives. Perhaps SD, not agreeableness, is the primary ingredient linking agreeableness to conflict choices. Graziano and Tobin (2002; Study 3) collected converging experimental and correlational data and ruled out this artifact explanation. Identifying and explicitly testing for third variable explanations like social desirability biases is especially important in conflict research because conflict is complexly determined and because in the absence of direct tests, such explanations will remain intuitively plausible until diffused. Collecting both personality data and conflict data from the same source is also problematic. When assessments of personality are not independent from assessments of conflict behavior, artificially inflated associations among the
328
LAURI A. JENSEN-CAMPBELL AND WILLIAM G. GRAZIANO
constructs can be found (Kenny & Kashy 1992). Thus, it is important to establish these relations with multiple, non-redundant raters. Many of our studies assess the link between personality and conflict behavior using multiple sources of information. For example, we have asked parents, teachers, and children all to provide independent assessments of personality and adjustment. Results from Jensen-Campbell and Graziano (2001) found that the endorsement of both negotiation and physical force were related to individual differences in agreeableness regardless of the source of the personality assessment (See Table 3). Gleason, Jensen-Campbell, and Richardson (2004) also assessed the relative importance of personality on the use of aggression in children’s peer relations. To avoid problems associated with shared method variance, they collected self-reported personality and peer-reported measures of aggression. By doing so, they could conclude more confidently that personality processes tied to agreeableness are indeed related to the use of aggression in children’s peer relations. Finally, researchers often asked participants to recall “actual” conflict events, ranging from the proximal (e.g., previous day) to the distal (e.g., childhood; see Halverson 1988). The use of self-reports for retrospective reporting is problematic because it requires participants to filter and to aggregate events that occur over an extended period. Henry, Moffitt, Caspi, Langley, and Silva (1994) conducted a longitudinal study on the retrospective method and found that even when there were significant relations between prospective and retrospective reports, the absolute agreement was poor. Moreover, psychosocial variables like family processes associated with conflict produced the lowest level of agreement. Reis and Wheeler (1991) have suggested that asking participants to recall actual conflicts is problematic because participants must deal with selection, recall, and aggregation issues. Selection poses several serious problems if the goal of the research is to understand actual conflict behavior. First, without specific instructions participants create their own criteria for including events (Reis & Wheeler 1991). Moreover, individuals may have difficulty storing and retrieving detailed information in long-term memory. It is probable that certain events (e.g., intense arguments) are more salient than other events (minor disagreements) (Halverson 1988; Reis & Wheeler 1991). Extraneous factors such as the person’s mood or personality (e.g., conscientiousness) may influence the selection and recall of events (Blaney 1986; Gilligan & Bower 1984; Isen 1984; Reis & Wheeler 1991; Schwarz & Clore 1983). A final methodological issue related to self-report measures is the aggregation process necessary for global measures (e.g., describe your conflicts over the last six months). Reis and Wheeler note that there is little empirical evidence on how individuals combine multiple interactions into a single rating dimension.
METHODOLOGIES FOR STUDYING PERSONALITY PROCESSES
329
Event-by-Event Recordings Event-by-event recordings eliminate some of these weaknesses by allowing the individual to evaluate specific, molecular events. Several relatively new methods for examining the nature and impact of daily life experience are useful for studying the impact of personality on daily interpersonal conflict as well. In other words, if the goal is to understand daily interpersonal conflict in everyday life, a method is needed that actually probes daily interactions. Interval-contingent recording, signal-contingent recording, and event-contingent recording are all event-by-event recording methods that meet this goal. In interval-contingent recording, persons are asked to record activities at chosen intervals. These intervals usually have some theoretical or logical basis (e.g., bedtime). In signal-contingent recording, participants must record their activities whenever they are signaled to do so by the researcher. These signals usually occur through telephone calls, alarms on PDAs, or the use of beepers. These signals can be at fixed times or they can be randomly generated. For example, Csikszenmihalyi and Larson (1984) used this methodology to study general adolescent experience. This method reduces the likelihood of getting reappraisal by requiring RPs to complete reports as soon to the signal as possible. This method, however, is less useful when researchers are interested in a specific class of events that occur infrequently because the chance that signals will coincide with the event becomes less likely. For example, young adolescents reported only 1.5 interpersonal conflicts per day in a diary study (Jensen-Campbell & Graziano 2000). Other studies have also shown that adolescents report few conflicts (e.g., 7.36 conflicts) on average each day (Laursen 1993). Thus, this method is less useful for studying the variation within interpersonal conflicts (Reis & Wheeler 1991). In event-contingent recording, individuals must report every event that meets some predetermined definition (e.g., behavioral opposition). A clear, precise definition is critical so that all events that meet a specific criterion are described. Like signal-contingent recording, this method reduces the likelihood of reappraisal. Event sampling, however, is best applied to discrete events such as conflict episodes. Lastly, event records are preferable when it is theoretically important to obtain a large number of events so that variation within a given category (e.g., type of relationship partner) may be studied. Jensen-Campbell and Graziano (2001) employed event sampling to assess the link between personality processes and conflict behavior in adolescence. An age-appropriate version of the Rochester Interaction Record (RIR) suitable for young adolescents was adapted from procedures used by Reis and Wheeler (1991). We created a record that measured daily ratings of every interaction lasting longer than ten minutes or more and any conflict that
330
LAURI A. JENSEN-CAMPBELL AND WILLIAM G. GRAZIANO
occurred in briefer interactions. Children were first assessed on their personality. Approximately eleven months later, we assessed perceptions of every conflict within each day across a two-week period using the event-contingent method of the RIR. Substantively, we found that the personality dimension of agreeableness, as well as the chronic use of destructive tactics in conflict, was significantly related to evaluations of the individuals’ adjustment by knowledgeable raters. In sum, event-by-event recording were explicitly designed to examine naturally occurring daily life experiences (Reis & Wheeler 1991; Suls, Martin, & David 1998). These methods allow researchers to supplement the usual paper-and-pencil ratings of conflict with basic, quasi-ethological reports about frequency and patterns of interpersonal conflict during daily social life.
Behavioral Observation Another approach that has been used in conflict research is structured behavioral observation in the laboratory. For example, Jensen-Campbell, Gleason, Adams, and Malcolm (2003) observed the relations between personality and conflict in late childhood by using a board game (that created conflict) and then unobtrusively videotaping and transcribing the events that occurred within the fixed situation. Laboratory observations can also involve less structured observations. For example, Ickes and his colleagues (1986, 1990) have developed a method to study unobtrusively naturally occurring interactions among dyads. Recently, Schweinle, Ickes, and Bernstein (2002) used a modified version of this paradigm to assess personality’s contribution (i.e., empathic accuracy) to husband-to-wife aggression. These methods are important and provide a rich body of empirical evidence in psychology. The data collected with these approaches, however, are usually constrained by the environment (e.g., laboratory) and often involve fixed activities. They do not allow for probing the full extent of everyday interpersonal conflicts in adolescence. In other words, the conflict behavior is mandated into existence by definition. Similarly, the obtrusiveness of the observation may influence the behaviors displayed. In addition, task demands constrain conflict resolution strategies (Laursen & Collins 1994). Disengagement is usually not possible in laboratory settings. For example, participants in the JensenCampbell et al. (2003) study were told that if they had a disagreement they must work it out. Their choices for conflict resolution were constrained; they could not seek the researcher’s help or leave the experimental session. Thus, the children could not withdraw from the situation before the experiment was finished and could not seek out any third-party interventions.
METHODOLOGIES FOR STUDYING PERSONALITY PROCESSES
331
Interpersonal conflict is a complex social phenomenon. To gain a better understanding of conflict processes, researchers developed statistical methodologies to look past simple main effects models (Campbell & Kashy 2002; Cohen, Cohen, West, & Aiken 2003; Finch & Graziano 2001; Graziano, et al. 1997; Kenny 1994). The remainder of this paper describes briefly some of these statistical methods and how they can be useful for studying personality’s contribution to interpersonal conflict. These statistical techniques include moderated multiple regression (MMR), structural equation modeling, multi-level modeling, the Social Relations Model, and the Actor-Partner Independence (APIM) Model (Bryk & Raudenbush 1991; Campbell & Kashy 2002; Cohen et al. 2003; Kenny 1994, 1996; Mendoza & Graziano 1982).
Moderated Multiple Regression Researchers are becoming increasingly interested in how individual differences might moderate the link between a predictor/independent variable and dependent variable. For example, research often exhibits a stronger link between agreeableness and social behavior in females than in males (JensenCampbell & Graziano 2001; Jensen-Campbell & Graziano 2000; Tobin, Graziano, Vanman, & Tassinary 2000). To study moderation, investigators typically examine the interaction between the predictor variable (e.g., personality characteristic) and the moderator (e.g., sex) to compare the equivalence of relations across groups (Baron & Kenny 1986). According to Cohen, Cohen, West, and Aiken (2003), an interaction refers to “an interplay among predictors that produces an effect on the outcome Y that is different from the sum of the effects of the individual predictors” (255). Given most personality variables are continuous variables, it is recommended that researchers study these interactions using moderated multiple regression procedures (MMR) rather than dichotomizing the variables and using an ANOVA approach. For example, Jensen-Campbell and Graziano (in press) used MMR procedures to assess personality’s contribution to self-regulatory processes in children. They found that for children with low levels of conscientiousness, agreeableness impacted self-regulatory behaviors. In other words, at lower levels of conscientiousness, children with higher agreeableness exhibited better self-regulation than persons lower on agreeableness. When there were medium or high levels of conscientiousness, there was no evidence agreeableness affected selfregulation. This outcome suggests some kind of compensatory processes between the two related personality dimensions of agreeableness and conscientiousness in children, as anticipated by developmental theory (Ahadi & Rothbart 1994). Presumably, their more advanced self-regulatory processes help high
332
LAURI A. JENSEN-CAMPBELL AND WILLIAM G. GRAZIANO
agreeable persons generate more positive responses than their peers to interpersonal conflict. Discussing the specifics of moderated regression is beyond the scope of this paper, but an excellent review of these procedures can be found in both Aiken and West (1994) and Cohen et al. (2003).
