Advances in Group Processes [1 ed.] 9780857243300, 9780857243294

Citation preview

ADVANCES IN GROUP PROCESSES

ADVANCES IN GROUP PROCESSES Series Editors: Edward J. Lawler and Shane R. Thye Recent Volumes: Volumes 1–17:

Edited by Edward J. Lawler

Volume 18:

Edited by Edward J. Lawler and Shane R. Thye

Volume 19:

Group Cohesion, Trust and Solidarity – Edited by Edward J. Lawler and Shane R. Thye

Volume 20:

Power and Status – Edited by Shane R. Thye and John Skvoretz

Volume 21:

Theory and Research on Human Emotions – Edited by Jonathan H. Turner

Volume 22:

Social Identification in Groups – Edited by Shane R. Thye and Edward J. Lawler

Volume 23:

Social Psychology of the Workplace – Edited by Shane R. Thye and Edward J. Lawler

Volume 24:

Social Psychology of Gender – Edited by Shelley Correll

Volume 25:

Justice – Edited by Karen A. Hegtvedt and Jody Clay-Warner

Volume 26:

Altruism and Prosocial Behavior in Groups – Edited by Shane R. Thye and Edward J. Lawler

ADVANCES IN GROUP PROCESSES VOLUME 27

ADVANCES IN GROUP PROCESSES EDITED BY

SHANE R. THYE Department of Sociology, University of South Carolina, USA

EDWARD J. LAWLER School of Industrial and Labor Relations, and Department of Sociology, Cornell University, USA

United Kingdom – North America – Japan India – Malaysia – China

Emerald Group Publishing Limited Howard House, Wagon Lane, Bingley BD16 1WA, UK First edition 2010 Copyright r 2010 Emerald Group Publishing Limited Reprints and permission service Contact: [email protected] No part of this book may be reproduced, stored in a retrieval system, transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without either the prior written permission of the publisher or a licence permitting restricted copying issued in the UK by The Copyright Licensing Agency and in the USA by The Copyright Clearance Center. No responsibility is accepted for the accuracy of information contained in the text, illustrations or advertisements. The opinions expressed in these chapters are not necessarily those of the Editor or the publisher. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN: 978-0-85724-329-4 ISSN: 0882-6145 (Series)

Emerald Group Publishing Limited, Howard House, Environmental Management System has been certified by ISOQAR to ISO 14001:2004 standards Awarded in recognition of Emerald’s production department’s adherence to quality systems and processes when preparing scholarly journals for print

CONTENTS LIST OF CONTRIBUTORS

vii

PREFACE

ix

STATUS, NETWORKS, AND OPINIONS: A MODULAR INTEGRATION OF TWO THEORIES Will Kalkhoff, Noah E. Friedkin and Eugene C. Johnsen

1

COLLECTIVE EVENTS, RITUALS, AND EMOTIONS J. David Knottnerus

39

GROUP REFLEXIVITY AND PERFORMANCE Richard L. Moreland and Jamie G. McMinn

63

TRUST AS AN EXPRESSIVE RATHER THAN AN INSTRUMENTAL ACT David Dunning and Detlef Fetchenhauer

97

BEING DIFFERENT OR BEING BETTER?: DISENTANGLING THE EFFECTS OF INDEPENDENCE AND COMPETITION ON GROUP CREATIVITY Jack A. Goncalo and Verena Krause

129

APPLYING A STATUS PERSPECTIVE TO RACIAL/ ETHNIC MISCLASSIFICATION: IMPLICATIONS FOR HEALTH Irena Stepanikova

159

v

vi

CONTENTS

COMPARISON PROCESSES IN SOCIAL EXCHANGE NETWORKS David R. Schaefer and Olga Kornienko

185

THE INCIDENCE OF STRONG POWER IN COMPLEX EXCHANGE NETWORKS Marcel A. L. M. van Assen

205

MULTIPLEX EXCHANGE RELATIONS Ko Kuwabara, Jiao Luo and Oliver Sheldon

239

CORRUPTION AS SOCIAL EXCHANGE Edward J. Lawler and Lena Hipp

269

LIST OF CONTRIBUTORS David Dunning

Department of Psychology, Cornell University, Ithaca, NY, USA

Detlef Fetchenhauer

Department of Economic and Social Psychology, University of Cologne, Cologne, Germany

