Ebook Carnevale, P. Ed., de Dreu, C.K.W. Ed. Methods of Negotiation Research International Negotiation Series

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), 20042005
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 ...................................... 1

Peter J. Carnevale and Carsten K.W. De Dreu
Methods of Negotiation Research II .......................................................... 5
Chapter 1
The Joys of Field Research ........................................................................ 7
James A. Wall, Jr.
Chapter 2
How Much Do We Know About Real Negotiations?
Problems in Constructing Case Studies .................................................... 23
David Matz
Chapter 3
Studying Negotiations in Context: An Ethnographic Approach .............. 39
Ray Friedman
Chapter 4
The Problem-Solving Workshop as a Method of Research .................... 49
Ronald J. Fisher
Chapter 5
Time-Series Designs and Analyses ............................................................ 61
Daniel Druckman
Chapter 6
Social Research and the Study of Mediation: Designing and
Implementing Systematic Archival Research ............................................ 79
Jacob Bercovitch
Chapter 7
Reections on Simulation and Experimentation in the Study of
Negotiation .................................................................................................. 93
Jonathan Wilkenfeld
vi CONTENTS
Chapter 8
Quantitative Coding of Negotiation Behavior ...................................... 105
Laurie R. Weingart, Mara Olekalns, and Philip L. Smith
Chapter 9
The Use of Questionnaires in Conict Research .................................. 121
Aukje Nauta and Esther Kluwer
Chapter 10
A Multilevel Approach to Investigating Cross-National Differences in
Negotiation Processes .............................................................................. 135
Xu Huang and Evert Van de Vliert
Chapter 11
Methodological Challenges in the Study of Negotiator Affect ............ 149
Bruce Barry and Ingrid Smithey Fulmer
Chapter 12
Comparative Case Studies ...................................................................... 165
I. William Zartman
Chapter 13
Discourse Analysis: Mucking Around with Negotiation Data ............ 177
Linda Putnam
Chapter 14
Field Experiments on Social Conict .................................................... 193
Dean G. Pruitt
Chapter 15
Laboratory Experiments on Social Conict .......................................... 211
Chapter 16
Managing Conict in the Literature: Meta-analysis as a Research
Method .................................................................................................... 227
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 Conict .................. 239
Robin S. Pinkley, Michele J. Gelfand, and Lili Duan
CONTENTS vii
Chapter 18
Markov Chain Models of Communication Processes in
Negotiation .............................................................................................. 257
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 ................................................................................ 273
Yeow Siah Cha
Chapter 20
The Method of Experimental Economics .............................................. 289
Rachel Croson
Chapter 21
Empirical Research in Law and Negotiation ........................................ 307
Rebecca Hollander-Blumoff
Chapter 22
Methodologies for Studying Personality Processes in
Interpersonal Conict .............................................................................. 323
Lauri A. Jensen-Campbell and William G. Graziano
Chapter 23
The Heart of Darkness: Advice on Navigating Cross-Cultural
Research .................................................................................................. 341
Catherine H. Tinsley
Chapter 24
Disparate Methods and Common Findings in the Study of
Negotiation .............................................................................................. 351
Carsten K.W. De Dreu and Peter J. Carnevale
Index ........................................................................................................ 361

Contributors
Bruce Barry:
Owen Graduate School of Management, Vanderbilt University, Nashville, TN
37203 USA
E-mail: bruce.barry@vanderbilt.edu
Jacob Bercovitch:
Department of Political Science, University of Canterbury, Private Bag 4800,
Christchurch, New Zealand
E-mail: jacob.bercovitch@canterbury.ac.nz
Peter J. Carnevale:
Department of Psychology, New York University, 6 Washington Place, Room
577, New York, NY 10003 USA
E-mail: peter.carnevale@nyu.edu
Rachel Croson:
The Wharton School, University of Pennsylvania, 567 Huntsman Hall, 3730
Walnut Street, Philadelphia, PA 191046340 USA
E-mail: crosonr@wharton.upenn.edu
Carsten K.W. de Dreu:

Organizational Psychology, University of Amsterdam, Roetersstraat 15, 1018
WB, Amsterdam, The Netherlands
E-mail c.k.w.dedreu@uva.nl
Daniel Druckman:
Institute for Conict Analysis and Resolution (ICAR), George Mason Univer-
sity, Fairfax, VA 22030 USA
E-mail: ddruckma@gmu.edu
Lili Duan:
Department of Psychology, University of Maryland, College Park, MD 20742
USA
E-mail: lduan@psyc.umd.edu)
x CONTRIBUTORS
Ronald J. Fisher:
International Peace and Conict Resolution Program, School of International
Service, American University, 4400 Massachusetts Avenue, NW, Washington,
DC 20016 USA
E-mail: rsher@american.edu
Ray Friedman:
Owen Graduate School of Management, Vanderbilt University, Nashville, TN
37203 USA
E-mail: ray.friedman@owen.vanderbilt.edu
Michele J. Gelfand:
Department of Psychology, University of Maryland, College Park, MD 20742
USA
E-mail: mgelfand@psyc.umd.edu
Treena L. Gillespie:
Department of Management, California State University-Fullerton, P.O. Box
6848, Fullerton, CA 928346848 USA
E-mail: tgillespie@fullerton.edu
Rebecca Hollander-Blumoff:
New York University School of Law, Lawyering Program, 245 Sullivan Street,
New York, NY 100121301 USA
E-mail: hollande@juris.law.nyu.edu
Xu Huang:
Department of Management & Marketing, The Hong Kong Polytechnic Uni-
versity, Hung Hom, Kowloon, Hong Kong
E-mail: mshuangx@polyu.edu.hk
Esther Kluwer:
Department of Social and Organisational Psychology, Utrecht University, P.O.
Box 80, 140 NL 3508 TC Utrecht, The Netherlands
E-mail: E.S.Kluwer@fss.uu.nl
David Matz:
Graduate Programs in Dispute Resolution, University of Massachusetts Boston,
100 Morrissey Boulevard, Boston, MA 02125 USA
E-mail: David.Matz@Umb.edu
CONTRIBUTORS xi
Aukje Nauta:
TNO Work and Employment, P.O. Box 718, 2130 AS Hoofddorp, the Nether-
lands E-mail: a.nauta@arbeid.tno.nl
Mara Olekalns:
Melbourne Business School, University of Melbourne, 200 Leicester Street
Carlton VIC 3053 Australia
E-mail: m.olekalns@mbs.edu
Robin L. Pinkley:
Director of the American Airlines Center for Labor Relations and Conict
Resolution, Edwin L. Cox School of Business, Southern Methodist University,
Fincher Building, P.O. Box 750333, Dallas, TX 75275 USA
E-mail: rpinkley@mail.cox.smu.edu
Dean G. Pruitt:
Institute for Conict Analysis and Resolution, George Mason University, 9006
Friars Road Bethesda, Maryland 20817 USA
E-mail: dean@pruittfamily.com
Linda L. Putnam:
Department of Communication, Texas A&M University, 4234 TAMU, College
Station, TX 778434234 USA
E-mail: lputnam@tamu.edu
Chayeow Siah:
Department of Social Work and Psychology, National University of Singa-
pore, 11 Law Link, Singapore 117570
E-mail: swkchays@nus.edu.sg
Philip L. Smith:
Department of Psychology, University of Melbourne, Victoria, 3010 Australia
E-mail: philipls@unimelb.edu.au
Ingrid Smithey Fulmer:

Eli Broad Graduate School of Management, Michigan State University, East
Lansing, MI 48824
USA E-mail: fulmer@msu.edu
xii CONTRIBUTORS
Alice F. Stuhlmacher:
Department of Psychology, DePaul University, 2219 N. Kenmore Avenue,
Chicago, IL 60614
USA (E-mail: astuhlma@depaul.edu)
Catherine H. Tinsley:
The McDonough School of Business, Georgetown University, Washington,
DC 20057 USA
E-mail: Tinsleyc@georgetown.edu
Evert van de Vliert:

Social and Organizational Psychology, University of Groningen, Grote Kruisstraat
2/I, 9712 TS Groningen, The Netherlands
E-mail: e.van.de.vliert@ppsw.rug.nl
James A. Wall, Jr.:

College of Business, University of Missouri, Columbia, 506 Cornell Hall,
Columbia, MO 652112600 USA
E-mail: wall@missouri.edu
Laurie R. Weingart:
David A. Tepper School of Business Carnegie Mellon University, 5000 Forbes
Avenue, Pittsburgh, PA 15213 USA
E-mail: weingart@cmu.edu
Jonathan Wilkenfeld:
Department of Government and Politics, University of Maryland, College
Park, Maryland 20742 USA
E-mail: Jwilkenf@gvpt.umd.edu
I. William Zartman:
School of Advanced International Studies, the Johns Hopkins University, 1740
Massachusetts Ave, Washington, DC 20036 USA
E-mail: zartman@jhu.edu
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
eld. 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
conict span all levels of society, including interactions between nation states,
small groups and organizations, people in close relationships, and even chil-
dren on a playground. Its study reects work in elds 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 eld. Yet practitioners and scholars, on occasion, may wonder if a partic-
ular 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, 13
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 eld
forward. With these articles, we make the point that there is a wealth of
methodological tools that negotiation and conict researchers have at hand,
and each has strengths and weaknesses. Our specic objectives include the
following: provide an introduction to a variety of methods and their utility;
identify issues and controversies with various methods; increase the accessi-
bility 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 conict 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 high-
lights common features of effects in negotiation that have been obtained with
diverse methodological tools. For example, many studies show that a forgiv-
ing strategy in negotiation can be exploited, and a tough strategy can backre
by producing a competitive response, and this has been obtained in both lab-
oratory 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 reects Don Campbells 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 identied (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 Authors Task
Each author was asked to write a paper about a specic 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 con-
ducting research in the given domain the trials and tribulations and the
tricks of the trade, so to speak, covering things like gut feelings about train-
ing coders for content analysis, implicit theories about which order of ques-
tionnaires 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 negotia-
tion and social conict. Many people helped make this method special issue
possible. We are grateful to the following individuals who provided reviews:
Andy Schotter, Kathleen OConnor, 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 1:1941.
Campbell, D.T., & Fiske, D.W. (1959). Convergent and discriminant validation by the multi-
trait-multimethod matrix, Psychological Bulletin 56:81105.
Methods of Negotiation Research II
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 conict. The 25 papers cover a lot
of ground: general methodological techniques and approaches eld 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,
reect well on the application of the scientic approach to understanding
conict and negotiation. But space limitations led us to leave some areas out;
we hope the reader will forgive us our omissions.
One difculty 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 rst, our inclination was to order the
papers randomly to avoid the imposition of an articial 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 eld), analy-
sis 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 negotia-
tion and social conict. 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 eld.
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

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 OConnor, 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 fathers B-17 squadron had put
up 13 planes. Over Germany, the ghters came out of the sun and in 10 sec-
onds eight planes 72 airmen went down, like burning leaves.
The current mission had gone better; fewer comrades had been lost; and
Dads bomber was somewhere back over England, but in a cloud bank, iso-
lated 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 mis-
sion, and the pilot was focused on the low fuel gauge.
The compass wasnt 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; were home. Then, in relief, he threw up.
Because of several experiences like this, my father does not like being lost,
but his rstborn 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 eld research. It, along with mistakes, scientism, the par-
ticipants, stuckness, and discovery are what I refer to as the joys of eld
research.
Lost
When I refer to being lost in eld research, I mean physically lost as well as
conceptually lost. My rst recognition that I liked lostness came on a blaz-
ing August afternoon in Nanjing during the mid-1980s. My translator and
I were pedaling up and down countless hills through a laoconian 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

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 dri-
ven across the U.S.-Canadian border on an expressway, traveling to Kingston,
to nd some Canadian peacekeepers. On the return trip, I could not nd my
way back into the U.S. This is stupid, I thought. If I came across the bor-
der on a four-lane highway, surely I can nd 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 ne physical representation a metaphor
for being lost in eld research.
Usually, a eld researcher is to some extent lost. Youre not always certain
what the process is that youre studying. An independent variable seems to be
missing. There are a confusing number of dependent variables. Cause and
effect are difcult to tease out. Results and conclusions are frequently con-
tradictory. 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 conict
in business-to-business e-commerce. But I am having fun.
Mistakes
Mistakes are somewhat akin to being lost. To me, conducting eld research
can be likened to rowing a single, a skiff. You are rowing backwards; there-
fore, you cannot see where you are going. If you attempt to overcome this
inconvenience by looking over your shoulder, you will ip, 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 nd yourself in the cool water. That is, you make mistakes. It
is a lot of fun.
In eld research, mistakes are not only fun; they also are instructive. A
heart transplant surgeon emphasized this to me when I was interviewing lead-
ers for my book Bosses (Wall 1986). In my eld, 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 specic mistake teaches an individual lesson; moreover, mis-
takes in general are quite helpful because they eradicate or at least lower
researcher hubris. In eld research, arrogance is the researchers worst enemy.
If you are overly condent, 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 nd yourself with a
years 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 eld 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 rie 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 red, it sounds like a child rumbling through the kitchen pots and
pans. But it always res, 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 t 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 re 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 re when they became dirty or rusty or frigid weapons that may have
been too complicated to quickly disassemble and reassemble.
For eld 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 rst 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 rst 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 win-
dows and sitting on the oor; we were up to 25 spectators. Rising to the
occasion, the mediator began mediating with gusto. As the oor 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
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 eld 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 nd it quite useful to focus, rst, 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 com-
munity mediation in India, I was examining the mediation techniques used by
a formal mediation group, the panchayat (group of ve). One of the tech-
niques 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
gure 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 rul-
ing that covered both parties (e.g., farmer A is to quit throwing dead rats into
farmer Bs irrigation ditch), but still the 1.3 seemed too low.
As I thought about this nding, it occurred to me that ideas about the set-
tlement might be coming from the disputants as well as from the mediator
that self-direction might be supplementing mediator direction. So I examined
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 pro-
ceedings, seemed to fall into two distinct categories. In four of these, the con-
cessions were totally self-imposed:
A father paid a self-imposed ne of 2100 rupees to the temple because of
his sons 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, rein-
stalled 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 eld 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
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 defendants attor-
ney prepare the case for trial. Here points of agreement are noted, disagree-
ments 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 deter-
mine 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 ghting 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 inter-
vention.) 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 loom-
ing darkness had any effect on their mediations. Some replied afrmatively,
which was consistent with the theory and our previous eld observations.
Others said there was no effect because they and their subordinates were
trained to operate 24/7 and were outtted with night-vision equipment to do
so. Most informative was the third group, who held that darkness would elim-
inate the need to mediate. One peacekeeper from this group noted that dark-
ness, rain, cold, wind and snow reduce the pressure. When these come, dis-
putants go home and there is no conict 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. Stuck-
ness, 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 Pirsigs logic, I have attempted to develop a taste for stuckness.
While it has improved my eld 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 ow. 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 mun-
dane comparison and then one from eld research. First, the mundane. A few
weeks ago my wife and I locked ourselves out of our house (the responsibil-
ity 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 nd 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
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 Pirsigs 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 wifes mentor.)
Now to an analogous comparison in eld research. For me, the sensations
of being lost and stuck both center upon the same set of variables: the tech-
niques 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 indepen-
dent 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 ve years, I have been lost but it has been great
fun and currently I have some tentative answers. The rst 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 ve) 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 supersti-
tion. Because about 80 percent of the mediations in the U.S. result in agree-
ment, mediators here are reinforced for whatever approach they take. There-
fore, they continue to employ the same techniques from mediation to media-
tion (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 nd myself in the latter state when I treat
the mediators techniques as the independent variable and attempt to investi-
gate its effects on other variables (e.g., on rate of agreement). Specically, I
am most stuck and irritated that I have been unable to relate any single tech-
nique 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 benets in eld research. It reduces ones 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 eld
research.
Discovery
Field research furnishes insights, perhaps because it allows me to observe

people, places, events, and solid objects in ve 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 bot-
tle, 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 writ-
ten form, very few can solve it quickly. But when I haul in a lot of wine bot-
tles 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 eld. 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 base-
ment ofce, or conduct laboratory studies about it (e.g., Arunachalem, Wall &
Lytle 2000). Yet, I feel my best insights come when I am in the eld. There
I glean insights as I discuss their mediations with the mediators themselves,
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 peacekeepers M-16, ddle 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 wifes 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 sulfate-
reducing 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 cool-
ing towers, and for subterranean water and oil pipes.
While trying to nd a way to eradicate this irritating characteristic from the
sulfate reducer, my wife considered removing the gene responsible for the
behavior from the microbes DNA. Good idea, but not the most impressive
insight.
Here is the insight: while laboring at this task, she reasoned, instead of try-
ing to eliminate this corrosion afnity, why not transfer it to a bacterium that
attaches itself to a metal we want to corrode, such as uranium. (When ura-
nium 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 com-
parison, 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. peace-
keepers 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 exibility 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 pat-
tern nally comes into focus. Its somewhat like watching the crystallization
of a supersaturated solution or watching ice forming a glaze on your wind-
shield 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
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 con-
fused. 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 benecial 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. Specically, 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 judges actions, the clients and attor-
neys paid substantial deference to him, in that they addressed him as judge
(not by his rst 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
judges behavior, rather than vice versa. The specic 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 drivers 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
best is neutral, but who teams up with the opposing lawyers against him or
her. Therefore, the client does not want the mediators advice and their attor-
neys must proffer the offers and concessions. Noting this, the mediator con-
cludes the attorney is in the drivers seat and addresses him or her.
Conclusion
A balanced article would follow the discussion of the joys of eld research
with a delineation of the costs, subsequently compare the two, and announce
the net benets or costs. As you might expect, this is not to be a balanced
article. I do not make cost-benet analyses when considering eld research,
and I am in good company: Michelangelo didnt consider the costs and
benets 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 Guns for
Hire. Al Pachino probably did not conduct a cost-benet 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 benets.
And John LeCarre, just as Stephen King, writes because of the passion for
the art, not because the benets outweigh the costs.
Likewise, I have a passion for eld research. When I rst 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 ddle with a real-life phe-
nomenon that I can see, hear, smell, and feel, interacting with participants
who are similarly excited that I am studying them. They want a better under-
standing 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 ow toward a passion as my waywardness, mistakes,
and stuckness give way to discovery and progress.
Yet, experience has shown me that this zeal for eld 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 veried
my conclusion. Im 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 ne, because
there is a possibility that the unimportant will become important. Are we not
glad, for example, that Flemmings 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 xed sched-
ule in the ofce is a major luxury of our trade. Personally, I enjoy study-
ing 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 phenome-
non is important simply because you are excited about it. To engage in such
delusion sacrices ones 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 eld 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
complex or so simple that its 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
eld 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 eld project.
4) As a nal 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 con-
vinced. Their conversion requires that you get their attention, explain why the
phenomena under study is important, present lucid details on what you did nd
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 Psycho-
logy, 31: 951980.
Pirsig, Robert (1974). Zen and the Art of Motorcycle Maintenance. London: Head.
Sherman, Michael (2002). The shamans of scientism. Scientic American, June, p. 35.
Wall, James (1981). Mediation: An analysis, review, and proposed research. Journal of
Conict Resolution, 25: 157180.
Wall, James (1986). Bosses. Lexington, Mass.; D.C. Heath.
Wall, James and Adams, Stacy (1974). Some variables affecting a constituents evaluations of
and behavior toward a boundary role occupant. Organizational Behavior and Human
Performance, 11: 390408.
Wall, James and Callister, Ronda (1999). Malaysian community mediation. Journal of
Wall, James and Lynn, Ann (1993). Mediation: A current review. Journal of Conict
Resolution, 36: 160194.
Wall, James and Druckman, Daniel (2003). Mediation in peacekeeping missions. Journal of
Wall, James, Stark, John and Standifer, Rhetta (2001). Mediation: A current review and the-
ory development. Journal of Conict Resolution, 45: 370391.
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 labora-

tory-generated data about negotiations are at some remove from what occurs
in the eld (Pruitt 1981). But do we know enough about behavior in the eld
to understand the size of that remove? It is common understanding that our
methods of gathering eld data on any topic do not promise complete accu-
racy (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 negoti-
ation presents problems even beyond the usual, and that as a result our teach-
ing 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 Baraks 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 negoti-
ation. 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

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 learn-
ing 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 difcult. (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 cross-
checking 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 negotiators constituencies or audiences, groups that may hold a
range of views about the negotiation, from full support to full hostility. Pri-
vacy blunts the capacity of these constituents to shoot down a component of
an agreement before it can be presented in the attering light of a full agree-
ment. 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 commu-
nication 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 specic 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, rec-
ollections 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 fur-
ther.) Whatever the cause, a perusal of negotiator memoirs makes clear that
competition does not end with the negotiation (see Kissinger 1982). Negotia-
tors memoirs often exhibit an apparent need to show superiority to the other
side and frequently to ones 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 difcult for authors to recapture after the event. The bold judg-
ment (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 signicant ow of communication at the table.
On some occasions the competition between negotiating teams is subordi-
nated 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 falsication 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 inuenced 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 bound-
aries 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 mat-
ter 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 ques-
tions, 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 sug-
gestions 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 inconsisten-
cies in the sources are frequent, and thus unsurprisingly one is rarely free of
the need to use a standard of plausibility as well as ones 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.
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 vis-
ited 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 (Baraks chief of staff, and one of the Israeli heads of the nego-
tiation delegation) describes a dinner on Wednesday evening at which he, two
other members of the Israeli delegation (both former members of Israeli secu-
rity 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 sh 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 y 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 rst proposed, as a somewhat empty negoti-
ating effort (too little too late), at best an opportunity to freeze dry the posi-
tions 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 ofcials present at the dinner. Even before my rst question, Hasson
volunteered to tell me of the dinner, but never mentioned Shers 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
conict 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 justies some scrutiny.
There are, I think, three possible interpretations of these accounts. One is,
that the Israelis invented Abu Alas admission altogether in order to blame
Arafat should the negotiation fail. A second is that, the Israelis may have
asked a question about Arafats 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 pre-
sents a blas account of his own expectation of Taba and of his own perfor-
mance there, on Monday evening Sher saw that some agreements at Taba
were possible. He called Barak and told him so. (On Tuesday some Palestin-
ians were also seeing agreement possibilities and called their Ramallah head-
quarters 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 inten-
sity of the Palestinian negotiators at Taba (commented on by several Israeli
negotiators at the time) reected 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
Arafats 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 play-
ers 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 pro-
fessional, would do so. Also, why would Sher and Hasson keep such a strik-
ing admission secret from the Israeli negotiating team for the rest of the
week, and from the press even after Taba was over? Considering the impor-
tance of blame in the negotiations, this silence is striking. The story showed
up in public only in Shers memoirs published 8 or 9 months after Taba.
My view is that at the dinner in the desert the Israelis opened the topic of
Arafats 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 Shers 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 Baraks failure to men-
tion 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 nego-
tiate, could you have reached a framework agreement on your topic? On all
issues, except two very important ones, the answer was afrmative. (The two
exceptions were control of the holy places in Jerusalem, and the process for
ending the conict.) This assessment was created by matching each sides
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 ac-
knowledge several qualications. A negotiators inference from the behavior
of his opposite number that an agreement was possible leaves open the pos-
sibility 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 cor-
rectly, 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 signicant 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 nego-
tiating. Since the record is consistent that the negotiation ended on Saturday
at the initiative of the Israelis with the uncomplaining concurrence of the Palestin-
ians, 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 actors intention. It is a base for measuring a
negotiators success. More generally, it is a major source from which we cre-
ate meaning and the foundation for understanding what might otherwise seem
an arbitrary set of negotiation moves. Knowing a negotiators 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 ques-
tion indeed is an invitation into quicksand. Intentions are always many, multi-
leveled, morphing, oating, dissolving. A serious reection 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 signicant 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 opponents inten-
tion 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 ones opposite num-
ber would make the decoding of intent more plausible. But when Yossi Beilin,
among the most pro-negotiation of the Israeli leadership, entered the rst ple-
nary 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
the emphasis on individual Palestinian applications for compensation wasnt

really a way to keep the refugee question open forever.
Thus there is a deep conict 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, satis-
fying the readers 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 sum-
mer 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 pre-
sent 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 ve days before
they had to. I began by interviewing Barak. I explained my purpose and nally
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 doesnt
make any difference why I ended it. It had to end because it wasnt going
anywhere. When I told him that my piecing together of interviews and mem-
oirs suggested that a framework agreement was at least possible with a few
more days of negotiating, Barak simply denied it. This of course may repre-
sent 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 nego-
tiators 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 signicant
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 Shers memoir, he was not aware
of it at all.
Believing that there was more to learn about Baraks 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 nally 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 break-
through was possible. He instead focused on negotiation only as an elec-
toral ploy.
He had a history of getting cold feet when an agreement started to look
possible. Thats 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 opportu-
nity. (This item has many examples and subcategories.)
He had wanted an agreement at Taba, but having learned (on Wednesday
night) that Arafat didnt want a deal, he changed his focus to one of mak-
ing the world understand Arafats 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 con-
vinced 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
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 cre-
ate 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 rst place. He seems never to have given his negotiating team
one clear mandate, and his various utterances were thus interpreted by differ-
ent 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 negotia-
tors 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 Baraks 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 skepti-
cism 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 extraordi-
nary strength, willing to take larger risks, with something larger than Baraks
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 pru-
dent to adopt the conventional campaign stance: assert that he had done
everything possible for peace.
About Arafats 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 through-
out 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 nego-
tiation. On the other hand, we have the puzzling account of the Wednesday
night dinner that might suggest Arafats lack of commitment. And we have,
nally, the ready acquiescence by the Palestinian leadership on Saturday to
the Israeli decision to end the negotiation, suggesting that the Palestinians
too whether reecting Arafat or not did not choose to negotiate further.
Arafats vicious speech in Switzerland the day after Taba ended attacked
Barak and the Israelis, and made no mention of Taba. When I asked Pales-
tinians 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 inten-
tions. 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 understand-
ing 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 Conict (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, Kissingers
careful analysis came after he was told by an Israeli aid that the solution
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 nego-
tiation was, as described by Golan and The Times, a four way power ght.
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 rst
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 st ghts 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 humanitar-
ian 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 g leaf as the barely visible appearance of a g leaf,
and can, with no undue cynicism, be seen as a conspiracy among the Egypt-
ian, Israeli, and US governments to fool the Egyptian people. The term
bridging solution suggests that each side used problem solving technique to
satisfy its interests. I think it accurate to say that this bridging solution
was a thin cover that reected 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), con-
trast 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, Israels being security, Egypts being sovereignty. The difculty
here is that the basic principles that satised 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. Signicant modications 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 Carters mediating team before the summit began, and it is difcult
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 prob-
lem solving and an exploration of interests. It was taking an obvious solution
staring the parties in the face and modifying it to t 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 trafc 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. Unofcially 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 ofcial Palestinian position
refused Israeli annexation of Maale Adumim, but unofcially 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 conrms Kissingers 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 Kissingers 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 Amis 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.
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 19942001. (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 Conict: Escalation, Stalemate,
and Settlement, second edition. New York. McGraw-Hill.
Schiff, Zeev (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 ethno-
graphic methods that Walton and McKersie (1965) developed their insight that
there are four dimensions of labor-management negotiations: integrative bar-
gaining, distributive bargaining, attitudinal restructuring, and intra-organizational
bargaining. In my work, ethnographic observations led me to see the impor-
tance 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 en-
abled 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. Merrys (1989) use of anthro-
pological 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 experi-
ments are designed to isolate specic phenomenon of interest and con-
trol all other factors, ethnographic research looks at behavior in the total
negotiation context. While experiments can be done with nave 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 inuence 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, com-
plex situations. Nor do experiments, as Hopmann (2002) points out, capture

40 RAY FRIEDMAN
some of the most important, albeit subtle aspects of negotiation [such as]. . . . iden-
tify formulation (p. 73). For ve years, I studied labor negotiations by sit-
ting in negotiation sessions, attending caucuses on both sides, debrieng lead
bargainers after negotiation sessions, interviewing participants, and follow-
ing the negotiations over weeks and months (Friedman 1994). This pro-
cess has several signicant 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 difcult con-
stituents (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 test-
ing 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 mem-
bership support there is for those issues. The other reason is to remind the
employer that this is not my agenda, it is our members 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 nd that most rank and le people dont 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 wont. 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 tac-
tics 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 Ill 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 com-
pany he puts on the table as a nal proposal. This is a guy I would trust implic-
itly. We will not bull shit each other.
Friedrich and Lamberts CEO is Tom OBrian. If he says this is what we agreed
on, Id sign it tomorrow. If I have a question about a clause I can call him and
hell x it. He has integrity.
Relationships can be a source of leverage, and provide alternative avenues for
communication. One union lawyer reported, discussing a particular manage-
ment 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
dont 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 rela-
tionships, and how negotiators depend on and use good relationships in nego-
tiations. 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 ethnog-
raphy. Ethnography has four features (Friedman & McDaniel 1998). First, ethnog-
raphy requires direct observation of the activity being studied in the case
of negotiations, sitting in on some or all of the preparation meetings, negoti-
ations, 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 nego-
tiations 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 condentiality in
negotiations). If you were to reveal to the other side what you hear in cau-
cuses, you could cause enormous damage to the negotiations. While reveal-
ing information will certainly not be a researchers intention, there are usually
social opportunities (such as when everyone gathers to get coffee before nego-
tiations begin). On such occasions, an inadvertent comment could reveal
something, or ones 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 condentiality and
ones 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 rst met
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 companys lawyers did not agree and insisted that I not be
allowed to return. In another case, the companys CEO approved the project
but would not force any individual to participate. In that case, one plant man-
ager refused to let me enter, fearing that my asking questions about a previ-
ous bitter negotiation would increase anger among employees. Entry into
negotiations is never really settled it requires constant renegotiation.
Given natural sensitivities to the ow of information, access will usually be
on one side or the other (I was able to access both sides 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-tak-
ing. 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 benets, 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 record-
ing 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
difcult issue for the moment. He is suddenly taking a new approach that is
different than his plan expressed in caucus maybe the unions prior com-
ment 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 ll out your notes right away. As you take notes, you
will inevitably have sentences that are not complete, or ideas that are scrib-
bled 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 ll 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 cre-
ated new ones. Spend some time each day actually writing out your evolving con-
ceptual 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 con-
cepts. For example, for my book I had one le lled 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 inter-
view it came from, and what page in my notes it came from. Writing was
often based around these conceptually-organized les (most computer pro-
grams designed for qualitative analysis, such as Aquad, have functional equiv-
alents of these les).
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 bar-
gaining 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 bar-
gainers. 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 conict between key
people in management that was affecting who in the bargaining team had
more power and inuence. 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 sides 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 sup-
port for key issues. These personal interviews were invaluable as a comple-
ment to actual eld observations.
While I tried to get greater depth of understanding from additional meet-
ings 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 bar-
gaining 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 across-
the-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 connes of university teaching schedules or summers. Also, since access
can be problematic, you may only be able to access a negotiation in a differ-
ent city, which makes the logistics even harder (and the costs even greater).
Reporting Results
Unlike more traditional research methods, there are few established conven-
tions 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 ndings 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 nding 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 eld 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.
Kolbs (1983) study builds towards a model explaining factors that inuence
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 ethno-
graphic data than survey data. There has been some publishing of ethno-
graphic 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 pow-
erful, but that research has been criticized for decontextualizing conict
(Barley 1991). Ethnographic research offers an alternative approach which can
provide insights into the types of complex situations that negotiators really
face. This approach is not easy it can be more time consuming, costly, and
difcult 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): 345359.
Barley, Stephen R. (1991) Contextualizing conict: 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, 165199.
Barley, Stephen R. (1996) Technicians in the workplace: Ethnographic evidence for bringing
work into organization studies. Administrative Science Quarterly, 41(3): 404441.
Carnevale, Peter J. (1985) Accountability of group representatives and intergroup relations.
In E.J. Lawler (Ed.), Advances in Group Processes, 227248, Greenwich, CN: JAI Press.
DeDreu, Carsten, Laurie Weingart, and Seungwoo Kwon. (2000) Inuence of social motives
on integrative negotiation: A meta-analytic review and test of two theories. Journal of Personality
and Social Psychology. 75(5): 889905.
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 negotia-
tions, Journal of Applied Behavioral Science, 27(3): 356378.
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. Whiteld (Eds.) Researching the World of Work:
Strategies, Methods, and Critical Views. Ithaca: ILR Press, 113126.
* Garnkel, 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), 721736.
* 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 ethno-
graphic texts convince. Organization Science. 4(4): 595616.
* Hammersley, M. and Atkinson, P. (1995) Ethnography: Principles in Practice. London: Routledge.
Hopmann, P.T. (2002) Negotiating data: Reections on the qualitative and quantitative analy-
sis of negotiation processes. International Negotiation, 7:6785.
Kolb, D. (1983) The Mediators. Cambridge, MA: The MIT Press.
Merry, S.E. (1989) Mediation in nonindustrial societies. Pages 6890 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 overcondence
on bargaining behaviors and outcomes. Academy of Management Journal, 28(1): 3449.
* 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: 732.
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): 489510.
* 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. 3756 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 eld of conict resolution. Indeed, the
problem-solving workshop was one of the rst innovations that helped to
dene and operationalize the eld 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 denition 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 unofcial yet inuential representatives of parties (states or
groups) involved in destructive conict. The parties are brought into intense,
face-to-face discussions, facilitated by a third party team, that focus on an
analysis of the conict 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 conict 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 eld identied its potential role in learning about the
phenomenon of conict and in testing concepts and models of conict causa-
tion, escalation, and resolution. Subsequently, the problem-solving workshop
has also been identied as an action research intervention, and as a com-
ponent of a wider program of action research on violent and protracted con-
ict. Unfortunately, the use of the problem-solving workshop as a method of
Authors Note: The author wishes to thank Herbert Kelman and an anonymous reviewer for
helpful comments on an earlier draft of this article.