Multi-level Modeling Often data that assesses personality’s influence on daily conflicts is also structured hierarchically. In other words, there are clusters or levels within the data set. Consider the conflicts reported by an individual participant using the event-by-event recording method. We would expect some association between the resolutions of conflicts across relationship classes and situations within the same individual. For example, individuals high in agreeableness will consistently resolve their conflict episodes more constructively with parents, peers, and teachers than will their peers. Moreover, conflicts that are reported for members of the same or similar conflict dyad(s) (e.g., mother/child; High/ High Agreeable dyads) should also be more highly correlated than other pairings. When data are collected at multiple levels, the data should be treated as a multi-level or hierarchical model and the use of random coefficient regression should be used to retain the data’s complexity (Cohen, et al. 2003). In Jensen-Campbell and Graziano (2001), both individuals and their interactions were treated as units of measurement because individual conflicts were collected from a participant across a two-week period. For example, we first calculated regression equations that estimated how much each participant reported systematic ratings with his/her partners based on the sex of the partner. In the second step, the first-step regression coefficients were regressed on participant-agreeableness, sex of participant, and the cross product of Sex X Agreeableness (see Bosker & Snijders 1999; Bryk & Raudenbush 2002; Cohen et al. 2003; Kenny et al. 1998; Kenny, Bolger, & Kashy 2002). By using multi-level modeling instead of aggregating our data by participant, we could assess the within-individual information instead of assuming that each conflict or conflict partner for the individual were equivalent. In other words, we could assess more accurately the contribution of an individual’s personality while assessing effects due to the interaction and interaction partner. We specifically constructed a two-level model, with variations in change parameters among interactions within an individual at level one (e.g., partner’s traits) and variation among participants as level two (e.g., participant’s traits). It is possible, however, to have more than two levels in a given data set. In sum, multilevel procedures treat both lower- or micro-levels (e.g.,
METHODOLOGIES FOR STUDYING PERSONALITY PROCESSES
333
interactions/partners) and upper- or macro-levels (e.g., persons) as sampling units so inferences can be made about persons, partners, and interactions and the complexity of the data remains in tact (Bryk & Raudenbush 1992, 2002; Cohen et al. 2003).
Mediational Analyses There is also increasing concern among personality researchers about the process underlying these individual differences. These concerns are reflected in studies that examine mediation or the processes that underlie the association between two variables. Gleason et al. (2004) used mediational procedures to examine the link between personality and adjustment. They found that the use of destructive tactics mediated at least partially the link between the personality dimension of agreeableness and adjustment in school. These analyses helped elucidate why the agreeableness-adjustment relationship might occur (Baron & Kenny 1986). Mediational analyses can be done using either OLS procedures or latent variable analyses such as structural equation modeling (SEM). For example, SEM allows an investigator to constrain each model parameter to equal some pre-specified value or to leave that parameter free to be estimated. Because SEM allows for the a priori specification of constraints, any specific pattern of structural relations specified by the investigator reflects a hypothesized model that can be empirically tested for its adequacy. SEM also allows any or all of the estimated model parameters to be estimated between groups. Moreover, SEM can be used to examine individual differences as well as reciprocal relationships in longitudinal data (e.g., Farrell 1994). Discussing the specific benefits and limitations of these statistical procedures are beyond the scope of this paper (see Bentler 1990; Cohen et al. 2003; DeShon 1998; McDonald & Ho 2002; and West & Aiken 1997).
Social Relations Model Analyses Another statistical method that can be used to study how personality influences perceptions of conflict in groups (e.g., triads) involves decomposing self- and other data into perceiver and target effects using Kenny’s (1994) Social Relation Model (SRM). Groups can be formed and asked to perform a conflict-inducing task or group decision-making task that creates conflict. Because each member in the group interacts and rates all other members in the group (i.e., a round-robin design), SRM analyses allow for personality effects of the individual to be partitioned from partner and relationship effects.
334
LAURI A. JENSEN-CAMPBELL AND WILLIAM G. GRAZIANO
For example, Graziano, Hair, and Finch (1997) manipulated simultaneously group goal structures and group composition to study college student’s interactions. They assigned participants to triads based on their own and their partner’s scores on personality, specifically agreeableness. The composition of the groups included either one high-agreeable participant and two low-agreeable participants (A+A−A−) or one low-agreeable participant and two high-agreeable participants (A−A+A+). According to this model, the actor/perceiver effect assesses variance that is in the perception of the observer. It is possible that individuals who rate themselves higher on agreeableness also rate others as more likeable and agreeable (regardless of their partners’ personal qualities). In other words, the actor effect provides information on a perceiver’s average level of the given behavior in the presence of a variety of partners. For example, Graziano, Hair, and Finch (1997) found that agreeable persons rated themselves as working less in opposition to others, regardless of their interaction partner. The target/partner effect, on the other hand, represents how much raters agree in rating targets on a trait. In other words, this effect specifies the extent to which an individual elicits consistent reactions across interaction partners. A significant target effect indicates that there is some convergence on which individuals in a group are seen as high or low on a personality variable when involved in a conflict. It is possible that agreeable persons will be rated as more cooperative and less argumentative consistently across interaction partners. For example, Graziano, Hair, and Finch (1997) found that compared to their peers, men high in agreeableness were rated as less likely to resist team’s efforts across interaction partners. Using SRM, Graziano and Tobin (2002) found that agreeableness was stronger as a rater effect than as a target effect. Compared with their peers, high agreeable persons tend to perceive positive characteristics in all of their interaction partners, projecting desirable attributes onto others. Finally, the relationship effect in SRM measures the unique perceptions between two people after controlling for actor effects and partner effects. One of the limitations of the Social Relations Model, however, is that it does not allow for the analysis of interaction effects.
Dyadic Analysis Contrary views of personality such as Shadel and Cervone (1993) argue that it is inappropriate to study personality as a basic unit of analysis because interpersonal behavior needs to be defined within context. To address these concerns, we used a multivariate approach for statistical analysis of dyadic
METHODOLOGIES FOR STUDYING PERSONALITY PROCESSES
335
social behavior presented in Mendoza and Graziano (1982). These procedures yield a simultaneous test of participant, partner, and interaction effects for agreeableness. For example, we found that nonverbal behaviors were influenced by both the participants and partners level of agreeableness (Graziano et al. 1996, Study 2). Arm crossing, a measure of disagreement, was one of these nonverbal behaviors (Mehrabian 1969). Agreeable participants in the H – L group crossed their arms significantly less when they were with a low agreeableness partner than when they were paired with a high agreeable partner. Low-agreeable participants did not differ in behavior across groups. A similar approach involves the use of the Actor-Partner Interdependence Model (APIM) to analyze dyadic data (Campbell & Kashy 2002; Kashy & Kenny 2000). In this model, the major unit of analysis is the dyad and variance is again partitioned into actor, partner, and the actor X partner interaction. Since dyads involve naturally occurring hierarchies, APIM can be estimated using both HLM and PROC MIXED in SAS (see Campbell & Kashy 2002 for a review). Given that conflict is defined in terms of mutual opposition (Laursen & Collins 1994; Shantz & Hartup 1992), we focused on both personality and the interpersonal aspects of conflict when discussing potential methodologies. We demonstrated that interpersonal conflicts contain emergent elements, and may produce phenomena not easily predicted from characteristics of either the person or the relationship in isolation (e.g., Graziano et al. 1996, Study 2). To gain a better understanding of conflict processes, future research must utilize multiple, converging methods in their studies. Due to differences in attitudes and life histories, individuals will have different perspectives on the same conflict episodes. We need to appreciate these differences by including personality as well as situations in research. We need to look past simple main effects models toward models that consider the links among personality, situated motives, and overt behavior (Finch & Graziano 2001; Graziano, JensenCampbell, & Finch 1997; Graziano, Hair, & Finch 1997). These approaches will lead to a better understanding of the causes, processes, and outcomes of interpersonal conflict.
Acknowledgements We thank Deborah A. Kashy, Gary J. Lautenschlager, Jorge Mendoza, Harry T. Reis, and Stephen Gilberto West for their help and expertise with our data collection and analyses over the years. We also wish to thank Shaun D. Campbell for his suggestions and comments on earlier versions of this manuscript. We thank all of the graduate and undergraduate students who have help
336
LAURI A. JENSEN-CAMPBELL AND WILLIAM G. GRAZIANO
in completing the data collection on the research we describe in this paper. This work was supported in part by the National Science Foundation Grant SBR 92–12201 NIMH grant R01 MH50069 to W.G. Graziano and a NIMH B/Start Grant to L.A. Jensen-Campbell.
References Ahadi, S.A. & Rothbart, M.K. (1994). “Temperament, development, and the Big Five,” in C.F. Halverson, Jr., G.A. Kohnstamm, & R.P. Martin, editors, The developing structure of temperament and personality from infancy to adulthood. Hillsdale, NJ: Lawrence Erlbaum Associates. Aiken, L.S., & West, S.G. (1991). Multiple regression: Testing and interpreting interactions. Newbury Park, NJ: SAGE Publications. Aronson, E., & Carlsmith, J.M. (1968). “Experimentation in social psychology,” in G. Lindzey & E. Aronson, editors, Handbook of social psychology, 2: 1–79. Reading, MA: AddisonWesley. 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. Barrick, M.R., & Mount, M.K. (1991). “The big five dimensions and job performance: A meta analysis.” Personnel Psychology, 44, 1: 1–26. Barrick, M.R., Mount, M.K. & Judge, T.A. (2001). “Personality and performance at the beginning of the new millennium: What do we know and where do we go next?” International Journal of Selection & Assessment, 9, 1–2: 9–30. Barrick, M.R., Stewart, G.I., Neubert, M.J., & Mount, M.K. (1998). “Relating member ability and personality to work-team processes and team effectiveness.” Journal of Applied Psychology, 83: 377–391. Barry, B., & Friedman, R.A. (1998). “Bargainer characteristics in distributive and integrative negotiation.” Journal of Personality and Social Psychology, 74: 345–359. Bentler, P.M. (1990). “Comparative fit indexes in structural models.” Psychological Bulletin, 107: 238–246. Biesanz, J., West, S.G., & Kwok, O.M. (2003). “Personality over time: Methodological approaches top the study of short-term and long-term development and change.” Journal of Personality, 71: 905–941. Blaney, P.H. (1986). “Affect and Memory: A review.” Psychological Bulletin, 99: 229–246. Bosker, R.J., & Snijders, T.A.B. (1999). Multilevel analysis: An introduction to basic and advanced multilevel modeling. Newbury Park, NJ: SAGE Publications. Bryk, A.S., & Raudenbush, S.W. (1992). Hierarchical Linear Models: Applications and data analysis methods. Newbury Park, CA: SAGE. Campbell, L., & Kashy, D.A. (2002). “Estimating actor, partner, and interaction effects for dyadic data using PROC MIXED and HLM: A user-friendly guide.” Personal Relationships, 9: 327–342. Cantor, N. & Kihlstrom, J.F. (1987). Personality and social intelligence. Englewood Cliffs, NJ: Prentice Hall. Cohen, J., Cohen, P. West, S.G., & Aiken, L.S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences (3rd ed.). Mahwah, NJ: Lawrence Erlbaum Associates.