Noah E. Friedkin

Department of Sociology, University of California at Santa Barbara, Santa Barbara, CA, USA

Jack A. Goncalo

School of Industrial and Labor Relations, Cornell University, Ithaca, NY, USA

Lena Hipp

School of Industrial and Labor Relations, Cornell University, Ithaca, NY, USA

Eugene C. Johnsen

Department of Mathematics, University of California at Santa Barbara, Santa Barbara, CA, USA

Will Kalkhoff

Department of Sociology, Kent State University, Kent, OH, USA

Verena Krause

School of Industrial and Labor Relations, Cornell University, Ithaca, NY, USA

J. David Knottnerus

Department of Sociology, Oklahoma State University, Stillwater, OK, USA

Olga Kornienko

School of Social and Family Dynamics, Arizona State University, Tempe, AZ, USA

Ko Kuwabara

Columbia Business School, Columbia University, New York, NY, USA

Edward J. Lawler

School of Industrial and Labor Relations, and Department of Sociology, Cornell University, Ithaca, NY, USA vii

viii

LIST OF CONTRIBUTORS

Jiao Luo

Columbia Business School, Columbia University, New York, NY, USA

Jamie G. McMinn

Department of Psychology, Westminster College, New Wilmington, PA, USA

Richard L. Moreland

Department of Psychology, University of Pittsburgh, Pittsburgh, PA, USA

David R. Schaefer

School of Social and Family Dynamics, Arizona State University, Tempe, AZ, USA

Oliver Sheldon

Management and Global Business, Rutgers University, Newark and New Brunswick, NJ, USA

Irena Stepanikova

Department of Sociology, University of South Carolina, Columbia, SC, USA

Marcel A. L. M. van Assen

Department of Methodology and Statistics, Tilburg University, The Netherlands

PREFACE Advances in Group Processes publishes theoretical analyses, reviews, and theory-based empirical chapters on group phenomena. The series adopts a broad conception of ‘‘group processes.’’ This includes work on groups ranging from the very small to the very large and on classic and contemporary topics such as status, power, trust, justice, influence, decision making, intergroup relations, and social networks. Previous contributors have included scholars from diverse fields including sociology, psychology, political science, business, philosophy, computer science, mathematics, and organizational behavior. A number of years ago, we began a new trend in the series. Our goal then was to publish a set of interrelated volumes that examine core issues or fundamental themes in the group processes arena. Each volume was to be organized around a particular problem, substantive area, or topic of study, broadly defined to include a range of methodological and theoretical orientations. Previous volumes have included:        

Group Cohesion, Trust, and Solidarity (v. 19) Power and Status (v. 20) Human Emotions (v. 21) Social Identification in Groups (v. 22) Social Psychology of the Workplace (v. 23) Social Psychology of Gender (v. 24) Justice (v. 25) Altruism and Prosocial Behavior in Groups (v. 26)

With volume 27 we return to a format more traditional to the Advances series – one that includes a wide range of chapters addressing diverse theoretical and empirical issues. Volume 27 opens with two chapters that advance important areas of theory in the social sciences. In ‘‘Status, Networks, and Opinions: A Modular Integration of Two Theories,’’ Will Kalkhoff, Noah E. Friedkin, and Eugene C. Johnsen theoretically merge social influence network theory with status characteristics theory. The resulting theory of social influence captures opinion change fueled by salient status distinctions in multiactor task-oriented groups. Not only does this work represent what may be the most broadly applicable theory of social influence to date, it is a textbook illustration of theory growth through ix