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) identied the method as con-

trolled communication. This was in order to emphasize the necessity of the
third party inuencing the interaction of the participants toward an open, ana-
lytical discussion of their conict, leading to increased mutual understanding,
followed by the joint creation of ideas for resolution, which would be injected
into ofcial 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 consis-
tently by Kelman, who served as a third party panel member in one of
Burtons rst workshops, and went on to develop his own approach of inter-
active problem solving (Kelman 1972, 1986). My own contribution was to
dene this method as a form of third party consultation in order to empha-
size 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 ve) to engage in low risk, noncom-
mittal, off-the-record discussions over a period of three to ve days in a neu-
tral and secluded setting conducive to a relaxed atmosphere and devoid of
intrusions. The participants are typically inuential individuals in their com-
munities who are not in ofcial policy-making roles, but have access to the
political leadership. Some variations involve ofcials, but in a private, un-
ofcial 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 conict. The objective is to create an infor-
mal atmosphere in which participants can freely express their views while respect-
ing those of the other side. The atmosphere also encourages them to move
from adversarial debate to a joint analysis of the conict 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 analy-
sis 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 partici-
pants 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 conict ana-
lyst 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 investiga-
tive potential for validating and creating theoretical notions about protracted
and violent conicts, 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
conict 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 conict 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 eld testing of the theoretical notions applied, allowing the third party to
judge their validity and scope of applicability. Thus, for example, in Burtons
early Cyprus workshop, the third party injected the conict spiral model of
escalation to help illuminate the development of the conict. However, one
participant questioned its validity in explaining the conict, because the par-
ties 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 conict organized by the author (Fisher
1997), the concept of the security dilemma was applied to the conict 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 identied by Kelman and Cohen
(1976) is the provision of theoretical inputs to enrich the analysis, a function
identied as diagnosing the conict in my model of third party consultation
(Fisher 1972). Moreover, one of the objectives in this initial model was the
study of conict, by which the members of the third party team could learn
about the nature of escalated and destructive conict, 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 conict. 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 conict 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 identica-
tion. 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 conict occurred because participants in workshops repeatedly
revealed such needs when asked to engage in conict analysis. In other
words, when asked to go beneath the surface positions and interests in the
conict, participants would typically move to a discussion of needs for secu-
rity, 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, conicts offered
by theorists such as Burton (1990) and Azar (1990). Human needs theory is
now regarded as an essential element of conict theory and a common tool
for conict analysis in problem-solving workshops.
Problem-solving workshops have been seen from the beginning as a useful
complement to negotiations between conicting parties, assuming that trans-
fer effects occur between the unofcial and ofcial tracks. Kelman and Cohen
(1976) initially outlined the ways in which workshops could make contribu-
tions at all stages of negotiation from encouraging parties to enter negotia-
tions, to working on issues that are hampering negotiations, to discussing long
range issues that need to be addressed once a settlement is reached. Within

this context, workshops have been regarded primarily as a prenegotiation sup-
plement 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 expe-
rience of the past forty years is summarized by Kelman (2002):
Binding agreements can be achieved only through ofcial negotiations. The very
binding character of ofcial negotiations, however, makes it very difcult for cer-
tain other things to happen in that context such as the exploration and dis-
covery 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 conict
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 cog-
nitive and emotional barriers and resistances that hamper parties from enter-
ing into negotiations or negotiating effectively when they do so. Black and
white mirror images, selective and distorted perceptions, self-serving attribu-
tions, blame, hostility, and mistrust lie embedded in the parties thinking and
their relationship. Surfacing and working through these residues of destructive
conict 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 com-
parison to unilateral attempts to prevail and punish. Therefore, the problem-
solving 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 Israeli-
Palestinian conict, which has documented the resistances to negotiation in
that conict 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 oth-
ers, 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 interven-
tion is planned and carried out to address the problem, and further informa-
tion is then collected to evaluate the intervention and plan future ones (Fisher
1982). This form of research is carried out in eld settings and is directed
toward social change, which the researcher and his/her collaborators deem is
desirable in order to address the identied problem.
To illustrate how the problem-solving workshop serves as a form of action
research, I will rst draw on my experience in organizing a series of work-
shops on the Cyprus conict in the early 1990s through the now defunct
Canadian Institute for International Peace and Security (Fisher 1997). CIIPS
became interested in the conict through Canadas 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 move-
ment toward a settlement difcult. At the last seminar, interest was shown by
inuential Cypriots in ofcial roles for a follow-up project that would take a
problem-solving approach to the conict. To assess interest in possible work-
shops 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 inuential Cypriots
on the two sides of the Green Line. There was generally wide support for
such an unofcial, low-risk, exploratory venture, and tacit approval from the
two leaderships was forthcoming. Further discussions worked to clarify expec-
tations 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
conict. 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 inuential but unofcial participants
from each side working with a four-member third party team. Given the intractable
nature of the dispute, the focus was on the issues that seemed to be sus-
taining the conict, both in terms of making it very difcult 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, acknowl-
edgements 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 conict 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 indi-
cated that they had increased their understanding of the perspectives, ratio-
nales, 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 difcult issues in the conict. Pre and post-workshop ques-
tionnaires 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 continu-
ing 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 con-
text of the conict. It also assisted in supporting the importance of evalua-
tion, 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, dis-
cussions 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 conict 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 conict (Kelman 1979). In this context, the
goal of the workshops is to understand and overcome the psychological bar-
riers 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 conict resolution in the Middle East with unique oppor-
tunities to observe and learn about the dynamics of the Middle East conict
and of international conict in general. Our action is designed to contribute to
conict resolution by creating opportunities for communication between inuen-
tial 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 attend-
ing conferences, and other activities to learn about the conict and to estab-
lish 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 orga-
nizing, implementing, and evaluating workshops are given priority, because
these ensure the viability of workshops and their contribution to conict res-
olution. 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 conicting parties to inter-
act with each other under auspices in ways that deviate from the norms gener-
ally 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 oppor-
tunity to make rich, detailed observations of ongoing processes of conict and
conict 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 exibility, 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
own actions. These kinds of lessons are clear contributors to the prenegotia-
tion 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 system-

atically as a method of research to study the complex phenomenon of social
conict. As the eld has developed, the predominant institutional base of
conict resolution work has shifted from academia to non-governmental orga-
nizations (Fisher, in press). While this shift entails advantages in terms of
greater exibility and variety of funding sources, it does tend to support prac-
titioners who have less interest and expertise in the application and induction
of theoretical ideas about social conict. Thus, the use of the method to study
conict is compromised, although some practitioners demonstrate commend-
able 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 Con-
ference task force on regional conicts (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 conict 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 tech-
niques 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 conict, but the application of
more objective and comprehensive measures is woefully absent. A fuller
application of the action research model would encourage third party inter-
venors 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 cre-
ativity and investment, the action research model could be better imple-
mented. 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 Conict. Hampshire, UK: Dartmouth
Publishing.
Burton, John W. (1969). Conict and Communication: The Use of Controlled Communication
in International Relations. London: MacMillan.
Burton, John W. (1987). Resolving Deep-Rooted Conict: A Handbook. Lanham, MD: Univer-
sity Press of America.
Burton, John W. (ed.) (1990). Conict: Human Needs Theory. New York: St. Martins Press.
Chufrin, Gennady I. and Saunders, Harold H. (1993). A Public Peace Process. Negotiation
Journal 9, 2:155177.
de Reuck, Anthony V.S. (1974). Controlled Communication: Rationale and Dynamics. The
Human Context 6, 1:6480.
Diamond, Louise and Fisher, Ronald J. (1995). Integrating Conict Resolution Training and
Consultation: A Cyprus Example. Negotiation Journal 11, 3: 287301.
Doob, Leonard W. (ed.) (1970). Resolving Conict 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
Conict. Journal of Conict Resolution 16, 1:6794.
Fisher, Ronald J. (1982). Social Psychology: An Applied Approach. New York: St. Martins
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 Conict Resolution. Syracuse, NY: Syracuse University
Press.
Fisher, Ronald J. (In Press). Interactive Conict Resolution. In I. William Zartman, editor,
Peacemaking in International Conict (rev. ed.). Washington, DC: United States Institute of
Peace.
Kelman, Herbert C. (1972). The Problem-Solving Workshop in Conict 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, 1:162186.
Kelman, Herbert C. (1979). An Interactional Approach to Conict Resolution and its
Application to Israeli-Palestinian Relations. International Interactions 6, 2:99122.
Kelman, Herbert C. (1982). Creating the Conditions for Israeli-Palestinian Negotiations.
Journal of Conict Resolution 26, 1:3975.
Kelman, Herbert C. (1986). Interactive Problem Solving: A Social-Psychological Approach to
Conict Resolution. In William Klassen, editor, Dialogue Toward Inter-Faith Understand-
ing. Jerusalem: Ecu-menical Institute for Theological Research.
Kelman, Herbert C. (1987). The Political Psychology of the Israeli-Palestinian Conict: How
Can We Overcome the Barriers to a Negotiated Solution? Political Psychology 8, 3:347
363.
Kelman, Herbert C. (1995). Contributions of an Unofcial Conict Resolution Effort to the
Israeli-Palestinian Breakthrough. Negotiation Journal 11, 1:1927.
Kelman, Herbert C. (2000). The Role of the Scholar-Practitioner in International Conict
Resolution. International Studies Perspectives 1, 3: 273288.
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 Conict Transformation. Lanham, MD: Rowman & Littleeld.
Kelman, Herbert C. and Cohen, Stephen P. (1976). The Problem-Solving Workshop: A Social-
Psychological Contribution to the Resolution of International Conict. Journal of Peace
Research 13, 2:7990.
Kelman, Herbert C. and Cohen, Stephen P. (1986). Resolution of International Conict: 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 Conicts. New York: Harper.
Mitchell, Christopher (2001). From Controlled Communication to Problem Solving: The
Origins of Facilitated Conict Resolution. International Journal of Peace Studies 6,
1:5967.
Mitchell, Christopher and Banks, Michael (1996). Handbook of Conict Resolution: The
Analytical Problem-Solving Approach. New York/London: Pinter.
Ross, Lee and Stillinger, Constance (1991). Barriers to Conict Resolution. Negotiation
Journal 7, 4:389404.
Rouhana, Nadim N. (2000). Interactive Conict Resolution: Issues in Theory, Methodology,
and Evaluation. In Paul C. Stern and Daniel Druckman, editors, International Conict 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 Conicts. New York: St. Martins 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 Conicts in Eastern
Africa. Journal of Applied Behavioral Science 6, 5:453489.
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 dened 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 others 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 con-
cessions or decisions, evaluating impacts of mediation and other conict-
settlement 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 ini-
tial 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 conict
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 fea-
tures 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 mea-
suring subjects responses over time in experiments. A difference, however, is
that within-subject responses (or repeated measures) are part of an experi-
mental 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 pri-
marily on the way the analysis was done in the context of the overall study

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 rst section shows
how time-series data can be used for post-diction with correlation and regres-
sion 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 nal section provides
examples of several qualitative time-series analyses of conict-resolving pro-
cesses. 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 rea-
sons. 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 re-
create 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 con-
trol 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 ve-year period. These talks were regarded by U.S. policy-
makers to be routine. Previous agreements provided the framework for the
terms of a new agreement. However, the negotiation process deed these expecta-
tions. It was not at all smooth. Spains demands were extraordinary includ-
ing 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
rst and third questions.
Statements made by both delegations during the sessions were coded into
categories reecting 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 aggre-
gate 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 state-
ments, 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 (19691979) of talks between the
Soviet Union and the U.S. on strategic arms limitations (SALT). Performed
with a deductive approach, these investigators evaluated the t of three alter-
native 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 species that negotiators
will either match (a concession for a concession) or mismatch (a retraction in
response to a concession) the others 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 others 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 assump-
tions 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 1969May 1972), at Vladivostok (June 1972November 1974), in
SALT II (January 1977June 1979), and across the entire set of rounds from
1969 to 1979. The results showed that the comparative model provided the best
t 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 del-
egation). 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 hypo-
theses (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 mod-
els of reciprocity were evaluated with time series of round-by-round moves
in seven negotiations. The directional model was compared to ve versions of
the trend and four versions of the comparative model: The ve 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 be-
tween 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 specied 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 others concession
from t 1 to t for the trend model.
Goodness-of-t 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 signicant). Thus, the comparative model produced
a better t to the data than the 9 alternative models, including the various
combinations of comparative and trend models. These time-series ndings

corroborate and attest to the generality of the results obtained in the two 1986
studies described above. They tell us that professional negotiators are atten-
tive 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 corre-
lations 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 pre-
cisely the inuence between any pair or trio of nations controlling for the
effects of any other national delegation or delegations in the talks. We dis-
covered, for example, that the effect of moves made by the leader of one alli-
ance had a stronger impact on the moves made by the leader of the other
alliance in the next round when the inuence 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 allys inuence was controlled through the use of a multiple-partial cor-
relation. The importance of these sorts of analyses is that they parse out inu-
ences 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 relation-
ships 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 negotia-
tion 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 signicance tests rather than with correlational techniques. A data set of
events collected during a six-year period of the conict between Armenia and
Azerbaijan over the territory known as Nagorno-Karabakh provides an exam-
ple of the way an interrupted time-series design is implemented.
The case provides an opportunity to examine whether various attempts to
mediate a serious conict were effective. Hypotheses about timing and readi-
ness 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 conict. In addi-
tion, 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 inde-
pendent variables, the conicting 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 before-
and-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. Signicance
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 ana-
lyzed from a laboratory experiment. A difference however is that it is possi-
ble 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.
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 ceasere. On the other hand,
the combat did provide the condition for reduced violence and a ceasere
agreement. Escalation proceeded to the large-scale use of military force, re-
sulting 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 theo-
retical and practical implications, it is a weak experimental design. The dif-
ferences 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 ndings would be more plausible if a com-
parable case without the interruptions (mediations, combat) was included in
the design; for example, de-escalation occurs only following combat in time-
series 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 col-
lapse, a natural disaster such as an earthquake, the insertion of a peacekeeping
force, a summit conference among national leaders, an inux of foreign eco-
nomic aid, or an entirely new insight into the source of a conict.
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 quasi-
experimental 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 esti-
mates of probabilities for future (unknown) events.
68 DANIEL DRUCKMAN
Bayesian Analysis
Another type of analysis that captures change over time, invented by Rev-
erend 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 Trans-
actions 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 rst in thinking on
the subject of it was, to nd out a method by which we might judge concern-
ing the probability that an event has to happen, in given circumstances, upon
supposition that we know nothing concerning it but that, under the same cir-
cumstances, 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 occur-
rence (Hi) or non-occurrence (Hj) of an event, for example, whether or not a
coup will occur within a dened period such as one year. The rst 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 reshufing (.4 that coup will occur, .3 that it will not occur if there
is reshufing), 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 proba-
bilities 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 reshufing
(.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 reshufing is taken into account (from .25 to .31).
Similar calculations can be made for each of the other symptoms, free elec-
tions 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 ve 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 ve factors that affect the longevity of agree-
ments. They assigned conditional probabilities to the factors as follows: fac-
tions 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 lon-
gevity, from .33 to .63. And, arms availability reduces the chances from .33
to .17. When the conditional probabilities of all ve 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 jeopar-
dizing 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 circum-
stances. It can also be understood as an alternative to regression for fore-
casting 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 lin-
ear extrapolation of past trends, considered as the initial probabilities or pri-
ors 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 updat-
ing can be continuous as a result of monitoring situations on a regular basis.
Expert judgments can also be rened 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, focus-
ing attention on changes which have implications for forecasts. Despite its appeal,
Bayesian techniques have rarely been used in addressing problems in negoti-
ation or conict 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 proba-
bilities, 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 time-
series analysis. We turn now to a variety of qualitative analyses performed to
depict changes more difcult to capture with numerical codes.
Qualitative Time-Series Analysis
Qualitative approaches can provide insight into patterns of interaction over

time that are difcult to measure with precision. An example is Lepgold and
Shambaughs (1998) analysis of Sino-American relations from 1969 to 1997.
These researchers attempted to explain the causes and consequences of reci-
procal exchanges between the two nations during this period. Their typology
of four distinct patterns of exchanges was used to depict changes in the bilat-
eral relationship. The types were dened in terms of expectations about time
horizons and the degree to which national decision-makers anticipate a reli-
able stream of benets 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 benets.
The results showed that when American and Chinese perceptions of their
time horizons and the likelihood of reliable streams of benets diverged, the
country that had a longer time horizon or perceived more benets from the
relationship was able to drag out negotiations until a preferable outcome had
been achieved. These variables are also shown to inuence bargaining strate-
gies, for example whether the parties engage in specic reciprocity (exchanges
are precisely matched in terms of equivalence or contingency), diffuse reci-
procity (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.
Lengs (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
each rivalry in terms of the dominant inuence 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,
194849 (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 lead-
ing 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
ndings 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 authors earlier ndings
based on a larger sample of crises occurring between 1816 and 1980 (Leng
1993). These are particularly interesting ndings 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
condence in ndings 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 Lengs analysis or from negotiation or mediation
interactions. The analysis is facilitated when the choice of events is guided by
a framework as in Druckmans (2001) study of turning points. In that study,
a path of events that occurred in each of 34 cases of negotiation was devel-
oped in terms of a three-part framework: precipitants, departures, and conse-
quences. 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 Unions nuclear arsenal approached parity with the United
States. Departures are changes that occur in the process; they may be rela-
tively 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 de-
escalatory 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 tak-
ing the form of European Community support for a U.S. plan. The conse-
quence 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 typ-
ical 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 de-
escalatory consequences at time t de-escalatory consequences at time t +1.
Another application of process tracing comes from my research on situational
levers of exibility. Using experimental techniques, an attempt was made to
identity the situational variables that had the strongest impact on exibility in
each of four stages of a simulated environmental conference on ozone deple-
tion. 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
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 (set-
ting-the-stage) limited media coverage (bargaining stage) limited media
coverage (endgame stage).
An example of a stalemate path is as follows:
Being the delegations 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 cover-
age. They appear in stages two, three, and four of both paths. The variables
identied in each stage inuence a process toward an outcome. Together,
these inuences 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 back-
ward tracing from a particular event or outcome through a series of earlier
decisions or events arranged chronologically. This can be done with docu-
mented cases by developing a chronology of events. Examples of chronolo-
gies 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. Carstarphens (2003) analysis of a dia-
logue about racial prejudice provides an illustration. The dialogue occurs
among a group of men from different backgrounds. It is presented in the well-
known documentary lm, 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 Wahs question to David (D): What keeps you from believing?
Ds lack of understanding produces an impasse in the discussion facil-
itator intervention D reects on his own experience and resistance to
change Ds reframing Ds new self awareness Ds acknowledge-
ment of his own prejudices and that racism exists other members changed
74 DANIEL DRUCKMAN
attitudes toward D African-American member acknowledges Ds acknowl-

edgement group shift in Ds relationship with the other members (focal
event).
This path illuminates the groups 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 conict-
resolution intervention, that lead to a particular outcome. It is a more complex,
process-oriented, time series of causation than many correlational analyses involv-
ing 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 Pruitts (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 ofcial and unofcial 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 rst ve 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 Professors Israeli diplomats
Israeli Foreign Minister
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 Israeli diplomats Israeli
Foreign Minister Rabin
Stage IV is a nal 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
unofcial (Track II) diplomacy involving the mediator and professors, was
replaced by ofcial (Track I) diplomacy, involving government ofcials on
both sides.
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 tech-
nique 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 conict 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 difculty 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
conict: 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 inter-
play among the actors, activities, and processes, including recursive relation-
ships 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 inuence
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 non-
linear activities intended to address ongoing conicts. 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 cross-
sectional (same point in time) comparisons and complement the quantitative
analyses discussed earlier. This concludes our treatment of the family of time-
series 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 ques-
tion of interest is whether the known outcome can be predicted from infor-
mation 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. Soph-
isticated 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 eld-) experimen-
tal 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 analysts attention on events that
can alter a process in important ways. The process-altering features of partic-
ular 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 analy-
ses of time series data are qualitative approaches. These approaches are par-
ticularly 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 inuence strategies, and the kinds of events and decisions that precede
turning points in negotiation or shifts in group processes as well as the con-
sequences 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 analy-
ses. It would seem that these advantages of qualitative time-series analysis
complement those of the quantitative techniques.
Further analytical specicity 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 pro-
posal 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
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 exam-
ple of a multi-method analysis of negotiation process-outcome relationships.)
References
Allen Nan, S. (1999). Complementarity and coordination of conict resolution efforts in the
conicts over Abkhasia, South Ossetia, and Transdniestria. Unpublished doctoral disserta-
tion, 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 conicts. 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 Mifin.
Druckman, D. (2001). Turning points in international negotiation: A comparative analysis,
Journal of Conict Resolution 45:519544.
Druckman, D. (1993). The situational levers of negotiating exibility, Journal of Conict
Resolution 37:236276.
Druckman, D. (1986). Stages, turning points, and crises: Negotiating military base rights,
Spain and the United States, Journal of Conict Resolution 30:327360.
Druckman, D., and Bonoma, T.V. (1976). Determinants of bargaining behavior in a bilateral
monopoly situation II: Opponents concession rate and similarity, Behavioral Science
21:252262.
Druckman D., and Druckman, J.N. (1996). Visibility and negotiating exibility, Journal of
Social Psychology 136:117120.
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 Conict Resolution 34:234251.
Faure, A.M. (1994). Some methodological problems in comparative politics, Journal of Theoretical
Politics 6:307322.
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 conict.
Unpublished doctoral dissertation, GeorgeMason University.
Leng, R.J. (1998). Reciprocity in recurring crises, International Negotiation 3:197226.
Leng, R.J. (1993). Interstate Crisis Behavior, 18161980: Realism versus Reciprocity. Cam-
bridge: Cambridge University Press.
Lepgold, J., and Shambaugh, G. (1998). Rethinking the notion of reciprocal exchange in inter-
national negotiation: Sino-American relations, 19691997, International Negotiation 3:227252.
78 DANIEL DRUCKMAN
Mooradian, M., and Druckman, D. (1999). Hurting stalemate or mediation?: The conict over
Nagorno-Karabakh, 199095. Journal of Peace Research 36:709727.
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 inter-
national conict resolution, International Studies Review 2:3363.
Stoll, R.J., and McAndrew, W. (1986). Negotiating strategic arms control, 19691979.
Journal of Conict Resolution 30:315326.
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 Conict 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 signicant 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 eld 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 conict resolution in general and mediation in
particular.
When it came to conict resolution there was a widespread feeling that
only by gaining an intimate and personal knowledge about a conict, can one
say anything meaningful about its management. Conict was seen as so aber-
rant 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 quasi-
realistic frames. Little wonder that the phenomenon remained so poorly under-
stood for so long. The difculties 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

80 JACOB BERCOVITCH
reports of real events, by real mediators and negotiators, in real conict situ-
ations, 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 elds, ranging from domestic disputes
through legal conicts and community disputes to international conicts. 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 nd
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 identiable 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 identied. A signicant part
of the literature dealing with mediation and purporting to answer these ques-
tions 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, difcult to draw general inferences
from these studies, no matter how carefully crafted they may have been. Des-
criptive approaches make good sense of idiosyncratic situations, they encour-
age 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 specic case and why mediation failed or succeeded. To answer the ques-
tions 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 vari-
able. 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 pheno-
menon can be studied scientically, 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 informa-
tion about the world. We usually know how the data is recorded and gener-
ated. 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 articiality that inevitably characterizes these studies. In other words,
the data is real, but the real world is so very different form the data. To con-
duct 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) valid-
ity is at best doubtful.
82 JACOB BERCOVITCH
Archival Data Analysis; Some Possible Benets
The reality of negotiation and mediation is made up of numerous particular

events; particular negotiators, particular individual mediators, particular con-
ict 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 bet-
ter understanding of some general acts pertaining to human behavior in
conict. 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 rst hand account by the parties directly involved. In compari-
son 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 ofcial 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 informa-
tion. 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 natural-
ness in the design of research.
Archival data enable a researcher to study contextual information and fea-
tures of larger social units (experiments of eld studies rarely study the
group as the unit of analysis).
Archival data is spontaneous. By this I mean it was produced without sub-
jects or participants being requested to do so.
Archival sources provide high quality information that can be easily sum-
marized 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 tra-
ditions. 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
conicts and numerous mediators. The larger the sample of cases, the
greater our condence in our ndings and interpretations.
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 exible and power-
ful 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 eld 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 the-

oretical framework and a proper research design. What I propose to do is to
examine all cases of international conict since 1945, and all the cases of
international mediation in these conicts. 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 conict 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, time-
dependent 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 inter-
action 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.
Dening the Relevant Population
The rst challenge in undertaking archival research is to identify the popula-

tion of cases to be studied. To start with, we examined existing sources of
84 JACOB BERCOVITCH
information on international conicts, (e.g., Small & Singer 1982: Butterworth

1976: Tillema 1991; Wallensteen 2002). International conicts are relatively
rare events, but most of these publicly available sources had different num-
bers of conicts since 1945. Where each project uses different dimensions to
dene conict, different thresholds (that is, number of fatalities incurred
before an event can be described as an international conict), and focuses on
different categories of conict (e.g., within or between states), neither com-
parability nor generalizations are possible. An approach to data collection on
international conict which does not dene conict primarily in terms of vio-
lence, and is more concerned with conict 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 conict even in the
absence of 100 or 1000 combat fatalities. We dene conict 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 denition in mid, we scanned numerous archival sources (e.g. existing
data, newspapers accounts, World Factbooks, etc.) and identied 309 conicts
from 1945 to 1995 (see Bercovitch & Jackson 1997). Each conict was stud-
ied for its history, duration, issues at stake, intensity, and manner of termina-
tion. Information about the parties in each conict (e.g., nature of political
system, power resources, etc.) was also collected.
Our interest of course has been with the occurrence, trends and character-
istics of mediation events in the real world. Very little systematic information
was ever collected on mediation events. Such information that does exist con-
tains 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 identied our population of conicts, we studied each and every one of
them in terms of the conict management methods used in these conicts and
the outcomes of each method. Each conict management method was studied
for such dimensions as timing, manner of performance, and impact.
Of the 309 conicts, 190 experienced some form of formal mediation.
Some conicts, such as the one in the Middle East or the conict in the for-
mer Yugoslavia (186 different mediation efforts from 19891995) experienced
many mediation efforts by different mediators. Other conicts had very few,
if any, mediation efforts. In these 190 conicts 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
efforts. These ranged from individuals to groups, to regional organization, ofcials

of a state or representatives of international organizations. Each mediation
effort was studied in terms of many independent variables to capture the fea-
tures of their identity, composition, resources, strategy etc. Table 1 below
gives some idea of the range and distribution of conict management events
in international conict.
Table 1. Conict management activities; 19451995
No conict management activity in 61 out of 309 international conicts (20%)
Conict Management Strategy Number of times used
Mediation 2194
(58.7%)
Negotiation 1249
(33.4%)
Arbitration 29
(0.8%)
Referral to an International or Regional Organization 166
(4.4%)
Multilateral Conference 38
(1.0%)
Total Number of Conict Management Strategies 3737
(100%)
Information on each conict management activity was collected from four

main archival sources dealing with events data; The New York Times Index,
The Times Index (London), Reuters Online News Service, and Keesings Archives
(renamed Keesings Record of World Events.) These news sources are widely
regarded as offering the best and most comprehensive information on politi-
cal 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 identied 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 conict 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
disputants territory);
strategies perceived to be in use by the mediator (e.g., communication-facil-
itation, procedural or directive strategies);
the outcome of each mediation attempt (e.g., agreements, cease-res, full
settlements and the durability of such outcomes); and,
numerous contextual observations (e.g., presence of ongoing hostilities, system
era, dispute issues and the disputants 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, conict 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 Characteristics of the Parties Conict Management

the Dispute Variables
Dispute Start/End Date Party Initiator in Previous Disputes Start/End Date

Duration Identity Party A & B 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 Strategies-Primary
budget etc.)
Hostility Level Parties Previous Relations Supplementary Strategies
Table 2 (cont.)
Characteristics of the Characteristics of the Conict Management

Dispute Parties Variables
Reciprocity History Disputes with Other Previous Relationship of

Parties Parties With Mediator
System Period Dispute Party Political System, Previous Attempts
Start Openness/Democracy, Med/Neg this Dispute
Closedness/Autocracy, Polity Type
and Change
Geographic Region Leadership change during the Previous Attempts this
dispute Mediator
Geographical Proximity Number Parties Supporting Timing (grouped)
of Parties the Disputants
Issue One Characteristics of Previous Timing (raw)
Disputes between the parties
Initiator, Issues, Balance of
Power, Highest Hostility Level,
Outcome, Temporal Proximity to
previous dispute, United Nations
Operations
Issue Two Dispute Phase Conict
Management
Issue Three Initiated by
Final Outcome Environment
Re-emergence of Dispute Outcome
Type of Conict Hostilities Reported
Number Mediation Efforts Durability Outcome
UN Involvement Rank Negotiator Party A
Type of Superpower Rank Negotiator Party B
Involvement
UN Peacekeeping/ Number of Mediators
Sanctions/Embargos Acting
Enduring Rivalries Functional Mediator
Identity
Characteristics of Mediation Type
Ethnic Conict Ethnic
Groups, Ethnic War,
Revolutionary War,
Abrupt/Disruptive
Regime Change, Genocide/
Politicide, Ethnic Issues,
Origin and Internationaliza-
tion of Ethnic Rivalries
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 conicts as is possible, using many archival resources.
Subsequently the project sought to encompass not just mediation but other
forms of conict management such as: negotiation, arbitration, multilateral
conferences, and referrals to international organizations. Information was col-
lected on (a) factors relating to the dispute, parties and issues in conict, (b)
factors relating to mediation performance and identity, and (c) factors describ-
ing the consequences or outcomes of mediation. Initially the project begun in
the mid 1980s with an examination of 72 international conicts and 210
cases of mediation (see Bercovitch 1986). Today the project contains signi-
cant information on 309 international conicts and 3737 cases of conict
management.
Some Findings
Making use of available data, we now have close to one million observations
on mediation and other forms of conict management over a fty 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 difcult. In
trying to understand the reality of mediation, we do, I believe, encounter
fewer difculties 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 codied, the data are arranged in the form of a large
matrix, with rows containing information about our units of analysis, media-
tion 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 quan-
titative or qualitative procedures to examine distribution of the data, relation-
ships between two or more variables, and develop causal inferences and
models, and answer the questions or puzzles posed above. Some of the most
important ndings of the project are highlighted below:
Nature of Conict
Many studies posit a relationship between a given conict and mediation out-
comes. We found that conict intensity impacts adversely on mediation effec-
tiveness (Bercovitch 1986). High fatalities and high intensity conicts are not
really amenable to mediation. Issues in conict 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). Conict complexity, exam-
ined as numbers of issues per conict, 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 conict is likely to be mediated. Democratic parties
are more likely to use mediation and achieve a positive outcome than non-
democratic parties (Bercovitch & Houston 1996). Similar patterns were found
between economically developed countries and mediation. Domestic obser-
vance of human rights and respect for liberty correlate very highly with suc-
cessful 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 conicts 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, experi-
ence and resources (Bercovitch & Houston 2000). We were also able to show
that the best time to initiate mediation is some months after a conict had
manifested itself (Bercovitch 1986). The most prominent mediators in inter-
national conicts are not regional or international organizations, but states,
and larger states at that. The most successful mediators are not private indi-
viduals 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 ndings,

the number of its practitioners, its capacity to impact, and its acceptance of
different paradigms and research methods. Conict and its resolution encom-
pass an array of methods, ndings, and paradigms that are not easily grouped
into a single set of beliefs or theories. At times we may be critical of each
others ideas, approaches, or ndings, 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 negotia-
tion by using available, public information. The sources of such information
may include vital national statistics, UN documents, data archives, newspa-
pers, letters, and survey reports. Archival research is best suited to the analy-
sis 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 schol-
ars. 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 conict 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 eld of study, mediation has developed over
the last three decades diverse philosophical underpinnings and different frame-
works 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 consid-
erations. 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 media-
tion. To understand something about the parties decisional calculus and
choices of behavior we need to pose different questions and use different
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 beneted considerably from the comments of Peter Carnevale and
an anonymous reviewer.
References
Bercovitch, J. (1986). International Mediation: A Study of Incidence, Strategies and Condi-
tions of Successful Outcomes, Cooperation and Conict 21(3):155168.
Bercovitch J., and Houston A. (1996). The Study of International Mediation: Theoretical
Issues and Empirical Evidence. In J. Bercovitch (Ed.) Resolving International Conicts. Boulder.
Westview Press.
Bercovitch, J., and Jackson, R. (1997). International Conict: Encyclopedia of Conicts and
their Management, 19451995. Washington, D.C.: Congressional Quarterly.
Bercovitch, J., and Houston, A. (2000). Why Do They Do It Like This? An Analysis of the
Factors Inuencing Mediation Behavior in International Conicts, Journal of Conict Resolution
44:170202.
Bercovitch J., and Schneider, G. (2000). Who Mediates: The Political Economy of Inter-
national Conict Management, Journal of Peace Research 37:145165.
Brookmire, D., and Sistrunk, F. (1980). The effects of perceived ability and impartiality of
mediator and time pressure on negotiation, Journal of Conict Resolution 24:31127.
Butterworth, R.L. (1976). Managing Interstate Conict, 19451974. University Center for
International Studies, University of Pittsburgh, PA, USA.
Carnevale, P. (1986). Strategic choice in mediation, Negotiation Journal 2:4156.
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: Jossey-
Bass, Inc.
Kochan, T.A., and Jick, T. (1978). A theory of public sector mediation process, Journal of
Conict Resolution 22:209240.
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 14:239246.
Princen, T. (1992). Intermediaries in International Conict. New Jersey: Princeton University
Press.
Rubin, J.Z. (1980). Experimental research on third party intervention in conict, Psycho-
logical Bulletin 87:379391.
Small, M., and David Singer, J.D. (1982). Resort to Arms Beverly Hills: Sage.
92 JACOB BERCOVITCH
Tillema, H.K. (1991). International Armed Conict Since 1945. Boulder: Westview Press.
Touval, S. (1982). The peace brokers Mediators in the Arab-Israeli conict 19481979. 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 Conict Resolution. London: Sage.
Zacher, M. (1979). International Conict and Collective Security. New York: Praeger.
Reections 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 net-
work-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 increas-
ingly involved the use of simulation and experimental approaches for the
study of foreign policy decision-making in general, and negotiation and medi-
ation 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 back-
ground on experimentation in political science in general, and in international
politics in particular. This will be followed by the story of how I became in-
volved with this approach in the rst place. Finally, in order to examine what
experimentation can and cannot do for us, I will present some preliminary
ndings that extend our knowledge on one particular phenomenon in inter-
national 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 eld. Kinder and Palfrey (1993: 1) offer a treatise on why experi-
mentation 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

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 qual-
ity that makes experimentation valuable: It is the most reliable means of
examining and understanding the nature of a causal relationship, of demon-
strating the internal validity of a theory in political science (Laponce 1972;
Bositis 1990; McGraw 1993).
Experiment-based research provides insights into relationships that are difcult,
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 science-
building. Because the results of experiments can enlighten analysts under-
standing of how and why a relationship exists and functions, this type of
research allows for the renement of general or preliminary theories in a way
distinct from examinations of large-n (quantitative) or small-n (qualitative) em-
pirical 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 benets 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 exper-
imentation 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 exper-
imental simulation work can be traced to the early work of Geutzkow and his
associates beginning in the late 1950s (Guetzkow et al. 1963). His Inter-
Nation 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 com-
puter scientist specializing in Distributed Articial Intelligence (DAI). (I had
REFLECTIONS ON SIMULATION AND EXPERIMENTATION OF NEGOTIATION 95
barely heard of Articial Intelligence at that point). Kraus had been working
on developing an automated agent capable of participating with other auto-
mated 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 Krauss case)
negotiated in difcult (crisis) situations in order to reach mutually benecial
agreements. Whereas my objective was to understand the conditions under
which negotiations ended without resort to violence, Krauss objective was to
understand how intelligent agents (machines) could be endowed with negoti-
ation 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 mini-
mize the amount of time spent on negotiation, and maximize the likelihood
of reaching mutually benecial 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 associ-
ated with various outcomes and hence enhance the subjects ability to maxi-
mize expected utility, even under conditions of incomplete information.3
Third, all communications among the participants have taken place in con-
trolled network environments, allowing for detailed analysis of the types of
tools they consulted as well as the content and form of their interactions dur-
ing 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 strate-
gic 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 shing
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) dene mediation as a process of

conict management where disputants seek the assistance of, or accept an
offer of help from, an individual, group, state, or organization to settle their
conict 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 denition is most useful because of its emphasis on
the key components of mediation the disputants, the third party, and the
specic conict 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 rst role is of the mediator
as communicator or facilitator (Touval & Zartman 1985; Burton 1984; Hop-
mann 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 ofces, or process facilitation (Hopmann 1996).
The second role dened by Touval and Zartman is mediator as formula-
tor. Here the mediator provides a substantive contribution to the negotiations
in order to assist the conicting 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 emo-
tions 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 solu-
tions, 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, inuence, 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 carrot-
and-stick measures (adding and subtracting benets to/from the proposed
solution) (Touval & Zartman 1985, Zartman & Touval 1996). Hopmann indi-
cates that only a powerful mediator can play this role, and he cites two modes
of mediator involvement as manipulator: Inuencing 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 sat-
isfaction 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 deci-
sion-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 conict that began in 1935 and ultimately terminated
in 1998 produced a total of ve 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 re, arbitration, and status quo. In part, we sought
to establish the reliability of our experimental environment through our abil-
ity 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 analy-
ses 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 gener-
ating agreements, and agreements that are considered satisfactory to negotia-
tors, only manipulative mediation has a positive effect on the level of benets
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 benecial outcome
than they could have otherwise secured, while facilitation may actually lower
average benets a situation that could lead to discontent among disputants
and, possibly, recurrent crises. In addition, the more rapid conclusion of a cri-
sis brought about only by manipulation is an essential component of cri-
sis management, given the relationship between prolonged crisis negotiations
and the likelihood of escalating violence and war. We conclude that manipu-
lative mediation, as compared with facilitation or no mediation, is an effec-
tive 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 ndings on negotiator satisfaction do not reveal
resentment of manipulation vis--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 ndings, 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 for-
eign policy decision-making process. On the surface, experimental work can
successfully address both of these issues. With a more or less unlimited sup-
ply of willing undergraduates as subjects, our n is virtually limitless. By care-
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 ocking 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 psy-
chologist seems natural requires of the political scientist a very real invest-
ment of time and energy, and a lot of trial and error. Thus, even with the
prospects of great advancements in knowledge seemingly at our ngertips, 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 evenings
indiscretions and plans for tonights, 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 partici-
pate in the experiment in the rst 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
difcult undertaking. In our own case, we have some evidence to suggest that
in a previous round of experiments involving a model of a shing dispute
roughly resembling the dispute between Canada and Spain in 1995, we not
only packed the model with excessive detail, but then superimposed an elab-
orate mediation process. The results showed that the subjects were not cop-
ing, 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
expense and time involved in the design and execution of a series of experi-
ments, it is not surprising that relatively few political scientists venture into
this domain.
Given this rather bleak picture, where do I see simulation and experimen-
tal design tting 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 approach-
ing 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 cri-
sis 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 com-
parative case studies, the experimental environment when properly speci-
ed can help us ll 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 gen-
eration of political scientists in simulation and experimental techniques. This
is not everyones cup of tea, nor should it be. It does offer an important alter-
native to the cross national and case study approaches that dominate the eld
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 convinc-
ingly 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-3595 and SG-5200). 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.
4. For details on our experimental procedures, see Santmire et al. (1998).
5. Hopmann (1996) qualies the notion of the manipulative mediator being a party to the dis-
pute by stating that this mediator Aalmost@ assumes the role of a disputant in the nego-
tiation 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 (forth-
coming).
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 rst 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 couldnt get war to break out. As it turned
out, the student playing one of the key roles was an avowed pacist, and simply wouldnt
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): 717.
Bercovitch, J., and Langley J. (1993). The nature of the dispute and the effectiveness of inter-
national mediation, Journal of Conict Resolution 37(4): 67099.
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 Inter-
national Politics. Princeton, NJ: Princeton University Press. 110128.
Burton, J.W. (1984). Global Conict: 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
Conict, Journal of Conict Resolution 16:1:6794.
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.
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 61:406416.
Hopmann, P.T. (1996). The Negotiation Process and the Resolution of International Conicts.
Columbia, SC: University of South Carolina Press.
Keashly, L., and Fisher, R.J. (1996). A Contingency Perspective on Conict Interventions:
Theoretical and Practical Considerations. In J. Bercovitch (Ed.) Resolving International Conicts:
The Theory and Practice of Mediation. Boulder, CO: Lynne Rienner Publishers. 235261.
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. Martins Press. 6496.
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. 139.
Kraus, S. (2001). Strategic Negotiation in Multiagent Environments. Cambridge: MIT Press.
Kraus, S., Wilkenfeld, J., and Zlotkin, G. (1995). Multiagent Negotiation under Time Con-
straints Journal of Articial Intelligence 75:297345.
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. 315.
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. 769786.
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 91:553566.
Princen, T. (1992). Intermediaries in International Conict. 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):721748.
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. 123.
Starkey, B. and Blake, E. (2001). Simulation in International Relations Education, Simulation
and Gaming 32(4): 537551.
Starkey, B. and Wilkenfeld, J. (1996). Project ICONS: Computer-assisted Negotiations for the
IR Classroom, International Studies Notes 21(1):2529.
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 56:1124.
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. Hills-
dale, NJ: Lawrence Erlbaum. 203219.
Torney-Purta, J. (1998). Evaluating Programs Designed to Teach International Content and
Negotiation Skills, International Negotiation: A Journal of Theory and Practice 3(1):
7797.
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. 717.
Wilkenfeld, J., and Kaufman, J. (1993). Political Science: Network Simulation in International
Politics, Social Science Computer Review 11(4):464476.
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 14:369391.
Wilkenfeld, J., Kraus, S., Santmire, Tara, and Frain, C. (1999). The Role of Mediation in
Conict 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 Conict 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 Conict. Washington, D.C.: United States Institute of Peace Press.
445461.
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? Per-
haps, 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 result-
ing 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 informa-
tion can be used to predict whether negotiators reach agreement as well as
the quality of their agreements, and to determine how negotiators can be bet-
ter 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 role-
play 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 cod-
ing 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 rst 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 inuence on ones ability to test and support hypotheses.