METHODOLOGIES FOR STUDYING PERSONALITY PROCESSES
337
Csikszentmihalyi, M., & Larsen, R. (1984). Being adolescent: Conflict and growth in the teenage years. New York: Basic Books. De Dreu, C.K.W., & Carnevale, P.J. (2003). “Motivational bases of information processing and strategy in negotiation and social conflict,” in M.P. Zanna, editor, Advances in Experimental Social Psychology, 35, New York: Academic Press. DeShon, R.P. (1998). “A cautionary note on measurement error corrections in structural equation models.” Psychological Methods, 3: 412–423. Farrell, A.D. (1994). “Structural equation modeling with longitudinal data: Strategies for examining group differences and reciprocal relationships.” Journal of Consulting and Clinical Psychology, 62: 477–487. Finch, J.F. & Graziano, W.G. (2001). “Predicting depression from temperament, personality, and patterns of social relations.” Journal of Personality, 69: 27–55. Furman, W. & McQuaid, E.L. (1992). “Intervention programs for the management of conflict.” in C.U. Shantz & W.W. Hartup, editors.) Conflict in child and adolescent development: 402–429. New York: Cambridge University Press. Gilligan, S.G. & Bower, G.H. (1984). “Cognitive consequences of emotional arousal.” in C.E. Izard, J. Kagan, & R.B. Zajonc, editors, Emotions, cognition, and behavior: 547–588. Hillsdale, NJ: Erlbaum. Gleason, K.A., Jensen-Campbell, L.A., & Richardson, D. (2004). “Agreeableness and aggression in adolescence.” Aggressive Behavior, 30: 43–61. Goldberg, L.R. (1992). “The development of markers of the Big Five factor structure.” Psychological Assessment, 4: 26–42. Graziano, W.G. (1987). “Lost in thought at the choice point: Cognition, context, and equity,” in J.C. Masters & W.P. Smith, editors, Social comparison, social justice, and relative deprivation: Theoretical, empirical and policy perspectives: 265–294. Hillsdale, NJ: Erlbaum. Graziano, W.G. (2003). “Personality development: An introduction. Toward process approaches to stability and change in persons.” Journal of Personality, 71, 6: 893–903. Graziano, W.G., Hair, E.C., & Finch, J.F. (1997). “Competitiveness mediates the link between personality and group performance.” Journal of Personality and Social Psychology, 73: 1394–1408. Graziano, W.G., Jensen-Campbell, L.A., & Finch, J.F. (1997). “The self as a mediator between personality and adjustment.” Journal of Personality and Social Psychology, 73: 392–404. Graziano, W.G., Jensen-Campbell, L.A., Hair, E.C. (1996). “Perceiving interpersonal conflict and reacting to it: The case for agreeableness.” Journal of Personality and Social Psychology, 70: 820–835. Graziano, W.G., Jensen-Campbell, L.A., Steele, R.G., & Hair, E.C. (1998). “Unknown words in self-reported personality: Lethargic and provincial in Texas.” Personality and Social Psychology Bulletin, 24: 893–905. Graziano, W.G., & Tobin, R.M. (2002). “Agreeableness: Dimension of personality or social desirability artifact?” Journal of Personality, 70: 695–727. Greenberg, J., & Folger, R. (1988). Controversial issues in social research methods. New York: Springer Verlag. Halverson, C.F. (1988). “Remembering your parents: Reflections on the retrospective method.” Journal of Personality, 56: 435–444. Henry, B., Moffitt, T.E., Caspi, A., Langley, J., Silva, P.A. (1994). On the “remembrance of things past”: A longitudinal evaluation of the retrospective method. Psychological Assessment, 6: 92–101. Hodges, E.V.E., Malone, M.J., Jr., & Perry, D.G. (1995, March). “Behavioral and social
338
LAURI A. JENSEN-CAMPBELL AND WILLIAM G. GRAZIANO
antecedents and consequences of victimization by peers,” in N.R. Crick, Recent trends in the study of peer victimization: Who is at risk and what are the consequences? Symposium conducted at the biennial meeting of the Society for Research in Child Development, Indianapolis, IN. Ickes, W., Robertson, E., Tooke, W., & Teng, G. (1986). “Naturalistic social cognition: Methodology, assessment, and validation.” Journal of Personality and Social Psychology, 51: 66–82. Ickes, W., Stinson, L., Bissonnette, V. (1990). “Naturalistic social cognition: Empathic accuracy in mixed-sex dyads.” Journal of Personality and Social Psychology, 59: 730–742. Isen, A.M. (1984). “Toward understanding the role of affect in cognition,” in R.S. Wyer, Jr., & T.K. Srull, editors, Handbook of social cognition: 179–236. Hillsdale, NJ: Erlbaum. Jensen-Campbell, L.A., Adams, R., Perry, D., Furdella, J.Q., Workman, K.A., & Egan, S. (2002). Agreeableness, Extraversion, and Peer Relations in Early Adolescence: Winning Friends and Deflecting Aggression. Journal of Research in Personality, 36: 224–251. Jensen-Campbell, L.A., Gleason, K.A., Adams, R., & Malcolm, K.T. (2003). Interpersonal conflict, agreeableness, and personality development. Journal of Personality, 71: 1059–1085. (Special Issue: Personality Development). Jensen-Campbell, L.A. & Graziano, W.G. (2000). Beyond the schoolyard: Relationships as moderators of daily interpersonal conflict. Personality and Social Psychology Bulletin, 26: 923–935. Jensen-Campbell, L.A., & Graziano, W.G. (2001). Agreeableness as a Moderator of Interpersonal Conflict. Journal of Personality, 69: 323–362. Jensen-Campbell, L.A., & Graziano, W.G. (in press). “Resisting Temptation: Differing Motives for Self-Control.” Merrill-Palmer Quarterly. Jensen-Campbell, L.A., Graziano, W.G., & Hair, E.C. (1996). “Personality and relationships as moderators of interpersonal conflict in adolescence.” Merrill-Palmer Quarterly, 42: 148–164. Kashy, D.A., & Kenny, D.A. (2000). “The analysis of data from dyads and groups.” in H.T. Reis & C.M. Judd, editors, Handbook of research methods in social and personality psychology: 451–477. New York: Cambridge University Press. Kenny, D.A. (1994). Interpersonal perception: A social relations analysis. New York: Guilford. Kenny, D.A. (1996). The design and analysis of social-interaction research. Annual Review of Psychology, 47: 59–86. Kenny, D.A., & Kashy, D.A. (1992). Analysis of the multi-trait-multi-method matrix by confirmatory factor analysis. Psychological Bulletin, 112: 165–172. Kenny, D.A., Kashy, D.A., & Bolger, N. (1998). “Data analysis in social psychology,” in D.T. Gilbert and S.T. Fiske, editors, The handbook of social psychology, 1: 233–265. New York: McGraw-Hill. Kenny, D.A., Bolger, N., & Kashy, D.A. (2002). “Traditional methods for estimating multilevel models,” in D.S. Moskowitz & S.L. Hershberger, editors, Modeling intraindividual variability with repeated measures data: Methods and applications. Multivariate applications book series: 1–24. Mahwah, NJ: Lawrence Erlbaum Associates. Kenny, D.A. & La Voie, L. (1985). “Separating group and individual level effects.” Journal of Personality and Social Psychology, 48: 339–348. Laursen, B. (1993). “The perceived impact of conflict on adolescent relationships.” MerrillPalmer Quarterly, 39: 535–550. Laursen, B., & Collins, W.A. (1994). “Interpersonal conflict during adolescence.” Psychological Bulletin, 115: 177–209. Lewin, K. (1935). A dynamic theory of personality. New York: McGraw-Hill.
METHODOLOGIES FOR STUDYING PERSONALITY PROCESSES
339
Mehrabian, A. (1969). “Methods and designs: Some referents and measures of nonverbal behaviors.” Behavioral Research Methods and Instruments, 1: 203–207. McAdams, D.P. (1995). “What do we know when we know a person?” Journal of Personality, 63: 365–396. McDonald, R.P. (2002). “Principles and practice in reporting structural equation analyses.” Psychological Methods, 7: 64–82. Mendoza, J., & Graziano, W. (1982). “The statistical analysis of dyadic social behavior: A multivariate approach.” Psychological Bulletin, 92: 532–540. Mischel, W. (1973). “Toward a cognitive social learning reconceptualization of personality.” Psychological Review, 80: 252–283. Patterson, G.R. (1982). Coercive family process. Eugene, OR: Castalia. Raudenbush, S.W., & Bryk, A.S. (2002). Hierarchical Linear Models: Applications and Data Analysis Methods (2nd ed.). Newbury Park, NJ: SAGE Publications. Reis, H.T., & Wheeler, L. (1991). “Studying social interaction with the Rochester Interaction Record,” in M.P. Zanna, editor, Advances in experimental social psychology, 24: 269–318. New York: Academic Press. Shadel, W.G., & Cervone, D. (1993). “The Big Five versus nobody?” American Psychologist, 48: 1300–1302. Shadish, W.R., Cook, T.D., & Campbell, D.T. (2001). Experimental and quasi-experimental designs for generalized causal inference. Boston: Houghton Mifflin. Shantz, C.U., & Hartup, W.W., editors, (1992). Conflict in child and adolescent development. New York: Cambridge University Press. Sherif, M. (1956). “Experiments in group conflict.” Scientific American, 195: 53–58. Schwarz, N., & Clore, G.L. (1983). “Mood, misattribution, and judgments of well-being: Informative and directive functions of affective states.” Journal of Personality and Social Psychology, 34: 513–523. Schwartz, D., Dodge, K.A., & Coie, J.D. (1993). “The emergence of chronic peer victimization in boys’ play groups.” Child Development, 6: 1755–1772. Schweinle, W.E., Ickes, W., & Bernstein, I.H. (2002). “Empathic inaccuracy in husband to wife aggression: The over attribution bias.” Personal Relationships, 9, 2: 141–158. Sternberg, R., & Soriano, (1984). “Styles of conflict resolution.” Journal of Personality and Social Psychology, 47: 115–126. Suls, J., Martin, R., David, J. (1998). “Person-Environment fit and its limits: Agreeableness, neuroticism, and emotional reactivity to interpersonal conflict.” Personality and Social Psychology Bulletin, 24: 88–98. Terhune, K.W. (1970). “The effects of personality in cooperation and conflict,” in P. Swingle, editor, The Structure of Conflict. New York: Academic Press. Tobin, R.M., Graziano, W.G., Vanman, E., & Tassinary, L.G. (2000). “Personality, emotional experience, and efforts to control emotions.” Journal of Personality and Social Psychology, 79: 656–669. West, S.G. & Aiken, L.S. (1997). “Toward understanding individual effects in multiple component prevention programs: Design and analysis strategies,” in K. Bryant, M. Windle, & S.G. West, editors, The science of prevention: Methodological advances from alcohol and substance abuse research: 167–209. Washington, DC: American Psychological Association.
The Heart of Darkness: Advice on Navigating Cross-Cultural Research CATHERINE H. TINSLEY
My apologies of course to Joseph Conrad for any implication that this article is as entertaining or as well written as The Heart of Darkness. Conrad’s novel tracks a British trader who descends deep into the jungle of Central Africa, where the lack of rules (or at least any structure that he recognizes) drives him crazy. Freedom obsesses him with power, and he becomes a psychotic despot. The story parallels one’s descent into cross-cultural research. In both instances, the path is murky, there are too many degrees of freedom for one to be truly comfortable, and the lack of structure leads researchers to establish methodological fiefdoms without oversight from others. This lack of structure or unifying paradigm for cross-cultural research explains in large measure why mainstream researchers have difficulty accepting cross-cultural studies or results, which are often perceived as descriptive or epiphenomenal findings with limited rigor or merit. Several years ago, a group of colleagues, including myself, wrote two chapters exploring cross-cultural research methods issues (Lytle, Brett, Barsness, Tinsley, & Janssens 1995; Brett, Tinsley, Janssens, Barsness, and Lytle 1997). We were beginning our own forays into cross-cultural research, and our goal was to put forth a methodology for conducting rigorous, confirmatory crosscultural research. These chapters have been generally accepted by other scholars doing cross-cultural research, and many of the specific suggestions for doing cross-cultural research remain useful, such as: modeling culture as a set of overlapping dimensions, studying both moderating and main effects of culture, assessing conceptual, structural, and functional equivalence, and constructing research designs that rule out alternative explanations for confirmatory results. However, as I reflect back on the overall gestalt of these chapters, I notice a certain innocence and optimism that is borne out of a theoretical rather than practical understanding of the issues surrounding cross-cultural research. The chapters lay out for the reader a rich array of choices in which to conduct rigorous cross-cultural research. For example, in the second chapter (Brett et al. 1997) we discuss the vast multi-dimensional space in which a researcher can position his or her cross-cultural endeavors. It is a space where International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 341–350 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
342
CATHERINE H. TINSLEY
a researcher can choose a positivist or interpretive epistemology, an emic or etic perspective, an inductive or deductive process, and use qualitative or quantitative methods. What I have experienced, in practice, however, is that this again allows too many degrees of freedom. When navigating through the heart of darkness, it is important to journey pretty closely to a pathway towards the center of our previously defined research space (Brett et al. 1997). To move towards either end of the dichotomies (positivist versus interpretive epistemologies, etic versus emic perspectives, inductive versus deductive processes) risks compromising the rigor and validity of one’s study. Indeed, five years later, I think of cross-cultural research as an endeavor devoted to managing the tensions created by these dichotomies. These dichotomies represent competing interests or paradigms, which are valid concerns, but can pull a researcher off the middle path. When embarking on a crosscultural research project, one is always striking a balance between competing interests, and continuously trying to find the middle road. Most social scientists recognize that there are tensions to be managed and tradeoffs to be made. The most oft recognized tension is found between induction versus deduction. Experimentalists claim to eschew a deductive approach. Starting with theory, experimentalists develop specific a priori hypotheses to operationalize the theory, and then gather data to either confirm or disconfirm their hypotheses. Ethnographers and case study researchers, on the other hand, claim to espouse an inductive approach. Beginning with observations, they synthesize these data to propose more generalizable patterns or theory. Yet, as Cattell (1988) observed the deductive-hypothetico and inductivehypothetico processes are really two parts of a larger whole: the inductivehypothetico-deductive-experimental (data)-inductive cycle. Thus while most researchers generally ascribe to one process or the other, in reality they iterate between both processes. Ethnographers gather data and begin to build theories, and then return to the field for additional observations to refine their theory as it develops (Spradley 1979; 1980). Similarly, deductive researchers often refine some of their conceptual categories as they learn more about the data (see Weingart 1997, for example, on how qualitative transcript data is often coded and analyzed.). While most researchers recognize the value of the deductive-hypothetico (analyzing a priori hypotheses for theory confirmation or disconfirmation) and inductive-hypothetico (conceptual re-evaluation to refine theory) processes, they generally purport to use only one process or the other in any single study. Most of the time, however, as social science researchers move throughout their explorations, they generally employ both induction and deduction and are trying to find the most appropriate balance between the two.
THE HEART OF DARKNESS
343
Cross-cultural researchers have not only the inductive-deductive tension to manage, but also that between positivist and interpretive epistemologies and emic versus etic perspectives. These two tensions are interrelated, as is the advice for navigating a balance. The remainder of this chapter defines these terms, discusses the tensions, and then concludes with concrete advice for the cross-cultural researcher who hopes to balance these tensions.
Tensions in Cross-Cultural Research Before expounding two fundamental tensions to be managed, I should express my assumptions, which are that the researcher is intent on conducting rigorous research to confirm cross-cultural similarities and differences in mid-range theories. Mid-range theories define a set of phenomena (or constructs), their interrelationships, and then propose a causal explanation for those relationships (James, Mulaik, & Brett 1981; Moore, Johns, & Pinder 1980). Examples would be a theory that motivation is caused by perceptions of equal inputs to outcomes across people (Adams 1963), or that self-centered attributions cause disputants to accept the same deal from a mediator that they would reject from the other party. I assume the researcher is interested in whether or not the causal relationships between the constructs hold across multiple cultures, or whether, for example, there are different variables that give rise to motivation or deal-acceptance across cultures. I also assume the researcher has some notion of whether this theory will hold or not based on cultural dimensions. That is, s/he has a model of culture as an interrelated set of dimensions that characterize the nature of any culture (such as individualism-collectivism; hierarchy-egalitarianism), as well as a proposed dimensional profile for each of the cultures examined. S/he is then interested in proposing what we call a “cultural explanation” (Brett et al., 1997) for any similarities or differences in the mid-range theory. An example, of a cultural explanation would be to expect differences in how the more collective Chinese versus the more individualistic Americans are motivated. Collective Chinese workers may be less motivated by equity (individuals’ reward commensurate with their inputs) than they are by an equal distribution of rewards across an entire work group, and for individualistic Americans the reverse may be true. With this goal in mind, of confirmatory research on cross-cultural similarities and differences based on an a priori cultural explanation, I now explain the two fundamental tensions that need to be managed.
344
CATHERINE H. TINSLEY
Positivist versus Interpretive Epistemology The positivist epistemology (or nature of knowledge) contends that reality is objective (Popper, 1959). It exists external to one’s perceptions and therefore can be measured precisely, and independent of any socially constructed theory. This is important for theory testing as it promotes the notion of neutral (i.e. theory free) observations that can be used to confirm or disconfirm theory. Post-positivists appear to subscribe to the notion of an existing objective reality, yet they have attacked the notion that observations of that reality are neutral (Hanson 1958; Kuhn 1962). They argue that is difficult (or impossible) to establish an independent test of theory, because a researcher’s measures will always be imbued with the theory s/he is trying to test. The interpretive epistemology goes even farther by arguing for multiple realities. Interpretivists espouse the notion of multiple socially-constructed realities. Therefore a researcher and subjects will define their own reality, unique to the participants and to the time of the study (Guba & Lincoln 1994). This perspective presents obvious difficulties to any theory generalization (generalizing what has been learned either to other populations or to other points in time). Modern social scientists have gotten around the issues raised by post-positivists and interpretivists by acknowledging the possibility of multiple realities and theory-laden observations. Yet, they argue that observations, while they may not be neutral, are nonetheless influenced by multiple theories (Duhem 1962; Cook & Campbell 1979). Single theory laden observations present problems for testing that theory, but observations that are multi-theory laden will not unduly influence the testing of any particular theory. Multitheory laden observations would be the situation analogous to random rather than systematic error. Moreover, they argue that while multiple realities may exist, there will be considerable overlap between people’s unique subjective realities. Hence there is a defining reality that can be captured by measuring the perceptions of multiple respondents (Cook & Campbell 1979). Unfortunately, for cross-cultural researchers these debates about whether observations are neutral or theory-laden and whether reality is singular and objective or multiple and subjective may not be as easily reconcilable. First, there may be less overlap between respondents’ realities, when a participant pool spans multiple cultures. The more heterogeneous the respondents, the less likely they are to share perceptions of reality. Second, observations may be multi-theory laden, yet these theories may not be culture-free, but rather may emanate from a singular dominant culture. So what is a cross-cultural researcher to do? How can s/he acknowledge multiple realities and theory laden observations, yet still hope to construct a
THE HEART OF DARKNESS
345
design that offers some independent confirmation or disconfirmation of a relevant theory? The best advice I can offer is to construct both theory and measurement with input from whatever cultures are to be studied. This is what we earlier called the N-way approach to research (Brett et al. 1997), where the researcher starts with a research question such as, “How will employees in different cultures be most motivated?” and a set of researchers or at least informants from every culture where the research question will be tested. For the theory construction of the research project the N-way approach should enable the researcher to determine how to best model the research question in all of the target cultures, to assess similarities and differences across the cultures, and to construct a cultural explanation for these similarities and differences (Brett et al. 1997). For example, representatives from each culture should determine all the relevant constructs for the research question and how these constructs are likely to interrelate. Once these nomological networks are determined for each culture, it will be easy to identify the similarities and differences across cultures, and together as the team questions why certain constructs interrelate in their cultures, they will formulate the cultural explanation for the cross-cultural similarities and differences. For measurement, each member of the team determines how to best operationalize each of the constructs for his or her particular culture, similarities and differences are noted, and a cultural explanation for any of these differences is formulated. Two words of caution are in order. First, in this team approach each researcher and/or informant represents his or her culture. Although anthropologists and other field researchers use “key informants” or focus groups to acquire in-depth information, these small samples are of course subject to biases not found in larger scale studies (Bazerman 2002). Thus, it would also be advisable for researchers to attempt to triangulate this key informant information with a broader foundation of knowledge from relevant literature. In this motivation example, the researchers would examine literature on motivation, rewards, and equity across all the cultural groups examined. Secondly, the process itself requires a balance between eliciting and modeling detailed information from each culture (to remain “true” to how that culture operates) and seeking broader relationships that are likely to have some overlap with the models from the other cultures of interest. Similarly for the measurement issues, the process requires a balance between operationalizing each construct for reliability and internal validity to each culture and looking for indicators that will overlap with those used to operationalize the constructs in other cultures of interest. Managing this balance is essentially confronting the emic versus etic tension, to which I now turn.