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
inuenced the quality of agreements whereas information about preferences
and positions negatively (or neutrally) inuenced 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 negotia-
tion 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 data-
driven? What steps should we take to ensure reliability? For more general dis-
cussions 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 rel-
ative emphasis. Whereas coding schemes that evolved from negotiation the-
ory emphasize outcomes, those that evolved from communication theory
emphasize the relational aspects of strategy choice. Ultimately, there is con-
siderable 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 iden-
tication 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 spe-
cic tactics such as threats, offers, questions regarding preferences and prior-
QUANTITATIVE CODING OF NEGOTIATION BEHAVIOR 107
ities, creative solutions, and information seeking. These specic behaviors

can, however, be linked to one of two strategic orientations integrative or
distributive and so convey implicit information about the underlying rela-
tionship. 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 Category

Numbera
Integrative States issue preferences 3

Information States issue priorities 4
Asks questions about preferences 8
Asks questions about priorities 9
Off-task comments (relationship building) 33
Create Value Makes multi-issue offer 2
Shows insight 13
Notes general differences 17
Notes general similarities 18
Makes positive comments 21
Suggests compromise 24
Suggests package trade-offs 25
Other process suggestions 30
Distributive Facts 7
Information Asks for bottom line 10
Asks about others substantiation (attack arguments) 11
Asks miscellaneous task related questions 12
Notes differences 15
Negative reactions 22
Suggests discuss one issue 23
Claim Value Makes single issue offer 1
Refers to bottom line 5
Substantiates Position 6
Refers to mutual interests to inuence other party 14
Threats 19
Refers to power 20
Suggests creative solutions to meet own interests 31
Push to closure Time checks 29
Notes similarities 16
Process Suggests using reciprocity 26
management Suggests vote 27
Suggests to move on 28
a
Numbers correspond to those in Figure 1.
Figure 1. Towers II Correspondence Analysis Results
2.5 Distributive information

15 23
2
16
1.5 Push to closure
12
7,10 29
1 22
0.5 11
17 21 20 6,1,5
0 2 19 31
25 14
0.5 24
13 Claim Value
1 30 18 33,9
Create Value 4,3
1.5 Integrative 8 26 28
information
2 27
Process Management
2.5
2.5 2 1.5 1 0.5 0 0.5 1 1.5 2
a
Numbers correspond to those in Table 1.
Communication theory based schemes start by considering the relational func-

tions served by communication: either to highlight similarities and bring peo-
ple together or to highlight differences and to push people apart. In the
context of negotiations, this has led to the identication of three categories of
negotiation behaviors, integrating, attacking and defending, and a set of
specic tactics that t 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 com-
munication can serve two functions: one is to provide a response to the other
party; the other is to cue the other partys 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.
Developing and rening 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 observa-
tion of negotiations themselves (refer to Putnams chapter in same volume).
However, the distinction between traditional theory-driven approaches and
more data-driven approaches is necessarily blurred in practice. Human behav-
ior is typically too complex for a researcher to be able to anticipate all rele-
vant behaviors without some direct experience of how people interact. Coding
schemes and their associated manuals should be considered living documents,
open to revision and clarication 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 benets of both approaches. For example, when we ini-
tially 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-specic 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 rening the coding scheme. This is best done by
having your coders independently code sections of transcripts and then check-
ing 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 classied),
and others that are too narrow or too similar to other categories (when decid-
ing 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 prob-
lems in training the coders or in the design of the scheme?
Coding data
In parallel with developing this coding scheme we developed a coding man-

ual, that is, a set of application rules/guidelines for the coders, including
identication of the unit to be coded (thought, sentence, speaking turn), category
labels, denitions, and rules of thumb for distinguishing between categories

and using context (i.e., statements surrounding the unit of interest) to inter-
pret meaning. Our coding manuals typically include general rules about uni-
tizing and coding as well as denitions 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 difcult to re-
solve, then the problem may reside in the design of the coding scheme itself.
Categories might overlap or be missing. These denitional 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 rst 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 oor, 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, identiable 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 identiable idea, but
also retained information about speaking turns to maximize our exibility.
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 analy-
sis 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 indi-
vidual statements convey a different meaning to that conveyed by a speaking
turn. We can also add redundancy as a result of separating immediate restate-
ments, that which simply repeat previous tactics rather than adding new infor-
mation. 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
provide rules of thumb for these decisions (e.g., Weingart et al. 1993).
Alternatively, the rst 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 inter-
pretive reliability. In both cases, the level of reliability places an upper bound
on a coding schemes predictive ability. Unitizing reliability, the degree of
agreement regarding identication 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
Guetzkows U, which calculates the difference between the number of units
identied by an independent coder and the true number of units (the aver-
age 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 identied 75
units and coder 2 identied 85 units in the same transcript, Guetzkows U =
(85 75)/(85 + 75) = .062, meaning that there was a 6% discrepancy between
the number of units identied 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 con-
sistency. Thus, consistency of interpretation is assessed in terms of the reli-
able 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 Cohens
Kappa (1960), which determines the level of agreement corrected for agree-
ments 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 classication, 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 Cohens original article (1960). More detailed discussions of
Kappa can be found in Fleiss (1971) and Folger et al. (1984).
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 (identied the speaking turns) slightly dif-
ferently 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 ow of conversation and so unitize around them
whereas others do not. This problem can be avoided by completing the uni-
tizing rst, 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) observer-
privileged: 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 ask-
ing. In this section, we consider three related issues: aggregating data, level
of analysis and time segmentation.
Aggregating data
The rst 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
important to know how many times each negotiator in a dyad or group uses
a specic 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 negotia-
tion? 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 descrip-
tion 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 negotia-
tors 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 mani-
pulated the composition of negotiating dyads and groups. At the level of the
dyad, this means that two negotiators can have similar attributes, for exam-
ple, both are cooperatively-motivated, or they can have mixed attributes, that
is, one negotiator is cooperatively motivated while the other is individualisti-
cally 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 ques-
tion again depends on the level of your theory whether global or specic
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 nd 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 dif-
ferent 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 fre-
quency 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
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 analy-
sis thus provides an important test of our theoretical assumptions about the
function and use of specic 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 com-
bined 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 analy-
sis (Greenacre 1993). This analysis establishes relationships between tactics,
based on their frequency of use. It is conceptually similar to factor analysis,
but developed specically 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 identied and
interpreted. There are no hard and fast rules for the interpretation of this out-
put. Visual inspection determines clusters based on (a) physical proximity and
(b) thematic t. 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 distribu-
tive 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 analy-
sis (see Brett, Weingart and Olekalns, 2004; Weingart & Olekalns, 2004). Our
other article, to be published in the next issue of International Negotiation
(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 negotia-
tion 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 specic 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 strate-
gies are used tells us about the global approach used by the negotiator(s). The
major shortcoming of this approach is that it reects a static view of negoti-
ation 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 fac-
tors and strategy choices, as well as between strategy choices and outcomes,
we cannot learn about the patterning of strategies over time. Analyzing fre-
quencies helps us to answer questions such as: Does the negotiating context
affect the kind of strategies that are used? Are specic strategies linked to dif-
ferent kinds of outcomes?
There are two approaches that allow us to better capture the dynamic
nature of negotiations. The rst 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 others behaviors.
This tells us whether negotiators are attempting to sustain or redirect the pre-
vailing 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 nego-
tiators break off distributive spirals or derail integrative processes. This
approach also helps us to develop more complex models of negotiators inter-
actions: It provides insights into how negotiators constrain each others
actions and into how they direct the negotiation towards obtaining their pre-
ferred outcomes. One important insight that we have gained from this type of
analysis is that sequences that are inuential 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).
Although sequences incorporate a temporal dimension, the focus on imme-

diate actions and reactions means that the analysis is of short-term effects.
Analyzing negotiation phases also captures the temporal aspect of negotia-
tions. However, the time frame for examining the consequences of strategies
is a longer-term one, allowing consideration of the ebb and ow of the nego-
tiation over time. There are two approaches to analyzing negotiation phases.
The rst, 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 identiable interaction sequences
with explicit beginnings and ends (Baxter 1982). They are dened on the
basis of runs of strategies (i.e., the continuous use of a single strategy) with
transitions between phases identied on the basis of interruptions to extended
runs of the same strategy. In the following section we consider the methodolog-
ical 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 rst is how many stages in which to divide the negotiation.
Negotiation theory identies anywhere between three and 12 stages. In our
research, we have used six stages as the number of segments that provides a
reasonably ne-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, ve 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 nd yourself com-
paring strategies that occur in the 1015 minute interval and which represent
the midpoint for one negotiating dyad but the endgame for another. We there-
fore use segmentation based on speaking turns. We prefer this approach as it

means that we are comparing comparable segments of the negotiation: mid-
point 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 nd 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). Conict management: An episodic approach, Small Group Behavior
13:2342.
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
40:111120.
Cohen, Jacob (1960). A coefcient of agreement for nominal scales. Educational and
Psychological Measurement, 20, 3746.
Donohue, W.A., Diez, M.E., and Hamilton, M. (1984). Coding naturalistic negotiation inter-
action, Human Communication Research 10:403425.
Fleiss, J.L. (1971). Measuring nominal scale agreement among many raters, Psychological
Bulletin 76:378382.
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. 115161). Norwood:
Ablex.
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 30:255272.
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 6:4758.
Hewes, D.E. (1979). The sequential analysis of social interaction, Quarterly Journal of
Speech 65:5673.
Hyder, E.B., Prietula, M.J., and Weingart, L.R. (2000). Getting to best: Efciency versus opti-
mality in negotiation, Cognitive Science 24(2):169204.
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 com-
petitive negotiations, Personality and Social Psychology Bulletin 25:657668.
Olekalns, M., and Smith, P.L. (2000). Negotiating optimal outcomes: The role of strategic
sequences in competitive negotiations, Human Communication Research 24:528560.
Olekalns, M., and Smith, P.L. (2003). Testing the relationships among negotiators motiva-
tional orientations, strategy choices, and outcomes. Journal of Experimental Social
Psychology.
Olekalns, M., Smith, P.L., and Walsh, T. (1996). The process of negotiating: Strategies, tim-
ing and outcomes, Organizational Behavior and Human Decision Processes 67:6177.
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 communica-
tion research (pp. 220256). 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 conict. Pacic Grove: Brooks/
Cole.
Pruitt, D.G., and Lewis, S.A. (1975). Development of integrative solutions in bilateral nego-
tiation, Journal of Personality and Social Psychology 31:621633.
Putnam, L.L., and Jones, T.S. (1982). The role of communication in bargaining, Communi-
cation Monographs 3:262282.
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 17:371394.
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. 184239). 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 Psy-
chology 78(3):504517.
Weingart, L.R., Brett, J.M., and Olekalns, M. (2002). Conicting social motives in negotiating
groups. Paper presented at the 62nd Annual meeting of the Academy of Management,
Denver, Colorado.
Weingart, L.R., Hyder, E.B., and Prietula, M.J. (1996). Knowledge matters: The effect of tac-
tical descriptions on negotiation behavior and outcome, Journal of Personality and Social
Psychology 70:12051217.
Weingart, L.R., and Olekalns, M. (2004). Communication Processes in Conict and Negotia-
tion. 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 Conict Management 1:731.
The Use of Questionnaires in Conict Research
AUKJE NAUTA and ESTHER KLUWER
Conict is an important theme to study, both in organizations and in close

relationships. In organizations, conict may hinder productivity and job satis-
faction. In close relationships, conict can be a threat to relationship satis-
faction and even to the endurance of the relationship. It is therefore important
to study conict 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 conicts
better and prevent conicts from escalating into destructive outcomes.
Unfortunately, it is far from easy to study conict empirically. It is espe-
cially difcult to organize settings in which conicts are acted out naturally,
either in eld settings or in the laboratory. Field settings are often too threat-
ening for participants and too uncontrollable for researchers, whereas labora-
tory settings have limits regarding external validity (i.e., the extent to which
experimental results can be generalized to real-life settings; Cook and Camp-
bell 1979). In addition, observing real conict is time-consuming and expensive.
An alternative to studying real-life conict is the use of questionnaires.
Questionnaires whether administered as paper-and-pencil surveys or as
(telephone) interviews can be useful alternatives for laboratory or observa-
tional studies on conict when researchers know how to avoid possible pit-
falls. 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 medi-
ation (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.

P. Carnevale and C.K.W de Dreu (eds.) Methods of Negotiation Research, 121133
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 question-
naire items are short (20 or fewer words), simple, direct, clear, specic, 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 rst), 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 specic challenges encoun-
tered when studying conict. In this article, we will therefore discuss the
strengths and weaknesses of using questionnaires in conict 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 conict in organizations and in close
relationships.
The main focus is on measuring conict behavior, although there are many
other relevant aspects to conict, such as conict issues, causes, and outcomes
(De Dreu, Harinck & Van Vianen 1999; Van de Vliert 1998), as well as the
motives, emotions, and cognitions that accompany conict (De Dreu & Carne-
vale, 2003). We believe conict behavior is a key factor of the conict pro-
cess because it largely determines the (positive or negative) consequences of
conict. At the same time, conict behavior may be one of the most difcult
conict features to measure through questionnaires, because questionnaires by
denition measure perceptions of current or past behavior instead of the
actual behavior itself.
We will address ve issues concerning the use of questionnaires in conict
research. The rst is that conict is a highly sensitive topic, which has con-
sequences for the acquisition of participants and the response to surveys. The
second issue concerns the validity of conict 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 specic threat to validity that
deserves special attention. Fourth, we will address the fact that conict
surveys usually entail correlational, and often cross-sectional, designs that
prohibit conclusions about causality. The fth and last issue concerns the fact
that data on conict behavior are by denition 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 conict in organizations and
in close relationships and include recommendations for doing questionnaire
research.
Sensitivity
Conict is a sensitive subject. People generally do not like to admit that they
have conicts with others because it is associated with negative consequences,
as Nauta experienced when interviewing employees in organizations about con-
ict and negotiation (Nauta & Sanders 2000, 2001; Nauta, De Vries, and
Wijngaard 2001). Because of this sensitivity, people are often reluctant to par-
ticipate in research on conict. For example, Nauta (2003) tried to recruit four
different Dutch organizations for a study on organizational determinants and
consequences of conict. 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 conict in the introduction of the study, Nauta
therefore chose to frame the research more positively by stressing how orga-
nizations could improve interpersonal and inter-team collaboration. This pre-
vented participants from experiencing some of the unpleasant connotations
that may accompany conict such as ghting and violence and aggres-
sion, facilitating the decision to participate in the study.
The fact that conict is a sensitive topic may cause another research pro-
blem 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 orga-
nizational unit. Some participants remarked that their anonymity was not
guaranteed by asking these specic background questions. For other partici-
pants, 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 conicts were less likely to return the questionnaire,
resulting in a selective sample of participants.
When using a questionnaire to study organizational conict, 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
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 study-
ing a delicate subject. For example, in research on close relationships, ques-
tionnaires are often chosen deliberately to study topics that are sensitive, such
as relationship problems, sex, and conict (e.g., Kluwer, Heesink, and Van de
Vliert 1996). People generally feel more comfortable lling 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 conict

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 ques-
tionnaires that measure the ve-part typology of conict 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 reach-
ing a solution that meets others but not own goals and interests), avoiding
(withdrawing from the conict, either temporarily or denitely), problem-
solving (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 mea-
sure these types of conict behavior. The oldest is the Conict Management
Survey (Hall 1969). Although the CMS is still very popular in conict man-
agement 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 Conict Inven-
tory (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 Conict Instrument (OCCI;
Putnam & Wilson 1982) and the Thomas-Kilmann Conict 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 ll
it out for free and directly receive the results on their preferred mode of
conict management, compared to a reference base.
A questionnaire that is widely used in the Netherlands is the Dutch Test
for Conict 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 corre-
lated highly with both ratings of the opponent and observer ratings. However,
the correlations for avoiding behavior were not signicant. Avoiding behavior
is difcult to observe, especially in the laboratory, because participants are not
likely to avoid conict when they are asked to act out a conict. In addition,
avoiding is ambiguous behavior open to multiple attributions. For example,
someone who consistently downplays the importance of the conict 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
ones way, to buy time, and to impose ones will on others (i.e., forcing).
Perhaps avoidance more than any other conict behavior involves acts
that are difcult 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, conver-
gence between self-reports and other-reports of avoiding is likely to be low.
Nevertheless, it is still a question whether conict 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 role-
play was over. Participants thus reected upon the behavior they had just
exhibited. However, in eld studies, the DUTCH is generally used to measure
the behavior one usually or generally shows when engaged in conict. In
this setting, the DUTCH appears to measure a general intention to certain
conict behavior or even a general style, instead of actual behavior. A possi-
ble solution to this problem is the critical incident method (Flanagan 1954;
see also Kluwer 2000), in which participants are asked to recall a recent
conict and answer questions about this particular incident, including behav-
ioral measures. This yields a more valid measure of actual conict behavior
and participants nd it easier to answer questions about their behavior in a
specic real-life situation than questions about their general behavioral style
(see for example, Kluwer 2000). Nevertheless, one must realize that the
answers may be inuenced by the specic context in which the critical inci-
dent 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
Table 1. The Dutch Test for Conict Handling (DUTCH)
When I have a conict 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 nd a compromise solution.
I insist we both give in a little.
I strive whenever possible towards a fty-fty compromise.
Forcing
I push my own point of view.
I search for gains.
I ght for a good outcome for myself.
I do everything to win.
Problem-solving
I examine issues until I nd a solution that really satises me and the other party.
I stand for my own and others goals and interests.
I examine ideas from both sides to nd a mutually optimal solution.
I work out a solution that serves my own as well as others 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 ve-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 conict between the participant and an opponent. After reading

the scenario, participants were asked to what extent they would use coopera-
tive tactics (e.g., try to collaborate in each others interest) and competitive
tactics (e.g., show distrust). Although these are still intentions to evince one
type of conict behavior or another rather than actual conict behavior, those
intentions are likely to be associated with real behavior because participants
had to imagine themselves in a specic situation rather than a generalized
conict situation. Furthermore, the use of vignettes enables researchers to
manipulate certain characteristics of the conict. For example, in their
research on relationship conict over the division of labor, Kluwer (1998) and
Kluwer, Heesink, and Van de Vliert (2000) used vignettes to manipulate the
topic of the conict (housework, paid work, or child care) and each partners
role in the conict (complainant versus defender of the status quo; see Table
2 for an example).
Table 2. Example of a Conict Scenario (Kluwer, Heesink, and Van de Vliert 2000)
You are dissatised with the time your spouse spends on household tasks. For exam-
ple, you are dissatised 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 satised with the
situation as it is.
Self-serving and Social Desirability Biases
Whenever researchers measure conict behavior, chances are high that they
will nd some sort of self-serving bias, that is, a tendency to see ones own
conict behavior as more constructive and less destructive than the conict
behavior of ones opponent (De Dreu, Nauta & Van de Vliert 1995; Kluwer
et al. 1998; Nauta and Van Sloten 2003). Although it is an interesting theo-
retical phenomenon, self-serving bias is a threat to validity, because partici-
pants report biased perceptions of their own and their opponents behaviors.
Self-serving bias may also imply that the measures are inuenced by a gen-
eral tendency towards social desirability, which is more likely to occur when
one rates ones own behavior than when one rates others behavior. This may
also occur when measuring conict frequency, for example in close relation-
ships. People are generally reluctant to admit that they have relationship
conicts, because it might imply that their relationship is in trouble. In addi-
tion, happy couples tend to distort their appraisal of their relationship in pos-
itive ways (e.g., Murray & Holmes 1997). Hence, people may not admit that
they have conicts with their partner due to social desirability biases or they
may not report conict due to positive illusions they hold about their rela-
tionship. For both reasons, measures of absolute conict frequency are likely
to be unreliable.
A possible solution to biased measurements of conict behavior and con-
ict frequency is to measure these concepts from different perspectives: Self
reports, opponent reports and observer reports of conict behavior and conict
frequency. When interrater reliability is high, it can be assumed that the
average of these three perspectives is an accurate (intersubjective) measure.
Cross-sectional Data
Conict 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 conicts. Although they
used multiple methods (i.e., questionnaires and oral interviews), it was still
difcult 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 respon-
dents 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 eld, 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. How-
ever, 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
conicts about working relationships between employees and their superiors
in ten different organizations, Nauta and Van Sloten (2004) used question-
naires, together with data from personnel les about absenteeism. Moreover,
they distributed surveys among both employees and their superiors, which enabled
them to aggregate data of superiors about determinants of conict behavior to
the organizational level, and then examine whether these organizational vari-
ables (like communication culture, perceived common ground) predicted the
conict behaviors of employees.
Dyadic Data
The last issue that we address concerns the dyadic nature of conict data. By
denition, conict is an interpersonal concept: Conict 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 conict. However, in orga-
nizations, it is often difcult to couple data of two conict parties while at
the same time guaranteeing anonymity. Therefore, studies of dyadic conict
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 inuence each other on various dimen-
sions. 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 rst assumes a partner effect: A characteristic of partner
A inuences another characteristic of partner B. For example, partner As
conict behavior inuences partner Bs report of relationship satisfaction and
vice versa. The second model assumes mutual inuence: A characteristic of
partner A inuences the same characteristic of partner B. For example, part-
ner As reported conict behavior inuences partner Bs reported conict
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 conict
behavior of the couple inuences their mutual report of relationship satisfac-
tion (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 conict interaction patterns in close relationships. They measured
both spouses perceptions of several conict 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 signicant variation on the marital interaction
variables (Kenny & La Voie 1985). They then computed the mean of the
husbands and the wifes 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 conict interaction patterns are couple constructs, as
opposed to husbands and wives reports of their own, individual conict
behaviors. This approach improves the reliability or robustness of conict
measures, because assessments stem from two sources instead of one.
Although computing the mean of both conict parties perceptions of conict
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
than observing actual interaction. It simply measures both parties joint per-
ceptions regarding interaction patterns that occur during conict. 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 vari-
able 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 conict behavior)
and variables at the individual level (e.g., individual measures of conict
behavior).
In sum, we recommend that, when it is theoretically necessary to couple
data of both sides of a conict, 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 conict through the use of questionnaires.
We discussed the sensitivity of conict 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 conict questionnaires and the fact that conict behavior is eas-
ily confounded with behavioral intentions. Specically, we noted that ques-
tionnaire measures of conict 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 conict data are often dyadic in nature, which leads
to specic methodological problems due to the dependency of dyad members.
We would like to conclude with a few recommendations to conict re-
searchers who want to use questionnaires. First, conict research generally needs
careful introduction, implementation and reporting to participants. In order

to get enough participants, it may be helpful to avoid using the word con-
ict and stress that one wants to examine correlates of collaboration and
interpersonal relationships. Second, validity problems may be solved by using
existing conict 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 question-
naires, we believe that questionnaires can be a very appropriate method to
study conict. It enables researchers to approach large samples, both in the
eld and in the laboratory, and when administered appropriately, it is possi-
ble to gather reliable and valid data on conict in real-life settings.
References
Acitelli, L.K., and Antonucci, T.C. (1994). Gender differences in the link between marital sup-
port and satisfaction in older couples, Journal of Personality and Social Psychology 67:688698.
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 conict: The demand/with-
draw interaction pattern. In S. Oskamp and M. Constanzo (Eds.), Gender issues in con-
temporary society. London: Sage. 113141.
Cook, T.D., and Campbell, D.T. (1979). Quasi-experimentation. Design and analysis issues for
eld settings. Boston: Houghton Mifin.
De Dreu, C.K.W., Evers, A., Beersma, B., Kluwer, E.S., and Nauta, A. (2001). A theory based
measure of conict management strategies in the work place, Journal of Organizational
Behavior 22:645668.
De Dreu, C.K.W., and Carnevale, P.J. (2003). Motivational bases of information processing
and strategy in conict and negotiation. In M.P. Zanna (Ed.), Advances in Experimental
Social Psychology. Vol. 35. New York: Academic Press 235291.
De Dreu, C.K.W., Harinck, F., and Van Vianen, A.E.M. (1999). Conict 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. 376405.
De Dreu, C.K.W., Nauta, A., and Van de Vliert, E. (1995). Self-serving evaluation of conict
behavior and escalation of the dispute, Journal of Applied Social Psychology, 25, 20492066.
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 hirarchische conicten
[Behavior and escalation in hierarchical conicts], Toegepaste Sociale Psychologie, 4:2841.
Flanagan, J.C. (1954). The critical incident technique, Psychological Bulletin 51:327359.
Hall, J. (1969). Conict management survey. Conroe, Tex.: Teleometrics International.

Janssen, O., and Van de Vliert, E. (1996). Concern for others goals: Key to de-escalation of
conict, International Journal of Conict Management 7:99120.
Kenny, D.A. (1995). The effect of nonindependence on signicance testing in dyadic
research, Personal Relationships 2:6775.
Kenny, D.A. (1996). Models of non-independence in dyadic research, Journal of Social and
Personal Relationships 13:279294.
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. 275285.
Kenny, D.A., and La Voie, L. (1985). Separating individual and group effects, Journal of
Personality and Social Psychology 48:339348.
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 conict
handling behavior, Educational and Psychological Measurement 37:309325.
Kluwer, E.S. (1998). Responses to gender inequality in the division of family work: The sta-
tus quo effect, Social Justice Research 11:337357.
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, JuneJuly.
Kluwer, E.S., De Dreu, C.K.W., and Buunk, B.P. (1998). Conict in intimate versus non-
intimate relationships: When gender role stereotyping overrides biased self-other judgment,
Journal of Social and Personal Relationships 15:637650.
Kluwer, E.S., Heesink, J.A.M., and Van de Vliert, E. (1996). Marital conict about the divi-
sion of household labor and paid work, Journal of Marriage and the Family, 58, 958969.
Kluwer, E.S. Heesink, J.A.M., and Van de Vliert, E. (1997). The marital dynamics of conict
over the division of labor, Journal of Marriage and the Family 59:635653.
Kluwer, E.S. Heesink, J.A.M., and Van de Vliert, E. (2000). The division of labor in close
relationships: An asymmetrical conict issue, Personal Relationships 7:263282.
Landy, F. (1978). Conict management survey. In O.K. Buros (Ed.), Eight mental measure-
ment yearbook. Vol. 2. Highland Park, N.J.: Gryphon. 11731174.
Murray, S.L., and Holmes, J.G. (1997). A leap of faith? Positive illusions in romantic rela-
tionships. Personality and Social Psychology Bulletin 23: 586604.
Nauta, A. (2003). Perceived inuence in team decisions: The more, the better for conict man-
agement 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 Conict Management 11:135161.
Nauta, A., and Sanders, K. (2001). Causes and consequences of perceived goal differences
between departments within manufacturing organizations, Journal of Occupational and
Organizational Psychology 74:321342.
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, orga-
nizational goal concerns and interdepartmental problem-solving behavior. Journal of
Organizational Behavior, 23, 199213.
Nauta, A., De Vries, J., and Wijngaard, J. (2001). Power and biased perceptions of interdepart-
mental negotiation behavior, Group Processes and Intergroup Relations 4:263270.
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 17:313330.
Putnam, L.L., and Wilson, C.E. (1982). Communicative strategies in organizational conicts:
Reliability and validity of a measurement scale. In M. Burgoon (Ed.), Communication Yearbook.
Vol. 6. Beverly Hills, CA: Sage. 629652.
Rahim, A., and Magner, N.R. (1995). Conrmatory factor analysis of the styles of handling
interpersonal conict: First-order factor model and its invariance across groups, Journal of
Applied Psychology 80:122132.
Raudenbush, S.W., and Bryk, A.S. (1986). A hierarchical model for studying school effects,
Sociology of Education 59:117.
Rubin, J.Z., Pruitt, D.G., and Kim, S. (1994). Social conict: 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, 957959.
Shaughnessy, J.J., Zechmeister, E.B., and Zechmeister, J.S. (2000). Research methods in psy-
chology. Boston: McGraw Hill.
Shockley-Zalabak, P. (1988). Assessing the Hall Conict Management Survey, Management
Communication Quarterly 1:302320.
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). Conict 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. 651717.
Thomas, K.W., and Killman, R.H. (1974). The Thomas-Kilmann conict mode instrument.
Tuxedo, N.Y.: Xicom.
Thomas, K.W., and Killman, R.H. (1978). Comparison of ve instruments measuring conict
behavior, Psychological Reports 42:11391145.
Van de Vliert, E. (1997). Complex interpersonal conict behaviour. Theoretical frontiers. Hove:
Psychology Press.
Van de Vliert, E. Conict and conict 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. 351376.
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 nego-
tiating 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 understand-
ing 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 conict 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 nego-
tiators (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 communi-
cation) on the outcomes of negotiations systematically differ from country to
country. They have proposed several theoretical models examining the mod-
erating 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 individu-
alism 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).

136 XU HUANG AND EVERT VAN DE VLIERT
Consequently, it remains problematic to generalize the individual-level nd-

ings to an explanation of cross-national differences in negotiation processes.
To tackle this problem, we propose a new approach to research on cross-
national 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 individual-
level relationships. As a result, we may develop a better understanding of why
the psychological states of the negotiators, the social conditions of the nego-
tiations, and the behavior of the negotiators impact differently on negotia-
tion outcomes across countries.
In the following sections, we rst briey 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 cross-
national research projects that used multilevel modeling, we delve into how
recent methodological developments contribute to a more rened cross-level
approach to investigating the moderating effects of national contextual vari-
ables. 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 inter-
esting, scientically they are of no great help as they do not offer theory-
based explanation of such cross-national variations.
In order to address such variations in negotiation processes on a theoreti-
cal 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 ori-
entation to dominate of the opponents was positively linked with joint out-
comes among Americans, but negatively linked with joint outcomes among
Norwegians. Using Hofstedes (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 behav-

ior of their opponents. Although Shapiro and Rogness 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 explana-
tions (e.g., Lytle, Brett, Barsness, Tinsley, and Janssens 1995).
Gelfand and Realos (1999) recent work was among the rst 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-modera-
tor model. Their ndings showed that the well-known link between account-
ability 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 prot among indi-
vidualists, but positively related to joint prot among collectivists. The
authors reasoned that, among individualists, accountability activated competi-
tive construals and behavior, and resulted in lower outcomes. By contrast,
among collectivists, accountability activated cooperative construals and behav-
ior, and resulted in higher outcomes. However, cultural individualism and col-
lectivism 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 Realos
(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 liv-
ing in the same country are inuenced by the same national context, which
may be reected 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 over-
estimation 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 individ-
uals who share the same culture, and leads to an underestimation of stan-
dard errors. Third, rival national moderating variables cannot be statistically
controlled. Therefore, in the following section, based on a series of cross-

national 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 volun-

teer work have contributed to a more sophisticated analytical model that per-
mits 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 mod-
erator 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 analy-
sis, country-level analysis and individual-level analysis. Statistically, the inte-
gration 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 multi-
level 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 coefcients
in ordinary regression analysis, and their signicance level can be tested using
t-distribution tests. Moreover, Mlwin produces a statistic called the deviance,
which indicates how well a given model ts the data. If two models are
nested, the difference in the deviances of the two models has a chi-square dis-
tribution 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 ts signicantly better than the sim-
pler 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, indepen-
dent variables from the individual level and country level can be entered into
the model to explain the residual variances at both levels, respectively.
These residual variances will decrease if the independent variables from the
two levels have a signicant 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 signicant
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 specic theo-
retical 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 Hofstedes
(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 respon-
dents. The response rate was about 75 percent. The sample contained 52 per-
cent 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 ndings 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 conrmed the expectation that job level (blue-collar workers
versus white-collar workers) is positively related to job satisfaction in indi-
vidualistic countries, but is not signicantly associated with job satisfaction in
collectivistic countries. We reasoned that more intrinsically motivating white-

collar work is more likely to engender more job satisfaction in individualis-
tic 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 blue-
collar 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 con-
trolled 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
rst 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 ones 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 inuencing 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 abil-
ities on job satisfaction. One of the interesting ndings 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 lit-
tle opportunity to use ones 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 gener-
ally have a higher level of education, and thus be more easily inuenced 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 ones
skills and abilities more than do blue-collar workers. Apparently, the cross-
level nation-as-moderator model seems very useful in quantitatively testing cross-
national variations in even more complex individual-level relationships.
Country Level 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
Little Opportunity to Use Skills and Much Opportunity to Use Skills and
Abilities Abilities
4 4
Individualism
Job Satisfaction
Job Satisfaction
3.5 3.5
Collectivism
Collectivism
3 3
Individualism
2.5 2.5
1.5 0.5 0.5 1.5 1.5 1 0.5 0 0.5 1 1.5
Blue-Collar White-Collar Blue-Collar White-Collar
Figure 2. Three-Way Interactive Effect of Cultural Individualism, Blue-Collar versus White-

Collar 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 19901993
World Values Survey (World Values Study Group 1994). The surveys were
carried out through face to face interviews by professional researchers, with
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 motiva-
tions might well be positively linked because self-sacricing is a way of
establishing ones self-identity, conrming ones 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 inuence on the cooperative or competitive nature of the complex
motivation to engage in volunteer work. Depicted in Figure 3 is the path dia-
gram of the variant of the cross-level nation-as-moderator model that repre-
sents those expectations.
A stepwise multilevel analysis was conducted to predict other-serving moti-
vations 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 ones 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 con-
tained average climatic temperature, temperature-squared and gross national
income per capita. The two-way interactions followed in step 5, and the three-
way 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 consider-
ably from one nation to another (step 3). The most important nding, in step
6, represented in Figure 4, showed that a voluntary workers self-serving and
other-serving motivations tend to be positively linked in high-income regions
Country Level National Wealth Climatic Temperature
Individual Level Self-Serving Motives 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 cli-
mates (e.g., West Africa around the Gulf of Guinea), and tend to be nega-
tively linked in low-income regions with cold climates (e.g., Baltic States).
We concluded that, in depressed countries where the nancial resources are
minimal and the climatic demands maximal, volunteering seems to be expe-
rienced 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 country-
level 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 coun-
tries the size of the correlation coefcients correlated .53 with cultural indi-
vidualism. 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, rst at the individual level and then at
the country level, they demonstrated that the cross-national variations in a
Cold Climate Temperate Climate Hot Climate
4.5 4.5 4.5

Other-Serving Motivation
3.5 3.5 3.5
Poor Countries
2.5 2.5 2.5 Poor Countries
1.5 1.5 Rich Countries 1.5
0.5 0.5 0.5
0.5 0.5 0.5
1.5 1.5 Poor Countries 1.5
2.5 Rich Countries 2.5 2.5
3.5 3.5 Rich Countries
3.5
4.5 4.5 4.5
1.5 0.5 0.5 1.5 1.5 0.5 0.5 1.5 1.5 0.5 0.5 1.5
Self-Serving Motivation Self-Serving Motivation Self-Serving Motivation
Figure 4. National Wealth and Thermoclimate Jointly Inuence the Link between
Self-Serving and Other-Serving Motivations
particular psychological relationship, the link between self-esteem and life sat-
isfaction, 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 con-
siderably across the nations under investigation. Oishi, Diener, Lucas, and Suh
(1999) pointed out that the number of respondents in Diener and Dieners
(1995) study ranged from 29 to 985. Consequently, the correlation coefcient
computed in Cameroon with 29 respondents is less reliable than the correla-
tion coefcient 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 coefcients drawn from sep-
arate countries to the national characteristics of these countries. In multilevel
analysis, the differential reliability becomes less problematic because regres-
sion lines instead of correlation coefcients are used as dependent variables.
In short, the studies on job satisfaction and motives for volunteer work dis-
cussed 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 specic
issues in cross-national research. More importantly, by linking national char-
acteristics 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 multi-
level modeling, researchers can control for a number of rival explanatory vari-
ables at the country level, and can even examine the joint effects of different
country-level variables on individual-level relationships.
Implications for Research on Negotiations
In this article, we challenged the methodological approaches used to investi-

gate and explain cross-national differences in the links between the psycho-
logical states of the negotiators, the social conditions of the negotiations and
the behavior of the negotiators, on the one hand, and outcomes of negotia-
tions, 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-as-
moderator 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 accu-
rate 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 indi-
vidualism is just one of the national contextual variables that may exert a
signicant inuence on the negotiation processes. Other national contextual
variables such as political rights, civil liberty, and dominant religion may also
inuence processes of negotiation (Bercovitch & Elgstrm 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 partys goals, and that this interde-
pendence 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 real-
ity, individuals are subject to the inuence 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 inuence individual-level negotiation processes.
Despite all of these virtues, the application of multilevel modeling to cross-
national 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
Country Level National Contextual Variables

e.g., Individualism
Power Distance
Political System
Individual Level Negotiators Psychological States Negotiation Outcomes

Social Conditions of Negotiation
Negotiators Behavior
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 indi-
vidual-level relationships can be tested. As shown in the studies reported
above (Huang & Van de Vliert, forthcoming; Van de Vliert et al., 2004), mul-
tilevel modeling only allows for tests of the moderating effects of national
contextual variables on a single individual-level relationship or on an indi-
vidual-level two-way interaction effect. As a rule, negotiation models pro-
posed by various researchers are characterized by a rather complex structure
(Gelfand & Dyer 2000; Morris & Fu 2001). It is difcult to test these nego-
tiation 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 psy-
chological states, social conditions of negotiation, and behavior of the nego-
tiators on negotiation outcomes may systematically vary across nations. In
some nations, some strategies may have a positive effect on outcomes; in oth-
ers, there may be no effect at all. There may even be a negative effect in cer-
tain nations. These cross-national variations may exist in a systematic manner.
And this systematic variation may well be explained by some national con-
textual 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
call this higher level of generalization of theories, a meta-theoretical per-

spective. In a nutshell, the cross-level approach discussed here may not only
offer an alternative and a more statistically rigorous way of testing cross-
national 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):
97104.
Brett, J.M., and Okumura, T. (1998). Inter and intracultural negotiations: US and Japanese
negotiators, Academy of Management Journal 41: 495510.
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: Scientic 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: 118121.
Chiu, L.H. (1993). Self-esteem in American and Chinese (Taiwanese) children, Current
Psychology: Research and Reviews 11: 309313.
Diener, E., and Diener, M. (1995). Cross-cultural correlates of life satisfaction and self-
esteem, Journal of Personality and Social Psychology 68(4): 653663.
Diener, E., Diener, M., and Diener, C. (1995). Factors predicting the subjective well-being of
nations, Journal of Personality and Social Psychology 69(5): 851864.
Fu, H.Y. and Morris, M.W. (2000). Need for closure fosters adherence to cultural norms:
Evidence from cross-cultural studies of conict resolution choices, Stanford GSB Research
Paper, #1649.
Gelfand, M.J., and Realo, A. (1999). Individualism-collectivism and accountability in inter-
group negotiations, Journal of Applied Psychology 84(5): 721736.
Goldstein, H., Rasbash, J., Plewis, I., Draper, D., Browne, W., Yang, M., Woodhouse, G., and
Healy, M. (1998) A users guide to Mlwin. University of London.
Graham, J.L., Kim, D.K., Lin, C.Y., and Robinson, M. (1988). Buyer-seller negotiations
around the Pacic Rim: Differences in fundamental exchange processes, Journal of
Consumer Research 15: 4854.
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: 329348.
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.
Lytle, A.L., Brett, J.M., Barsness, Z.I., Tinsley, C.H., and Janssens, M. (1995). A paradigm
for conrmatory cross-cultural research in organizational behavior, Research in Organi-
zational Behavior 17: 167214.
Maslow, A.H. (1954). Motivation and Personality. New York: Harper & Row.
McCusker, C.R. (1994). Individualism-collectivism and relationships in distributive negotia-
tion: An experimental analysis, Unpublished Dissertation. University of Illinois at Urbana-
Champaign.
Oishi, S., Diener, E.F., Lucas, R.E., and Suh, E.M. (1999). Cross-cultural variations in pre-
dictors of life satisfaction: Perspectives from needs and values, Personality and Social
Psychology Bulletin 25(8): 980990.
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: 12591270.
Van de Vijver, F.J.R. and Poortinga, Y.H. (2002). Structural equivalence in multilevel research,
Journal of Cross-Cultural Psychology 33: 141156.
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:
6273.
World Values Study Group (1994). World values survey, 19811984 and 19901993. 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 liter-

ature on conict 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 vol-
ume of completed research in this area is sufciently 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 inuence 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 experi-
ential 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 inuence the other partys perceptions and/or actions
within the encounter (Barry 1999). Both perspectives raise challenging meth-
odological issues.