346
CATHERINE H. TINSLEY
Emic versus Etic Perspectives Many researchers discuss the emic versus etic perspective as a contrast between an insider’s view and an outsider’s view (Triandis 1972; Berry 1980). These terms were first delineated by Pike (1966) a cultural linguist who noticed that some sounds phonemics are unique to a particular (set of) culture(s), where as other sounds phonetics are universal across culture. Therefore, in Pike’s delineation, an emic perspective is not only that of the “native insider,” but also represents culturally specific phenomena. Emic researchers examine human behavior within a particular cultural system, seeking meaning and causal explanations from within that cultural system, which are likely to be unique to that system. They generally espouse an interpretive epistemology, that there are multiple realities, each unique to a particular group of participants at a particular point in time. Emic research, such as the “thick descriptions” of Geertz (1983), generally focus on a particular culture, striving for a deep internal analysis of that culture’s unique meaning and social structure. Etic researchers, on the other hand, examine human behavior from an external or outsider’s perspective, and they tend to study the same phenomena across multiple cultures. Rather than discovering a culture’s emergent structure or model for a phenomenon, etic researchers generally develop and impose a theory, structure, or model to understand a phenomenon. They generally espouse a more positivist epistemology, so that the constructs and measures are considered objective and universal. Obviously, the etic perspective is more conducive to cross-cultural comparisons. Indeed with a purely emic perspective, equivalent comparisons become impossible. For the etic perspective, though, the researcher formulates a mid-range causal theory for a phenomenon, and assumes that the constructs in this causal theory have the same meaning across cultures and can be captured with the same measures. The structure of the nomological network is presumed stable (that is, the constructs will interrelate in the same way), although the strength of the causal relationships may be moderated by culture (Brett et al. 1997). Finally, the cultural explanation for similarities and differences is assumed to be meaningful in all cultures as well as measurable with the same indicators. These assumptions make it very easy for the researchers to test for cross-cultural similarities and differences (in either mean levels of constructs or strengths of causal relationships), and to test the cultural explanation for these similarities and differences. The common complaint about the etic perspective is that scholars generally do not buy into these fundamental assumptions of generalizability (of theory
THE HEART OF DARKNESS
347
constructs, measurement, or the cultural explanation). This perspective is criticized for being insensitive to the nuances of any culture being studied (Triandis 1992; Berry 1989). I actually find this an unfair criticism, since the goal of a purely etic perspective is cultural generality – the comparison of equivalent phenomena to see whether mean levels of these phenomena or their interrelationship differ across cultural samples. The problem with etic research, as I see it, is that researchers are never purely etic. Instead they are comparing their own culture (for which they have deep insider knowledge) to other “arm’s length” cultures. The “imposed” mid-range causal theory and cultural explanation are not imposed from outside of all cultures in the research design. Therefore the study is emic for one culture and etic for others. Although I will advocate below that research designs incorporate emic and etic perspectives, a design that “blocks” these perspectives by culture is of course likely to bias results. How does the researcher who is interested in equivalent comparisons be as sensitive as possible to the nuances of all cultures being studied? The solution is to have as much emic “insider” information possible for all the cultures under investigation, but for the researchers to consciously consider how to transform that emic “culturally sensitive” knowledge into etic “culturally universal” constructs, models, and measures. Again this prescription calls for a team of researchers (or informants) and the N-way approach (Brett et al. 1997) for the problems of both theory construction and measurement. For theory construction, the team of researchers begins with their research question and each member develops a mid-range causal model to address that research question in his or her culture. This is the emic part of the theory construction. Now it is necessary to somehow transform these emic and unique models into an overall etic or generalizable model which will allow for testing in all cultures. Here I depart from the advice in Brett et al. (1997) that advocates forming a union model incorporating all the relevant constructs from each of the cultures being investigated. Instead, I suggest extracting only the intersection model, which incorporates those constructs that are part of each culture’s emic model. Hence the etic model is the intersection of all the culture’s emic models. The risk here is under-specification of the phenomenon in a particular culture of study (James et al. 1982). Yet, this may outweigh the risks of the overspecified union model that incorporates constructs irrelevant in some cultures. Aside from the statistical issues associated with trying to compare beta weights across cultures whose models have different (or different numbers of) constructs, there is a conceptual issue of measuring a culturally irrelevant construct. More constructs measured and correlated leads to a greater chance of
348
CATHERINE H. TINSLEY
type I error or finding a relationship where one does not truly exist (Cook & Campbell 1979). This seems to be a more dangerous risk than the risk of type II error, which would characterize the underspecified intersection model, where a researcher fails to find a relationship that truly does exist (Cook & Campbell 1979). Since science progresses as a body of knowledge that is modified over time, scientists generally adopt a conservative approach, so that type II error is considered more benign (Cook & Campbell 1979). Similarly, it may be less dangerous to design cross-cultural comparisons using the intersection model of all cultures emic models, rather than the union model. The researcher risks losing nuances of a particular culture (and hence risks missing a relationship that truly does exist), but minimizes the risks of concluding a relationship with spurious variables. A similar logic applies to the measurement issues, when the same construct may have a different domain across the various cultures of interest. The approach has been to capture this domain using culturally sensitive (emic) indicators when necessary (Brett et al. 1997; Hui & Triandis 1985; Triandis 1992). Thus a construct may be operationalized with measures A, B, and C in one culture and with measures A, B, C, and D in another culture. The bias seems to be towards cultural sensitivity, hence being as true to a culture’s emic knowledge as possible. Yet, this raises obvious issues of conceptual and measurement equivalence. If the domains are substantially different, are they really the same construct across cultures? And if they are measured with different indicators, then there are likely to be differences in correlations of this construct with others that have more to do with measurement differences than with cultural differences. Thus, I now think it is best to keep emic indicators to a minimum. My advice would be for the researcher to use the intersection set of all emic indicators, that is, to use only those items that are valid in all the cultures. Although this may fail to capture the entire domain of a construct in a particular culture, the researcher is more assured of equivalent comparisons. Moreover, the discussion section can explain to the reader the broader domain of a construct in a particular culture and speculate as to how that limits findings.
Putting It All Together After participating in several cross-cultural research projects, I have become humbled by the issue of how to construct rigorous cross-cultural comparisons. I still believe in testing mid-range theory across cultures, using a quantitative, confirmatory approach that tests for similarities and differences with an a priori cultural explanation. I still very much champion the N-way approach where
THE HEART OF DARKNESS
349
research questions are operationalized by a team of researchers or key informants. Yet, for managing the positivist/interpretive epistemologies and etic/ emic perspectives I am now less idealistic about the potential alternatives for conducting cross-cultural research. I see the pathway to rigor as more constrained, and my approach reflects a conservative bias. I suppose it is akin to the old political adage – to be young and not be idealistic is to have no heart, but to be seasoned and not be conservative is to have no head. Rigorous crosscultural research projects are possible. Yet, I believe researchers should be very cautious about culturally specific constructs or measures in any final research design. Although emic knowledge is invaluable for ensuring a valid operationalization of the research question for any one culture, this needs to be tempered by the need for equivalent models and measures across cultures. The goal of each researcher or informant should be to transform their emic knowledge into an etic model. Ultimately I would sacrifice internal validity for generalizability, in order to ensure equivalence for measurement and testing.
References Adams, J.S. (1963). “Toward an Understanding of Inequity.” Journal of Abnormal and Social Psychology, 67: 422–36. Bazerman, M.H. (2002). Judgment in Managerial Decision Making, 5th edition. New York: John Wiley & Sons. Berry, J.W. (1980). “Introduction to Methodology,” in H.C. Triandis & J.W. Berry, editors, Handbook of Cross-Cultural Psychology, volume 2. Needham Heights, MA: Allyn & Bacon. Berry, J.W. (1989). “Imposed Etics-Emics- and Derived Etics: The Operationalizations of a Compelling Idea.” International Journal of Psychology, 24: 721–735. Brett, J.M., C.H. Tinsley, M. Janssens, Z.I. Barnsness & A.L. Lytle. (1997). “New Approaches to the Study of Culture in I/O Psychology,” in P. Christopher Earley and Miriam Erez, editors, New Perspectives on I/O Psychology: 75–129. San Francisco, CA: Jossey-Bass, Inc. Cattell, R.B. (1988). “Multivariate Method and Theory Construction,” in J.R. Nesselroade. Y.R.B. Cattell, editors, Handbook of Multivariate Experimental Psychology (2nd edition): 1–20. New York: Plenum. Cook, T.D. & Campbell, D.T. (1979). Quasi Experimentation: Design and Analysis Issues for Field Settings. Boston: Hougton-Mifflin. Duhem, P. (1962). The Aim and Structure of Physical Theory. New York: Atheneum. Guba, E.G. & Lincoln, Y.S. (1994). “Competing Paradigms in Qualitative Research,” in N.K. Denzin & Y.S. Lincoln, editors, Handbook of Qualitative Research, 105–117. Thousand Oaks, CA: Sage. Hanson, N.R. (1958). Patterns of Discovery: An Inquiry into Conceptual Foundations of Science. Cambridge, England: Cambridge University Press. Hui, C.H. & Triandis, H.C. (1985). “Measurement in Cross-Cultural Psychology: A Review and Comparison of Strategies.” Journal of Cross-Cultural Psychology, 16 (2): 131–152. James, L.R., Mulaik, S., & Brett, J.M. (1982). Causal Analysis: Assumptions, Models, and Data. Thousand Oaks, CA: Sage.
350
CATHERINE H. TINSLEY
Kuhn, T.S. (1962). The Structure of Scientific Revolutions. Chicago: University of Chicago Press. Moore, J.F., Johns, G., & Pinder, C.C. (1980). “Towards Middle Range Theory: An Overview and Perspective,” in C.C. Pinder & J.F. Moore, editors, Middle Range Theory and the Study of Organizations, 1–16. Boston: Nijhoff. Lytle, A.L., Brett, J.M., Barnsness, Z.I., Tinsley, C.H. & Janssens, M. (1995). “A Paradigm for Quantitative Cross-Cultural Research in Organization Behavior,” in B.M. Staw and L.L. Cummings, editors, Research in Organizational Behavior, 17: 167–214. Popper, K.R. (1959). The Logic of Scientific Discovery. New York: Basic Books. Spradley, J.P. (1979). The Ethnographic Interview. Chicago: Holt, Rinehart & Winston, Co. Spradley, J.P. (1980). Participant Observation. New York: Holt, Rinehart, & Winston, Co. Triandis, H.C. (1972). “Major Theoretical and Methodological Issues in Cross-Cultural Psychology,” in W.J. Lonner Walters, editor, Readings in Cross-Cultural Psychology, pp. 26–38. Triandis, H.C. (1992). “Cross-Cultural Research in Social Psychology,” in D. Granberg & G. Sarup (Eds.), Social Judgments and Intergroup Relations: Essays in Honor of Muzafer Sherif. New York: Springer-Verlag. Weingart, L.R. (1997). “How did they do that? The Ways and Means of Studying Group Processes,” in L.L. Cummings & B.M. Staw, editors, Research in Organizational Behavior, Vol. 19: 189–239. Greenwich, CT: JAI Press.
Disparate Methods and Common Findings in the Study of Negotiation CARSTEN K.W. DE DREU and PETER J. CARNEVALE
Conflict and negotiation lie at the very heart of many different types of social interaction, be it among chimpanzees, businessmen, diplomats, spouses, or drug dealers and their junkies. Accordingly, conflict and negotiation is studied in many disciplines within the social sciences, including social and personality psychology, organizational behavior, communication sciences, economics, and political science. Yet at the same time, it seems that within each of these academic disciplines conflict and negotiation is studied in very different ways, that some methods and techniques are commonplace in some areas while unknown or even frowned upon in adjacent areas. An example is the debate about the usefulness of laboratory experiments in the study of conflict and negotiation. Proponents of the laboratory experiment argue that it provides high internal validity and allows one to detect causal relationships (Pruitt 1981). Opponents counter that laboratory experiments lead to a science of 18-year old psychology students performing abstract tasks of short duration and only because they have to. How could this provide any information about, for example, peace negotiations between Israel and the PLO, or an ongoing conflict between a manager and her subordinates? The problem with this type of debate is that it may lead scholars to discount research findings obtained with the “wrong” method. When some methods are more popular and more accepted in some academic disciplines than in others, this could reduce the extent to which insights about conflict and negotiation transcend academic boundaries. An alternative would be to assume that no method or technique is perfect and able to answer all questions at once. Triangulation – the use of different methods to study the same phenomenon – offers a solution to deal with deficiencies of each method and technique (Campbell & Stanley 1966). In this article, we address two tasks. First, we analyze the conflict and negotiation literature to provide an empirical profile of the variety of methods and techniques in the various disciplines; we suspect that some methods are more popular in some disciplines than others. Second, we selectively review
International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 351–360 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
352
CARSTEN K.W. DE DREU AND PETER J. CARNEVALE
effects in conflict and negotiation that have been studied with different methods to see whether triangulation indeed works and provides useful insight.
Disparate Methods in the Study of Conflict and Negotiation As is often the case when research is driven by problems surrounding real-life phenomena, conflict and negotiation is studied with different methods, including comparative case studies, laboratory experiments, cross-sectional or longitudinal field research using surveys and questionnaires, participant observation, computer simulation, mathematical modeling, and so on. Data are analyzed with a variety of techniques including statistical models that derive from the General Linear Model (e.g., analysis of variance, multiple regression), multidimensional scaling, time series analysis, or Markov-Chain analysis. The casual reader may feel that some areas in the social sciences tend to use certain methods and techniques much more than others. Researchers in the organizational behavior tradition may be more inclined to conduct field research involving professionals in organizations, whereas those in social and personality psychology are more likely to design laboratory experiments. Qualitative and comparative case studies or ethnographic approaches, seem more common in political science. However, as far as we know, there is no empirical foundation for these assumptions. To redress this situation, and to provide a more quantitative insight into the popularity of methods and techniques used in the study of conflict and negotiation, we coded research articles concerned with conflict or negotiation published in a particular scientific field. We distinguished five scientific areas: political science, economics, communication science, organizational behavior, and social and personality psychology. Within each of these areas, we selected two or three journals based on their general reputation and impact, and the fact that they tend to publish work on negotiation and related issues. Because we wanted to cover a full five-year period (1997–2001), a final criterion was that the journal existed throughout this entire period. The following journals were selected: (1) Political Science: American Political Science Review [APSR], Journal of Conflict Resolution [JCR]; (2) Economics: Econometrica [ECA], Experimental Economics [EE], Journal of Economic Behavior & Organization [JEBO]; (3) Communication Science: Human Communication Research [HCR], Communication Research [CR], Communication Monographs [CM]; (4) Organizational Behavior: Academy of Management Journal [AMJ], Journal of Applied Psychology [JAP], Organizational Behavior and Human Decision Processes [OBHDP], and (5)
DISPARATE METHODS AND COMMON FINDINGS IN THE STUDY OF NEGOTIATION
353
Social and Personality Psychology: Journal of Personality and Social Psychology [JPSP], Personality and Social Psychology Bulletin [PSPB], and Journal of Experimental Social Psychology [JESP]. In addition, we added two journals that are highly relevant and thematic, but did not clearly fall within one of the above scientific disciplines: International Journal of Conflict Management [IJCM], and Group Decision and Negotiation [GDN]. Obviously, this is a representative but not comprehensive list of academic outlets for the study of negotiation and social conflict, and it excludes many relevant journals. The selection should, however, allow us to detect possible trends in the use of methods and techniques across the social sciences. Two research assistants independently coded all articles on (a) conflict, (b) negotiation, or (c) experimental game in terms of the method and/or technique used (see also Table 2). The checklist covers all the methods and techniques considered in the various articles in this special issue and some others that emerged while coding was undertaken. Inter-rater reliabilities were satisfactory (all Cohen’s K > .80), and discrepancies were solved through discussion, with the first author serving as mediator/arbitrator. A total of 345 articles were coded. Table 1 provides the total number of articles on conflict and negotiation for each of the five social science disciplines, as well as the percentage of the total number of articles published in that journal in the entire five-year period. As can be seen, both in absolute and relative terms, conflict and negotiation assume a more important role in political science and economics, than in any of the other areas we analyzed. That not all papers in the topical journals were coded as dealing with conflict and negotiation is because some work published in these journals is conceptual in nature, dealing with group decision-making and support systems or reviews published books. Table 1. Articles Published and Articles on Conflict Published in Top-tier Journals in Five Social Sciences Discipline Between 1997 and 2001 Articles Area Organizational Behavior Social & Personality Psychology Political Sciences Communication Sciences Economics Topical Journals
Conflict and Negotiation 44 28 75 13 107 78
Percentage of Total 2.1 2.1 16.5 8.6 14.5 46.2
354
CARSTEN K.W. DE DREU AND PETER J. CARNEVALE
In Table 2, we tabulate all methods and techniques across disciplines. The statistics reveal the relative popularity of certain methods and the rarity of others. The reader should bear in mind that some studies included at least two different methods and techniques so that the numbers in Table 2 exceed the number of articles listed in Table 1. As can be seen in Table 2, about half of the studies coded included laboratory experiments, and about one-third included some form of mathematical modeling or included surveys and questionnaires. Over forty per cent included an outcome measure. Far less common are methods like multi-dimensional scaling, quantitative coding of behavioral sequences, meta-analysis, multi-level analysis, time series analysis, or participant observation. Because the latter type of methods and techniques are less common outside the domain of conflict and negotiation as well, the low numbers observed here may reflect lack of training in these techniques or their labor intense character. Table 2. Methods and Techniques in the Study of Negotiation and Social Conflict
Use of Mathematical Modeling Computer Simulation Use of Games for Model Testing Laboratory Experiment Use of Outcome Measures Coding of Communication Processes Quantitative Coding of Behavioral Sequences Use of Surveys and Questionnaires Markov-Chain Analysis Methods of Culture Comparison Multi-dimensional Scaling Multi-Level Analysis Meta-Analysis True Field Experiment Field Research Analysis of Archival Data Time Series Analysis Workshop as Research Tool Participant Observation Comparative Case Analysis Internet as a Method Methodology for Personality Processes
Observed Number
Percentage
124 2 82 161 14 41 3 134 3 18 1 1 4 1 35 52 3 5 1 14 1 19
35.9 0.6 23.8 46.7 140.9 11.9 0.9 38.8 0.9 5.2 0.3 0.3 1.2 0.3 10.1 15.1 0.9 1.4 0.3 4.1 0.3 5.5
Note: Articles can be coded in multiple categories when multiple methods and techniques are used (e.g., “laboratory experiment” and “surveys and questionnaires”). Total number of articles coded = 345, with 1 article not fitting into any of the above categories.