150 BRUCE BARRY AND INGRID SMITHEY FULMER
In the pages that follow, we rst briey 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 stud-
ies 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 inter-
action 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, exibility,
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 humor-
ous cartoons and small gifts (Carnevale & Isen 1986), and humorous video-
tapes (Kramer, Newton, and Pommerenke 1993), nding 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 condence 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. Nego-
tiators 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 difcult
test. In the negative mood condition, participants believed that they had per-
formed poorly on an easy test. A control group received no cues about test
difculty 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, nding that negotiators in pos-
itive moods elicited more cooperative behavior and post-negotiation compli-
ance 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 nego-
tiation opponents 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 opponents interests, diminished accuracy
of their judgments of the other partys interests, lower joint gains, and dimin-
ished 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 conict resolution behav-
iors. Rhoades, Arnold, and Jay (2001) had employed individuals complete
measures of mood and affective disposition, and document experiences of
workplace conict 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 conicts. Negative mood and negative affectivity were associ-
ated 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 intra-

negotiation emotional reactions experienced by negotiators. As with studies
described above exploring moods as predictors, the prevailing empirical par-

adigm is the use of laboratory negotiation simulations. OConnor 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-efcacy moderated this associ-
ation: more condent negotiators experienced less negative emotion following
impasse. In a study of e-mail-based negotiation, Moore, Kurtzberg, Thompson,
and Morris (1999) manipulated group afliation and self-disclosure. Although
this study was not oriented around emotion per se, ndings indicated that a
context favoring a personal relationship (negotiation with someone in ones
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 ones
own outcome increased satisfaction and decreased disappointment and re-
sentment; outcome fairness was, however, also positively related to guilt.
(Perceptions of the fairness of the other partys 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 sub-
stantially related to subsequent anger, particularly when individuals respond-
ing 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 benecial offers (1996: 220).
A few studies have examined negotiator affect in non-simulation contexts.
Scanzoni and Godwin (1990) investigated marital negotiations through retro-
spective recall of couple decision-making (a critical incident technique). After
recalling and reconstructing incident details, husbands and wives each sepa-
rately 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 liter-
ature, an exploratory study of crisis negotiation utilized a combined measure
of language intensity and message valence as a marker for negotiator affect
(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
Barrys (1999) investigation of the relative perceived ethical appropriateness
of emotionally deceptive negotiation tactics compared with other (cognitive)
deceptive tactics such as blufng or misrepresentation. Emotionally deceptive
tactics include the false expression or suppression of either positive or nega-
tive emotion. Although results suggest that people are substantially more
approving of and express more self-efcacy in their ability to deploy emo-
tionally deceptive tactics as compared to other tactics of informational decep-
tion, methodological challenges (to be discussed at length later) have limited
researchers ability to study these tactics in the laboratory or classroom.
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 stud-
ies 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 emo-
tion 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, conces-
sions 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 identies
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 con-
tagion (the transfer of emotion from one party to another) and emotional
tuning (the construction of messages designed to control or regulate the
other partys emotional responses.) Allred (1999) proposed a model of retal-
iatory conict describing how parties perceive each others harmful behavior,
and either retaliate or defuse tensions, depending on the nature of cognitive
appraisal and attribution.
Taken together, the models just identied 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
kinds of models. In the following sections, we discuss methodological con-

straints 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 pay-
ments) generate sufcient mental engagement among study participants to jus-
tify the generalizability of laboratory ndings conducted in the cognitive
domain. But when it comes to the study of emotion in laboratory-type set-
tings, . . . 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 nancial risks, are
often conducted with people with whom one shares a signicant ongoing per-
sonal relationship, and sometimes are related to major life events (divorce,
child custody, judicial proceedings, etc.), all of which heighten outcome signicance
and potentially result in intensied emotional responses. Furthermore, pro-
tracted real-world negotiations may introduce an element of time that could
either serve to dampen emotions or, conversely, facilitate the intensication of
emotions via the opportunity for rumination. And not least of all, contextual-
ized negotiations often involve the expectations and emotions of multiple
negotiators and constituencies simultaneously. Such scenarios are difcult 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 set-
tings 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 eld 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 emo-
tional, 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 ip side, although
most university human subjects committees frown on the use of coercion to
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 negoti-
ate 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 cir-
cumstances 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 expres-
sion (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 difcult but probably nec-
essary in order to study their effects on negotiation processes. As difcult 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 con-
text-rich negotiation is not only inevitable, but also crucial to the advance-
ment of knowledge about the role of emotion in negotiation. The discussion
that follows offers examples of several approaches that have been utilized suc-
cessfully 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 difcult to implement is to place a greater emphasis on the observational
study of live (not simulated) negotiations. As with all eld studies, gaining
access to observe such negotiations is half the battle. Some targets for obser-
vational 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
centers, judicial settings (e.g., plea negotiation conferences, small claims court
mediation), and other nonprot or governmental agencies where disputes are
frequently negotiated or mediated.
For the observational approach to bear fruit, one must accept that a re-
searcher can accurately detect and reliably classify the experience of emo-
tion by interactants. As we mentioned above, there is evidence that observers
can reliably assess visible emotional expression but expression and experi-
ence 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 en-
hanced 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 sam-
pling methodology (ESM) (also referred to as ecological momentary assess-
ment 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, nding that average mood contributed
to job satisfaction over and above job beliefs and dispositional tendencies
toward happiness. Ilies and Judge (2002) explored the longitudinal interrela-
tionships 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, per-
sonal digital assistants or palmtop computers are being used to collect ques-
tionnaire 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.
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 check-
lists that solicit responses to predened discrete emotional states. While con-
venient, the use of questionnaires potentially constrains the range of emotions
reported to those that t 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 inven-
tories (1998: 596).
ESM approaches in conjunction with more open-ended emotion diaries (Oatley
& Duncan 1992), although too time-consuming and intrusive for some nego-
tiation 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 Duncans 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 open-
ended questions. ESM utilizing such emotion diaries could prove especially
useful in conjunction with corroborating data (facial observation and/or phys-
iological 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 vari-
ous elds 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
which the traders were responding, and found, among other things, prelimi-
nary 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 inated and took blood pressure readings every 45 minutes during the
day for six days; when the cuff inated, respondents also responded to ques-
tionnaires (Diary of Ambulatory Behavioral States) on palm-top computers.
Although the effects of negative affect and emotional arousal on blood pres-
sure were generally modest, researchers noted substantial individual differ-
ences in blood pressure responsiveness to emotion.1
Realistically, the collection of physiological data is probably not feasible in
all eld studies of emotion, but as the data collection process becomes tech-
nically 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 distinguish-
ing 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 physio-
logically, which could prove especially valuable in detecting the use of sup-
pressive 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 expres-
sion, 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, profan-
ity statements, qualiers, etc.) and valence may reect negotiator affective
patterns and be related to crisis negotiation outcomes.
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 expec-
tations about the information that can be acquired and the variables that can
be examined (Becker 1996/2001). Emotion is arguably more socially and cul-
turally embedded, and more contextually inuenced than cognitive process-
ing and so may lend itself more to qualitative methods in order to obtain
sufciently rich or thick description (Geertz 1973) about how emotion man-
ifests itself in vivo. Ethnographic methods such as interviews and participant-
observation 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. Friedmans (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 inter-
action although again, one is mindful that the range of emotional experi-
ence in the lab may be constrained by the articiality 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 quan-
titatively oriented approaches such as self-report measures and/or physiologi-
cal 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 con-
currently, both quantitative methods and qualitative approaches (see Creswell
2003, for a review of mixed methods research strategies).
Conclusion
The position taken in this article can be distilled to six observations: (1) emo-
tion 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 ts into
negotiation are far out in front of empirical investigation in terms of sophis-
tication 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 elds 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 tech-
niques 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 con-
junction with diary studies.
References
Alliger, G.M., and Williams, K.J. (1993). Using signal-contingent experience sampling
methodology to study work in the eld: A discussion and illustration examining task per-
ceptions and mood, Personnel Psychology 46:525549.
Allred, K.G. (1999). Anger and retaliation: Toward an understanding of impassioned conict
in organizations. In R.J. Bies, R.J. Lewicki, and B.H. Sheppard, (Eds.) Research on nego-
tiation in organizations, Vol. 7. Stamford, CT: JAI Press.
Allred, K.G., Mallozzi, J.S., Matsui, F., and Raia, C.P. (1997). The inuence of anger and
compassion on negotiation performance, Organizational Behavior and Human Decision
Processes 70:175187.
Baron, R.A. (1990). Environmentally induced positive affect: Its impact on self-efcacy, task
performance, negotiation, and conict, Journal of Applied Social Psychology 20:368384.
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.
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 67:127143.
Bazerman, M.H., Curhan, J.R., Moore, D.A., and Valley, K. (2000). Negotiation, Annual
Review of Psychology 51:279314.
Becker, H.S. (2001). The epistemology of qualitative research. In R. Emerson, (Ed.) Contemporary
eld 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 54:579616.
Brett, J.M., Shapiro, D.L., and Lytle, A.L. (1998). Breaking the bonds of reciprocity in nego-
tiations, Academy of Management Journal 41(4):410424.
Brief, A.P., and Weiss, H.M. (2002). Organizational behavior: Affect in the workplace.
Annual Review of Psychology 53:279307.
Cacioppo, J.T., Berntson, G.G., Larsen, J.T., Poehlmann, K.M., and Ito, T.A. (2000). The psy-
chophysiology 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:
191214.
Carnevale, P.J.D., and Isen, A.M. (1986). The inuence of positive affect and visual access
on the discovery of integrative solutions in bilateral negotiation, Organizational Behavior
and Human Decision Processes 37:113.
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, 1:7378.
Creswell, J.W. (2003). Research design: Qualitative, quantitative and mixed methods appro-
aches. 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, 4:579599.
Feldman Barrett, L., and Barrett, D.J. (2001). An introduction to computerized experience
sampling in psychology, Social Science Computer Review 19(2):175185.
Forgas, J.P. (1992). Affect in social judgments and decisions: A multiprocess model,
Advances in Experimental Social Psychology 25:227275.
Forgas, J.P. (1998). On feeling good and getting your way: Mood effects on negotiator
cognition and behavior, Journal of Personality and Social Psychology 74:565577.
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 Inter-
national Association for Conict Management, Melbourne, Australia.
Hagemann, D., Waldstein, S.R., and Thayer, J.F. (2003). Central and autonomic nervous sys-
tem integration in emotion. Brain and Cognition 52:7987.
Hegtved, K.A., and Killian, C. (1999). Fairness and emotions: Reactions to the process and
outcomes of negotiations, Social Forces 78:269303.
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 eld experience sampling study, Organizational Behavior and
Human Decision Processes, 89:11191139.
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 conict, and emotional activation on ambulatory
cardiovascular activity: Daily life consequences of recurring stress in a multiethnic adult
sample, Health Psychology 17:1729.
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: 922.
Kramer, R.M., Newton, E., and Pommerenke, P.L. (1993). Self-enhancement biases and nego-
tiator judgment: Effects of self-esteem and mood, Organizational Behavior and Human Decision
Kring, A.M. and Gordon, A.H. (1998). Sex differences in emotion: Expression, experience,
and physiology, Journal of Personality and Social Psychology 74:686703.
Kubey, R., Larson, R., and Csikszentmihalyi, M. (1996). Experience sampling method appli-
cations to communication research questions, Journal of Communication 46:99120.
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 58:164171.
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 strati-
cation in private law rms, Work and Occupations 27(1):3263.
Lo, A.W., and Repin, D.V. (2002). The psychophysiology of real-time nancial risk process-
ing, Journal of Cognitive Neuroscience 14(3):323339.
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 afliations and good vibra-
tions, Organizational Behavior and Human Decision Processes 77:2243.
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.
OConnor, K.M., and Arnold, J.A. (2001). Distributive spirals: Negotiation impasses and the
moderating role of disputant self-efcacy, Organizational Behavior and Human Decision
Park, O.S., Sims, H.P., and Motowidlo, S.J. (1986). Affect in organizations: How feelings and
emotions inuence 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 nat-
ural language use: Our words, our selves, Annual Review of Psychology 54:547577.
Pillutla, M.M., and Murnighan, J.K. (1996). Unfairness, anger, and spite: Emotional rejec-
tions of ultimatum offers. Organizational Behavior and Human Decision Processes 68:208224.
Rafaeli, A., and Sutton, R.J. (1991). Emotional inuence strategies as means of social
inuence: Lessons from criminal interrogators and bill collectors, Academy of Management
Journal 34: 749775.
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 conict, Journal of
Organizational Behavior 22:329345.
Rogan, R.G., and Hammer, M.R. (1995). Assessing message affect in crisis negotiations: An
exploratory study, Human Communication Research 21(4):553574.
Russell, J.A., Bachorowski, J., and Fernandez-Dols, J. (2003). Facial and vocal expressions of
emotion, Annual Review of Psychology 54:32949.
Scanzoni, J., and Godwin, D.D. (1990). Negotiation effectiveness and acceptable outcomes,
Social Psychology Quarterly 53:239251.
Thoits, P.A. (1996). Managing the emotions of others, Symbolic Interaction 19(2):85109.
Thompson, L.L., Nadler, J., and Kim, P.H. (1999). Some like it hot: The case for the emo-
tional negotiator. In L.L. Thompson, J.M. Levine, and D.M. Messick, (Eds.) Shared cogni-
tion in organizations: The management of knowledge. Mahwah: Erlbaum.
Watson, D., Clark, L.A., and Tellegen, A. (1988). Development and validation of brief mea-
sures of positive and negative affect: The PANAS scales. Journal of Personality and Social
Psychology 54: 10631070.
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 78:504517.
Weingart, L.R., Hyder, E.B., and Prietula, M.J. (1996). Knowledge matters: The effect of tac-
tical descriptions on negotiation behavior and outcome, Journal of Personality and Social
Psychology 70:12051217.
Weingart, L.R., Thompson, L.L., Bazerman, M.H., and Carroll, J.S. (1990). Tactical behavior
and negotiation outcomes, International Journal of Conict Management 1:731.
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 experi-
ences over time, Organizational Behavior and Human Decision Processes 78(1):124.
Comparative Case Studies
I. WILLIAM ZARTMAN
Introduction
Case studies are one of, if not the, most frequently used methods for con-
ducting research on negotiation. They can vary from purely historical studies
that seek to establish all the relevant facts of the encounter to analytical stud-
ies chosen to illustrate specic theoretical propositions. The most successful
cases for the purposes of generating useful knowledge are those located some-
where in the middle of that spectrum, leaning toward the latter side of cen-
ter. Case is used here to refer to the story of negotiations on a single
conict 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 num-
ber 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 understand-
ing 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 conrmatory, 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.
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 innite, unique data; leaves need to be aggre-
gated 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 aggre-
gated 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
verication to see whether they hold together and make sense, and an empir-
ical verication 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 rst.
Building theory depends on prior identication of its components and pur-
poses, 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 regu-
larities 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 rst 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 pro-
duced important conceptual development and cumulative knowledge. Old
effects since the process of negotiation itself is millennia old have been
identied 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 exibility, prenegotiation, BATNA
and security points, integration and distribution, toughness dilemma, value
making and value taking, among many others that have made the under-
standing 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 gments 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 negotia-
tions 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 verication but work merely to establish
a good record and understanding of the event itself, nding an explanation of
outcomes in the events (and often personalities) of the case itself. Fine exam-
ples 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 rst three names cited) or as a diligent researcher, and has devel-
oped 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 con-
text and meaning and achieve a Weberian understanding of its dynamics
(Skocpol and Somers 1980/1994).
If the theory ts 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 veried; 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 out-
come from the available concepts and theories. In single case studies, the
choice is generally directed by considerations external to the concept; the pur-
pose is to nd 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 estab-
lished 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 Rubins
(1981) collection of interpretations of the Kissinger shuttle negotiations in the
Arab-Israeli conict and Pruitts (1997) collection of analyses of the Oslo
negotiations on the Palestine-Israeli conict; 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 ction
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
the better (knowing, as noted, that the generality can never be proved or
expected to be universal). Zartman (1994) and Hampson (1995) are the rst
works to attempt a theoretical analysis of multilateral negotiations, each using
cases somewhat differently; the rst subjects two cases (the Single Europe Act
and the GATT Uruguay Round) to a competition among six different theoret-
ical approaches (game, decision, small group, leadership, coalition, and orga-
nization theories) to try to draw out a common analytical perspective, whereas
the second examines nine cases from which to draw conceptual characteris-
tics 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 argu-
ment 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 applic-
ability 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 identied, a con-
dition 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 signicantly
increases the difculty 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 suc-
cess in the same country in a comparative analysis that uses specic concepts
to provide an explanation of outcomes, as was well and explicitly done in
Touval (1982) examining nine attempts to mediate the Arab-Israeli conict,
and in Stedman (1991) comparing three failures and one success in mediat-
ing the Zimbabwean anti-colonial war. Interestingly but only coincidentally,
both works focus on ripeness as a major explanatory variable, making impor-
tant contributions to an understanding of the concept. Another comparative
single-case analysis can be provided by examining the role of various com-
ponents of a single set of negotiations, as is done in the rst study of
European integration as negotiation (Meerts & Cede 2004), which analyzes
negotiations in each of the major institutions of the European Union. If mul-
tiple instances or segments within a single case are not available or chosen,
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 difcult 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 big-
ger 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 nego-
tiating African civil wars and Zartman (1995, 11 cases in 353 pages) on the
difculties of negotiating centralist and regionalist civil conict. Yet, the abil-
ity 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 scientically,
but by the criteria of saliency and interest already mentioned. Most cases
were signicant 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 conclu-
sion of the current phase of the conict, only to be followed by a new phase.
The cases further divided into conicts for control of the central government
and conicts for control of a region. A conceptual framework, built on con-
trasting notions of stalemate and on characteristics of internal wars asym-
metry, phases, agendas, escalation and mediator tactics, was set up at the
beginning and rened inductively in interaction with the case authors, pro-
viding 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
a nal 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 pro-
posed, 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 t between the gen-
eralizations 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 com-
parative 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 worlds most dangerous substance on the process
of negotiation (Avenhaus, Kremenyuk & Sjstedt 2002), economic issues in
order to compare the strength of economic vs. negotiation explanations of out-
comes (Sjstedt 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 (Sjstedt 1993), and a broad range of historic and contemporary encoun-
ters in order to answer some major conundrums about resolution vs. transfor-
mation 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, specically; 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
didnt 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
in aggregate data, large collections of precisely identied 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, rm
generalizations and regularities can be established, both for knowledge and for
appropriate policymaking. Usually, there is no unanimity of support for a the-
ory or concept, but merely statistically signicant correlations among a num-
ber 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 signicance tests.
This research uses aggregate data on the largest number of cases possible
either to test deductive propositions or generate inductive ndings through
correlation or factoring. Despite careful coding, it needs to group large num-
bers of diverse cases together into types, and is more interested in showing
statistically signicant correlations than in nding 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 con-
cern, the question of durability of negotiated agreements. Hampson (1999,
ve 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 signicant method-
ological 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 pre-
liminary 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, power-
division, elections) was the signicant procedural provision when more than
one occurred? When was a ceasere a ceasere? When was an instance of
conict 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 inexibly quantitative denitions.
Second, not only do the statistical data turn difcult 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-
tion when the evidence is not given, and an assurance of intercoder reliabil-
ity 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, quantiable data, and
so, because it has no feel for its subject, it has to rely on indicators or
proxies; subjective elements must be objectied 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 Hoefer 2002; Collier et al. 2003: 66).
Fourth, what cannot be measured or proxied is not analyzed. Since it is
difcult 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 identied as crucial elements in
negotiation.
Fifth, as a result of the above, data become no longer data but are them-
selves events squeezed into generalizations. They become shorthand for the
event and so enter into a tangling tautology: they become theoretical gener-
alizations required to test theoretical generalizations. If household incomes are
used to proxy grievance in order to test its role in causing conict, 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 analy-
sis. 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.
Conclusion
This is a rich list of problems; an equally full list can be made for case stud-
ies 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
satised to understand one case well and produce some lessons for someone
else to test on other cases, rather than nding 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 conict or through several negotiations of several problems or
conicts. 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 con-
cepts 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 renement 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 nego-
tiation process made available over the past four decades involves case stud-
ies 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 com-
parative case studies.
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 signicant 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., & Sjstedt, 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 & Hoefer, Anke (2002). Greed and Grievance in Civil War. Working Paper
200201. Oxford University, Centre for the Study of African Economies.
Collier, John, et al. (2004). Breaking the Conict 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: Ceasere Agreements and the Durability of Peace. Princeton,
NJ: Princeton University Press.
Genesis 18: 2032.
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 Decision-
Making, Advances in Information Processing 2: 2158.
Hampson, F.O. (1999). Nurturing Peace. Washington: United States Institute of Peace Press.
Jentleson, B., editor (2000). Opportunities Missed, Opportunities Seized. Lanham, MD: Row-
man and Littleeld.
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: Uni-
versity of Toronto Press.
Nicolson, H. (1946). The Congress of Vienna. New York: Harcourt Brace.

Parker, R. (1993). The Politics of Miscalculation in the Middle East. Bloomeld, 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: 2742.
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: 25979.
Skocpol, Theda and Somers, Margaret (1980/1994). The Uses of Comparative History in
Macrosocial Analysis. Comparative Studies in Society and History XXII, 2: 17497
(April), reprinted in Skocpol, Theda, editor, Social Revolutions in the Modern World.
Cambridge, UK: Cambridge University Press, 1994.
Sjstedt, G., editor (1993). International Environmental Negotiations. Thousand Oaks, CA:
Sage.
Sjstedt, 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 Implemen-
tation 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: Prince-
ton 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 Conict 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 & Littleeld.
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 ngers, 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 mes-
sages 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 (Wein-
gart, Olekalns and Smith 2004). In both instances, scholars examine language,
messages, talk, or written texts that emanate from a negotiation. Clearly, dis-
course 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 dis-
course 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 dis-
tinguish 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, difculties, and pitfalls of this approach.

178 LINDA L. PUTNAM
What Is Discourse Analysis and How Is It Used?
Discourse analysis is an extensive arena of research that crosses linguistic, lit-

erary, 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, partic-
ularly 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 reafrm a differential status between union and manage-
ment (ODonnell 1990), to initiate new concessions through statements of
reection (Walker 1995), and to shape offers and counteroffers through cul-
tural 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 negotia-
tion 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 negotia-
tion 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 justi-
cation to produce an account for ones 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 pro-
nouns, 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 perpetra-
tors 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 per-
suasion 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 defen-
sive reactions similar to bargaining strategies and tactics (Putnam and Jones
1992; Roloff, Tutzauer and Dailey 1989), as legitimating a bargainers claims
(Schuetz 1978), as ways of redening 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
gures of speech identify language patterns that have multiple and often con-
tradictory meanings within the larger bargaining context.
The nal type of rhetorical analysis stems from narrative or dramaturgical
texts. In narrative analyses, researchers focus on the way talk constructs sto-
ries, 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 nego-
tiation. The most informative analyses treat negotiation as more than simply
another type of conversation, but rather as a speech event in which the set-
ting, participants, norms, and history of the process inuence 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 ana-
lysts let the themes or patterns grow out of the transcripts as part of the
meanings that are co-developed during the negotiation. Researchers use lan-
guage 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 negotia-
tions, moving from standing for sections of the contract itself to signaling a
bargaining formula for packaging items, and nally 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 ts 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 negotia-
tion practices possible, ones such as legitimating positions and identities,
managing role conicts 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 Themes and Patterns Inferences from

Exemplars from Text Themes and Patterns
Phase 1 Language as synecdoche The use of the word, lan-

We need to gure out functions as the whole guage, serves as an orga-
the language here. Lets to refer to all the policy nizing device to cluster
set forth some language issues in the contract. issues in the contract.
on this. Money as synecdoche is a The use of the terms,
Lets talk about money. short hand word to refer money, cost, and funds
What is the cost of this to money items in the indicts overall cost of the
package and what are the contract. contract.
sources of these funds
Phase 2 Both sides begin to own, The terms, language and
Thats our language on claim, and exchange lan- money, shift in meaning
dues deduction. Hes guage through the use of to become commodities to
added our language. pronouns such as our be traded as a concession
We can give them some and theirs. strategy. Money refers to
language. Language and money dollars in the teachers
Table 1 (cont.)
Discourse Units and Themes and Patterns Inferences from

Exemplars from Text Themes and Patterns
Lets give on some become used with verbs pockets versus costs to
minor items and hold out to exchange, to give and the Board.
on our money. to claim.
Phase 3 Money becomes the coun- The terms form a duality
Teachers: The Board will terpart of language. The such that gains in one
not give both a good two possessions can be represents losses in the
money settlement and lan- pitted against each other other. Negotiators manage
guage items in a win-lose struggle. this dichotomy of
Board: This package To keep policy items out language versus money
does not contain your of the contract, the Board by separating the two are-
standard throwaway lan- can trade money to buy nas and shifting back and
guage. It will take serious out language items. forth between them.
effort to get them to let
go of this language. I
assume that means
money, not your
autographed picture
Overall Pattern It becomes a bargaining The language versus
The tacit norm for getting formula that both enables money duality gives the
a settlement in District 1 and constrains the process Board control over policy
was a dichotomy: trading and the two sides. issues in the contract and
language for money. 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 natu-
rally-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 researchers 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; ODonnell 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 researchers interests and philosophical posi-
tion guide the choice of texts and help a scholar work through the conundrum
of how broad or narrow, how inclusive or exclusive, and how ne 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 sufcient 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 sufce, but to determine how cultural dif-
ferences and preferences for commodity trades guide these patterns, re-
searchers might need interview data and copies of trade policies that impinge
on the interactions (Grindsted 1989). Hence, beginning with an explicit re-
search question provides a frame for the study and aids in making decisions
about data collection and analysis.
Another factor that inuences 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 nego-
tiation 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 ne-
grained 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-
estimate the importance of an interesting question. Interesting questions are

often determined by what a research project challenges rather than by what it
conrms 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 assump-
tive ground that threats are typically linked to conict 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 t 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 dene and redene 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 naturalis-
tic 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 reexive in that discourse units arise from the
talk itself at the same time the negotiation is being constructed. These units
draw from and reect back on interactions that inuence the bargaining in
184 LINDA L. PUTNAM
particular ways. For instance, an analysis of words and phrases that signies
drawing closer together or distancing from each other (e.g., we versus
they, that versus this) shows how hostage negotiators protect the per-
petrators 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 perpetrators use of these units and the hostage
negotiators response reveals the patterns or themes linked to distancing or
pulling together. The overall interpretations occur reexively 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 com-
puter programs such as ATLAS and NVIVO. These programs consolidate dif-
ferent 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 dis-
course 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 neg-
ative remarks about an accountant who handles nances 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 vil-
lain or as someone who makes bargaining difcult for all of them. Moreover,
when they mention this persons name at the table, it serves as a bonding
experience that leads them to sequential concessions and agreements on
difcult 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 veried 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 ndings 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
fashion. Both parties were aware of this bargaining formula, lamented the
way it constrained them, but felt it was essential for an effective settlement.
Reecting 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 benets. Both sides wanted to preserve their ways of con-
trolling 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 evi-
dence supports the claims, the plausibility of the ndings, and the insights gleaned
from the analytic approach (Phillips and Hardy 2002). A studys overall
coherence and the way it rings true to the reader are also important crite-
ria for evaluating this research. For example, Friedmans (1995) observations
that negotiations are both constrained and enabled by the networks that bar-
gaining 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, dis-
course 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 formu-
las 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 indi-
cated why they made it difcult 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 behav-
iors in negotiation. These connections stem from the way that discourse
enacts the process while it refers back to the larger context in which the bargain-
ing occurs. Thus, a third strength of discourse analysis is its reexivity or
using language to reect back on the bargaining context, how a study was
conducted, and how processes evolve over time (Holland 1999). Reexivity
also addresses how discourse reveals instrumental, relational, and identity lev-
els of negotiation simultaneously. In this way discourse analyses demonstrate
how bargaining about issues and positions also includes negotiating ones
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 dened 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 dis-
putants label a conict. 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 reexivity, and discovering new concepts to enrich
bargaining theory.
Discourse analysis also has its drawbacks and may be difcult to implement
in a number of settings. Researchers need texts, either transcripts of negotia-
tion interactions, documents, informal exchanges, interviews, eld notes, pro-
posals, or some form of verbal/written material to analyze. Its difculty 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
study on the way that negotiators positioned themselves in the bargaining talk.
Relying on detailed methods of conversational analysis, we examined posi-
tioning 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 sys-
tematic 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 negoti-
ated outcomes. Through a systematic tracing of different paths, the investiga-
tors examined links between these negotiation shifts and conict 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 difcult aspects of discourse analysis lies in drawing infer-
ences from the data. Researchers need to avoid inferences that are too gen-
eral 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, benets, 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 inuenced
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 inter-
preted throughout the state to make major claims about its effects.
A nal 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 se-
quence 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 rst 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 difcult 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 Friedmans (1992, 1994) investiga-
tion of the International Harvester negotiation, the idea of a dramaturgical
lens came from observing negotiation not only from the front stage perfor-
mance but also from the back-stage networking and interaction rituals that
inuenced front-stage scripts. Drama, as a metaphor of the process, arose
through asking questions about traditional scripts and the way negotiators var-
ied 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 com-
munication 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
working back and forth from the case to concepts in the literature, we broad-
ened our focus to cover a complex communication system that merged well
with literature on back-channel negotiations. In like manner, Cobbs (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 privi-
lege others. Similar to the rst 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 solv-
ing. Inconsistencies are contradictions that appear in the text and call for the
researcher to question the data. For example, illustrating how a particular pat-
tern of argument in negotiation both conceals and reveals information simul-
taneously leads the researcher to probe different questions about information
exchange (Putnam 1997). Friedmans (1992) discovery that intergroup bargaining
promotes exible 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 nal guideline for conducting discourse analysis is to dispute your own
interpretations. Qualitative data analysis involves discovering and assem-
bling pieces of a puzzle and drawing inferences to form a broad image. By
disputing interpretations of ndings, 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 inter-
pretations 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 activ-
ity, 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 litera-
ture 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 Dimen-
sions of International Negotiations. Negotiation Journal, 4, 3: 257266.
Cobb, Sara and Rifkin, Janet (1991). Practice and Paradox: Deconstructing Neutrality in Mediation.
Law & Social Inquiry, 16: 3562.
Donohue, William A. and Diez, Mary (1985). Directive Use in Negotiation Interaction.
Communication Monographs, 52: 305318.
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 Conict
Management, 2, 4: 257274.
Donohue, William A. and Roberto Anthony (1994). Relational Development as Negotiated in
Hostage Negotiation. Human Communication Research, 20, 2: 175198.
Druckman, Daniel (2001). Turning Points in International Negotiations: A Comparative
Analysis. Journal of Conict Resolution, 45: 519544.
Druckman, Daniel, Husbands, Jo L. and Johnston, Karin (1991). Turning Points in the INF
Negotiations. Negotiation Journal, 7, 1: 5567.
Fant, Lars M. (1989). Cultural Mismatch in Conversation: Spanish and Scandinavian
Communicative Behavior in Negotiation Settings. Hermes, Journal of Linguistics, 3:
247265.
Firth, Alan (1994). Accounts in Negotiation Discourse: A Single-Case Analysis. Journal of
Pragmatics, 23: 199226.
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:
5379.
Friedman, Raymond A. (1989). Interaction Norms as Carriers of Organizational Culture:

A Study of Labor Negotiations at International Harvester. Journal of Contemporary Ethnography,
18: 329.
Friedman, Raymond A. (1992). The Culture of Mediation: Private Understandings in the
Context of Public Conict, in Deborah M. Kolb and Jean M. Bartunek, editors, Hidden
Conict: 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: 267279.
Hamilton, Peter M. (1997). Rhetorical Discourse of Local Pay. Organization 4, 2: 229254.
Hamilton, Peter M. (2000). Attaining Agreement: A Rhetorical Analysis of an NHS Nego-
tiation. The International Journal of Public Sector Management, 13: 285300.
Holland, R. (1999). Reexivity. Human Relations, 52, 4: 463485.
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: 127146.
ODonnell, Katherine (1990). Difference and Dominance: How Labor and Management Talk
Conict, in A.D. Grimshaw, editor, Conict 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: 3553.
Putnam, Linda L. (1997). Productive Conict: Negotiation as Implicit Coordination, in
Carsten De Dreu and Evert Van De Vliert, editors, Using Conict in Organizations. London:
Sage.
Putnam, Linda L. and Carcasson, Martin (1997). Communication and the Oslo Negotiation:
Contacts, Patterns, and Modes. International Negotiation, 2, 2: 251278.
Putnam, Linda L. and Geist, Patricia (1985). Argument in Bargaining: An Analysis of the
Reasoning Process. The Southern Speech Communication Journal, 50: 225245.
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: 171191.
Putnam, Linda L. and Roloff, Michael. E. (1992). Communication Perspectives on Nego-
tiation, 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:
85103.
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 Conict:
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: 129152.
Rogan, Randall G. and Hammer, Mitch R. (1994). Crisis Negotiations: A Preliminary Investiga-
tion of Facework in Naturalistic Conict Discourse. Journal of Applied Communication Research
22: 216231.
Rogan, Randall G. and Hammer, Mitch R. (1995). Assessing Message Affect in Crisis Negotiations:
An Exploratory Study. Human Communication Research 21: 553574.
Roloff, Michael E., Tutzauer, Frank E. and Dailey, William O. (1989). The Role of Argu-
mentation in Distributive and Integrative Bargaining Contexts: Seeking Relative Advantage
But at What Cost? in M. Afzalur Rahim, editor, Managing Conict: An Interdisciplinary
Approach. New York: Praeger.
Schuetz, Jan (1978). Argumentative Competence and the Negotiation of Henry Kissinger. Journal
of the American Forensic Association 15: 116.
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: 441455.
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 Conict
DEAN G. PRUITT
Advancements in the study of social conict, as in other sectors of science,

depend on gathering clearly interpretable data. Field experiments are an
important means to this goal.
In eld 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 eld experiments, the investigator manipu-
lates the variables by randomly assigning participants (individuals, hospitals,
mediation cases, etc.) to the levels of these variables. True eld experiments
will be the topic of this article and will be referred to simply as eld exper-
iments. Good discussions of this methodology can be found in Boruch
(1997), Cook and Campbell (1979), and Mosteller and Boruch (2002).
An example of a eld experiment involving social conict 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 ran-
domly assigned to three conditions: med-arb (same), in which the same per-
son 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: lab-
oratory experiments and observational eld studies. Field experiments and
laboratory experiments are alike in that the investigator manipulates at least
one variable, but laboratory experiments take place in articial settings con-
structed by the investigator rather than in naturally occurring settings. For

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 hospitaliza-
tion benets 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 rst 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
rst condition than the latter two.
In observational eld 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-nding, mediation followed by fact-
nding 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-nding law. This nding 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 under-
stand the arguments for and against eld experiments, we should rst exam-
ine the pros and cons of conducting experiments (in either the laboratory or
the eld) versus those of observational studies. Subsequently, we will consider
the respective pros and cons for doing eld 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 vari-
able) inuences (has a causal impact on) Y; (2) Y inuences X; (3) some third
common factor, Z, inuences both X and Y, producing a spurious relation-
ship between them; (4) the relationship between X and Y is due to chance.
Statistical tests of signicance allow us to rule out explanation 4 at an accept-
able level of condence. 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 expla-
nations 2 and 3, arguing that the investigator is the source of variation in X
and hence neither Y nor Z can have inuenced that variable. This reasoning
leaves explanation 1 as the only plausible account, that the relationship was
produced by X (or an associated variable) inuencing 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 rst three
explanations would be plausible. The nature of the procedure could plausibly
be inuencing the ease of reaching agreement (explanation 1). But the typi-
cal ease of reaching agreement in a state could also plausibly have inuenced
what procedure was adopted (explanation 2). Furthermore, many third com-
mon factors, such as the availability of funds to pay public employees, could
plausibly be inuencing 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 rm conclusions about
cause and effect. Explanation 2, (that Y inuenced 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 environ-
mentally 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 obser-
vational studies. Still their name is a misnomer; they are not really experi-
ments 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 inuencing 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-nding 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 ndings.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 poten-
tial 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 media-
tor was powerful, anxiety about the relationship with the mediator, belief that
a denitive 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 partici-
pants in each condition, making it hard to interpret the data. Similar problems
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 dening feature of true experi-
ments 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 nding 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 chan-
neled easier, less conicted cases into mediation and tougher, more hostile
cases into adjudication, which could account for these ndings.
In observational studies such as this, investigators often try to rule out
selection problems logically or by statistical control after the data are gath-
ered. Thus McEwen and Maiman present two arguments against the notion
that their ndings were due to selection effects. They note that perception of
having freely chosen mediation was unrelated to whether the defendant sub-
sequently 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 condi-
tion 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 eld experiment would have provided more solid support.
Advantages of Observational Studies over Experiments
It is straightforward that experiments are not useful for studying all phenom-
ena. 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 lim-
ited 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 stud-
ies, causal sequences can be followed in still other ways, for example, by ask-
ing 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 reduc-
ing variation in the measured variables. This allows us to reach statistical
signicance if our variables are related to each other, but unnaturally magnies
the strength of those relationships.
Laboratory vs. Field Experiments
Both laboratory and eld settings are good places to observe new phenomena
and generate new hypotheses the laboratory because it isolates and mag-
nies relationships that might not be noticed in the eld, and the eld because
phenomena occur there that are unlikely to be programmed into the labora-
tory until they have been noticed. Thus, a mix of laboratory and eld 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 eld?
Advantages of Doing Experiments in the Laboratory Rather than the Field
Laboratory experiments are far more common than eld experiments in

research on social conict 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 eld
experiments in social service agencies.
Control. Laboratory settings usually allow greater control over the elements
of research than do eld settings. This makes it easier to conduct good exper-
iments. For example, randomization is usually more possible and more secure
in laboratory settings. In many eld 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 randomiza-
tion 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 eas-
ier to reach statistical signicance 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 eld
experiments, an issue we will return to toward the end of the article.
Range of manipulations and measures. Another reason for conducting labora-

tory experiments is that they usually allow a wider range of manipulations
and measures than do eld experiments. Investigators are king in the labora-
tory and, within ethical limits, can do almost anything they want. More con-
straints exist in the eld.
Constraints are especially severe when eld 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 con-
dition, in which they were only asked the small favor; and an exposure con-
dition, 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 con-
cession condition, but only 17% agreed in the no-concession condition and
only 25% in the exposure condition.6 It was not possible to interview partic-
ipants about why they made the decisions they did because this would have
blown the investigators cover.
Field experiments in social service agencies. Another type of eld experi-

ment assigns cases to different treatments in a social service agency. Parti-
cipants 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 sup-
porters as part of the agencys 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 agencys program. Two of the conditions were
part of the mediation centers 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 agencys program too much out of shape. We were also con-
strained 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 difcult to do than laboratory
experiments because they must be sold to busy agency ofcials, data gather-
ers must travel extensively to the agency headquarters, and operational prob-
lems are more likely to emerge.
Advantages of Doing Experiments in the Field Rather than the Laboratory
If laboratory experiments provide more control and exibility at lower cost,

why should anyone do a eld experiment? One answer to this question is that
some conditions and variables can only be realistically produced in the eld.
Thus, one must move to the eld in order to study whether arrest or coun-
seling is more effective at stopping men from abusing their wives. Sherman
and Berk (1984) did such an experiment and found only half as much recidi-
vism 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 eld because laboratory
effects are usually too weak to persist very long. The Emery study is a good
example of the strength of eld 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 childrens lives,
maintained more contact with their children, and had a greater inuence 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 nd-
ings can be generalized to real-life settings that involve the same variables.
Laboratory settings differ from eld 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 ndings.
Motivational and emotional impact. Disputants usually experience stronger

passions such as frustration, anger, and the desire for revenge in the eld
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 eld. It also means that effects that are
defeated by strong passions such as awareness of ones surroundings and
202 DEAN G. PRUITT
concern about the impression one is making on others will usually dimin-
ish as we move from the laboratory to the eld.
Our med-arb study (McGillicuddy et al., 1987) is an example of an effect
that was strengthened by moving into the eld. Before doing the eld exper-
iment, 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 exper-
iment yielded only weak trends in the direction later taken by the results of
our eld 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 eld. Hence, the participants in the med-arb condition were probably
less anxious about the prospect of having the mediator take over their deci-
sion making, and hence less motivated to reach agreement.
A case in which a robust laboratory nding 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 media-
tion 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 eld
setting and ran the same experiment in small claims court, this effect disap-
peared. This effect may have been partly drowned out by the greater error
variance that inevitable occurs in the eld. But it was probably also a victim
of the strong passions that were quite evident in the eld setting, where peo-
ple were dealing with frustrating and ego-involving issues from their own
lives. Heavy, emotion-laden involvement tends to block perception of back-
ground features of the environment. Hence, the conict and the adversary
probably stood out starkly while the rest of the scene including the pres-
ence or absence of a table probably receded into oblivion.
If I am right about the source of the difference between Druckman and
Arais two studies, it implies that their laboratory experiment was not a fail-
ure. The results of that study suggest that the absence of a table facilitates
agreement in low key, non-emotional conicts a subclass of all conicts.
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
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 real-
world bargaining long-term working relationships, parallel secret negotia-
tions, 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 eld,
and the laboratory ndings may lack external validity.
Other possible sources of articiality derive from the presence of an ex-
perimenter and the impact of such an authority gure 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 conict (such as that
between Israel and the Palestinians) and have them discuss or react to infor-
mation about that conict (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 experimenters control. A whole industry of
research on aggression has grown up around this paradigm (Berkowitz 1993;
Geen 2001), and many of the ndings 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
conict 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 improv-
ing external validity.
Problems Specic to Field Experiments and Some Suggested Solutions
Mortality
Mortality is a common problem in eld 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 eld 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 con-
dition, 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 med-
arb study. Ofcers 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 arbi-
trator always imposed a settlement.) They complained to the mediation cen-
ter employee who was doing the random assignment, and she changed her
procedure without notifying us. As each new case came into her ofce, 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 ofce 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 signicantly more participants have dropped out of one condition
than another, a sure sign of differential mortality. Mortality was a big prob-
lem 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 condi-
tions. Fortunately, the dropout rate was not signicantly greater in one condi-
tion than another. This was a comforting nding but it by no means ruled out
differential mortality.
If there is substantial mortality, it is also important to check whether dif-
ferent 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 ones ndings. 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 conict-
laden relationships were less likely to drop out of the mediation sample than
the litigation sample. However, this did not endanger their conclusions
because conict 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 mor-
tality, but they are seldom denitive. Even if ones ndings are exonerated by
all the variables one has measured, it is still possible that differential mortal-
ity is a problem on some other unmeasured variable. Hence, it is always bet-
ter 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 eld, require permission
from a human participants review board. But many eld 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 agencys 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 agencys freedom of action with its clients
and can provide an administrative burden to the agency. Because of this
difculty, 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).
Condentiality
Disputants often want to keep the details of their conicts secret for fear
of embarrassment, benet to an adversary, or legal liability. Hence, eld
experiments on social conict require particular attention to condentiality.
Concerns about condentiality 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
reect 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 difcult 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 tanta-
mount 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 eld experiments should be
employed more often then they are in the study of social conict. They are
more useful than observational studies for assessing the impact of novel con-
ditions, 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, eld experiments clearly have their limitations. Some variables
cannot be practically or ethically manipulated and require the use of observa-
tional methods. Observational methods are also more useful for looking at the
relationships among a large number of variables and for estimating the

strengths of relationships between variables. Furthermore, laboratory experi-
ments allow more control of conditions and greater exibility in designing manip-
ulations, and they are usually easier to implement than eld 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 address-
ing a given problem. For example, the conciliatory effect of the expectation
of arbitration was rst 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 eld experiment. As a result, we can be more condent about the
external validity of this effect, and more certain about the mechanisms under-
lying 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 eld studies.
3. The reader will recall that in the just prior section we said that the most plausible interpre-
tation of the Kochan and Jick nding 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 denitively 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 eld
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, 1:1941.
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, 31:206213.
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: 150178. 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, 69:323332.
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, 59:410418.
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: 1549. 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, 56:110.
Kochan, T.A. and Jick, T. (1978). The public sector mediation process: A theory and
empirical examination. Journal of Conict Resolution, 22:209240.
McEwen, C.A. and Maiman, R.J. (1984). Mediation in small claims court: Achieving
compliance through consent. Law and Society Review, 18:1149.
McGillicuddy, N.B., Welton, G.L. and Pruitt, D.G. (1987). Third party intervention: A eld
experiment comparing three different models. Journal of Personality and Social Psychology,
53:104112.
Mikolic, J.M., Parker, J.C. and Pruitt, D.G. (1997). Escalation in response to persistent annoy-
ance: Groups vs. individuals and gender effects, Journal of Personality and Social
Psychology, 72:151163.
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.
Zartman and G.O. Faure, editors, Escalation and Negotiation. Cambridge, England: Cam-
bridge 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, 17:313330.
Robinson, R.J., Keltner, D., Ward, A. and Ross, L. (1995). Actual versus assumed differences
in construal: Nave realism in intergroup perception and conict, Journal of Personality
and Social Psychology, 68:404414.
Sherman, L.W. and Berk, R.A. (1984). The specic deterrent effects of arrest for domestic
assault. American Sociological Review, 49:261272.
Vallone, R.P., Ross, L. and Lepper, M.R. (1985). The hostile media phenomenon: Biased per-
ception and perceptions of media bias in coverage of the Beirut massacre. Journal of Personality
and Social Psychology, 49:577585.
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 behav-
iors. Journal of Conict Resolution, 36:546572.
Laboratory Experiments on Negotiation and
Social Conict
This article addresses the laboratory experiment from the perspective of the
experimental social psychologist interested in making statements about social
conict and negotiation. The main concern is the validity of these statements
whether the conclusions or inferences about the data are justied (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 studys 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!).
OConnor and Carnevale (1997) wanted to demonstrate that such an issue of
common interest would be exploited by people to gain advantage in an agree-
ment (in one situation, such exploitation occurred in 53% of the negotia-
tions!). Both studies were conducted by employing naive university students
in conducting an articial task in an articial setting. One wonders if these
patterns would occur in other settings, with other kinds of people, other peri-
ods 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 psychologists laboratory, but they occur in the real
world as well (Mook 1983).

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 vari-
ables in a laboratory setting an articial 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 equal-
ity of conditions prior to any treatment. Often the manipulated variables (the
independent variables) are situations or conditions that surround the negotia-
tion (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 psycho-
logical states cognitions, motivations, and emotions. Sometimes researchers
assume that the connection between situation and outcome variables is medi-
ated 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, percep-
tual, 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 vari-
ance (e.g., make treatment conditions as distinct as possible), minimize error
variance (e.g., have accurate assessments of processes), and control extrane-
ous 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 clas-
sic. 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) pro-
vides great wisdom on the many choices facing the researcher. Aronson and
Carlsmiths (1968) review of experimental method inspired many young social
psychologists when they noted, in the last line of the paper, that the enter-
prise is fun.
Many social psychology experiments involving negotiation and social
conict arise from observations gleaned in natural settings. Triplets (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 Johnsons (1970) study of the impact mediators suggestions
have on negotiator concession-making built off Stevens (1963) real-world
mediation observations.
Laboratory Tasks for Negotiation and Social Conict
A central element of any laboratory study of negotiation and social conict is

the task. It represents the reward structure, the incentives, the set of alterna-
tives 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 deci-
sion 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, nding that about three-fourths maintained a par-
ticipant 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 simplied versions of reality, yet retain key elements of the struc-
ture or processes of negotiation.
Economics has inuenced 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
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 cards shown value (e.g.,
three for one person, six for the other). In the absence of agreement, each gets
zero (or some pre-specied minimum). Clearly a competitive incentive exists
(i.e., I would rather have six than three; or ve 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 identied by
Schelling (1960). Other single dimension of value tasks include the single-
issue 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. Managements 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 ve 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 xed sum of
value (in this case, 240), the task is considered xed-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 pos-
sessing 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 par-
ties. 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. Managements negotiator employs the issue chart
shown at the top of Table 2, and the union negotiator is assigned the issue
chart displayed at bottom. These charts list ve 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 val-
ued 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 indi-
vidually 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 expan-
sive social psychology of justice literature). They noted that distributive bar-
gaining is the process by which each party attempts to maximize his/her own
share in the context of xed-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 taxon-
omy 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 behav-
ioral 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 (Gth, 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 co-
ordination model individual decisions to distribute outcomes, usually over
time. These include, for example, the prisoners-dilemma and trust games.
Coordination games typically generate simple response variables (C or D
Table 1. The Singe Issue Negotiation: Issue Charts for Management and Union Negotiators
Management Issue Chart
Annual Raise (value)
15% (00)
12% (60)
9% (120)
6% (180)
3% (240)
Union Issue Chart
Annual Raise (value)
15% (240)
12% (180)
9% (120)
6% (60)
3% (00)
Note: All agreements sum to a constant for the dyad, 240, making the task xed-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, includ-
ing its joint value, degree of efciency, 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 reect ever-increasing levels of conict (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 articles rst author while he was a graduate
student working in Dean Pruitts laboratory in Buffalo, New York in the early
Table 2. Issue Charts for Management (Top) and Union (Bottom) Negotiators
Management Issue Chart
Salary Vacation Days Annual Raise Medical Coverage
70.000 (00) 3 weeks (00) 15% (00) 100% (00)

65.000 (15) 2.5 weeks (30) 12% (60) 80% (100)
60.000 (30) 2 weeks (60) 9% (120) 60% (200)
55.000 (45) 1.5 weeks (90) 6% (180) 40% (300)
50.000 (60) 1 week (120) 3% (240) 20% (400)
Union Issue Chart
Salary Vacation Days Annual Raise Medical Coverage
70.000 (400) 3 weeks (120) 15% (240) 100% (60)

65.000 (300) 2.5 weeks (90) 12% (180) 80% (45)
60.000 (200) 2 weeks (60) 9% (120) 60% (30)
55.000 (100) 1.5 week (30) 6% (60) 40% (15)
50.000 (00) 1 week (00) 3% (00) 20% (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

COORDINATION GAMES AGREEMENT GAMES
Prisoners Dilemma 1-Issue Negotiation

Chicken Ultimatum Bargaining
Single Dimension N-Person Dilemma
of Value Locomotion Games
Coalition Games
Trust Game
Etc.
Multiple Dimensions not done Integrative Bargaining
of Value Integrative Ultimatum
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 dened accountability in terms of social power and
as an aspect of the relationship between negotiator and constituent. One idea
posited that a mediators 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 integra-
tive 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 infer-
ences about mediating processes. The logic was straightforward: if the differ-
ence between high and low accountability was due to process X (in this case,
a trusting relationship between negotiators), then controlling X should mod-
erate the accountability effect. It soon became apparent that Pruitts design
was elegant and structurally reected Platts notion of strong inference (Platt
1964). To illustrate this type of design and others, what follows is a brief dis-
cussion 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 rep-
resented 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 con-
sisting 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 depen-
dent 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-
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 nding. 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 ndings 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 posi-
tive 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 asso-
ciated with the gift/cartoon manipulation of positive mood.
In path diagram notation, the Carnevale and Isen study might be repre-
sented 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 com-
posed 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 oblig-
ation 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 rene the experimen-
tal manipulation. This is often done by including one or more control groups
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 reects 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 specic 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, Car-
nevale 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 mediators beliefs about the likelihood of agreement (the
mediators perceived common ground) or the mediators concern for the
parties aspirations. Carnevale and Conlon constructed moderator variables
(Z) that controlled both possible mediating processes. The data were clear:
time pressure affected perceived common ground, not concern. When per-
ceived 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 unimpor-
tant, 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 benet 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 benets; for example, they allow us to test the-
ories 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
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 own, but wind
tunnel testing saves lives and cost. Of course, the wind tunnel needs to rep-
resent processes that occur in nature or at least in that part of nature where a
plane with that wing will y. Another benet is that laboratory designs allow
measurement and tracking of processes that would be impossible or at least
extremely difcult 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 difcult to generalize results from laboratory settings to natural settings,
a problem shared with eld 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: 179.
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 psy-
chological research: Conceptual, strategic, and statistical considerations. Journal of
Personality and Social Psychology, 51: 11731182.
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: 171201.
Bornstein, G. (2003). Intergroup conict: Individual, group, and collective interests. Per-
sonality and Social Psychology Review, 7: 129145.
Borsboom, D., Mellenbergh, G.J., & van Heerden, J. (2004). The concept of validity. Psychological
Review, 111: 10611071.
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: 5773.
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: 357361. [reply to J.K. Murnighan (1986), The structure of mediation and intra-
vention: Comments on Carnevales strategic choice model. Negotiation Journal, 2:
351356.]
Carnevale, P.J. (1992). The usefulness of mediation theory. Negotiation Journal, 8: 387390.
[reply to E. van de Vliert (1992), Questions about the strategic choice theory of mediation.
Negotiation Journal, 8: 379386.]
Carnevale, P.J., & Conlon, D. (1988). Time pressure and strategic choice in mediation.
Organizational Behavior and Human Decision Processes, 42: 111133.
Carnevale, P.J., & Isen, A.M. (1986). The inuence of positive affect and visual access on the
discovery of integrative solutions in bilateral negotiation. Organizational Behavior and
Human Decision Processes, 37: 113.
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: 387410.
Cook, T.D., & Campbell, D.T. (1979). Quasi-experimentation: Design and analysis issues for
eld 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 efciency of
two bargaining processes. Behavioral Science, 30: 180183.
Forgas, J.P. (1998). On feeling good and getting your way: Mood effects on negotiator cog-
nition and behavior. Journal of Personality and Social Psychology 74: 565577.
Gth, W., Schmittberger, R. & Schwarze, B. (1982). An experimental analysis of ultimatum
bargaining. Journal of Economic Behavior and Organization 3: 367388.
Hafner, E.M., & Presswood, S. (1965). Strong inference and weak interactions. Science, 149:
503510.
Hertwig, H., & Ortmann, A (2001). Experimental practices in economics: A methodological
challenge for psychologists? Behavioral and Brain Sciences, 24: 383451.
Kelley, H.H. (1966). A classroom study of the dilemmas in interpersonal negotiations. In
K. Archibald. in Strategic interaction and conict: 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: 36198.
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: 196217.
Latham, G.P., Erez, M., & Locke, E.A. (1988). Resolving scientic 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: 753772.
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: 543554.
Liebert, R.M., Smith, W.P., Hill, J.H., & Keiffer, M. (1968). The effects of information
and magnitude of initial offer on interpersonal negotiation. Journal of Experimental Social

Locke, E. (1986) (editor). Generalizing from laboratory to eld settings: Research ndings from
industrial-organizational psychology, organizational behavior, and human resource manage-
ment. 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:
69102.
Mikolic, J.M., Parker, J.C., & Pruitt, D.G. (1997). Escalation in response to persistent annoy-
ance: Groups versus individuals and gender effects. Journal of Personality and Social Psychology,
72: 151163.
Mook, D.G. (1983). In defense of external invalidity. American Psychologist, 38: 379387.
Murnighan, J.K. (1986). The structure of mediation and intravention: Comments on
Carnevales strategic choice model. Negotiation Journal, 4: 351356.
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.
OConnor, K.M., & Carnevale, P.J. (1997). A nasty but effective negotiation strategy:
Misrepresentation of a common-value issue. Personality and Social Psychology Bulletin, 23:
504515.
Platt, J. (1964). Strong inference. Science, 146: 347353.
Pruitt, D.G. (1970). Motivational processes in the decomposed prisoners dilemma game. Journal
of Personality and Social Psychology, 14: 22738.
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 negotia-
tion: Constituent surveillance and contentious behavior. Journal of Experimental Social
Pruitt, D.G., & Johnson, D.F. (1970). Mediation as an aid to face-saving in negotiation.
Journal of Personality and Social Psychology, 14: 23946.
Pruitt, D.G., & Lewis, S.A. (1975). Development of integrative solutions in bilateral negotia-
tion. Journal of Personality and Social Psychology, 31: 621633.
Rettinger, D.A., & Hastie, R. (2001). Content effects on decision making. Organizational
Behavior and Human Decision Processes, 85: 336359.
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: 17551758.
Schelling, T.C. (1960). The strategy of conict. Boston, MA: Harvard University Press.
Sieber, J.E. & Saks, M.J. (1989). A census of subject pool characteristics and policies.
American Psychologist, 44: 105361.
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: 218226.
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 eco-
nomics. Behavioral and Brain Sciences, 24: 428429.
Solomon, R.L. (1949). An extension of control group design. Psychological Bulletin, 46: 137150.
Stevens, C.M. (1963). Strategy and collective bargaining negotiation. New York: McGraw-Hill.
Thompson, L.L. and Hastie, R. (1990). Social perception in negotiation. Organizational Behavior
and Human Decision Processes, 47: 98123.
Triplett, N. (1897). The dynamogenic factors in pacemaking and competition. American
Journal of Psychology, 9: 507533.
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: 12051217.
Weingart, L.R., Olekalns, M., & Smith, P.L. (2005). Quantitative coding of negotiation behav-
ior. International Negotiation.
Managing Conict 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 20032004 revealed over 450 articles with meta-
analysis included in the title. This presence in academic publications illus-
trates the accessibility of meta-analysis to researchers and its acceptance
within the scientic community as a viable research methodology. Within the
realm of negotiation and social conict, use of this technique has emerged
(e.g., de Dreu & Weingart 2003; de Dreu, Weingart, & Kwon 2000; Druck-
man 1994; Stuhlmacher & Citera, in press; Stuhlmacher, Gillespie, & Cham-
pagne 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 tech-
nique and an interest in its application to the conict and negotiation eld. 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 condence in ndings. Provid-
ing an approach for quantitatively cumulating research study results, meta-
analysis 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 eld of
study. Thus, the picture painted by meta-analysis is believed to be subject to
less distortion than historical approaches to synthesizing research ndings (Cooper
1990a; Cooper & Rosenthal 1980; Glass, et al. 1981; Rosenthal 1991).

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, dened 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 signicance levels and counting the number
of studies with ndings in the expected direction versus those in an opposite
direction. This focus on signicance, 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 signicance lev-
els failed to provide information on how much impact or change was com-
mon 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 pub-
lishing 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 benet 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 synthesiz-
ing 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 signicance. 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 meta-
analyses address not only effect size (d) and probability ( p) estimates, but include
methodological guidance on issues such as assessing moderators and weight-
ing 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 re-
search domain from a sample of studies (Huffcutt, 2002). Rather than exam-
ining individuals as data points, meta-analysis examines studies as the indi-
vidual data points. The steps of meta-analysis follow the framework typical
for scientic inquiry: formulating the problem, collecting and evaluating data,
analyzing the data, and interpreting the results (Hedges, Shymansky, & Wood-
worth 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 difcult 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 identied 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 de-
nition 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, xed num-
ber 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.
Practical issues may also inuence 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 denition of the independent or dependent variables. Or, it
might be feasible to paint a broader picture of the construct by including mul-
tiple 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 con-
cerns 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 dened constructs, investigator judgment calls are neces-
sary. Hedges et al. (1989) warned, Treatments will not have been imple-
mented 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 conict research is marked by several research
traditions across multiple scholarly disciplines, thus creating some potential
misclassications. 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 syn-
thesizing the literature, and an example of how meta-analyses can clarify past
research efforts.
Similarly, because conict and negotiation are vast research literatures, a
common misperception is that large numbers of studies exist that can be com-
bined on a particular topic. We have been extremely surprised over the some-
times small number of studies that actually address a particular question and
report relevant data. The most startling nding was a search that resulted in
only 21 explicit negotiation studies (yielding 53 effect sizes) directly compar-
ing the prots men and women earned in explicit negotiation (Stuhlmacher &
Walters 1999). Despite several hundred articles being screened, we were puz-
zled that such a simple relationship was not reported in a way that allowed
effect size calculations. In this case, reviewers were equally puzzled, and
asked for validation of the search. It was certainly possible that gender com-
parisons 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 nal reports. In many cases, gender-prot
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-signicant
(possible given the sample size to detect a small effect) and publication pres-
sures resulted in omitting gender comparisons from the report. While it is
unsatisfying to nd 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 choos-
ing 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 aca-
demic libraries, presented at local conferences, or crammed into the dark
recesses of a laboratory le cabinet. Some of these studies are easily accessi-
ble (e.g., journal articles, books), while others require more diligent investi-
gation. Regardless, the comprehensive meta-analyst will search for pertinent
work by using a variety of tactics such as contacting prominent researchers
in the eld, 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
issues, or those variables that relate to the topic of study and may be hypoth-
esized 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 con-
federates). These variables can examine if differences in procedures inuence
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 captur-
ing the studies and to assist in interpreting the ndings.
Analyses. After the data are collected, they often are analyzed using tech-
niques associated with a methodological tradition. Again, this is a judgment
call. Interestingly, the approach may be highly inuenced by the academic
discipline of the meta-analyst. In an analysis of meta-analytic trends (Mohr,
Zickar, & Russell 2000), Schmidt and Hunters methods were employed in
85% of the meta-analyses by industrial/organizational psychologists, Glasss
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 inuence 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 inter-
ested 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, non-
randomly missing data impacts the validity of the inferences made from meta-
analytic results. The researcher then must determine how to deal with missing
data (Pigott 1994). In addition to the more traditional concerns of missing
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
prot, 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 method-
ological 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 nd 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, inuenc-
ing the project with the investigators biases. However, judgment calls are a
part of the process and can be managed professionally. One way of approach-
ing 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 nal 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 (opti-
mal vs. suboptimal goals, and goal vs. no goal) to tease out if different oper-
ational denitions 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 nal report. The nal report ideally
includes the judgments that were made throughout the course of the meta-
analysis; readers are thus alerted to judgment calls that limit the generaliz-
ability of the results or the inferences that can be made. For example, specic
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 inuences the resulting report and potentially future research.
As Cooper (1990b) explained, Thus, while a great deal of agreement might
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 cen-
tral 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 non-
independent data, and the problem of selection bias in the studies that are
more easily accessible to researchers. However, they countered these argu-
ments by showing the insignicance of the effects of these cited methodolog-
ical 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 appre-
ciate 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 difculty of such projects. As more sci-
entists become familiar with the technique, it will hopefully become less nec-
essary for the researcher to educate promotion or tenure committees, grant
agencies, and colleagues that meta-analyses involve signicant time, effort,
and resources but can yield a substantial contribution to the eld. Ultimately,
we nd that high-quality research requires time and a meta-analysis is no
different.
These challenges are less likely as meta-analyses are published which clar-
ify 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-conict can be benecial to team
performance was not supported. In fact, task-conict appeared to be as dys-
functional as relationship conict. Future research can clearly benet from the
research review that a meta-analysis can provide to a topic.
Future of Meta-analysis
The future for meta-analysis is bright, and advancing with increasing sophis-
tication. One way this can be seen is through an increase in software devel-
oped 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 stud-
ies 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 build-
ing 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).
Conict 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, conict. If it is true that the eld of conict resolution has
become, ironically, very competitive (Coleman 2001: 597), then this area is
fertile ground for meta-analytic work. Competition between ideas and tradi-
tions has always existed in negotiation research. These pressures may have
been intensied of late as negotiation has become more central to the cur-
riculum in many areas of study, as more researchers enter the eld, 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
conict.
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, edi-
tors, The Handbook of Conict 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 ndings. Psychological Bulletin, 87: 442449.
De Dreu, Carsten K., Weingart, Laurie R., & Kwon, Seungwoo (2000). Inuence of social
motives on integrative negotiation: A meta-analytic review and test of two theories. Journal
of Personality and Social Psychology, 78: 889905.
De Dreu, Carsten K., & Weingart, Laurie, R. (2003). Task versus relationship conict, team
performance, and team member satisfaction: A meta-analysis. Journal of Applied Psy-
chology, 88, 4: 741749.
Druckman, Daniel. (1994). Determinants of compromising behavior in negotiation: A meta-
analysis. Journal of Conict Resolution, 38: 507556.
Eagly, Alice H., & Johnson, Blair T. (1990). Gender and leadership style: A meta-analysis.
Psychological Bulletin, 108: 233256.
Eagly, Alice H., Karau, Steven J., & Makhijani, Mona (1995). Gender and the effectiveness
of leaders: A meta-analysis. Psychological Bulletin, 117: 125145.
Glass, Gene V. (1976). Primary, secondary and meta-analysis of research. Educational Researcher,
5: 38.
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: 447466.
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 ndings. 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: 429471.
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 meta-
analytic 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: 4765.
Smith, Mary L., & Glass, Gene V. (1977). Meta-analysis of psychotherapy outcome studies.
American Psychologist, 32, 9: 752760.
Stuhlmacher, Alice F., & Citera, Maryalice. (in press). Negotiator hostility and prot 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 Conict Management,
9: 97116.
Stuhlmacher, Alice F., & Walters, Amy E. (1999). Gender differences in negotiation outcome:
A meta-analysis. Personnel Psychology, 52: 653677.
Walters, Amy E., Stuhlmacher, Alice F., & Meyer, Lia L. (1998). Gender and negotiator
competitiveness: A meta-analysis. Organizational Behavior and Human Decision Processes,
76: 129.
Wanous, John P., Sullivan, Sherry E., & Malinak, Joyce (1989). The role of judgment calls in
meta-analysis. Journal of Applied Psychology, 74, 2: 259264.
Zetik, Deborah C., & Stuhlmacher, Alice F. (2002). Goal setting and negotiation performance:
A meta-analysis. Group Processes and Interpersonal Relations, 5: 3552.
When, Where and How: The Use of Multidimensional
Scaling Methods in the Study of Negotiation and
Social Conict
ROBIN L. PINKLEY, MICHELE J. GELFAND and LILI DUAN
Multidimensional scaling (MDS) like other multivariate procedures is a data-

reduction 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 denes 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 cor-
relations 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 sim-
ilarity between them. By producing a map of the evaluated stimuli, MDS
techniques are able to illuminate the hidden structure and thus, the cogni-
tive 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 phe-
nomena 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

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 fac-
tor 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
reecting attitudes and cognitions are nonmetric, MDS is ideally suited to the
study of conict and negotiation.
Although MDS has wide-ranging theoretical and applied appeal, it has been
highly underutilized in the conict 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 conict
and negotiation. We begin with a discussion of how MDS can be differenti-
ated from other multivariate techniques, such as factor analysis, illustrating its
distinct advantages to conict scholars. Next we provide a brief overview of
multidimensional scaling techniques highlighting the various methods avail-
able 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 conict
and negotiation literature, highlighting the promise this technique holds for
future research.
Multidimensional Scaling (MDS) versus Factor Analysis (FA)
Conict 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 impor-
tant 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 cate-
gories, they are all supervariables or latent variants that subsume a number
of variables (or stimuli) into fewer, broader variable classes. Thus, both fac-
tor 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 benecial, 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 inde-
pendent 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 subjects attribution is not sufcient for the applica-
tion of factor analysis. In contrast, MDS uses individuals as the unit of analy-
sis. This means that one respondents evaluation of stimuli is sufcient
(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 indi-
vidual 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 dis-
cussed 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 sim-
ilarity (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 conict stimuli into categories based on their similarity. In both stud-
ies, 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 pro-
vide 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-dened set of attributes. For example, Hensen, Sarma, & Collins (1999)
asked participants to rate their preferences on items reecting Hollands six
occupational themes. These preferences were later transformed into Euclidean
Distances for input into an MDS analysis. Importantly, MDS studies can use
either attribute-free or attribute-based approaches, whereas factor analysis can

only be applied to the attribute-based approaches. That is, FA requires sub-
jects to rate stimuli on some list of attributes provided by the researcher.
Accordingly, it retains a higher risk of contamination by the researchers own
criteria.
A second advantage of using MDS techniques is that they provide a mech-
anism for detecting, quantitatively categorizing and labeling peoples percep-
tions 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 specic 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 homo-
geneous), a traditional two-way matrix (sometimes referred to as in-group
scaling) should be used. This would be the case, for example, if an experi-
menter 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 differ-
ences 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 danger-
ous 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) per-
ception 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
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 dimen-
sion 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 afliated with research
institutions weight a dimension labeled rigor more heavily than those
afliated 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 per-
ceptions 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 t for
each respondent (or demographic type) stimulus conguration (i.e., the
amount of variance in respondent judgments explained by the multidimen-
sional 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 clus-
ter 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 met-
ric or nonmetric MDS and has individual differences scaling options and
thus can compare the differences of several individual or group matrices.
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 conver-
gence, a wide range of data transformations, and a number of different options
for tting 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, peo-

ple, behavior, events) the input for MDS techniques is the similarity data,
referred to as relational or proximity data. MDS handles all kinds of proxim-
ity 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 com-
pared (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 stim-
uli are also acceptable for input.
Interpreting the MDS Conguration and Labeling the Dimensions
There are two issues that scholars face when interpreting MDS congurations:
1) determining the number of dimensions that best represents the actual prox-
imities 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 opti-
mum number of dimensions needed to describe the stimulus space. While reliance
on these methods varies, experimenters typically use all three criteria for deter-
mining dimensionality.
The rst criterion used for determining dimensionality, is Kruskals (1964)
STRESS index or goodness of t, which indicates how the distances displayed
in the conguration reect 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 indi-
cates good t, with good t nearing zero and poor t 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 inuence of the inter-
pretation 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 stim-
ulus space (i.e., I > 4R).
A second criterion for evaluating the number of dimensions necessary and
sufcient 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 t and a
zero indicates no t at all.
To determine the appropriate number of dimensions, researchers typically
plot the rst 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 t of the MDS model to the
proximity data (Young & Hamer 1987). The elbow test is usually accompa-
nied by an assessment of how well the raw data ts the MDS model by
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 dimen-
sions 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 multi-
dimensional space (conguration). One common technique is to have the par-
ticipants who made the proximity judgments rate the stimuli on a number of
unidimensional attributes and then use multiple regression to regress the uni-
dimensional 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) signicant multiple correlation and F-value, indicating that the con-
guration explains the attribute well, and 2) signicant Beta weight (nor-
malized regression coefcient) 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 judg-
ments 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