DISPARATE METHODS AND COMMON FINDINGS IN THE STUDY OF NEGOTIATION
355
Table 3. Relative Frequency of Methods and Techniques in the Study of Negotiation as a Function of Social Sciences Discipline Social Science Discipline OB
SP
COMM ECON POLSCI X 2(4)
Method Use of Mathematical Modeling Computer Simulation Use of Games for Model Testing Laboratory Experiment Use of Outcome Measures Coding of Behavioral Sequences Coding of Communication Processes Use of Surveys and Questionnaires Markov-Chain Analysis Methods of Culture Comparison Multi-dimensional Scaling Multi-Level Analysis Meta-Analysis True Field Experiment Field Research Analysis of Archival Data Time Series Analysis Workshop as Research Tool Participant Observation Comparative Case Analysis Internet as a Method Methodology for Personality Processes
0.0 0.0 0.0 84.1 54.5 0.0 31.8 68.2 0.0 18.2 0.0 0.0 2.3 0.0 11.4 4.5 0.0 2.3 0.0 2.3 0.0 4.5
0.0 0.0 0.0 85.7 53.6 3.6 17.9 89.3 7.1 3.6 0.0 0.0 3.6 0.0 10.7 0.0 0.0 0.0 0.0 0.0 0.0 21.4
7.7 0.0 0.0 76.9 38.5 0.0 53.8 69.2 0.0 23.1 0.0 0.0 0.0 0.0 7.7 7.7 0.0 0.0 0.0 15.4 0.0 15.4
68.9 1.9 58.5 43.4 44.3 0.0 0.9 5.7 0.9 0.0 0.0 0.0 0.0 0.0 0.0 2.8 0.0 0.0 0.0 0.0 0.0 0.9
48.0 0.0 20.0 10.7 25.3 0.0 6.7 14.7 0.0 4.0 0.0 0.0 0.0 0.0 2.7 57.3 4.0 0.0 0.0 10.7 1.3 1.3
91.29*** 3.04 82.53*** 86.12*** 13.25** 8.53* 53.01*** 124.15*** 10.62** 27.36** NA NA 5.18 NA 14.52** 105.67*** 7.73 5.06 NA 18.39** 2.56 27.06**
Note: OB = Organizational Behavior; SP = Social Psychology; COMM = Communication Science; ECON = Economics; POLSCI = Political Science; NA = Not Available; * p < .10, ** p < .01, *** p < .001
In Table 3, we examine the relative popularity of methods and techniques as a function of scientific discipline. In this analysis, we excluded the articles published in cross-disciplinary journals. Consistent with common beliefs, within social and personality psychology the large majority of studies involved laboratory experiments and included surveys and questionnaires. Somewhat surprising is the finding that within the organizational behavior area, an equally large majority of studies involved laboratory experiments. When we compare the relative frequency across disciplines, many significant differences emerge. Mathematical modeling and the use of games for model testing is virtually absent in organizational behavior, social and personality psychology, and the communication sciences, but strongly relied upon in economics and political science. Surveys and questionnaires, the coding of behavior, and cross-cultural comparisons are frequently used in organizational
356
CARSTEN K.W. DE DREU AND PETER J. CARNEVALE
behavior, social and personality psychology, and the communication sciences, but rarely employed in economics and political science. Archival data are used more often and laboratory experiments less in political science than anywhere else.
Common Findings The above analysis supports the idea that some methods and techniques (e.g., laboratory experiments, surveys) are more frequently used than others (e.g., time series analysis, meta-analysis, coding of behavioral sequences). Substantial differences between social science disciplines emerge as well. Organizational behavior, social and personality psychology, and the communication sciences share a heavy reliance on laboratory experiments, the use of surveys, and the coding of communication processes. These methods and techniques are far less popular in economics and political science, where mathematical modeling, the use of games for model testing, and the use of archival data tend to dominate. To some extent the differences in methods and techniques reflect different research questions scholars in different areas pursue. Questions with regard to the economic correlates of actual war decisions cannot be answered with laboratory experiments, whereas questions with regard to neurological correlates of facing a competitive opponent require at this time a laboratory design. Sometimes, however, scholars within different areas of the social sciences pursue similar research questions, but with different methods and techniques. As mentioned, the use of different methods and techniques to study similar processes and phenomena may result in more confidence in the validity and robustness of these processes and phenomena (Campbell & Stanley 1966). Furthermore, using different methods and techniques allows one to view a particular phenomenon from different angles. To illustrate this, we discuss research on gain-loss framing conducted in organizational behavior and social and personality psychology on the one hand, and in political science on the other, and research on interaction sequences conducted in social psychology on the one hand, and the communication sciences on the other.
Gain-Loss Framing in Laboratory Negotiations and International Crises Inspired by Kahneman and Tversky’s (1979) prospect theory, Bazerman, Magliozzi and Neale (1985) conducted an experiment in which half of the negotiators were presented with payoffs presented as more than their walk-
DISPARATE METHODS AND COMMON FINDINGS IN THE STUDY OF NEGOTIATION
357
way value, and the other half were presented with payoffs presented as less than their highest possible outcome. Thus, although negotiators’ objective outcomes were the same in both conditions, half were led to view these as gains, and half were led to view these as losses. Consistent with the idea that lossframing makes people more risk-seeking and that losses are more painful than gains are pleasant, results showed that dyads with a loss frame more often reached an impasse and less often reached integrative, win-win solutions than dyads with a gain frame. Since this original experiment, multiple studies have replicated and extended these results, uncovering both underlying psychological processes for the effects (Bottom 1997; De Dreu, Carnevale, Emans & Van de Vliert 1994), as well as boundary conditions under which the framing effect is mitigated or even reversed (e.g., De Dreu & McCusker 1997). Jervis (1989), a political scientist, observed that states seem to make greater efforts to preserve the status quo against a threatened loss than to improve their position by a comparable amount. A state might be willing, for example, to fight and defend the same territory that it would not have been willing to fight to acquire, or to accept greater costs in order to maintain an international regime than to create it in the first place. Loss aversion – losses are more painful than equivalent gains are pleasurable – could explain these tendencies as much as it explains the laboratory findings on gain-loss framing reviewed above (Levy 1992). A series of in-depth case studies substantiates these claims. For example, Farnham (1992) demonstrates that loss aversion may well explain Roosevelt’s decision-making behavior during the Munich crisis in September 1938. McDermott uses prospect theory to explain President Carter’s decision to rescue the American hostages in Iran in 1980. He argues that Carter found himself in a loss-frame and took considerable risk to regain the status quo.
The Law of Reciprocation Reciprocation refers to the tendency to match another’s person’s behavior (Gouldner 1960). Outside the domain of conflict and negotiation, social psychologists and etiologists have discovered that such mimicry is often subconscious and automatic and that it manifests itself at the level of facial expressions, non-verbal behavior, and in explicit exchange of goods and services (e.g., Aureli & De Waal 2000; Van Balen et al. in press). Within conflict and negotiation, with its strong pressures to act strategically, similar tendencies to reciprocate have often been observed. For example, using the prisoner’s dilemma game or its variations, researchers in psychology (e.g.,
358
CARSTEN K.W. DE DREU AND PETER J. CARNEVALE
Deutsch 1949; Pruitt 1998) have demonstrated that individuals exhibit a strong tendency to reciprocate others’ cooperative or competitive choices, even when not reciprocating (i.e., responding competitively when the other makes a cooperative choice) would be in one’s own immediate self-interest. The strong tendency to reciprocate has been uncovered in other areas, with quite different methods and techniques. Communication scientists coded behavioral episodes during negotiation and, using Markov-Chain analyses, showed a strong tendency for negotiators to reciprocate each other’s cooperative (integrative) as well as competitive (distributive) behaviors (e.g., Weingart, Hyder, Genovese & Prietula 1999; Olekalns & Smith 2000). Political scientists including Leng (1993) examined archival data across multiple interstate crises and across extended periods in history. One important finding is that decision making by one state tends to be reciprocated by similar decisions made by the counter-part, and this strong inclination to reciprocate predicts the occurrence of war quite accurately. All this led Ostrom (1998: 12) to note: “. . . while individuals vary in their propensity to use reciprocity . . . a substantial proportion of the population . . . has sufficient trust that others are reciprocators to cooperate with them even in one-shot, no-communication experiments. Furthermore, a substantial proportion of the population is also willing to punish non-cooperators (or individuals who do not make fair offers) at a cost to themselves . . .” Likewise, Komorita, Parks, and Hulbert (1992: 608) note “. . . the reciprocity norm is relevant and important in a social dilemma situation . . . [it] prescribes that we should help those who have helped us in the past, and retaliate against those who have injured us . . . the reciprocity norm is the underlying basis of stable relationships, and theorists have postulated that reciprocity is a basic norm of social interaction . . .”.
Conclusion Our data on the use of various methods in general and across disciplines suggest two immediate observations: first, the study of negotiation and social conflict is a very small part of some domains (about 2% of published articles in social psychology as well as organizational behavior; about 8% in communication), and closer to the core of other domains (political science and economics). Of course, one reason for this may be the existence and attraction of the specialty outlets such as International Journal of Conflict Management, Group Decision and Negotiation, Journal of Conflict Resolution, Negotiation Journal, and of course, International Negotiation.
DISPARATE METHODS AND COMMON FINDINGS IN THE STUDY OF NEGOTIATION
359
Second, disciplines tend to maintain their own way of doing things: laboratory experiments that entail coding of behavior and self-report data using surveys are especially popular in psychology, organizational behavior, and communication sciences; mathematical modeling, the use of experimental games, and the use of archival data, are especially popular in economics, and political science. Diverse methods can be a source of convergent insights, yet they also can be a barrier to understanding if effects observed in one domain are not easily accessible to scholars in another. Such a barrier is less an issue when scholars working in different areas are aware of the various methods and techniques that exist and have some basic understanding of their virtues and limitations. The consistency in research findings with regard to gain-loss framing, and reciprocity, suggests that different methods and techniques allow one to study similar phenomena and to build a body of knowledge that would have been impossible had the research community limited itself to a narrow set of settings, populations, and research methods. If our effort to organize these two special issues of International Negotiation on method tells us anything, it is that researchers should adopt, or should continue to employ, triangulation as an approach to validity. In navigation, triangulation is the technique where two visible points are used to determine the location of a third point. Applied to validity and reliability of measurement, triangulation is the use of multiple, different indicators in such a way that errors can be excluded and underlying constructs can be identified (Campbell & Fiske 1959). The concept applies quite well not only to measurement, but to all aspects of method. When two or more methods or data sources converge on a construct, we have greater assurance that our conclusions are not driven by an error or artifact of any one procedure. Each method offers strengths and weaknesses, and to the extent they do not overlap, we can stand on more solid ground with conclusions about theory. An exciting direction is the development of collaborative enterprises such as the multi-disciplinary research team where scholars with complementary expertise work together on common problems, a successful model in the natural sciences (Hopmann 2002).