Conict and Negotiation
Several examples may further illustrate the inherent benets of using MDS
techniques in the study of conict and negotiation. Although we provide only
a couple examples here, a handful of scholars have used MDS techniques to
address such issues as negotiatior perceptions of conict situations (Pinkley
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 conicts
(Druckman 1997; Druckman, Martin, Nanand, & Yagcioglu 1999). A survey
of these studies will demonstrate the varied techniques for collecting proxim-
ity data, as well as, determining situations and labeling the dimensionality.
Example 1
Pinkley, Brittain, Neale, & Northcraft (1995) used MDS to conduct an induc-
tive 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
conict (i.e., managerial dispute intervention goals, dispute intensity, time pres-
sure, dispute importance, managerial power, and the relative power of the dis-
putants). To fulll this objective, the authors used a ve-step method to collect
and analyze the data. In step one, alumni from four universities lled out a
survey in which they provided a description of the last time they intervened in
a corporate conict, as well as, specics regarding the nature of the conict.
Of the 142 obtained descriptions, 40 were randomly chosen as step 2 stimulus
materials.
In step two, ten of the remaining 40 conict descriptions were randomly
selected as target descriptions. One hundred participants, were randomly sep-
arated into ten groups, each of which was assigned target description, such
that ten participants were given the rst target description, ten the second tar-
get 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 SASs alternative least squares scaling
(ALSCAL, see Takane, Young, & DeLeeuw 1977) implementation of the indi-
vidual differences scaling model (Carroll & Chang 1970). A new set of
allowed experimenters to test the hypothesized relationship between manage-
rial strategy selection (as dened by dimensionality) and the nature of
conict. Two criteria were used to determine the optimum number of dimen-
sions: 1) Kruskals (1964) STRESS index and 2) an elbow test accompanied
by an examination of the amount of variance accounted for by each dimen-
sional solution. Both procedures suggested that a six-dimensional solution did
not signicantly improve on the ve-dimensional solution, which accounted
for 95% of the variance.
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 partici-
pants to rank-order the intervention strategy descriptions, and 2) the third-
party 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 reected
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 ve dimensions and were thus, used
to label the conguration.
In step four, the authors used conrmatory analysis to verify the appropri-
ateness of each of the ten labels by asking ve trained raters (unaware of the
step one three results) to rate the step one intervention strategy descriptions
in terms of each label. Cronbachs alpha was used to assess interrater relia-
bility (found to be quite high at .85). The F values and beta weights found
that all ten of the selected labels loaded onto the ve-dimensional solution
with reliability ranging from .77 to .97. As a result, the ve 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 conict versus avoids con-
ict, and 5) Dispute handled publicly versus privately.
Finally, step ve used multiple regression to relate the ve-dimensional solu-
tion to the intervenor goals and perceptions of conict specied 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 conicts across cultures.
The purpose of the study was to discern if there are universal (or etic) dimen-
sions of conict construal and if there are culture-specic (emic) construals
that are consistent with prevailing cultural values and practices. This study
involved ve steps. In step one, students from the U.S. and Japan were asked
to write a description of a conict 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
degree of severity of the dispute. Participants were asked to describe the

conict situation in terms of the following two questions: 1) Briey, what is
the conict really about? and 2) What is at the heart of the conict? In both
countries, instructions were given in the native language, English or Japanese.
All materials were rst translated into Japanese, and then back-translated by
another translator into English to check for discrepancies.
In step two, 28 conicts in the U.S. and 28 conicts in Japan were ran-
domly selected for the MDS portion of the study. Selected episodes had to be
1) brief (23 sentences); 2) clear and unambiguous as to the exact nature of
the conict; and 3) relevant in both cultural contexts. Japanese conict
episodes were translated into English (for U.S. participants), and U.S. conict
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 conict situation. Participants in both countries were ran-
domly assigned to sort either the U.S. conict episodes or the Japanese
conict episodes, and were unaware of the source of the conict episodes
(i.e., Japan or U.S.). Participants were asked to sort the conict 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.
conicts; 2) Japanese cognitive representations of U.S. conicts; 3) American
cognitive representations of Japanese conicts; and 4) Japanese cognitive rep-
resentations of Japanese conicts. 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 con-
struals of identical conict episodes (in the spirit of an etic approach). It also
enabled the identication of strong universals of conict construal (i.e.,
dimensions of construal that are found regardless of the source of conict and
the cultural background of the participants).
In step four, participants rated the conicts on a number of unidimensional
items to assist in the labeling of the dimensions. These items were derived
from previous studies of conict construal conducted in the U.S. (i.e., Pinkley
1990) as well as from literature on conict in Japan and the U.S. Due to time
restrictions and the cognitive load of the ratings (28 conicts 21 ratings
would require 588 ratings), participants were randomly assigned to rate either
the rst 14 conicts or the last 14 conicts on the 21 unidimensional items.
In step ve, a conict episode by conict 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
program was used to analyze the matrices. An elbow test of Kruskals mea-
sure 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 com-
prehensible interpretations for all of the MDS solutions.
In step six, the dimensions were labeled based on a) an examination of the
conicts in the MDS spaces as well as b) multiple regression analyses which
examined how well the location of each conict on these unidimensional
items was predicted by its location in the multidimensional space. Items with
a signicant multiple correlation and signicant Beta weights indicate that the
conguration explains the item well. Items that load on multiple dimen-
sions, however, are not as useful for labeling the dimensions. The results
demonstrated that Japanese and American participants construed both U.S.
and Japanese conicts through a Compromise versus Win frame (Pinkley
1990), providing evidence of a universal dimension of conict construal. The
results also illustrated that Japanese perceived both sets of conicts to be
more Compromise focused, as compared to Americans. In addition, there were
unique dimensions of conict construal among Americans and Japanese (e.g.,
Infringements to Self and Giri Violations, respectively), suggesting that iden-
tical conict episodes can be perceived differently across cultures.
Example 3
Druckman, Martin, Nan, & Yagcioglu (1999) used MDS to examine the struc-
ture of actual cases of international negotiation. The objective of the study
was to test whether Ikls (1964) typology of international negotiation could
account for similarities among the actual negotiations. This typology distinguished
among ve objectives in international negotiations, including extension, nor-
malization, 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 Ikls 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 char-
acteristics, including region and type of issue (e.g., economic cases, security
issues, environmental, hostage negotiations). All cases were approximately ten
to fteen pages and of a common format, consisting of background informa-
tion, a discussion of the unfolding of the negotiation, and an analysis of the
processes and the outcomes of the negotiation. After selecting the cases, the
authors categorized the conicts in terms of Ikls ve objectives. Final deci-
sions were reached by a consensus between at least two people (the rst
author and another student familiar with Ikls theory). The authors found that
the cases that were chosen were representative of the ve objectives identied
by Ikls theory (eight innovation cases, ve normalization cases, ten redistri-
bution cases, ve 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 agree-
ment) and conditions surrounding the negotiation (e.g., time pressure and
media coverage). A smaller sample of cases was subject to inter-rater relia-
bility. There were high levels of agreement (generally over 90%) and dis-
agreements were resolved through a renement of the denitions 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 Ikls 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 multi-
lateral regimes, which they noted had not existed at the time of the publi-
cation of Ilks book. Next, they performed a Kruskal-Wallis ANOVA to test
for signicant differences among the clusters, and also performed a K-means
cluster analysis. This analysis revealed three clusters (multilateral and nor-
malization cases), innovation cases, and redistribution cases.
In step ve, the authors checked to see whether their initial categorization
of the cases could be veried. In order to do so, they performed discriminant
analyses that examined whether the sixteen coded variables could distinguish
among Ikls categories. These analyses illustrated that a high percentage of
cases were classied accurately.
Finally, in step six, the authors created proles of the clusters of types of
negotiations along the sixteen variables, providing a parsimonious under-
standing of the processes that characterize different types of international
negotiations. This analysis showed, for example, that normalization negotia-

tions 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 Pegnetters (1985), into as many mutually exclu-
sive categories as they wanted. After the sorting task, the mediators rated each
tactic on ve bipolar scales: friendly-unfriendly, assertive-passive, controlling-
uncontrolling, use frequently-use infrequently, and effective-ineffective. All mate-
rials 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 analy-
ses and clustering analyses. A two-way nonmetric MDS program, KYST2A
(Kruskal, Young, & Seery 1977) scaled the similarities data. Kruskals 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 three-
dimensional solution appeared to be closest to the true structure of mediation
tactics.
In the third step, unidimensional scales were regressed onto the MDS congura-
tion 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 reexive, 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
and MDS results were compatible with the results of factor analyses of medi-
ation tactics.
Conclusion
Multidimensional scaling is a powerful method to uncover the hidden struc-

tures that underlie peoples judgments about themselves and their environ-
ments. In the domain of conict 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 conict and negotiation, such as modeling disputant percep-
tions of conicts, characterizing the types of negotiation and mediation tactics
that are used to resolve social conict, and organizing the features or charac-
teristics of conicts and negotiations, among other issues. It also holds much
promise to illuminate how individual differences such as personality, edu-
cation, and gender, as well as organizational or cultural differences affect
conict and disputing. At the same time, compared to its cousin, factor analy-
sis, MDS has been highly underutilized in the eld. In this article, we hope
we have begun to show the unique benets that this technique can bring to
the science and practice of conict and negotiation.
References
Beardsworth, A., & Keil, T. (1992). The vegetarian option: Varieties, conversions, motives and
careers. Sociological Review 40: 253293.
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 multidimen-
sional scaling program for individual differences scaling with constraints, in W. Bandilla &
F. Faulbaum, editors, Softstat 97: Advances in Statistical Software, Stuttgart: Lucius &
Lucias, 6: 6774.
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: 6581.
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 Psycho-
metric Society, Stanford, CA.
Druckman, D. (1997). Dimensions of international negotiations: Structures, processes, and out-
comes. Group Decision & Negotiation 6, 5: 395420.
Druckman, D., Martin, J., Nanand, S.A., & Yagcioglu, D. (1999). Dimensions of international
negotiation: A test of Ikls typology. Group Decision & Negotiation 8: 89108.
Gelfand, M., Nishii, L., Holcombe, K., Dyer, N., Ohbuchi, K., & Fukuno, M. (2001). Cultural
Inuences on Cognitive Representations of Conict: Interpretations of Conict Episodes in
the U.S. and Japan. Journal of Applied Psychology 86, 6: 10591074.
Gelfand, M., Triandis, H., & Chan, D. (1996). Individualism versus collectivism or versus
authoritarianism? European Journal of Social Psychology 26, 3: 397410.
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,
fth edition. Prentice-Hall, Inc., Upper Saddle River, New Jersey.
Hansen, J.C., Sarma, Z.M., & Collins, R.C. (1999). An evaluation of Hollands model of
vocational interests for Chicana(o) and Latina(o) college students. Measurement and Evalua-
tion in Counseling & Development 32: 214.
Hirschberg, N., Jones, L., & Haggerty, M. (1978). Whats in a face: Individual differences in
face perception? Journal of Research in Personality 12, 4: 488499.
Jehn, K.A. (1994). A multimethod examination of the benets and detriments of conict.
Administrative Science Quarterly 40: 256282.
Johnson, C.S. (1995). The Rokeach value survey: Underlying structure and multidimensional
scaling. Journal of Psychology 129: 583597.
Jones, L. & Young, F. (1972). Structure of a social environment: Longitudinal individual dif-
ferences scaling of an intact group. Journal of Personality & Social Psychology 24, 1:
108121.
Kruskal, J.B. (1964). Nonmetric multidimensional scaling: A numerical method. Psycho-
metrika 29: 115129.
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. 07011. Beverly Hills, California: Sage.
Kruskal, J.B., Young, F.W., & Seery, J.B. (1977). How to use KYST2A: A very exible 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: 888913.
McLaughlin, M., Carnevale, P., & Lim, R. (1991). Professional mediators judgments of medi-
ation tactics: Multidimensional scaling and cluster analyses. Journal of Applied Psychology
76, 3: 465472.
Pinkley, R.L. (1990). Dimensions of conict frame: Disputant interpretations of conict.
Journal of Applied Psychology 75, 2: 117126.
Pinkley, R.L., Brittain, J., Neale, M.A. & Northcraft, G.B. (1995). Managerial third party dis-
pute intervention: The impact of situational factors on strategic choice. Journal of Applied
Psychology 80: 363402.
Robinson, S.L. & Bennett, R.J. (1995). A typology of deviant workplace behaviors: a multi-
dimensional scaling study. Academy of Management Journal 38, 2: 555572.
Takane, Y., Young, F.W. & De Leeuw, J. (1977). Nonmetric individual differences in multi-
dimensional scaling: An alternating least squares method with optimal scaling features.
Psychometrika 42: 767.
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: 361373.
Young, F.W. and Hamer, R.M. (1987). Multidimensional Scaling: History, Theory and Appli-
cation. Hillsday, New Jersey: Lawrence Erlbaum Associates.
Young, F.W. and Torgerson, W.S. (1968). A FORTRAN IV program for nonmetric multi-
dimensional 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 benecial out-
comes is affected by antecedent conditions. However, an understanding of the
impact of antecedent conditions does not tell us how they exert their inuence.
To understand this, we turn to analyses of negotiation processes, which pro-
vide 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 proper-
ties of negotiation. Because each negotiator reacts to the other party on a moment-
to-moment basis, we need to understand how strategies and tactics are
sequenced. In our research, we have used Markov chain analysis to try to cap-
ture 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 conse-
quences (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 recipro-
cated? Is the reciprocation of integrative tactics related to the quality of the
negotiated outcome? Does the relationship between outcomes and the recip-
rocation 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

258 PHILIP L. SMITH, MARA OLEKALNS AND LAURIE R. WEINGART
construct a contingency table that represents the dependencies among strate-

gies 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 con-
tribute to the overall model t 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 classied into a nite 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 tac-
tics, 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 schema-
tically 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 partic-
ular 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
xed number of the set of preceding X values. Specically, the process is
called a rst-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 pre-
viously used by the other party and on the negotiators own previous tactic.
In our work we have found that second-order chains are usually sufcient 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 reects the lengths of the sequences we wish to analyze. In the simplest
analysis, we might choose to code behavior as either integrative or distribu-
tive and to focus on a rst-order Markov chain (sequences of length 2). A
transition matrix for such rst-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 nding the dyad in a particular

state, given the preceding state or states. If we identify the states 0 and 1 with
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 sufce 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 nite 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, how-
ever, 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 repre-
sented 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 rst-order or second-order, etc.) may be assessed by determin-
ing 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 contin-
gency table are the tactics at consecutive time steps. The number of dimen-
sions 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 rst-order chain
requires a two-dimensional table; a second-order chain requires a three-dimen-
sional table, and so on. Because the order of the chain is an empirical ques-
tion, it is necessary to test for the nonsignicance of interaction terms that
represent chains of higher order. Thus to conclude that a given set of data
comes from a second-order chain, it is necessary to test for the nonsigni-

cance of the interaction terms associated with a third-order chain. As third-
order chains have dependencies among tactics at four consecutive time steps
these tests, which involve four-way interactions, must be performed on a four-
dimensional 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 three-
element 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 classication 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 classication 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 rst 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 Second-Order Data
X1 X2 X3 N X1 X2 X3 N
0 0 0 454 0 0 0 332
0 0 1 300 0 0 1 176
0 1 0 206 0 1 0 143
0 1 1 285 0 1 1 332
1 0 0 300 1 0 0 177
1 0 1 191 1 0 1 288
1 1 0 284 1 1 0 322
1 1 1 478 1 1 1 738
Note: The variables X 1, X 2, X 3 code the rst, second, and third element of each three-
element sequence. The variable N is the total number of three-element sequences of each type.
Both simulations were based on 2500 observations.
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 t rst- and second-order Markov chains, we gen-
erated two data sets (shown in Table 1) using Monte Carlo simulations of two
chains, each representing different underlying processes. The rst data set (left
side of Table 1) was based on the assumption that the underlying communi-
cation structure can be represented as a rst-order chain. This chain was gen-
erated 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 second-
order chain. This chain was generated using the probability model below:
P[Xn = 0|Xn-1 = 0, Xn-2 = 0] = .7
P[Xn = 0|Xn-1 = 0, Xn-2 = 1] = .5
P[Xn = 0|Xn-1 = 1, Xn-2 = 0] = .4
P[Xn = 0|Xn-1 = 1, Xn-2 = 1] = .2 (6)
Note that the rst-order transition probabilities for this chain, averaged over
values of Xn-2, are .6 and .3, which are the same as those for the rst-order
chain. However, the statistical structure of the two chains is very different.
Table 2 shows the result of tting Markov chain models to these simulated
data. These models are specied using the notation of Fienberg (1980; see
also Agresti, 1990). In Fienbergs notation, a loglinear model is specied as
a set of bracketed model terms. This notation provides a succinct way of indi-
cating 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 contin-
gency table with margins X1, X2, and X3. The rst 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 inde-
pendence in two-way tables. Psychologically, it asserts that there are no sta-
tistical dependencies in the tactics used in consecutive speaking turns in the
negotiation.
The second model, [X1, X2][X2, X3] is a rst-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 sufcient residual degrees of freedom to allow failures of
Table 2. Markov Models For Simulated Data

Model G2 df p DG2 Ddf p
First-order data, 3-way table (independence and rst-order models)
[X1][ X2][ X3] 233.77 4 0

[X1, X2][X2, X3] 2.84 2 .24 230.93 2 0
Second-order data, 3-way table (rst-order model)
[X1, X2][X2, X3] 50.03 2 0
Second-order data, 4-way table (rst- 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 5 0
First-order data with two subpopulations, 3-way table

(rst-order models with subpopulation dependencies)
[M] [X1, X2][X2, X3] 426.74 9 0

[M, X1] [M, X2][M, X3] 11.64 6 .07 415.10 3 0
[X1, X2][X2, X3]
[M, X1] [M, X2] [M, X3] 0.91 4 .92 10.73 2 .005
[X1, X2][X2, X3] [M, X1, X2]
[M, X2, X3]
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 like-
lihood ratio test statistic. It tests whether the indicated model provides a signicantly better
description of the data than the model that precedes it.
t to be detected. In the case of the rst-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
rst-order model to t 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 signicant, and the four-way inter-
action X1.X2.X3.X4 that represents the third-order effect is not.
Loglinear models like those described above may be tted in a generalized
linear modeling framework using a logarithmic link function and a Poisson
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 distribu-
tion in a more general form than that of classical regression models (see
McCullagh & Nelder, 1989; Venables & Ripley, 2002).
The rst part of Table 2 shows a t of independence and rst-order Markov
models to the rst-order data from Table 1. As can be seen, the data are best
tted by a rst-order order. The values of G2 demonstrate that the indepen-
dence model, [X1][X2][X3], fails to capture the structure of the data, whereas
the rst-order Markov model, [X1, X2][X2, X3], captures it accurately. Table 2
also shows conditional likelihood t statistics DG2. This statistic tests the improve-
ment in t (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 mov-
ing from an independence model to a model with rst-order sequential depen-
dencies. The value of DG2 for the rst-order Markov model in the top part of
the table shows that it provides a signicantly better description of the data
than does the independence model.
The second part of Table 2 shows a t of a rst-order Markov model to
the second-order data from Table 1. The failure of a rst-order model to t
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 ts of the second-
order chain reclassied into a four-dimensional contingency table (i.e., into
sequences of length four). The rst-order Markov model, which in this table
is [X1, X2][X2, X3] [X3, X4], again fails to t. 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 nonsignicance of the residual G2 statistic and the notable improve-
ment in t over the rst-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 dif-

ferent subpopulations sequence their tactics during the course of a negotiation.
These subpopulations may be dened prospectively, at the beginning of the
negotiation, by the attributes of the negotiators or by the experimental condi-
tions to which they are assigned. Structural power (high/low), motivational
orientation (proself/prosocial), and tactical knowledge (present/absent) are
examples of classication variables of this kind (Olekalns & Smith, 1999,
2001, 2003; Weingart et al., 1999). Alternatively, they may be dened retro-
Table 3. First-Order Data with Two Subpopulations

M X1 X2 X3 N
0 0 0 0 302
0 0 0 1 148
0 0 1 0 83
0 0 1 1 141
0 1 0 0 149
0 1 0 1 76
0 1 1 0 142
0 1 1 1 207
1 0 0 0 83
1 0 0 1 81
1 0 1 0 27
1 0 1 1 134
1 1 0 0 81
1 1 0 1 81
1 1 1 0 135
1 1 1 1 626
Note: The variables X 1, X 2, X 3 code the rst, second, and third element of each three-
element 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 negotia-
tors achieve. For example, Olekalns and Smith (2000, 2003) have considered
models in which outcomes from simulated negotiations are classied as dis-
tributive, suboptimal, optimal, or impasse, according to where the joint out-
come 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, conse-
quently, are viewed as qualitatively different from one another. However, in a
loglinear analysis framework, they are treated in an identical way. Specically,
hypotheses about any of the possible relationships among antecedent condi-
tions, communication behavior, or outcomes are tested via interaction terms
in a loglinear model.
Table 3 shows data from two simulated, rst-order, Markov chains from
subpopulations with different transition matrices. The transition matrix for the
rst subpopulation M = 0 was
0 1
[ ]
0 .7 .3 .
1 .4 .6
(7)
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 ts of three rst-order Markov mod-
els 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 probabili-
ties 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 orienta-
tion. In the model [M, X1][M, X2][M, X3] [X1, X2][X2, X3] only tactic frequen-
cies 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 improve-
ment in t, 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 reect the
symmetrical properties of the underlying contingency table. Thus, for exam-
ple, the model terms [M, X1], [M, X2] and [M, X3] all represent the same
effect, namely, subpopulation dependencies in tactic frequencies, which are
reected in each of three margins of the contingency table. Similar redun-
dancies exist elsewhere in Table 2 in tests of the order of the underlying
Markov chains. When testing such effects in empirical data, rather than test-
ing overall DG2 values for signicance, as we have done in Table 2, it is
preferable to base tests of improvement in t on only one of a set of redun-
dant terms to avoid inating 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 signicant subpopulation effects have been identied 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 litera-
ture, but for our purposes the most relevant and easily interpretable is the
standardized residual. The standardized residual, ei, for any cell i of the con-
tingency table is dened as
= ni mi ,

ei
mi
where ni is the observed frequency of the cell and m i is the frequency pre-
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 sin-
gle degree of freedom chi-square). As a rule of thumb, residuals of the order
of 2.0 can be viewed as making large or signicant contributions to the
t 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 par-
ticular model terms, like subpopulation dependencies in sequential effects,
signicantly improves the t. 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.
Table 4 shows observed and predicted frequencies, together with standard-

ized residuals, for the t of the model [M][X1, X2][X2, X3] to the subpopula-
tion 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 tran-
sition structure is also represented in the M.X2 . X3 table. Because the simula-
tions 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 reects the differences between
the transition matrices (7) and (8) that were used to generate the data. In par-
ticular, 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 classiable responses,
Table 4. Residuals for [M][X1, X2][X2, X3]

M X1 X2 ni i
m ei
0 0 0 450 307.0 8.16
0 0 1 224 192.5 2.27
0 1 0 225 193.5 2.26
0 1 1 3349 555.0 8.74
1 0 0 164 307.0 8.16
1 0 1 161 192.5 2.27
1 1 0 162 193.5 2.26
1 1 1 761 555.0 8.74
Note: Model [M][X1, X2][X2, X3] is a rst-order model in which the transition matrix is the
same for both subpopulations. The variables ni, m i and ei are the observed frequencies, predicted
frequencies, and residuals for each cell of the M.X1.X2 table, summing over levels of X3.
but with models of even moderate complexity, this may result in sparse con-
tingency tables, in which the number of observations is small relative to the
number of cells. Goodness-of-t statistics like G 2 require samples of a
sufcient size to allow them to be tested for signicance 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. 246250). This may result in overtting 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 undertting bias
can result.
For this reason, it is usually preferable to base inferences on the conditional
likelihood ratio tests, DG2, rather than on the t 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 mar-
ginal 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 t 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 statis-
tical analysis of Markov chains include Anderson and Goodman (1957), Bart-
lett (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).
Acknowledgements
Preparation of this article was supported by Australian Research Council Discovery

Grant DP0342967 and National Science Foundation grant SBR961671.
Appendix: Sample Transcript
integrative 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 pack-
age at six thousand for you and we have 60% removal costs,
that would be additional.
distributive No Id need more than 60% I was thinking 80% actually.
distributive What if we kept it at 60% and upped it to $8000 for the
package?
distributive I just want to get clear in my head about the vacation. Like I
said Im not prepared to relocate to Canberra unless I get more
than the two weeks.
integrative Well it depends how many years youre 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 youd get three weeks holiday.
distributive I would be prepared to go to Canberra for six years only for six
weeks vacation.
integrative So if we were to do say three weeks vacation, how many years
contract would you agree on?
integrative Perhaps if we get back to the package. Id be prepared to go to
Canberra for 6 years, if thats important to you. . . .
integrative Well thats related into vacation, and I know vacations impor-
tant 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.
integrative I think its a good starting point. It obviously needs some work.
My contract can be longer than four years; I think thats nego-
tiable. Canberra if its really important to you.
distributive The thing with Canberra is that weve got three positions there.
How open are you to challenge in new areas? Because we rec-
ognize that your strategic skills are applicable across any broad
range and you have demonstrated very high levels of achieve-
ment in the strategic areas. Were not looking at the strategic

abilities in HR, were looking at your strategic abilities as a sep-
arate set of skills and youve already demonstrated those abili-
ties in HR to a very high degree, which is why you are sitting
here having this conversation with me. Wed 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: 89109.
Bartlett, Maurice S. (1951). The frequency goodness of t test for Markov chains. Proceed-
ings of the Cambridge Philosophical Society 47: 8695.
Billingsley, Patrick (1961). Statistical methods in Markov chains. Annals of Mathematical
Statistics 32: 1240.
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 orien-
tations on negotiator cognition and behavior. Personality and Social Psychology Bulletin 21:
11781188.
De Dreu, Carsten K.W., Weingart, Laurie. R., & Kwon, Soon (2000). Inuence of social
motives on integrative negotiations: A meta-analytic review and test of two theories.
Journal of Personality and Social Psychology 78: 889905.
England, John L. (1973). Mathematical models of two-party negotiations. Behavioral Science
18: 189197.
Fienberg, Stephen E. (1980). The analysis of cross-classied, categorical data (2nd ed.) Cambridge,
MA: The MIT Press.
Hawes, Leonard C., & Foley, Joseph R. (1973). A Markov analysis of interview communica-
tion. Speech Monographs 40: 210219.
Hewes, Dean E. (1979). The sequential analysis of social interaction. The Quarterly Journal
of Speech 65: 5673.
Hoel, Paul G. (1954). A test for Markov chains. Biometrika 41: 430433.
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.
OConnor, Kathleen (1997). Motives and cognition in negotiation: A theoretical integration
and an empirical test. International Journal of Conict Management 8: 114131.
Olekalns, Mara, & Smith, Philip L. (2003). Testing a three-way relationship among negotia-
tors motivational orientations, strategy choices, and outcomes. Journal of Experimental
Social Psychology 39: 107117.
Olekalns, Mara, & Smith, Philip L. (2003). Social motives in negotiation: The relationship
between dyad composition, strategy sequences and outcomes. International Journal of
Conict Management 14: 233254.
Olekalns, Mara, & Smith, Philip L. (2000). Negotiating optimal outcomes: The role of strate-
gic sequences in competitive negotiations. Human Communication Research 24: 528560.
Olekalns, Mara, & Smith, Philip L. (1999). Social value orientations and strategy choices in
competitive negotiations. Personality and Social Psychology Bulletin 25: 657668.
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, negotia-
tor strategies and outcomes. European Journal of Social Psychology, 26: 299313.
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: 287296.
Thomas, Andrew, Roger, Derek, & Bull, Peter (1983). A sequential analysis of informal
dyadic conversation using Markov chains. British Journal of Social Psychology 22: 177188.
Ting-Toomey, Stella (1983). An analysis of verbal communication patterns in high and low
marital adjustment groups. Human Communication Research 9: 306319.
Wasserman, Stanley, & Iacobucci, Dawn (1986). Statistical analysis of discrete relational
data. British Journal of Mathematical and Statistical Psychology 39: 4164.
Wasserman, Staney, & Weaver. Sheila O. (1985). Statistical analysis of binary relational data:
Parameter estimation. Journal of Mathematical Psychology 29: 406427.
Weingart, Laurie R., Olekalns, Mara & Smith, Philip L. (2004) Quantitative coding of nego-
tiation 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: 366393.
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 notication 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 difculty
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 difculty. 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 ve hours of his/her time for
research participation, this averages to about 70 subject hours per project a
number that is hardly sufcient for a 2 2 between-subjects experimental
design!
Fortunately, there is now a real possibility that such woeful tales of in-
sufcient 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 simultane-
ous 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 gure of about 30 million people world-wide connected
to the internet in 1995 (GVUs 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 gure further
increased to 106.1 million by December 1999. This large pool of internet users

274 CHA YEOW SIAH
makes the medium practically a goldmine for those involved in the data col-
lection 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 depart-
ments found in many companies), or serious academic research purposes.
In addition to the virtually unlimited pool of potential participants on the
internet, the obvious benets 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 dis-
ability or special illnesses) through the help of newsgroups or listservs (auto-
mated mailing lists). This provides researchers the capability to conduct
research that they would otherwise nd 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 mate-
rials. Transcription errors are minimized, if not entirely eliminated, when one
can program for the transfer of data directly to a spreadsheet or statistical pro-
grams 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 rst month of the data col-
lection period. Nancarrow, Pallister and Brace (2001) succinctly summarized
the above obvious advantages as the speed, ease and cost appeals of con-
ducting 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 difculties of
implementing an internet-based study? What are the hidden threats to valid-
ity 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 ramications 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 tech-
niques 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 col-
lection techniques but to discuss the quality of data collected using the inter-
net 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 partici-
pants 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 benets 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 sur-
veys, followed by experiments. The discussion of problems related to inter-
net-based study will be made relative to these two research methods.
One reason why surveys tend to be the more commonly employed method-
ology 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, exper-
iments 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) measure-
ment 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 diver-
sity, certain groups may have been disproportionately excluded from the sam-
pling 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 discrim-
ination), 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 pre-
dominantly 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-representa-
tion of certain demographic groups is fast disappearing because of the rapid
growth of internet-access across all segments of the society. The sample ob-
tained 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 ade-
quate frame population for sampling based on phone or mail directories, there
are still tremendous difculties at the current stage to obtain an adequate
frame population for internet-based study from which to make statistical infer-
ences. 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 tar-
get population (to which one wants to make inference) from the frame popu-
lation, 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 esti-
mates to be made due to the unknown standard error of measurement.
Therefore, though the internet provides an efcient 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 het-
erogeneous sample seems persuasive considering that the most common crit-
icism 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 hetero-
geneity 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 sufcient 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 sufciently large sample size, data
could also be partitioned into different strata based on the demographic vari-
ables and analyzed separately.
Nonetheless, it is worthwhile to note that the argument of conducting
internet-based experiments for the benet 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 compro-
mise the data integrity.
The consequence of lying on demographic information in a survey is that
it causes erroneous classication that results in inaccurate generalization.
Though the problem also exists in conventional research setting (e.g., a res-
pondent may still lie about his age in a telephone survey), the danger is con-
siderably 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 curiositys 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 classied generally into proactive iden-
tication 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 limita-
tions because the repetition could be due to different participants using the
same machine in a computer laboratory to participate in the study. In addi-

tion, 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 detec-
tion task. Other strategies such as setting a cookie in the participants 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 identiers (such as participants ID or password
allocated prior to the start of the experiment) and to reject such submissions
when they are agged requires the allocation of password that raises partici-
pants concern that their responses are no longer condential.
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 alterna-
tive 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 cooper-
ation could be enhanced by informing them that their action of multiple sub-
missions 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 submis-
sion 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 character-
istics, experimenter expectation), the respondent (incomprehension, missing
responses, response set) or the instrument (poor wording or unclear instruc-
tions, unreliable measuring instruments). The nature of internet-based study
(anonymity, social distance from experimenter, program-controlled proce-
dures) 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 ver-
sions. 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 high-
light an improbable response that lies outside the reasonable range of a value
could also help to inform participants of their mistakes. In experiments con-
ducted 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 con-
cerns. For instance, Montgomery and Ritchie (2002) found during the testing
of their experiment program that as many as one-fth of their respondents
either failed to read the directions on the screen or simply could not gure
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 par-
ticipant who is unsure what he is supposed to do next will submit random
responses that are difcult 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 inuences on the data. For instance, the use of
differently congured 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
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 inuences such as those mentioned are
likely to contribute signicantly 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 inuences 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 psycho-
logical variables are at work when participants respond to survey on the inter-
net. For instance, Epstein and Kinenberg (2002) replicated an earlier paper-
and-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 respon-
dents 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 com-
fortable 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 inter-
net helps to induce more candid responses in surveys, especially those which
ask sensitive questions. This is welcoming news for researchers who study
conicts or investigate stigmatized or deviant behaviors that are traditionally
plagued with the problem of untruthful reporting. But to ensure that this qual-
ity 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 experi-

ments on the internet, it is important to consider the anonymity factor, and to
understand how it inuences 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 possi-
bly affect the process and outcome of social interactions. According to Social
Identity theory, whether an individual adheres or dees 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 inuence of an anony-
mous setting and how this is moderated by group-level identity. An under-
standing 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
conicts and their resolution. Substantively, trust (or the lack of ) in an on-
line setting opens up new and intriguing avenues of research for social
conicts researchers. Methodologically, trust could also affect studies con-
ducted on the internet via its inuence on the treatment effect.
For instance, in a recent investigation where I hoped to nd out whether
participants given a computerized social dilemmas task would behave differ-
ently if the task was carried out in the lab versus over the internet, I discov-
ered an interesting difference owing to the setting in which the two groups of
student participants completed their tasks. Specically, the task was deliber-
ately programmed to lead student participants to believe that they were con-
nected 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 computer-
ized 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 suspi-
cious 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
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 exper-
imenters. 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; ONeil
and Penrod 2001). It is no longer possible for researchers to provide a mini-
mally 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 back-
ground), and the kind of activities that participants may be carrying out con-
currently 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 signicant 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 benets of increasing self-
disclosure 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 dynam-
ics and different experiences within individual participants that would interact
in an unintended manner with the investigated variables. However, such issues
have not been given sufcient 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 eth-
ical 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 con-
duct 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 rep-
resents 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 specic newsgroups? What are the dos
and donts 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 ourish.
To this end, experiences shared by various researchers (e.g., Buchanan and
Smith 1999; Montgomery and Ritchie 2002; Pettit 1999) and related discus-
sions 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) lled 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 col-
lecting data via the internet. While some of the problems present considerable
challenges to the viability of conducting research on the internet, it is impor-

tant to take note that there is no lack of internet-based studies that have pro-
duced encouraging results.
For instance, McGraw, Tew, and Williams (2000) showed that it is possi-
ble 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 t 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 thor-
oughly completed than the traditional counterpart. In a similar vein, Pettit
(2002) identied no difference in response set effects between an on-line ver-
sion 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 sam-
ples 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 benets and costs asso-
ciated 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 benets outweigh the
cost of conducting research on the internet.
Finally, it is undeniable that internet-based studies will continue to pro-
liferate. The list of problems identied in this article should not be viewed
as insurmountable, but rather, as new opportunities for innovation and renement.
More important, it is an additional tool in an arsenal of research methods that
provide researchers with another means to triangulate their variables of inter-
est, 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 inter-
net-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 skep-
ticism 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 nd 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: 573590.
Berkowitz, Leonard and Donnerstein, Edward (1982). External validity is more than skin
deep: Some answers to criticisms of laboratory experiments. American Psychologist, 37, 3:
245257.
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: 803832.
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: 125144.
Cook, Thomas C. and Campbell, Donald T. (1979). Quasi-experimentation: Design and analy-
sis issues for eld settings. Chicago: Rand McNally.
Couper, Mick P. (2000). A Review of Issues and Approaches, in David de Vaus, editor,
Social Surveys 2: 149180. 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: 3347.
Gonzalez, J.E. (2002). Present Day Use of the Internet for Survey-Based Research. Journal
of Technology in Human Services, 19, 2/3: 1931.
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: 93104.
Graphic Visualization and Visibility Centres 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: 157159.
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: 186191.
John, Oliver. P. and Srivastava, Sanjay (1999). The Big Five Trait taxonomy: History, mea-
surement, 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 Web-
delivered Experiments: Can You Trust the Data? Psychological Science, 11: 502506.
McNemar, Q. (1942). Opinion-attitude methodology. Psychological Bulletin, 43: 289374.
Michalak, Erin E. and Szabo, Attila (1998). Guidelines for Internet Research: An Update.
European Psychologist, 3, 1: 7075.
Montgomery, Paul and Ritchie, David (2002). Kermitt: Conducting an Experiment on the
Web. Journal of Technology in Human Services, 19, 2/3: 135149.
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: 518.
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: 136149.
ONeil, Kevin M. and Penrod, Steven D. (2001). Methodological variables in Web-based
research that may affect results: Sample type, monetary incentives, and personal informa-
tion. Behavior Research Methods, Instruments & Computers, 33, 2: 226233.
Pettit, Frances A. (1999). Exploring the use of the World Wide Web as a psychology data col-
lection tool. Computers in Human Behavior, 15, 1: 6771.
Pettit, Frances A. (2002). A Comparison of World-Wide Web and paper-and-pencil personality
questionnaires. Behavior Research Methods, Instruments & Computers, 34, 1: 5054.
Piper, Allison I. (1998). Conducting Social Science Laboratory Experiments on the World
Wide Web. Library & Information Science Research, 20, 1: 521.
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: 161198. 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 collec-
tion on the internet. Behavior Research Methods, Instruments & Computers, 33: 201211.
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: 754775.
Smith, Michael A. and Leigh, Brant (1997). Virtual subjects: Using the internet as an alter-
native source of subjects and research environment. Behavior Research Methods, Instru-
ments, & Computers, 29, 4: 496505.
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: 4649.
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: 867873.
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 inuence of psycho-
logical research in that growth.1
There are many similarities between experimental economics and psycho-
logical research. Researchers in both elds are concerned with similar sub-
stantive 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 elds also share many methodological practices. Both
elds use convenient populations (like undergraduate students) as participants
in their experiments. Both elds elicit decisions or introspections from their
participants and use those to learn about the world. Both elds 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 the-
ories; psychology experiments are designed to address psychological theories.
This distinction may seem obvious at rst, but it has important and often
unforeseen implications for methodological differences in the two elds.
These methodological differences are the subject of this article. I will discuss
ve areas where experimental economists and experimental psychologists dif-
fer. They are ordered from most to least important for economists, although
readers may disagree with my ordering.
This is by no means the rst article on experimental economics methodol-
ogy. However, most previous articles have been aimed at traditional (non-
experimental) 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

290 RACHEL CROSON
(1994) have written the denitive text on experimental economics methodol-

ogy aimed at economists who want to begin running or consuming experi-
ments, and I highly recommend it for psychologists who want to see how
economists think about experimental methodology. A very few papers explic-
itly compare economic and psychological methodology, Cox and Isaac (1986)
present experimental economics methodology for an interdisciplinary audi-
ence. Hertwig and Ortmann (2001) provide a well-documented comparison of
the methodologies. Holt (1995) presents his perceptions of methodology in exper-
imental psychology for economists. Croson (forthcoming) compares the two
methodologies for law and economics scholars.
As many have noted, one main difference between the elds of economics
and psychology is the existence of a unied 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 dialec-
tic of the scientic 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 theorys assumptions. If the exper-
iment is not internally valid, the data it produces is not relevant to the the-
orys predictions. For instance, if the theory a researcher is trying to test
assumes that a game is innitely repeated, the experiment must implement
such a game.3 Similarly one cannot test one-shot theories using repeated inter-
action among the same players.4
Many (possibly all) of the ways in which experimental economics method-
ology 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 under-
stand and believe the relationship between their actions and their payoffs;
deception would endanger this belief and is almost nonexistent. Experimental
details are chosen to inuence participants perceptions of one or another of
these areas. The nal 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 elds choice of methodol-
ogy, 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 benet 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 infor-
mative. Economic theory makes no predictions of what individuals will say
they would do, only what they will actually do when faced with a given deci-
sion and the resulting payoffs. For example when studying ultimatum bar-
gaining 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 econom-
ics (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 at-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 rst identied the disjunction effect
(nonconsequential reasoning), their papers included data from surveys and
hypothetical questions (Shar and Tversky 1992; Tversky and Shar 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, pub-
lic 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, every-
one 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 individuals decisions and
argue (successfully) that the payments involved are consistent with the theo-
ries being tested. For example, in Boyce et al. (1992) the authors want to
elicit willingness-to-pay and willingness-to-accept values for specic environ-
mental 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 exper-
imenter will chop down if the damage must be incurred. This tree-killing
action captures nicely the theorys assumptions of incurring environmental
damage and is an excellent example of non-nancial rewards that are none-
theless contingent on the participants 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 nancial 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 nd that incentives decrease framing effects, bring
auction bids closer to optimality and eliminate preference reversals. They also
identify contexts when payment had no effect, specically condence judg-
ment and information acquisitions. For some tasks contingent payment does
not seem to affect the mean outcomes, while for others it does. There is gen-
eral agreement, however, that contingent payment reduces the variance of
responses one receives. Smith and Walker (1993) survey 31 experimental eco-
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 sufcient) 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 cho-
sen (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 partici-
pants, 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 sur-
vey or hypothetical responses. Many experiments in economics use a post-
experimental 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 rela-
tion 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 at fee, but an amount contingent on their decisions. This is particularly
important if the experiments results are inconsistent with economic theory. It
is far too easy (and common) for economists to simply disregard experimen-
tal results inconsistent with their theories because the participants werent
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 some-
thing in the real world (e.g. Croson and Mnookin 1997). Participants are not
told they are providing a public good (or, even more specically, 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 behav-
ior 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 signicant
differences between treatments of the experiment. Finally, and most impor-
tantly, context can add systematic bias or demand effects. For example, if par-
ticipants 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 everyones 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 signicantly change the con-
clusions 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 Wasons-task errors signicantly 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 methodologi-
cal deviation as a lack of incentives. However, papers aimed at economists
need to argue that the results are not being driven by the particular context
chosen or, even better, demonstrate the same effect in multiple (sufciently
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 behav-
ior 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 eco-
nomics 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 differ-
ences, especially gender differences. Eckel and Grossman (forthcoming b) and
Croson and Gueezy (2005) provide reviews of this literature. Other experi-
mental economics research has started to investigate cultural differences. For
example, Buchan, Croson and Dawes (forthcoming) use experimental eco-
nomics 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 psychol-
ogy, where students in introductory psychology courses participate in experi-
ments 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 deci-
sions 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 dilem-
mas (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 dif-
ferent than non-economists. Second, given this unresolved debate, the experi-
mental economics practice of using volunteers from all over the university as
participants in their experiments rather than students from their own (eco-
nomics) classes is a sensible and conservative one.
A nal 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 prob-
lematic 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 pro-
fessionals as participants makes sense. For example, Cummings Holt and
Laury (2002) use farmers to test competing designs for mechanisms to allo-
cate water rights in Georgia. Dyer, Kagel and Levin (1989) and Dyer and
Kagel (1996) test theories of auctions with contractors who submit competi-
tive bids for a living.
Once the researcher moves from students to professionals, the incentives
used in the experiment become both more important and more difcult. 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 signicantly 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 pro-
fessionals 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 incen-
tives 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 ones
own students as participants in experiments.
Deception
One methodological difference between experimental psychologists and econ-

omists which has recently received attention is the use of deception in exper-
iments. A series of papers examines (and criticizes) standard practice in both
elds, 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 de-
scribed above, the validity of an economic experiment rests on the link
between behavior and payoffs (incentives). If that link is weakened, the exper-
iment 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 at fee for their
participation, they will begin to distrust the experimenters 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 benets 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 experi-
ment 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 eld 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 any-
thing 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 econo-
mists, deception will prove to be a difcult barrier to overcome. If its 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 justication 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 collec-
tion of practices in the implementation of economics experiments, rather than
globally-accepted wisdom about experimental design. Many of these proce-
dures are commonly used both psychology and economics.
Experimental projects begin with an application to ones human subject
committee. Experimental economists tend to have relatively little trouble with
human subjects committees, as many of the usual ags are not present in
their designs. The experiments do not involve deception, there are rarely any
consequences other than nancial, and participants earn money for their par-
ticipation. 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
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 econ-
omists 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. Provid-
ing 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 partici-
pants 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 nan-
cial 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 con-
struct examples that are far outside the possible range of behavior. For exam-
ple, 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 exam-
ples, 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 ll in their own numbers
for their decisions. Then we ask participants to calculate their earnings (and
sometimes, their counterparts earnings). Experimental monitors circulate through-
out 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 partici-
pant 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 dont we just all cooperate so we can earn
more money?). These are attempts to inuence the actions of others dis-
guised as questions, and these the experimenter doesnt want asked and
answered publicly. One tactic I nd 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 nal issue that often arises during experiments involves randomization.
While it is now technologically very easy to use computers to generate ran-
dom numbers, participants are often skeptical about the unbiasedness of this
method. Thus economics experiments typically use actual randomizing de-
vices (e.g. dice as in Croson and Johnston 2000) to implement random out-
comes.
Once the experiment is over, participants receive their earnings (in cash)
and are asked to sign a receipt before they leave. In many economics exper-
iments it is important that these earnings be paid privately. Theories often
assume that individuals value money absolutely, without social inuence 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 con-
siderations 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 nal end-of-experiment consideration involves debrieng. As with our
rst topic (human subject committees) extensive debrieng 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
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 psy-
chologists (or even economists) to use in their implementations. No econo-
mist will object to a psychological experiment on the basis of the debrieng
procedure used, for example. Hopefully, however, this discussion will provide
some advice and insight into the nitty-gritty of running economics experi-
ments and perhaps even some ideas for psychologists to use.
Data Analysis
A nal methodological dimension along which the elds differ is the data
analyses they employ. In my publications in both economics and psychol-
ogy/management journals, I have found referees and editors surprisingly parochial
about their favored statistical methods.
Both sides use nonparametric statistics, although economists favor Wil-
coxon tests (Mann-Whitney U tests) after an inuential paper demonstrating
its power in ultimatum and dictator game data (Forsythe et al. 1994).6 In con-
trast 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 regres-
sions 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 objec-

tives of the groups. In psychology, a results 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 dont) illuminate the underlying cause of the result. In eco-
nomics, 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 nal methodological difference concerns the ex-ante and ex-post hypoth-
eses. 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 psy-
chology and sociology were asked how frequently they had personally ob-
served 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 unied body of theory. A hypothesis described as ex-
ante 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 dis-
proving the hypothesis is a perfectly acceptable and often sought-after out-
come. Kerr (1998) presents an outstanding discussion of HARKing and its
costs and benets for individuals and the profession.
Conclusion
There are many similarities between experimental economics and psycholog-

ical research. This article, however, was designed to illuminate some of the
differences. I have discussed ve 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 deci-
sions based on the objectives of their experiment, the methods currently used
in their eld, 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
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/mod-
eration analyses as methods to eliminate alternative explanations for demon-
strated phenomena.
I believe the questions that psychologists and economists are asking in their
research have the potential to inform each others elds deeply and pro-
foundly. Understanding and accepting methodological differences is an impor-
tant rst step toward generating surplus-creating gains from trade, and I hope
this article will help in taking that rst 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 elds 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 benets of making the theory more complex. The benets 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 asymmet-
ric you can capture different risk preferences in gains and losses, or by adding others pay-
offs into ones own utility function you can capture social factors like altruism, envy and
inequality-aversion). However, there are also costs to adding extra parameters. The parame-
ters 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 innitely repeated game in the lab is to derive the theorys pre-
dictions under a discount rate of d, then implement the discount rate by setting the proba-
bility 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. Shar and Tversky (1992) did include some studies on the prisoners dilemma game in
which participants were paid for their earnings. Morris, Sim and Girotto (1998) ran similar
experiments with a at fee.
6. This paper also demonstrated signicant differences between behavior in paid and unpaid
ultimatum and dictator games.
References
Andreoni, James (1988). Why Free Ride? Strategies and Learning in Public Goods Experi-
ments. Journal of Public Economics, 37: 291304.
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:
257264.
Binmore, Kenneth (1999). Why Experiment in Economics? Economic Journal, 109, 453: F1624.
Boles, Terry; Croson, Rachel; Murnighan, Keith (2000). Deception, Retribution, and the Negotiation
Process. Organizational Behavior and Human Decision Processes, 83: 235259.
Bonetti, Shane (1998a). Experimental Economics and Deception. Journal of Economic Psychology,
19, 3: 377395.
Bonetti, Shane (1998b). Experimental Economics and Deception: Response. Journal of Economic
Psychology, 19, 3: 411414.
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: 13661373.
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: 168206.
Camerer, Colin; Hogarth, Robin (1999). The effects of nancial incentives in experiments: A
review and capital-labor production theory. Journal of Risk and Uncertainty 19, 13: 742.
Carter, John; Irons, Michael (1991). Are Economists Different, and If So, Why? Journal of
Economic Perspectives, 5: 171177.
Cox, James; Isaac, Mark R. (1986). Experimental Economics and Experimental Psychology:
Ever the Twain Shall Meet? in Economic Psychology: Intersections in Theory and Applica-
tion, MacFadyen and MacFadyen, editors, North-Holland, Amsterdam: 647669.
Croson, Rachel (1996a). Information in Ultimatum Games: An Experimental Study. Journal
of Economic Behavior and Organization, 30: 197212.
Croson, Rachel (1996b). Partners and Strangers Revisited. Economics Letters, 53: 2532.
Croson, Rachel (1999). The Disjunction Effect and Reason-Based Choice in Games.
Organizational Behavior and Human Decision Processes, 80: 118133.
Croson, Rachel (2000). Thinking like a Game Theorist: Factors Affecting the Frequency of
Equilibrium Play. Journal of Economic Behavior and Organization, 41: 299314.
Croson, Rachel (2002). Why (and How To) Experiment: Methodologies from Experimental
Economics. University of Illinois Law Review, 2002: 921945.
Croson, Rachel; Boles, Terry; Murnighan Keith (2003). Cheap Talk in Bargaining Experi-
ments: Lying and Threats in Ultimatum Games. Journal of Economic Behavior and Organization,
51: 143159.
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: 5073.
Croson, Rachel; Mnookin, Robert. (1997). Does Disputing through Agents Enhance Coope-
ration? Experimental Evidence. Journal of Legal Studies, 26: 331345.
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: 108115.
Dyer, Douglas; Kagel, John (1996). Bidding in common value auctions: How the commer-
cial construction industry corrects for the winners curse. Management Science, 42, 10:
14631476.
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: 347369.
Frank, Robert; Gilovich, Thomas; Regan, Dennis (1993). Does studying economics inhibit
cooperation? Journal of Economic Perspectives 7: 159171.
Frank, Robert; Gilovich, Thomas; Regan, Dennis (1996). Do economists make bad citizens?
Journal of Economic Perspectives, 10: 187192.
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 Dont Pay At All. Quarterly Journal
of Economics, 791810.
Hertwig, Ralph; Andreas Ortmann (2001). Experimental Practices in Economics: A Methodological
Challenge for Psychologists? Behavior and Brain Sciences, 24, 383451.
Hey, John (1998). Experimental Economics and Deception: A Comment. Journal of
Economic Psychology, 19, 3: 397401.
Holt, Charles (1995). Psychology and Economics: A Discussion of Methodology. University
of Virginia, Discussion Paper.
Kamecke, Ulrich (1997). Rotations: Matching Schemes That Efciently Preserve the Best
Reply Structure of a One Shot Game: Note. International Journal of Game Theory, 26, 3:
409417.
Laband, David; Biel, Richard (1999). Are Economists More Selsh than Other Social Scientists?
Public Choice, 100, 12: 85101.
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: 295310.
McDaniel, Tanga; Starmer, Chris (1998). Experimental Economics and Deception: A Com-
ment. Journal of Economic Psychology, 19, 3: 403409.
Morris, Michael; Sim, Damien; V. Girotto (1998). Distinguishing Sources of Cooperation in
306 RACHEL CROSON
the One-Round Prisoners Dilemma: Evidence for Cooperation Decisions Based on the
Illusion of Control. Journal of Experimental Social Psychology, 34, 494512.
Ortmann, Andreas; Hertwig, Ralph (1997). Is Deception Acceptable? American Psychologist,
52, 7: 746747.
Ortmann, Andreas; Hertwig, Ralph (1998). The Question Remains: Is Deception Acceptable?
American Psychologist, 53, 7: 806807.
Ortmann, Andreas; Hertwig, Ralph (2002). The Costs of Deception: Evidence from
Psychology. Experimental Economics, 5: 111131.
Plott, Charles (1982). Industrial Organization Theory and Experimental Economics. Journal
of Economic Literature, 20: 14851527.
Plott, Charles (1991a). Will Economics Become an Experimental Science? Southern
Economic Journal, 57, 4: 901919.
Plott, Charles (1991b). Economics in 2090: The Views of an Experimentalist. Economic
Journal, 101, 404: 8893.
Plott, Charles (1994). Market Architectures, Institutional Landscapes and Testbed Experi-
ments. Economic Theory, 4: 310.
Roth, Alvin (1986). Laboratory Experimentation in Economics. Economics & Philosophy, 2,
2: 245273.
Roth, Alvin (1988). Laboratory Experimentation in Economics: A Methodological Overview.
The Economic Journal, 98, 393: 9741031.
Roth, Alvin (1991). Game Theory as a Part of Empirical Economics. Economic Journal, 101,
404: 107114.
Roth, Alvin (1994). Lets Keep the Con out of Experimental Econ: A Methodological Note.
Empirical Economics, 19, 2: 279289.
Shar, Eldar and Tversky, Amos (1992). Thinking through uncertainty: Nonconsequential rea-
soning and choice. Cognitive Psychology, 24, 4: 449474.
Smith, Vernon (1976). Experimental Economics: Induced Value Theory. American Economic
Review, 66, 2: 274279.
Smith, Vernon (1989). Theory, Experiment and Economics. Journal of Economic Perspectives,
3, 1: 151169.
Smith, Vernon (1994). Economics in the Laboratory. Journal of Economic Perspectives, 8, 1:
113131.
Smith, Vernon and Walker, James (1993). Monetary Rewards and Decision Cost in Experi-
mental Economics. Economic Inquiry, 31, 2: 245261.
Tversky, Amos and Shar, Eldar (1992). The Disjunction Effect in Choice under Uncertainty.
Psychological Science, 3: 305309.
Yezer, Tony; Goldfarb, Robert; and Poppen, Paul (1996). Does Studying Economics Dis-
courage Cooperation?: Watch What We Do, Not What We Say or How We Play. Journal
of Economic Perspectives, 10, 1: 171186.
Legal Research on Negotiation
REBECCA HOLLANDER-BLUMOFF
Lawyers and conict are a classic combination. Lawyers are engaged in the
resolution of conict on a constant basis, whether they are negotiating settle-
ments during litigation or crafting deals during corporate transactions. The
study of legal conict resolution is a natural area for legal scholarship to
address. This article offers a sketch of the methodology used by legal schol-
ars studying negotiation, describing the range of approaches used and sug-
gesting some of the benets of and obstacles to empirical work on legal
negotiation.
The rst 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 negotia-
tion. 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 nish 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 thought-
fully argued law review article that offers the authors 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.

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 schol-
ars 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 situ-
ation say, falling crime rates (Ehrenreich Brooks 2003: 2308) or the grow-
ing 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 profes-
sors in recent years, and articles discussing a problem from a law and eco-
nomics 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
benecial and why the case law, therefore, is or is not correctly decided
(Bodie 2004). That is not to say that the authors own experience and
thoughts are presented, transparently, as one persons 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 pre-
sents (in the third person, almost always) a picture of the landscape of a cer-
tain topic, what others have said and thought about it, and nally the way
that things should be seen. The author typically draws mostly on legal author-
ities 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 re-
search are woven into, sprinkled onto, or even used as the basis for a more
traditional law review article. Empirical ndings may support the authors
premise, giving the authors argument heft and making it more convincing. In
this category, for example, Birke and Fox (1999) use data gathered by psy-
chologists 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 lit-
igation. In a similar vein, Korobkin and Guthrie (2004) describe how com-
mon 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 advo-
cacy, 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 dis-
cussion of previous social science research about the status quo bias but also
by a tailored experiment of his own design that specically 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 pre-
fer 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 eld 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 accep-
tance, but empirical work in the social science vein still accounts for a tiny
minority of all articles published.
Legal academias institutional biases in favor of normative theoretical
analysis over empirical research pose a challenge to the study of legal nego-
tiation. Very little case law or statutes concern legal negotiation, and thus nor-
mative 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
critical component of the current legal system, and legal scholars have in-
creasingly turned their attention to the process of dispute resolution outside
the courts.
In studying how disputes are resolved through negotiation, then, legal aca-
demics face a core tension between the study of an important and pervasive
phenomenon in their eld 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 empir-
ical work in law scholarship, as described above, thus poses particular chal-
lenges 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 found-
ing 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 consid-
ers 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 exam-
ined 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
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 denition of a bargaining zone (zone denition)
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 litiga-
tion of an apology (Cohen 1999; Latif 2001; Bartels 2000/2001). Similarly,
many law review articles on negotiation explore a unique context say, nego-
tiation in a labor setting (Dau-Schmidt 1992), or negotiation in the context of
a hostile corporate takeover (Subramanian 2003; Oesterle 1986), or negotia-
tion over the sale of a car (Ayres 1991). Indeed, some commentators suggest
that any discussion of legal negotiation must be context-specic for it to be
meaningful (Neumann & Krieger 2003). Often these articles examine the
implications of legal rules or endowments on negotiation in an individual con-
text. Robert Mnookin and Lewis Kornhausers (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 exam-
ples, 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 nego-
tiation: 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) identied 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 prob-
lems 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 authors own
experience and thoughts, as well as a review of relevant literature in the eld.
There are also, however, more expressly empirical articles about legal negotiation.
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 effec-
tive 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 adver-
sary 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, nding 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 eld 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 nego-
tiation; for reasons described both above in both the preceding and the fol-
lowing 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 publica-
tions. Still others who study legal negotiation empirically may include psy-
chologists, sociologists, or business school academics whose works are
published outside of law reviews and legal journals and are thus less acces-
sible to the legal community.
Obstacles to Researching Legal Negotiation
Law differs from psychology and other disciplines that study conict, 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 behav-
ior, 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 eld of law, in contrast, encompasses both legal practice itself (by prac-
titioners) and the study of the legal rules and legal practice (by legal acade-
mics). This makes it especially ironic that legal research on negotiation often
does not delve into the substance of real behavior. I attribute the dearth of
such research to two different sources: rst, obstacles within the legal acad-
emy, 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 difcult. 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 quan-
titative analysis of any sort. Because legal academics are not trained in a tra-
dition of empirical research, they are less likely to approach the problems in
their eld 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 men-
tion 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 acad-
emy. 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 identied 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
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 nding limited and nar-
rowed 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 dis-
tinguish 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 eld 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 difcult. Nonetheless, many lawyers are delighted to tell
an interested listener all about their experiences that is to regale an audi-
ence 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 prose-
cutors 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
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 difcult to analyze in a systematic rather than an anecdotal way.
This was ethnographic-type data that was not susceptible to statistical analy-
sis 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 nego-
tiation 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 rst 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, per-
haps, more troubling to legal academics because they are not typically trained
to assess these factors. To better describe some of the methodological chal-
lenges 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 specic to the legal context: the problem of condentiality and
privilege.
1. Condentiality/Privilege
The attorney-client privilege protects the relationship between lawyers and clients.
It is likely to be very difcult 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 mal-
practice 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 orig-
inal legal matter? Not only might a researchers presence at a client meeting
be construed as a waiver of the attorney-client privilege, but sharing conden-
tial information outside of the relationship could be detrimental to the rela-
tionship itself.
Indeed, some attorneys are actually bound in what they can discuss by
their ofce policies. For example, I interviewed lawyers working in the
United States Attorneys Ofce; this ofce has strict rules on who can talk to
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 insti-
tutional 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 condentiality 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 negotia-
tion as would be helpful because of condentiality 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 aca-
demic study of and reection on their behavior might somehow be different
kinds of people with different styles and experiences of negotiating. My sam-
ple, 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 nal 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 pre-
sented themselves that could have skewed my results. As noted above, I net-
worked with other attorneys to nd my interviewees, and although I assured
them of condentiality, they may have been concerned about information get-
ting back to their peers.
4. External Validity
In my research described above, I sought to determine whether or not crimi-
nal plea bargaining could aptly be considered negotiation per se by explor-
ing 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 ndings about plea bar-
gaining t within literature about the process of negotiation. My empirical
research was unquestionably anecdotal. I did not have a sufcient sample size
to draw any scientic 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 signicant and dispositive in any scientic 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 Schneiders (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 ndings. 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 difcult to generalize
the ndings 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, eld research in legal negotiation cannot offer any possibility for
controlled experiments. Legal ethics (not to mention logistics) make assign-
ment 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 negotia-
tions, conducted by rst-year students at New York University law school, are
videotaped. Additionally, all participants in the negotiations ll out a detailed
survey immediately after completing the negotiation. The research centers on
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 re-
search also presents more straightforward challenges: coding the behavior of
individuals in a forty-ve 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 reects 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 nego-
tiating 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 gures make
decisions (Tyler & Lind 1992). When a third-party authority is involved, par-
ties have been shown to be more satised with and accepting of outcomes
when they are arrived at through the use of fair processes. Indeed, individu-
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 them-
selves 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 con-
servative, risk-averse eld, so that we are able to amass more research on real
attorneys and enable our empirical ndings 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.
Notes
1. I leave aside for this article the debate over what empirical means. First, a wide gulf sep-
arates qualitative from quantitative empiricism in the law; and then, there is hearty dis-
agreement over what may be considered qualitative empirical data. For the purpose of this
discussion, I dene 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
Westlaws law review database.
4. Russell Korobkin (2002: 1038) identied four potential sources of empirical data on con-
tract 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 rst category is not relevant to the negotiation context, as very few opin-
ions 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 circu-
lation and audience than more mainstream law reviews.
6. Korobkin raises a related point when he notes that a single empirical study might be less
fullling 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: 1112.
Ayres, Ian (1991). Fair driving: gender and race discrimination in retail car negotiations.
Harvard Law Review, 104: 817872.
Ayres, Ian (1995). Further evidence of discrimination in new car negotiations and estimates
of its cause. Michigan Law Review, 94: 109147.
Babcock, Linda & Loewenstein, George (1997). Explaining bargaining impasse: the role of
self-serving biases. Journal of Economic Perspectives, 11: 10926.
Babcock, Linda & Pogarsky, Greg (1999). Damage caps and settlement: a behavioral
approach. Journal of Legal Studies, 28: 341370.
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: 141157.
Birke, Richard & Fox, Craig R. (1999). Psychological principles in negotiating civil settle-
ments. Harvard Negotiation Law Review, 4: 157.
Blasi, Gary (2002). Advocacy against the stereotype: lessons from cognitive social psychol-
ogy. UCLA Law Review, 49: 12411281.
Bodie, Matthew T. (2004). Questions about the efciency of employment arbitration agree-
ments. Georgia Law Review, 39: 182.
Busch, Marc L. & Reinhardt, Eric (2000). Bargaining in the shadow of the law: early settle-
ment in GATT/WTO Disputes. Fordham International Law Journal, 24: 158172.
Cohen, Jonathan R. (1999). Advising clients to apologize. Southern California Law Review,
72: 10091069.
Dallas, Lynne L. (2003). The multiple roles of corporate boards of directors. San Diego Law
Review, 40: 781820.
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: 419514.
Dickinson, David L. (2004). A comparison of conventional, nal-offer, and combined arbi-
tration for dispute resolution. Industrial and Labor Relations Review, 57: 288301.
Ehrenreich Brooks, Rosa (2003). The new imperialism: violence, norms, and the rule of
law. Michigan Law Review, 101: 22752340, 2308.
Epstein, Lee & King, Gary (2002). Empirical research and the goals of legal scholarship: The
rules of inference. University of Chicago Law Review, 69: 1133.
Fisher, Roger & Jackson, William (1993). Teaching the skills of settlement. Southern
Methodist University Law Review, 46: 19851993.
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: pre-
serving client-centered advocacy in the negotiation context. UCLA Law Review, 34: 811862.
Guthrie, Chris (1999). Better settle than sorry: The regret aversion theory of litigation behav-
ior. University of Illinois Law Review, 1999: 4390.
Hoffmann, Joseph L. & Kahn, Marcy L. (2001). Plea bargaining in the shadow of death.
Fordham Law Review, 69: 23132391.
Hollander-Blumoff, Rebecca (1997). Getting to guilty: Plea bargaining as negotiation.
Harvard Negotiation Law Review, 2: 115148.
Jolls, Christine, Sunstein, Cass R., & Thaler, Richard (1998). A behavioral approach to law
and economics. Stanford Law Review, 50: 14711550.
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: 705734.
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: 105156.
Korobkin, Russell (1998a). Inertia and preference in contract negotiation: The psychological
power of default rules and form terms. Vanderbilt Law Review 51: 15831651.
Korobkin, Russell (1998b). The Status quo bias and contract default rules. Cornell Law
Review 83: 608687.
Korobkin, Russell (2000). A Positive Theory of Legal Negotiation. Georgetown Law Journal,
88: 17891831.
Korobkin, Russell (2001). A multi-disciplinary approach to legal scholarship: economics,
behavioral economics, and evolutionary psychology. Jurimetrics Journal, 41: 319336, 327.
Korobkin, Russell (2002). Empirical scholarship in contract law: possibilities and pitfalls.
University of Illinois Law Review, 2002: 10331066.
Korobkin, Russell & Guthrie, Chris (1994). Psychological barriers to litigation settlement: an
experimental approach. Michigan Law Review, 93: 107192.
Korobkin, Russell & Guthrie, Chris (1997). Psychology, economics, and settlement: a new
look at the role of the lawyer. Texas Law Review, 76: 77141.
Korobkin, Russell & Guthrie, Chris (2004). Heuristics and biases at the bargaining table.
Marquette Law Review, 87: 795808.
Latif, Elizabeth (2001). Apologetic justice: evaluating apologies tailored toward legal solu-
tions. Boston University Law Review, 81: 289320.
Mnookin, Robert H. & Kornhauser, Lewis (1979). Bargaining in the shadow of the law: the
case of divorce. Yale Law Journal 88: 950997.
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: 362383.
Nadler, Janice (2004). Rapport in legal negotiation: how small talk can facilitate e-mail deal
making. Harvard Negotiation Law Review, 9: 223251.
Neumann, Richard & Krieger, Stefan (2003). Empirical inquiry twenty-ve years after the
lawyering process. Clinical Law Review, 10: 349397.
Oesterle, Dale A. (1986). The negotiation model of tender offer defenses and the Delaware
Supreme Court. Cornell Law Review, 72: 117157.
Pogarsky, Greg & Babcock, Linda (2001). Damage caps, motivated anchoring, and bargaining
impasse. Journal of Legal Studies, 30: 143159.
Rachlinski, Jeffrey J. (1996). Gains, losses, and the psychology of litigation. Southern
California Law Review, 70: 113184.
Rossi, Jim (2001). Bargaining in the shadow of administrative procedure: the public interest
in rulemaking settlement. Duke Law Journal, 51: 10151058.
Schneider, Andrea Kupfer (2002). Shattering negotiation myths: empirical evidence on the
effectiveness of negotiation style. Harvard Negotiation Law Review 7: 143233.
Subramanian, Guhan (2003). Bargaining in the shadow of takeover defenses. Yale Law
Journal, 113: 621686.
Tyler, Tom R., & Lind, E. Allan (1992). A relational model of authority in groups, in Mark
Zanna, editor, Advances in Experimental Social Psychology, 25: 115191, New York:
Academic Press.
Wachter, Michael L. & Cohen, George M. (1988). The law and economics of collective bar-
gaining: an introduction and application to the problems of subcontracting, partial closure,
and relocation. University of Pennsylvania Law Review, 136: 13491417.
Methodologies for Studying Personality Processes in
Interpersonal Conict
LAURI A. JENSEN-CAMPBELL and WILLIAM G. GRAZIANO
Interpersonal conict is an inevitable part of life. In 1956, Muzafer Sherif

noted conict . . . has no simple cause, nor is mankind yet in sight of a cure
(54). The inevitability of conict stems from three apparently-panhuman psy-
chological tendencies: 1) People differ in their attitudes, beliefs, knowledge,
and life experiences; 2) Such differences induce people to be egocentric, and
often to have difculty 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 briey 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 cata-
lysts for conicts whenever people interact.
For this perspective, conict 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 dening interpersonal
conict (Laursen & Collins 1994; Shantz & Hartup 1992). This mutual oppo-
sition, like many relationship interactions, is typically dyadic in nature, but
may involve groups of persons. Interaction patterns present in certain rela-
tionships (e.g., parent/child) may be more likely to produce repeated conicts
than patterns in others (Furman & McQuaid 1992; Patterson 1982). Similarly,
certain relationships may evoke different patterns of conict 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 the-
ory 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 sit-
uation (e.g., personality) (Cantor & Kihlstrom; Jones 1985; Mischel 1973).

324 LAURI A. JENSEN-CAMPBELL AND WILLIAM G. GRAZIANO
Even recent conict theories have focused mainly on the surrounding contexts
involved in conict (e.g., Shantz & Hartup 1992).
Situational context provides only part of the conict story. Given that inter-
personal conict is dened as two or more individuals opposing one another,
personality differences should also contribute signicantly to the dyadic rela-
tion of individuals. Here is one major source of individual differences in atti-
tudes, beliefs, and life experiences discussed in the previous denition of
conict. It is too simplistic and atheoretical, however, to treat personality as
a non-interacting set of individual differences. Instead, personality may be
more usefully dened 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 ten-
dencies. 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 inuence within conict situations. His rst level consists of comparative
dimensions of personality like the Big Five personality traits that are not con-
ditional or dependent on the situation. In other words, Level I traits involve
what a person has and brings to the conict situation. The second level
involves context-dependent strategies, plans, and concerns that enable individuals
to resolve conicts. In other words, Level II speaks to what a person does
within a given conict 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 conict at Level II.
When two people are involved in conict, there is interdependence among
effects due to the person, effects due to the partner, and effects due to the rela-
tionship and situation (Kenny & LaVoie 1984; Lewin 1935). Thus, under-
standing the inuence of personality on social conict 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 personalitys contribution to conict behavior. We then focus on sta-
tistical advances that recognize the complex nature of interactions between
individual, their partners, and the situation when studying conict. The meth-
ods reported here are not intended as an exhaustive list for studying person-
alitys contribution to conict, but will provide the reader with a starting
point.
METHODOLOGIES FOR STUDYING PERSONALITY PROCESSES 325
Personality and Conict Behavior
We will also focus on agreeableness, a dimension in the Big Five factor

model of personality, and its relation to interpersonal conict to exemplify the
utility of various methodologies. Obviously, there are many personality
dimensions that are also worthy of study when discussing personalitys con-
tribution to conict 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
nding 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 negotia-
tion and conict behavior (e.g., Barry and Friedman, 1998), this article
focuses on methodologies for studying personality processes in interpersonal
conict. Thus, providing an exhaustive list of personality characteristics that
are related to interpersonal conict processes remains beyond the scope of
this article.
Hypothetical Situations/Role Playing
A classic laboratory technique for studying conict involves the hypothetical

situations/role playing. Aronson and Carlsmith (1968) describe this methodol-
ogy 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 conict situation. For example,
Sternberg and Soriano (1984) used this method to uncover cross-situational
consistencies in the perception and resolution of hypothetical conicts. We
also used this methodology to assess the link between the Big Five personal-
ity dimension of agreeableness and conict 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
Goldbergs (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 cre-
ated stimulus stories representing conict situations. Stories were divided
equally among relationship types (e.g., mother, sibling) (see Jensen-Campbell
et al. 2001 for actual vignettes). The adolescents task was to tell us how
good or bad each choice was in solving conicts in the different situations/
relationships. As predicted, the personality dimension of agreeableness
uniquely predicted the endorsement of both constructive and destructive
conict 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 conict situation presented and to avoid some ethi-
cal problems associated with deception and manipulation of participants per-
ceptions of conict situations. There are also several potential weaknesses
with using these role-playing methodologies (Graziano 1987; Greenberg & Folger
1988: 3960). 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 tech-
nique biased empirical results in the conict literature. Specically, he found
that adolescents responded to hypothetical conicts with ideal resolution
choices (i.e., negotiation). In actual conicts, however, power assertion and
disengagement were used more frequently than negotiation. As a general rule,
responses to hypothetical conicts may be biased toward responses perceived
to be desirable, but social desirability processes themselves are complex, elu-
sive phenomena (Graziano & Tobin 2002). Furthermore, some studies report
convergence. Using a multi-method format that collected data for both hypo-
thetical vignettes and daily diary records across a two-week period, Jensen-
Campbell and Graziano (2001) found convergence between hypothetical conicts
and diary records of actual daily interpersonal conicts. Even so, our out-
comes suggest that reactions to vignettes may have an underlying structure
different from accounts of in-vivo conict behaviors (Graziano 1987: 284
287). In other words, the hypothetical vignette method relies heavily on
simplistic, singular, deterministic views of personality on conict behavior
and fails to attend to contextual factors (e.g., conict partner) that probably
inuence the behavior. (For a related discussion of personality and social
desirability artifacts in conict research, see Graziano & Tobin 2002).
Self-Report Methodologies
Self-report questionnaires typically ask persons to report information about

their conicts during a specic period. Such measures may be useful as
descriptions of individuals global perceptions of their conict. For example,
Jensen-Campbell and her colleagues (2003) asked adolescents to complete
personality measures and the Styles of Conict Resolution (SCR) question-

naire. The personality dimension of agreeableness again predicted global per-
ceptions of appropriate conict behavior.
Potential threats to the validity of self-report summary measures in assess-
ing personalitys contribution to conict 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 con-
tribute to more comprehensive explanations.
First, to establish that personality is systematically related to effects in
social conict, the researcher must show temporality. A study that collects all
measures contemporaneously does not provide for an adequate test of causal-
ity. Prospective longitudinal studies provide clearer evidence of causality by
establishing temporality (Shadish, Cook & Campbell 2001). For example, Jensen-
Campbell and her colleagues (2002) assessed the associations between per-
sonality 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 condence
in the predictive effects of personality on conict behavior than does exam-
ining 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
conict 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 conict resolution tactics
might be an artifact of a correlation between agreeableness and the third vari-
able of social desirability (SD) motives. Perhaps SD, not agreeableness, is the
primary ingredient linking agreeableness to conict 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 conict research because conict is complexly determined and
because in the absence of direct tests, such explanations will remain intu-
itively plausible until diffused.
Collecting both personality data and conict data from the same source is
also problematic. When assessments of personality are not independent from
assessments of conict behavior, articially inated associations among the
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 conict behavior using multi-
ple sources of information. For example, we have asked parents, teachers, and
children all to provide independent assessments of personality and adjust-
ment. Results from Jensen-Campbell and Graziano (2001) found that the en-
dorsement of both negotiation and physical force were related to individual
differences in agreeableness regardless of the source of the personality assess-
ment (See Table 3). Gleason, Jensen-Campbell, and Richardson (2004) also
assessed the relative importance of personality on the use of aggression in
childrens 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 condently that per-
sonality processes tied to agreeableness are indeed related to the use of
aggression in childrens peer relations.
Finally, researchers often asked participants to recall actual conict
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 lter and to aggre-
gate events that occur over an extended period. Henry, Moftt, Caspi,
Langley, and Silva (1994) conducted a longitudinal study on the retrospective
method and found that even when there were signicant relations between
prospective and retrospective reports, the absolute agreement was poor. Moreover,
psychosocial variables like family processes associated with conict produced
the lowest level of agreement. Reis and Wheeler (1991) have suggested that
asking participants to recall actual conicts is problematic because partici-
pants must deal with selection, recall, and aggregation issues.
Selection poses several serious problems if the goal of the research is to
understand actual conict behavior. First, without specic instructions partic-
ipants create their own criteria for including events (Reis & Wheeler 1991).
Moreover, individuals may have difculty 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 per-
sons mood or personality (e.g., conscientiousness) may inuence the selec-
tion and recall of events (Blaney 1986; Gilligan & Bower 1984; Isen 1984;
Reis & Wheeler 1991; Schwarz & Clore 1983). A nal methodological issue
related to self-report measures is the aggregation process necessary for global
measures (e.g., describe your conicts over the last six months). Reis and Wheeler
note that there is little empirical evidence on how individuals combine mul-
tiple interactions into a single rating dimension.
Event-by-Event Recordings
Event-by-event recordings eliminate some of these weaknesses by allowing

the individual to evaluate specic, molecular events. Several relatively new
methods for examining the nature and impact of daily life experience are use-
ful for studying the impact of personality on daily interpersonal conict as
well. In other words, if the goal is to understand daily interpersonal conict
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 sig-
nals usually occur through telephone calls, alarms on PDAs, or the use of
beepers. These signals can be at xed times or they can be randomly gener-
ated. 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 inter-
ested in a specic 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 conicts per day in a diary
study (Jensen-Campbell & Graziano 2000). Other studies have also shown
that adolescents report few conicts (e.g., 7.36 conicts) on average each day
(Laursen 1993). Thus, this method is less useful for studying the variation
within interpersonal conicts (Reis & Wheeler 1991).
In event-contingent recording, individuals must report every event that
meets some predetermined denition (e.g., behavioral opposition). A clear,
precise denition is critical so that all events that meet a specic criterion are
described. Like signal-contingent recording, this method reduces the likeli-
hood of reappraisal. Event sampling, however, is best applied to discrete
events such as conict 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 conict 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 conict that
occurred in briefer interactions. Children were rst assessed on their person-

ality. Approximately eleven months later, we assessed perceptions of every
conict 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 conict, was
signicantly related to evaluations of the individuals adjustment by knowl-
edgeable raters.
In sum, event-by-event recording were explicitly designed to examine nat-
urally occurring daily life experiences (Reis & Wheeler 1991; Suls, Martin,
& David 1998). These methods allow researchers to supplement the usual
paper-and-pencil ratings of conict with basic, quasi-ethological reports about
frequency and patterns of interpersonal conict during daily social life.
Behavioral Observation
Another approach that has been used in conict research is structured behav-
ioral observation in the laboratory. For example, Jensen-Campbell, Gleason,
Adams, and Malcolm (2003) observed the relations between personality and
conict in late childhood by using a board game (that created conict) and
then unobtrusively videotaping and transcribing the events that occurred
within the xed situation. Laboratory observations can also involve less struc-
tured 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 modied ver-
sion of this paradigm to assess personalitys contribution (i.e., empathic accu-
racy) 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 xed activ-
ities. They do not allow for probing the full extent of everyday interpersonal
conicts in adolescence. In other words, the conict behavior is mandated into
existence by denition. Similarly, the obtrusiveness of the observation may
inuence the behaviors displayed. In addition, task demands constrain conict
resolution strategies (Laursen & Collins 1994). Disengagement is usually not
possible in laboratory settings. For example, participants in the Jensen-
Campbell et al. (2003) study were told that if they had a disagreement they
must work it out. Their choices for conict resolution were constrained; they
could not seek the researchers help or leave the experimental session. Thus,
the children could not withdraw from the situation before the experiment was
nished and could not seek out any third-party interventions.
Interpersonal conict is a complex social phenomenon. To gain a better

understanding of conict processes, researchers developed statistical method-
ologies 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 briey some
of these statistical methods and how they can be useful for studying person-
alitys contribution to interpersonal conict. 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 differ-

ences 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 (Jensen-
Campbell & 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 char-
acteristic) and the moderator (e.g., sex) to compare the equivalence of rela-
tions 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 personalitys 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 consci-
entiousness, children with higher agreeableness exhibited better self-regulation
than persons lower on agreeableness. When there were medium or high lev-
els of conscientiousness, there was no evidence agreeableness affected self-
regulation. This outcome suggests some kind of compensatory processes
between the two related personality dimensions of agreeableness and consci-
entiousness in children, as anticipated by developmental theory (Ahadi & Rothbart
1994). Presumably, their more advanced self-regulatory processes help high
agreeable persons generate more positive responses than their peers to inter-
personal conict. Discussing the specics 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 personalitys inuence on daily conicts is also struc-
tured hierarchically. In other words, there are clusters or levels within the data
set. Consider the conicts reported by an individual participant using the
event-by-event recording method. We would expect some association between
the resolutions of conicts across relationship classes and situations within the
same individual. For example, individuals high in agreeableness will consis-
tently resolve their conict episodes more constructively with parents, peers,
and teachers than will their peers. Moreover, conicts that are reported for
members of the same or similar conict dyad(s) (e.g., mother/child; High/
High Agreeable dyads) should also be more highly correlated than other pair-
ings. When data are collected at multiple levels, the data should be treated as
a multi-level or hierarchical model and the use of random coefcient regres-
sion should be used to retain the datas complexity (Cohen, et al. 2003).
In Jensen-Campbell and Graziano (2001), both individuals and their inter-
actions were treated as units of measurement because individual conicts
were collected from a participant across a two-week period. For example, we
rst calculated regression equations that estimated how much each participant
reported systematic ratings with his/her partners based on the sex of the part-
ner. In the second step, the rst-step regression coefcients 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 partici-
pant, we could assess the within-individual information instead of assuming
that each conict or conict partner for the individual were equivalent. In
other words, we could assess more accurately the contribution of an individ-
uals personality while assessing effects due to the interaction and interaction
partner. We specically constructed a two-level model, with variations in
change parameters among interactions within an individual at level one (e.g.,
partners traits) and variation among participants as level two (e.g., partici-
pants 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.,
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 reected
in studies that examine mediation or the processes that underlie the associa-
tion 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 mod-
eling (SEM). For example, SEM allows an investigator to constrain each
model parameter to equal some pre-specied value or to leave that parameter
free to be estimated. Because SEM allows for the a priori specication of
constraints, any specic pattern of structural relations specied by the inves-
tigator reects a hypothesized model that can be empirically tested for its ade-
quacy. SEM also allows any or all of the estimated model parameters to be
estimated between groups. Moreover, SEM can be used to examine individ-
ual differences as well as reciprocal relationships in longitudinal data (e.g., Farrell
1994). Discussing the specic benets and limitations of these statistical pro-
cedures 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 inuences
perceptions of conict in groups (e.g., triads) involves decomposing self- and
other data into perceiver and target effects using Kennys (1994) Social
Relation Model (SRM). Groups can be formed and asked to perform a
conict-inducing task or group decision-making task that creates conict. 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.
For example, Graziano, Hair, and Finch (1997) manipulated simultaneously