References Aureli, F., & De Waal, F.B.M. (2000). (Editors). Natural conflict resolution. Berkeley: University of California Press. Bazerman, M.H., Magliozzi, T., & Neale, M.A. (1985). “Integrative bargaining in a competitive market.” Organizational Behavior and Human Decision Processes, 35: 294–313.
360
CARSTEN K.W. DE DREU AND PETER J. CARNEVALE
Bottom, W.P. (1998). “Negotiator risk: Sources of uncertainty and the impact of reference points on negotiated agreements.” Organizational Behavior and Human Decision Processes, 76: 89–112. Campbell, D.T., & Fiske, D.W. (1959). “Convergent and discriminant validation by the multitrait-multimethod matrix.” Psychological Bulletin, 56: 81–105. De Dreu, C.K.W., & McCusker, C. (1997). “Effects of gain-loss frames on cooperation in twoperson social dilemmas: A transformational analysis.” Journal of Personality and Social Psychology, 72: 1093–1106. De Dreu, C.K.W., Carnevale, P.J.D., Emans, B.J.M., & Van de Vliert, E. (1994). “Effects of gain-loss frames in negotiation: Loss aversion, mismatching, and frame adoption.” Organizational Behavior and Human Decision Processes, 60: 90–107. Deutsch, M. (1949). “A theory of cooperation and competition.” Human Relations, 2: 199–231. Farnham, B. (1992). “Roosevelt and the Munich Crises: Insights from Prospect Theory.” Political Psychology, 13: 205–235. Jervis, R. (1989). The meaning of the nuclear revolution. Ithaca, NY: Cornell University Press. Gouldner, A.W. (1960). “The norm of reciprocity: A preliminary statement.” American Sociological Review, 25: 161–178. Hopmann, P.T. (2002). “Negotiating data: Reflections on the qualitative and quantitative analysis of negotiation processes.” International Negotiation, 7: 67–85. Kahneman, D., & Tversky, A. (1979). “Prospect theory: An analysis of decision under risk.” Econometrica, 47: 263–291. Komorita, S.S., Parks, C., & Hulbert, L. (1992). “Reciprocity and the induction of cooperation in social dilemmas.” Journal of Personality and Social Psychology, 62: 607–617. Leng, R.J. (1993). Interstate crisis behavior, 1816–1980: Realism versus reciprocity. New York, NY: Cambridge University Press. Levinger, G., & Rubin, J.Z. (1994). “Bridges and barriers to a more general theory of conflict.” Negotiation Journal, (July): 201–215. Levy, J.S. (1992). “Prospect theory and international relations: theoretical applications and analytical problems.” Political Psychology, 13: 283–311. McDermott, R. (1992). “The failed rescue mission in Iran: An application of prospect theory.” Political Psychology, 13: 16–29. Olekalns, M. & Smith, P. (2000). “Understanding optimal outcomes: The role of strategy sequences in competitive negotiations.” Human Communication Research, 26: 527–557. Ostrom, E. (1998). “A behavioral approach to the rational choice theory of collective action.” The American Political Science Review, 92: 1–22. Pruitt, D.G. (1981). Negotiation. New York: Academic Press. Pruitt, D.G. (1998). “Social conflict,” in D. Gilbert, S.T. Fiske, & G. Lindzey (Editors). Handbook of social psychology (4th ed.), 2. New York: McGraw-Hill: 89–150. Weingart, L.R., Prietula, M.J., Hyder, E.B., & Genovese, C.R. (1999). “Knowledge and the sequential processes of negotiation: A Markov Chain analysis of response-in-kind.” Journal of Experimental Social Psychology, 35: 366–393.
INDEX Approach Cross-culture approach 135–136 Crosslevel approach 136, 147 Multilevel approach 135, 143 Multimethod approaches 2 Bayesian analysis 68–69, 76 Bargaining context 179, 186, 314 Case study 23, 57, 80, 100, 165, 167, 173, 187, 342 Coding 63, 76, 85–86, 105–107, 109–113, 117, 156, 172, 185, 206, 216, 231–232, 235, 318–319, 353–356, 359 Communication communication processes 177, 257, 356 Comparative case studies 100, 165, 169, 174, 352 Conflict Conflict data 128, 130, 327 Conflict resolution 49, 53, 55–57, 69, 79, 90, 96, 151, 235, 307, 323, 325–327, 330, 352, 358 Interactive conflict resolution 55, 56 Marital conflict 124, 126–127, 129 Organizational conflict 123–124, 151 Context 24, 33, 39, 41, 53, 55–56, 61, 67, 96, 105, 108, 110, 113, 115, 125, 135, 137–138, 145, 152, 156, 160–161, 167–168, 177–179, 182–188, 190, 214–215, 232, 249, 290, 292, 294–295, 303, 307, 311, 314–315, 317, 320, 323–324, 334 Crisis 63, 71, 94–95, 97–100, 152–153, 159, 357 Crosslevel approach: see Approach Cross-culture approach: see Approach Data analysis 1, 42, 82, 105, 183, 189, 234, 267, 269, 278, 290, 301, 313 Deception 153, 158, 290, 293, 297–300, 326
Decision support systems 95, 97, 100–101 Discovery 7, 15–16, 18, 53, 64, 94, 186, 189, 270 Discourse analysis 5, 177–179, 181–183, 185–190 Dyadic analysis 1 Dynamic processes 61 Emotion 13, 96, 122, 149–161, 202, 212 Empirical research 90, 154, 283, 307, 309, 312–313, 317 Ethnography 41–42, 160 Experimental design 61, 66–67, 93, 99–100, 128, 196, 198, 206, 212, 218–220, 273, 289, 298–301 Experimental economics 289–291, 295–298, 302–303, 352 Field work 314 Field experiment 193–194, 197–203, 205–207 Incentives 25–26, 30, 155, 160, 211, 213–214, 221, 278–279, 290–294, 296–297, 299, 316 Independent variable 8, 14–15, 20, 66, 81, 85–86, 88, 94, 130, 138–139, 168, 196, 201, 212, 218, 220, 241, 301, 331 Interrupted time-series: see Time series Laboratory experiment 1, 66, 193–194, 199–202, 206–207, 211–213, 216, 221, 351–352, 354–356, 359 Language 1, 25, 41, 139, 152, 159, 177–180, 182–184, 186–190, 249, 274–275 Markov chain analysis 257–258, 352 Mediation Mediation style 97 Meta-analysis 227–229, 231, 233–236, 281, 354, 356
International Negotiation Series 2: P. Carnevale and C.K.W. de Dreu (eds.) Methods of Negotiation Research, 361–362 © 2006 Koninklijke Brill NV. Printed in the Netherlands.
362
INDEX
Methods Internet-based methodology 274–280, 283–285 Qualitative method 79, 160, 177 Research methods 1, 5, 39, 46–47, 50, 54, 90, 122, 130, 177, 227, 275, 285, 313, 315, 324, 341, 359 Self-report methodologies 326 Statistical methods 113, 174, 301, 313, 331, 333 Multidimensional scaling 5, 239–240, 243–244, 252–253, 352 Multilevel approach: see Approach Multimethod approaches: see Approach Multivariate techniques 240
Archival research 79, 82–83, 88, 90 Field research 5, 7–8, 10–11, 13–15, 18–20, 198, 222, 317, 352 Mediation research 96–97 Negotiation research 1, 5, 19, 39, 46, 136, 149, 177, 215, 231, 233, 236, 319 Qualitative research 160 Research in natural settings 41, 222 Research methods 1, 5, 39, 46–47, 50, 54, 90, 122, 130, 177, 227, 275, 285, 313, 315, 324, 341, 359 Research synthesis 228, 235 Social conflict research 3, 5, 57, 199, 213, 227, 236, 253, 282, 324, 327 Reflexivity 186
Negotiation International negotiation 1, 5, 63–64, 72, 114, 135–136, 171, 250, 286, 358–359 Israeli-Palestinian Negotiation 28 Legal negotiation 307–312, 315, 317–320 Negotiation processes 24–25, 43, 45–46, 55–56, 61–62, 66, 71, 74, 77, 95–97, 101, 105–106, 113, 115–117, 135–137, 145–146, 156, 167, 174, 177–178, 182, 220, 252, 257, 259, 286, 310, 319–320 Negotiation texts 179, 188
Simulation 93–94, 97–101, 152, 155, 197, 216, 262, 268, 318, 320, 352 Social conflict 1, 3, 5, 34, 57, 193, 199, 206, 211, 213, 227, 236, 239, 253, 282, 324, 327, 353, 358 Social context 186 Subject pool 273, 276, 290, 295–296 Surveys 1, 46, 90, 121–124, 128, 139, 141, 161, 198, 213, 247–248, 252, 273–278, 280–281, 283, 285, 291–293, 296, 302, 307, 311–314, 317, 320, 352, 354–356, 359
Personality 5, 39, 157, 253, 278, 285, 323–335, 351–353, 355–356 Post-dicting events 62 Prenegotiation 49, 53, 57, 167 Problem-solving workshop 49–54, 56–57 Process tracing 71-72, 74–76 Psychometrics 124 Qualitative method 79, 160, 177 Questionnaire 3, 55, 121–125, 128, 130–131, 139, 152, 157–159, 199–200, 220, 293, 317, 326–327, 352, 354–355 Research Action research 49, 51, 54–57
Taba 23–34, 36 Test-construction 345 Time series Time-series designs 61, 66, 70, 75 Interrupted time-series 62, 65–66, 76 Validity Validity threats 327
81, 122, 127, 274, 279,
web experiments 277 web survey 276