group goal structures and group composition to study college students inter-
actions. They assigned participants to triads based on their own and their part-
ners scores on personality, specically agreeableness. The composition of the
groups included either one high-agreeable participant and two low-agreeable
participants (A+AA) or one low-agreeable participant and two high-agree-
able participants (AA+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 them-
selves 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 perceivers average level of the given behav-
ior 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 species the extent
to which an individual elicits consistent reactions across interaction partners.
A signicant 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 conict. It is possible that agreeable persons will be rated as
more cooperative and less argumentative consistently across interaction part-
ners. For example, Graziano, Hair, and Finch (1997) found that compared to
their peers, men high in agreeableness were rated as less likely to resist
teams 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 dened within context. To address these
concerns, we used a multivariate approach for statistical analysis of dyadic
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
inuenced 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 partici-
pants in the H L group crossed their arms signicantly 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 Inter-
dependence 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 conict is dened in terms of mutual opposition (Laursen &
Collins 1994; Shantz & Hartup 1992), we focused on both personality and the
interpersonal aspects of conict when discussing potential methodologies. We
demonstrated that interpersonal conicts contain emergent elements, and may
produce phenomena not easily predicted from characteristics of either the per-
son or the relationship in isolation (e.g., Graziano et al. 1996, Study 2). To
gain a better understanding of conict 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
conict episodes. We need to appreciate these differences by including per-
sonality as well as situations in research. We need to look past simple main
effects models toward models that consider the links among personality, situ-
ated motives, and overt behavior (Finch & Graziano 2001; Graziano, Jensen-
Campbell, & Finch 1997; Graziano, Hair, & Finch 1997). These approaches
will lead to a better understanding of the causes, processes, and outcomes of
interpersonal conict.
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 manu-
script. We thank all of the graduate and undergraduate students who have help
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 9212201 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: 179. 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
Barrick, M.R., & Mount, M.K. (1991). The big ve dimensions and job performance: A meta
analysis. Personnel Psychology, 44, 1: 126.
Barrick, M.R., Mount, M.K. & Judge, T.A. (2001). Personality and performance at the begin-
ning of the new millennium: What do we know and where do we go next? International
Journal of Selection & Assessment, 9, 12: 930.
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
Barry, B., & Friedman, R.A. (1998). Bargainer characteristics in distributive and integrative
negotiation. Journal of Personality and Social Psychology, 74: 345359.
Bentler, P.M. (1990). Comparative t indexes in structural models. Psychological Bulletin,
107: 238246.
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: 905941.
Blaney, P.H. (1986). Affect and Memory: A review. Psychological Bulletin, 99: 229246.
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: 327342.
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.
Csikszentmihalyi, M., & Larsen, R. (1984). Being adolescent: Conict 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 conict, 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 equa-
tion models. Psychological Methods, 3: 412423.
Farrell, A.D. (1994). Structural equation modeling with longitudinal data: Strategies for exam-
ining group differences and reciprocal relationships. Journal of Consulting and Clinical
Finch, J.F. & Graziano, W.G. (2001). Predicting depression from temperament, personality,
and patterns of social relations. Journal of Personality, 69: 2755.
Furman, W. & McQuaid, E.L. (1992). Intervention programs for the management of conict.
in C.U. Shantz & W.W. Hartup, editors.) Conict in child and adolescent development:
402429. 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: 547588.
Hillsdale, NJ: Erlbaum.
Gleason, K.A., Jensen-Campbell, L.A., & Richardson, D. (2004). Agreeableness and aggres-
sion in adolescence. Aggressive Behavior, 30: 4361.
Goldberg, L.R. (1992). The development of markers of the Big Five factor structure. Psychological
Assessment, 4: 2642.
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 depri-
vation: Theoretical, empirical and policy perspectives: 265294. 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: 893903.
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:
13941408.
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: 392404.
Graziano, W.G., Jensen-Campbell, L.A., Hair, E.C. (1996). Perceiving interpersonal conict
and reacting to it: The case for agreeableness. Journal of Personality and Social
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: 893905.
Graziano, W.G., & Tobin, R.M. (2002). Agreeableness: Dimension of personality or social
desirability artifact? Journal of Personality, 70: 695727.
Greenberg, J., & Folger, R. (1988). Controversial issues in social research methods. New York:
Springer Verlag.
Halverson, C.F. (1988). Remembering your parents: Reections on the retrospective method.
Journal of Personality, 56: 435444.
Henry, B., Moftt, T.E., Caspi, A., Langley, J., Silva, P.A. (1994). On the remembrance of
things past: A longitudinal evaluation of the retrospective method. Psychological Assess-
ment, 6: 92101.
Hodges, E.V.E., Malone, M.J., Jr., & Perry, D.G. (1995, March). Behavioral and social
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 con-
ducted at the biennial meeting of the Society for Research in Child Development, India-
napolis, IN.
Ickes, W., Robertson, E., Tooke, W., & Teng, G. (1986). Naturalistic social cognition: Methodology,
assessment, and validation. Journal of Personality and Social Psychology, 51: 6682.
Ickes, W., Stinson, L., Bissonnette, V. (1990). Naturalistic social cognition: Empathic accuracy
in mixed-sex dyads. Journal of Personality and Social Psychology, 59: 730742.
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: 179236. 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 Deecting Aggression. Journal of Research in Personality, 36: 224251.
Jensen-Campbell, L.A., Gleason, K.A., Adams, R., & Malcolm, K.T. (2003). Interpersonal
conict, agreeableness, and personality development. Journal of Personality, 71: 10591085.
(Special Issue: Personality Development).
Jensen-Campbell, L.A. & Graziano, W.G. (2000). Beyond the schoolyard: Relationships as
moderators of daily interpersonal conict. Personality and Social Psychology Bulletin, 26:
923935.
Jensen-Campbell, L.A., & Graziano, W.G. (2001). Agreeableness as a Moderator of Inter-
personal Conict. Journal of Personality, 69: 323362.
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 conict in adolescence. Merrill-Palmer Quarterly, 42: 148164.
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 psy-
chology: 451477. 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
Kenny, D.A., & Kashy, D.A. (1992). Analysis of the multi-trait-multi-method matrix by
conrmatory factor analysis. Psychological Bulletin, 112: 165172.
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: 233265. 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 variabil-
ity with repeated measures data: Methods and applications. Multivariate applications book
series: 124. Mahwah, NJ: Lawrence Erlbaum Associates.
Kenny, D.A. & La Voie, L. (1985). Separating group and individual level effects. Journal of
Laursen, B. (1993). The perceived impact of conict on adolescent relationships. Merrill-
Palmer Quarterly, 39: 535550.
Laursen, B., & Collins, W.A. (1994). Interpersonal conict during adolescence. Psychological
Bulletin, 115: 177209.
Lewin, K. (1935). A dynamic theory of personality. New York: McGraw-Hill.
Mehrabian, A. (1969). Methods and designs: Some referents and measures of nonverbal
behaviors. Behavioral Research Methods and Instruments, 1: 203207.
McAdams, D.P. (1995). What do we know when we know a person? Journal of Personality,
63: 365396.
McDonald, R.P. (2002). Principles and practice in reporting structural equation analyses.
Psychological Methods, 7: 6482.
Mendoza, J., & Graziano, W. (1982). The statistical analysis of dyadic social behavior: A mul-
tivariate approach. Psychological Bulletin, 92: 532540.
Mischel, W. (1973). Toward a cognitive social learning reconceptualization of personality.
Psychological Review, 80: 252283.
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: 269318.
New York: Academic Press.
Shadel, W.G., & Cervone, D. (1993). The Big Five versus nobody? American Psychologist,
48: 13001302.
Shadish, W.R., Cook, T.D., & Campbell, D.T. (2001). Experimental and quasi-experimental designs
for generalized causal inference. Boston: Houghton Mifin.
Shantz, C.U., & Hartup, W.W., editors, (1992). Conict in child and adolescent development.
New York: Cambridge University Press.
Sherif, M. (1956). Experiments in group conict. Scientic American, 195: 5358.
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: 513523.
Schwartz, D., Dodge, K.A., & Coie, J.D. (1993). The emergence of chronic peer victimiza-
tion in boys play groups. Child Development, 6: 17551772.
Schweinle, W.E., Ickes, W., & Bernstein, I.H. (2002). Empathic inaccuracy in husband to wife
aggression: The over attribution bias. Personal Relationships, 9, 2: 141158.
Sternberg, R., & Soriano, (1984). Styles of conict resolution. Journal of Personality and
Social Psychology, 47: 115126.
Suls, J., Martin, R., David, J. (1998). Person-Environment t and its limits: Agreeableness,
neuroticism, and emotional reactivity to interpersonal conict. Personality and Social Psychology
Bulletin, 24: 8898.
Terhune, K.W. (1970). The effects of personality in cooperation and conict, in P. Swingle,
editor, The Structure of Conict. 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: 656669.
West, S.G. & Aiken, L.S. (1997). Toward understanding individual effects in multiple com-
ponent 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: 167209. 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 arti-
cle is as entertaining or as well written as The Heart of Darkness. Conrads
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 psy-
chotic despot. The story parallels ones 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 efdoms without oversight from others. This lack of
structure or unifying paradigm for cross-cultural research explains in large
measure why mainstream researchers have difculty accepting cross-cultural
studies or results, which are often perceived as descriptive or epiphenomenal
ndings with limited rigor or merit.
Several years ago, a group of colleagues, including myself, wrote two chap-
ters 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, conrmatory cross-
cultural research. These chapters have been generally accepted by other schol-
ars doing cross-cultural research, and many of the specic 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 cul-
ture, assessing conceptual, structural, and functional equivalence, and constructing
research designs that rule out alternative explanations for conrmatory results.
However, as I reect 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 con-
duct 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

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 dened research space (Brett et al.
1997). To move towards either end of the dichotomies (positivist versus inter-
pretive epistemologies, etic versus emic perspectives, inductive versus deduc-
tive processes) risks compromising the rigor and validity of ones study.
Indeed, ve years later, I think of cross-cultural research as an endeavor de-
voted to managing the tensions created by these dichotomies. These dicho-
tomies represent competing interests or paradigms, which are valid concerns,
but can pull a researcher off the middle path. When embarking on a cross-
cultural research project, one is always striking a balance between competing
interests, and continuously trying to nd 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 specic a priori hypothe-
ses to operationalize the theory, and then gather data to either conrm or dis-
conrm 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 inductive-
hypothetico processes are really two parts of a larger whole: the inductive-
hypothetico-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 theo-
ries, and then return to the eld for additional observations to rene their the-
ory as it develops (Spradley 1979; 1980). Similarly, deductive researchers
often rene 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 conrmation
or disconrmation) and inductive-hypothetico (conceptual re-evaluation to
rene 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 nd 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 denes 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 rigor-
ous research to conrm cross-cultural similarities and differences in mid-range
theories. Mid-range theories dene a set of phenomena (or constructs), their
interrelationships, and then propose a causal explanation for those relation-
ships (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 moti-
vation or deal-acceptance across cultures. I also assume the researcher has
some notion of whether this theory will hold or not based on cultural dimen-
sions. 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 prole 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 indi-
vidualistic 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 conrmatory research on cross-cultural similarities and differences based
on an a priori cultural explanation, I now explain the two fundamental ten-
sions that need to be managed.
Positivist versus Interpretive Epistemology
The positivist epistemology (or nature of knowledge) contends that reality is

objective (Popper, 1959). It exists external to ones 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 conrm or disconrm theory.
Post-positivists appear to subscribe to the notion of an existing objective real-
ity, yet they have attacked the notion that observations of that reality are neu-
tral (Hanson 1958; Kuhn 1962). They argue that is difcult (or impossible) to
establish an independent test of theory, because a researchers measures will
always be imbued with the theory s/he is trying to test. The interpretive epis-
temology goes even farther by arguing for multiple realities. Interpretivists espouse
the notion of multiple socially-constructed realities. Therefore a researcher and
subjects will dene their own reality, unique to the participants and to the time
of the study (Guba & Lincoln 1994). This perspective presents obvious
difculties 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-posi-
tivists and interpretivists by acknowledging the possibility of multiple reali-
ties and theory-laden observations. Yet, they argue that observations, while
they may not be neutral, are nonetheless inuenced by multiple theories
(Duhem 1962; Cook & Campbell 1979). Single theory laden observations pre-
sent problems for testing that theory, but observations that are multi-theory
laden will not unduly inuence the testing of any particular theory. Multi-
theory 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 peoples unique subjective
realities. Hence there is a dening reality that can be captured by measuring
the perceptions of multiple respondents (Cook & Campbell 1979).
Unfortunately, for cross-cultural researchers these debates about whether obser-
vations are neutral or theory-laden and whether reality is singular and objec-
tive 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
design that offers some independent conrmation or disconrmation of a rel-

evant theory? The best advice I can offer is to construct both theory and mea-
surement 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 dif-
ferences (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 dif-
ferences across cultures, and together as the team questions why certain con-
structs 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 con-
structs 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 anthropol-
ogists and other eld 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 infor-
mation with a broader foundation of knowledge from relevant literature. In
this motivation example, the researchers would examine literature on motiva-
tion, 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 cul-
tures of interest. Managing this balance is essentially confronting the emic
versus etic tension, to which I now turn.
Emic versus Etic Perspectives
Many researchers discuss the emic versus etic perspective as a contrast

between an insiders view and an outsiders view (Triandis 1972; Berry
1980). These terms were rst 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 Pikes delineation, an emic perspective is not only that of the native
insider, but also represents culturally specic 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 epistemol-
ogy, that there are multiple realities, each unique to a particular group of par-
ticipants at a particular point in time. Emic research, such as the thick
descriptions of Geertz (1983), generally focus on a particular culture, striv-
ing for a deep internal analysis of that cultures unique meaning and social
structure.
Etic researchers, on the other hand, examine human behavior from an
external or outsiders perspective, and they tend to study the same phenom-
ena across multiple cultures. Rather than discovering a cultures emergent
structure or model for a phenomenon, etic researchers generally develop and
impose a theory, structure, or model to understand a phenomenon. They gen-
erally espouse a more positivist epistemology, so that the constructs and mea-
sures are considered objective and universal.
Obviously, the etic perspective is more conducive to cross-cultural com-
parisons. 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 cap-
tured 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 differ-
ences 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 expla-
nation 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
constructs, measurement, or the cultural explanation). This perspective is crit-

icized for being insensitive to the nuances of any culture being studied
(Triandis 1992; Berry 1989). I actually nd 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 arms 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 oth-
ers. 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 solu-
tion is to have as much emic insider information possible for all the cul-
tures 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 con-
struction. Now it is necessary to somehow transform these emic and unique
models into an overall etic or generalizable model which will allow for test-
ing 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 cultures emic model. Hence the etic model is the intersection of all the
cultures emic models.
The risk here is under-specication of the phenomenon in a particular cul-
ture of study (James et al. 1982). Yet, this may outweigh the risks of the over-
specied 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 con-
struct. More constructs measured and correlated leads to a greater chance of
type I error or nding 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 underspecied intersection model,
where a researcher fails to nd a relationship that truly does exist (Cook &
Campbell 1979). Since science progresses as a body of knowledge that is modied
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 intersec-
tion 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 rela-
tionship that truly does exist), but minimizes the risks of concluding a rela-
tionship 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 cultures
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 dif-
ferent 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 con-
struct 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 ndings.
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,
conrmatory approach that tests for similarities and differences with an a pri-
ori cultural explanation. I still very much champion the N-way approach where
research questions are operationalized by a team of researchers or key infor-

mants. 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 con-
strained, and my approach reects 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 cross-
cultural research projects are possible. Yet, I believe researchers should be very
cautious about culturally specic constructs or measures in any nal research
design. Although emic knowledge is invaluable for ensuring a valid opera-
tionalization of the research question for any one culture, this needs to be tem-
pered by the need for equivalent models and measures across cultures. The
goal of each researcher or informant should be to transform their emic knowl-
edge into an etic model. Ultimately I would sacrice internal validity for gen-
eralizability, in order to ensure equivalence for measurement and testing.
References
Adams, J.S. (1963). Toward an Understanding of Inequity. Journal of Abnormal and Social
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: 721735.
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: 75129. 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):
120. New York: Plenum.
Cook, T.D. & Campbell, D.T. (1979). Quasi Experimentation: Design and Analysis Issues for
Field Settings. Boston: Hougton-Mifin.
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, 105117. 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): 131152.
James, L.R., Mulaik, S., & Brett, J.M. (1982). Causal Analysis: Assumptions, Models, and
Data. Thousand Oaks, CA: Sage.
Kuhn, T.S. (1962). The Structure of Scientic 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, 116. 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: 167214.
Popper, K.R. (1959). The Logic of Scientic 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. 2638.
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: 189239. Greenwich, CT: JAI Press.
Disparate Methods and Common Findings in the
Study of Negotiation
CARSTEN K.W. DE DREU and PETER J. CARNEVALE
Conict 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, conict and negotiation is studied
in many disciplines within the social sciences, including social and personal-
ity psychology, organizational behavior, communication sciences, economics,
and political science. Yet at the same time, it seems that within each of these
academic disciplines conict 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 conict 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 psychol-
ogy 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 conict between
a manager and her subordinates?
The problem with this type of debate is that it may lead scholars to dis-
count research ndings obtained with the wrong method. When some meth-
ods are more popular and more accepted in some academic disciplines
than in others, this could reduce the extent to which insights about conict
and negotiation transcend academic boundaries. An alternative would be to
assume that no method or technique is perfect and able to answer all ques-
tions at once. Triangulation the use of different methods to study the same
phenomenon offers a solution to deal with deciencies of each method and
technique (Campbell & Stanley 1966).
In this article, we address two tasks. First, we analyze the conict and
negotiation literature to provide an empirical prole 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

352 CARSTEN K.W. DE DREU AND PETER J. CARNEVALE
effects in conict and negotiation that have been studied with different meth-
ods to see whether triangulation indeed works and provides useful insight.
Disparate Methods in the Study of Conict and Negotiation
As is often the case when research is driven by problems surrounding real-life

phenomena, conict and negotiation is studied with different methods, includ-
ing comparative case studies, laboratory experiments, cross-sectional or longi-
tudinal eld 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), multi-
dimensional 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 eld
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 tech-
niques used in the study of conict and negotiation, we coded research arti-
cles concerned with conict or negotiation published in a particular scientic
eld. We distinguished ve scientic areas: political science, economics, com-
munication science, organizational behavior, and social and personality psy-
chology. 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
ve-year period (19972001), a nal 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 Conict 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], Com-
munication 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 scientic disciplines: International Journal of Conict Management
[IJCM], and Group Decision and Negotiation [GDN]. Obviously, this is a rep-
resentative but not comprehensive list of academic outlets for the study of
negotiation and social conict, 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) conict, (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 Cohens K > .80), and discrepancies were solved through discussion, with
the rst author serving as mediator/arbitrator.
A total of 345 articles were coded. Table 1 provides the total number of
articles on conict and negotiation for each of the ve social science disci-
plines, as well as the percentage of the total number of articles published in
that journal in the entire ve-year period. As can be seen, both in absolute
and relative terms, conict 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 conict
and negotiation is because some work published in these journals is concep-
tual in nature, dealing with group decision-making and support systems or reviews
published books.
Table 1. Articles Published and Articles on Conict Published in Top-tier Journals in

Five Social Sciences Discipline Between 1997 and 2001
Articles
Area Conict and Negotiation Percentage of Total
Organizational Behavior 44 2.1
Social & Personality Psychology 28 2.1
Political Sciences 75 16.5
Communication Sciences 13 8.6
Economics 107 14.5
Topical Journals 78 46.2
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 ques-
tionnaires. Over forty per cent included an outcome measure. Far less com-
mon are methods like multi-dimensional scaling, quantitative coding of behavioral
sequences, meta-analysis, multi-level analysis, time series analysis, or partic-
ipant observation. Because the latter type of methods and techniques are less
common outside the domain of conict and negotiation as well, the low num-
bers observed here may reect lack of training in these techniques or their
labor intense character.
Table 2. Methods and Techniques in the Study of Negotiation and Social Conict
Observed Number Percentage
Use of Mathematical Modeling 124 35.9
Computer Simulation 2 0.6
Use of Games for Model Testing 82 23.8
Laboratory Experiment 161 46.7
Use of Outcome Measures 14 140.9
Coding of Communication Processes 41 11.9
Quantitative Coding of Behavioral Sequences 3 0.9
Use of Surveys and Questionnaires 134 38.8
Markov-Chain Analysis 3 0.9
Methods of Culture Comparison 18 5.2
Multi-dimensional Scaling 1 0.3
Multi-Level Analysis 1 0.3
Meta-Analysis 4 1.2
True Field Experiment 1 0.3
Field Research 35 10.1
Analysis of Archival Data 52 15.1
Time Series Analysis 3 0.9
Workshop as Research Tool 5 1.4
Participant Observation 1 0.3
Comparative Case Analysis 14 4.1
Internet as a Method 1 0.3
Methodology for Personality Processes 19 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 tting into any of the above categories.
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
Method X 2(4)
Use of Mathematical Modeling 0.0 0.0 7.7 68.9 48.0 91.29***

Computer Simulation 0.0 0.0 0.0 1.9 0.0 3.04
Use of Games for Model Testing 0.0 0.0 0.0 58.5 20.0 82.53***
Laboratory Experiment 84.1 85.7 76.9 43.4 10.7 86.12***
Use of Outcome Measures 54.5 53.6 38.5 44.3 25.3 13.25**
Coding of Behavioral Sequences 0.0 3.6 0.0 0.0 0.0 8.53*
Coding of Communication Processes 31.8 17.9 53.8 0.9 6.7 53.01***
Use of Surveys and Questionnaires 68.2 89.3 69.2 5.7 14.7 124.15***
Markov-Chain Analysis 0.0 7.1 0.0 0.9 0.0 10.62**
Methods of Culture Comparison 18.2 3.6 23.1 0.0 4.0 27.36**
Multi-dimensional Scaling 0.0 0.0 0.0 0.0 0.0 NA
Multi-Level Analysis 0.0 0.0 0.0 0.0 0.0 NA
Meta-Analysis 2.3 3.6 0.0 0.0 0.0 5.18
True Field Experiment 0.0 0.0 0.0 0.0 0.0 NA
Field Research 11.4 10.7 7.7 0.0 2.7 14.52**
Analysis of Archival Data 4.5 0.0 7.7 2.8 57.3 105.67***
Time Series Analysis 0.0 0.0 0.0 0.0 4.0 7.73
Workshop as Research Tool 2.3 0.0 0.0 0.0 0.0 5.06
Participant Observation 0.0 0.0 0.0 0.0 0.0 NA
Comparative Case Analysis 2.3 0.0 15.4 0.0 10.7 18.39**
Internet as a Method 0.0 0.0 0.0 0.0 1.3 2.56
Methodology for Personality Processes 4.5 21.4 15.4 0.9 1.3 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 scientic 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 nding that within the organizational behavior
area, an equally large majority of studies involved laboratory experiments.
When we compare the relative frequency across disciplines, many signi-
cant differences emerge. Mathematical modeling and the use of games for
model testing is virtually absent in organizational behavior, social and per-
sonality 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
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 any-
where 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 commu-
nication 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 reect different
research questions scholars in different areas pursue. Questions with regard to
the economic correlates of actual war decisions cannot be answered with lab-
oratory 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 condence 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 per-
sonality 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 Tverskys (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-
way value, and the other half were presented with payoffs presented as less
than their highest possible outcome. Thus, although negotiators objective out-
comes 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 loss-
framing 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 psycholog-
ical 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 exam-
ple, to ght and defend the same territory that it would not have been will-
ing to ght to acquire, or to accept greater costs in order to maintain an
international regime than to create it in the rst place. Loss aversion losses
are more painful than equivalent gains are pleasurable could explain these
tendencies as much as it explains the laboratory ndings on gain-loss fram-
ing reviewed above (Levy 1992). A series of in-depth case studies substanti-
ates these claims. For example, Farnham (1992) demonstrates that loss
aversion may well explain Roosevelts decision-making behavior during the
Munich crisis in September 1938. McDermott uses prospect theory to explain
President Carters 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 anothers persons behavior

(Gouldner 1960). Outside the domain of conict and negotiation, social
psychologists and etiologists have discovered that such mimicry is often sub-
conscious 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
conict and negotiation, with its strong pressures to act strategically, similar
tendencies to reciprocate have often been observed. For example, using the
prisoners dilemma game or its variations, researchers in psychology (e.g.,
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 ones 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 others cooperative (inte-
grative) as well as competitive (distributive) behaviors (e.g., Weingart, Hyder,
Genovese & Prietula 1999; Olekalns & Smith 2000). Political scientists in-
cluding Leng (1993) examined archival data across multiple interstate crises
and across extended periods in history. One important nding 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 occur-
rence 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 sufcient trust that others are reciprocators to cooperate with them even
in one-shot, no-communication experiments. Furthermore, a substantial pro-
portion of the population is also willing to punish non-cooperators (or indi-
viduals 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 sta-
ble 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 sug-
gest two immediate observations: rst, the study of negotiation and social
conict is a very small part of some domains (about 2% of published articles
in social psychology as well as organizational behavior; about 8% in com-
munication), and closer to the core of other domains (political science and
economics). Of course, one reason for this may be the existence and attrac-
tion of the specialty outlets such as International Journal of Conict Management,
Group Decision and Negotiation, Journal of Conict Resolution, Negotiation
Journal, and of course, International Negotiation.
Second, disciplines tend to maintain their own way of doing things: labo-
ratory 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 politi-
cal 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 schol-
ars working in different areas are aware of the various methods and tech-
niques that exist and have some basic understanding of their virtues and limitations.
The consistency in research ndings 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 set-
tings, populations, and research methods.
If our effort to organize these two special issues of International Nego-
tiation on method tells us anything, it is that researchers should adopt, or
should continue to employ, triangulation as an approach to validity. In navi-
gation, 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 identied
(Campbell & Fiske 1959). The concept applies quite well not only to mea-
surement, but to all aspects of method. When two or more methods or data
sources converge on a construct, we have greater assurance that our conclu-
sions 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 direc-
tion is the development of collaborative enterprises such as the multi-disci-
plinary 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 conict resolution. Berkeley: Univer-
sity of California Press.
Bazerman, M.H., Magliozzi, T., & Neale, M.A. (1985). Integrative bargaining in a competi-
tive market. Organizational Behavior and Human Decision Processes, 35: 294313.
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: 89112.
Campbell, D.T., & Fiske, D.W. (1959). Convergent and discriminant validation by the multi-
trait-multimethod matrix. Psychological Bulletin, 56: 81105.
De Dreu, C.K.W., & McCusker, C. (1997). Effects of gain-loss frames on cooperation in two-
person social dilemmas: A transformational analysis. Journal of Personality and Social
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: 90107.
Deutsch, M. (1949). A theory of cooperation and competition. Human Relations, 2: 199231.
Farnham, B. (1992). Roosevelt and the Munich Crises: Insights from Prospect Theory.
Political Psychology, 13: 205235.
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: 161178.
Hopmann, P.T. (2002). Negotiating data: Reections on the qualitative and quantitative analy-
sis of negotiation processes. International Negotiation, 7: 6785.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk.
Econometrica, 47: 263291.
Komorita, S.S., Parks, C., & Hulbert, L. (1992). Reciprocity and the induction of cooperation
in social dilemmas. Journal of Personality and Social Psychology, 62: 607617.
Leng, R.J. (1993). Interstate crisis behavior, 18161980: Realism versus reciprocity. New York,
NY: Cambridge University Press.
Levinger, G., & Rubin, J.Z. (1994). Bridges and barriers to a more general theory of conict.
Negotiation Journal, (July): 201215.
Levy, J.S. (1992). Prospect theory and international relations: theoretical applications and ana-
lytical problems. Political Psychology, 13: 283311.
McDermott, R. (1992). The failed rescue mission in Iran: An application of prospect theory.
Political Psychology, 13: 1629.
Olekalns, M. & Smith, P. (2000). Understanding optimal outcomes: The role of strategy
sequences in competitive negotiations. Human Communication Research, 26: 527557.
Ostrom, E. (1998). A behavioral approach to the rational choice theory of collective action.
The American Political Science Review, 92: 122.
Pruitt, D.G. (1981). Negotiation. New York: Academic Press.
Pruitt, D.G. (1998). Social conict, in D. Gilbert, S.T. Fiske, & G. Lindzey (Editors).
Handbook of social psychology (4th ed.), 2. New York: McGraw-Hill: 89150.
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: 366393.
INDEX
Approach Decision support systems 95, 97, 100101

Cross-culture approach 135136 Discovery 7, 1516, 18, 53, 64, 94, 186,
Crosslevel approach 136, 147 189, 270
Multilevel approach 135, 143 Discourse analysis 5, 177179, 181183,
Multimethod approaches 2 185190
Dyadic analysis 1
Bayesian analysis 6869, 76 Dynamic processes 61
Bargaining context 179, 186, 314
Emotion 13, 96, 122, 149161, 202, 212
Case study 23, 57, 80, 100, 165, 167, 173, Empirical research 90, 154, 283, 307, 309,
187, 342 312313, 317
Coding 63, 76, 8586, 105107, 109113, Ethnography 4142, 160
117, 156, 172, 185, 206, 216, 231232, Experimental design 61, 6667, 93,
235, 318319, 353356, 359 99100, 128, 196, 198, 206, 212,
Communication 218220, 273, 289, 298301
communication processes 177, 257, 356 Experimental economics 289291, 295298,
Comparative case studies 100, 165, 169, 302303, 352
174, 352
Conict Field work 314
Conict data 128, 130, 327 Field experiment 193194, 197203, 205207
Conict resolution 49, 53, 5557, 69, 79,
90, 96, 151, 235, 307, 323, 325327, Incentives 2526, 30, 155, 160, 211,
330, 352, 358 213214, 221, 278279, 290294,
Interactive conict resolution 55, 56 296297, 299, 316
Marital conict 124, 126127, 129 Independent variable 8, 1415, 20, 66, 81,
Organizational conict 123124, 151 8586, 88, 94, 130, 138139, 168, 196,
Context 24, 33, 39, 41, 53, 5556, 61, 67, 201, 212, 218, 220, 241, 301, 331
96, 105, 108, 110, 113, 115, 125, 135, Interrupted time-series: see Time series
137138, 145, 152, 156, 160161,
167168, 177179, 182188, 190, Laboratory experiment 1, 66, 193194,
214215, 232, 249, 290, 292, 294295, 199202, 206207, 211213, 216, 221,
303, 307, 311, 314315, 317, 320, 351352, 354356, 359
323324, 334 Language 1, 25, 41, 139, 152, 159,
Crisis 63, 71, 9495, 97100, 152153, 159, 177180, 182184, 186190, 249,
357 274275
Crosslevel approach: see Approach
Cross-culture approach: see Approach Markov chain analysis 257258, 352
Mediation
Data analysis 1, 42, 82, 105, 183, 189, Mediation style 97
234, 267, 269, 278, 290, 301, 313 Meta-analysis 227229, 231, 233236, 281,
Deception 153, 158, 290, 293, 297300, 326 354, 356

362 INDEX
Methods Archival research 79, 8283, 88, 90

Internet-based methodology 274280, Field research 5, 78, 1011, 1315,
283285 1820, 198, 222, 317, 352
Qualitative method 79, 160, 177 Mediation research 9697
Research methods 1, 5, 39, 4647, 50, Negotiation research 1, 5, 19, 39, 46,
54, 90, 122, 130, 177, 227, 275, 285, 136, 149, 177, 215, 231, 233, 236, 319
313, 315, 324, 341, 359 Qualitative research 160
Self-report methodologies 326 Research in natural settings 41, 222
Statistical methods 113, 174, 301, 313, Research methods 1, 5, 39, 4647, 50,
331, 333 54, 90, 122, 130, 177, 227, 275, 285,
Multidimensional scaling 5, 239240, 313, 315, 324, 341, 359
243244, 252253, 352 Research synthesis 228, 235
Multilevel approach: see Approach Social conict research 3, 5, 57, 199,
Multimethod approaches: see Approach 213, 227, 236, 253, 282, 324, 327
Multivariate techniques 240 Reexivity 186
Negotiation Simulation 9394, 97101, 152, 155, 197,

International negotiation 1, 5, 6364, 72, 216, 262, 268, 318, 320, 352
114, 135136, 171, 250, 286, 358359 Social conict 1, 3, 5, 34, 57, 193, 199,
Israeli-Palestinian Negotiation 28 206, 211, 213, 227, 236, 239, 253, 282,
Legal negotiation 307312, 315, 317320 324, 327, 353, 358
Negotiation processes 2425, 43, 4546, Social context 186
5556, 6162, 66, 71, 74, 77, 9597, Subject pool 273, 276, 290, 295296
101, 105106, 113, 115117, 135137, Surveys 1, 46, 90, 121124, 128, 139, 141,
145146, 156, 167, 174, 177178, 182, 161, 198, 213, 247248, 252, 273278,
220, 252, 257, 259, 286, 310, 319320 280281, 283, 285, 291293, 296, 302,
Negotiation texts 179, 188 307, 311314, 317, 320, 352, 354356,
359
Personality 5, 39, 157, 253, 278, 285,
323335, 351353, 355356 Taba 2334, 36
Post-dicting events 62 Test-construction 345
Prenegotiation 49, 53, 57, 167 Time series
Problem-solving workshop 4954, 5657 Time-series designs 61, 66, 70, 75
Process tracing 71-72, 7476 Interrupted time-series 62, 6566, 76
Psychometrics 124
Validity
Qualitative method 79, 160, 177 Validity threats 81, 122, 127, 274, 279,
Questionnaire 3, 55, 121125, 128, 327
130131, 139, 152, 157159, 199200,
220, 293, 317, 326327, 352, 354355 web experiments 277
web survey 276
Research
Action research 49, 51, 5457

Ebook Carnevale, P. Ed., de Dreu, C.K.W. Ed. Methods of Negotiation Research International Negotiation Series

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ebook Carnevale, P. Ed., de Dreu, C.K.W. Ed. Methods of Negotiation Research International Negotiation Series

Uploaded by

Copyright:

Available Formats

METHODS OF NEGOTIATION RESEARCH

International Negotiation Series

Vol. 1 I.W. Zartman (ed.)

Vol. 2 P. Carnevale and C.K.W. de Dreu (eds.)

P. Carnevale and C.K.W. de Dreu

MARTINUS NIJHOFF PUBLISHERS

Authorization to photocopy items for internal or personal use is

PRINTED AND BOUND IN THE NETHERLANDS

Methods of Negotiation Research: Introduction ...................................... 1

Index ........................................................................................................ 361

Carsten K.W. de Dreu:

Ingrid Smithey Fulmer:

Evert van de Vliert:

James A. Wall, Jr.:

PETER J. CARNEVALE and CARSTEN K.W. DE DREU

Main Entry: method

A close look at the many methodological practices in the study of negotiation

International Negotiation Series 2:

The Authors Task

Each author was asked to write a paper about a specic methodological

PETER J. CARNEVALE and CARSTEN K.W. DE DREU

One aim of these special issues of International Negotiation on negotiation

International Negotiation Series 2:

JAMES A. WALL, JR.

International Negotiation Series 2:

the behaviors of the disputants instead of focusing solely on the mediation

Field research furnishes insights, perhaps because it allows me to observe

ask mediators to recall their mediations, observe mediations, review maps of

How do we know what occurs in a real negotiation? We know that labora-

International Negotiation Series 2:

Why It Is Hard To Learn What Really Happens In Negotiations

Crosschecking and Detective Work: Some Examples

Inferences About Intentions

Indeed, in studying negotiation, the question we frequently ask is Why? We

the emphasis on individual Palestinian applications for compensation wasnt

International Negotiation Series 2:

Research Steps and Strategies

additional members of the bargaining team. In two cases, a union negotiator

During observation, the researcher monitors two parallel levels of behavior.

Tracking What You Observe

The biggest challenge of ethnographic research is to keep track of the data

The Scope of the Study

Nearly all of the focus in traditional experimental studies of negotiation is on

be a logistical problem, since real world negotiations do not happen within

Barry, B. and Friedman, R. (1998) Bargainer characteristics in distributive and integrative

The problem-solving workshop is one of the best known methods in the

International Negotiation Series 2:

The Workshop Method

The initial conceptualization by Burton (1969) identied the method as con-

A Method for Applying and Generating Theory

The academic base of early problem-solving workshops was compatible with

range issues that need to be addressed once a settlement is reached. Within

Workshops as a Form of Action Research

If the problem-solving workshop is cast as a method of social research, it

It is unfortunate that the problem-solving workshop is not used more system-

Negotiation is a dynamic process that can be analyzed as a time-series of

International Negotiation Series 2:

Post-dicting Events with Time Series Data

Perhaps the most popular application of time-series techniques is to predict

previous increase or decrease by their opponent. The comparative model

combinations of comparative and trend models. These time-series ndings

Interrupted Time-Series Analysis

The variables used in the correlational analyses above are measured as