x

PREFACE

‘‘modular’’ connections across theory domains. The next chapter entitled ‘‘Collective Events, Rituals, and Emotions,’’ by J. David Knottnerus, offers a new theory that specifies how collective ritual events operate to impact emotional reactions and group commitment. Importantly, this line of work catalogues and specifies the kinds of social encounters that tend to trigger emotional states, group solidarity, and commitment. This new theory will undoubtedly inspire new research in the sociology of emotions and group commitment. The next two chapters scrutinize and examine the scientific evidence for popular beliefs in psychology and economics. Richard L. Moreland and Jamie G. McMinn question the scientific evidence for the presumed (but often untested) benefits associated with ‘‘group reflexivity’’ in their chapter entitled ‘‘Group Reflexivity and Performance.’’ After critically reviewing the evidence both for and against the benefits of group reflexivity on various measures of performance, they report the results from their own line of scientific testing. Overall, this work questions the conceptual clarity and presumed benefits of having a group reflect on its performance. In a similar vein, in ‘‘Trust as an Expressive Rather Than an Instrumental Act,’’ David Dunning and Detlef Fetchenhauer question whether acts of trust occur as an instrumental act, as often presumed in psychology and economics, or as an expressive act. They report a line of experimental evidence that suggests people trust expressively because of the dynamics that surround the act of trust rather than the instrumental outcomes. Both chapters shed critical new light on longstanding beliefs surrounding these phenomena. Working at the intersection of diverse theoretical traditions, the next two chapters examine the interplay between group dynamics and positive outcomes, in the form of group creativity, physical health, and mental well-being. Jack A. Goncalo and Verena Krause ask whether individualism boosts group creativity because of independence or competition in ‘‘Being Different or Being Better?: Disentangling the Effects of Independence and Competition on Group Creativity.’’ This work highlights the distinction between these two constructs and derives a set of testable propositions that explain how independence and competition uniquely impact the creative process in groups. This insightful chapter undoubtedly sparks much future research on the creative process in groups across the social sciences. Next Irena Stepanikova offers a new taxonomy of racial/ethnic misclassification in ‘‘Applying a Status Perspective to Racial/Ethnic Misclassification: Implications for Health.’’ This approach utilizes relative shifts in status to explore how racial/ethnic misclassifications impact physical and emotional health. Reporting data from the Behavioral Risk Factor Surveillance System, she

Preface

xi

finds that the odds of reporting physical and emotional problems increase dramatically for those who suffer racial/ethnic loss. This important chapter has wide reaching implications for not only those who study status dynamics in groups but scientists and practitioners who seek to understand the etiology of physical and emotional well-being. Although unplanned, the volume closes with four chapters that build on or address some facet of social exchange theory. In ‘‘Comparison Processes in Social Exchange Networks,’’ David R. Schaefer and Olga Kornienko extend and explore the more general notion of Thibaut and Kelley’s comparison level, that is, when actors compare actual rewards to their expectations for rewards. They assert that when relations are not independent, network structure can impact an actor’s comparison level and subsequent use of power. A new laboratory experiment offers initial support for this idea. This chapter offers a new way to think about alternatives and promises many new avenues of research for those interested in network power dynamics. Next, Marcel A. L. M. van Assen pushes network exchange theory in new directions in ‘‘The Incidence of Strong Power in Complex Exchange Networks.’’ In this chapter, he examines different large-scale structures, such as small world and tree networks, by simulating networks that contain up to 144 nodes. The primary finding is the incidence of strong power turns critically on network density. This chapter clearly inspire more work laboratory and field work in the areas of power, network dynamics, complexity, and exchange. Finally, the volume closes with two exemplary chapters that pose fresh questions for social exchange theorists and social scientists more generally. In ‘‘Multiplex Exchange Relations,’’ Ko Kuwabara, Jiao Luo, and Oliver Sheldon explore how overlapping relations – as when coworkers become friends or vested business associates – affect the nature of social relations more generally. After providing a comprehensive review of the scant literature on multiplex relations, the authors illustrate how established social exchange principles and dynamics can deepen our understanding of this phenomenon. The creative contribution of this work is to use social exchange theory to provide key questions and issues to guide the theoretical study of multiplex relations. Finally, Edward J. Lawler and Lena Hipp focus the lens of social exchange theory on a new empirical problem in ‘‘Corruption as Social Exchange.’’ Building upon classic work regarding the forms of social exchange, they explain why ‘‘reciprocal forms’’ of corruption often evolve into more secure ‘‘negotiated’’ or ‘‘productive’’ forms of exchange. Ironically, however, this evolution tends to leave the

xii

PREFACE

corrupt exchange more open to detection. In the end the authors show how corrupt exchanges are intrinsically unstable. An important aspect of this work is that it applies both theoretical and empirical principles of exchange to the universal problem of corruption. Shane R. Thye Edward J. Lawler Volume Co-Editors

STATUS, NETWORKS, AND OPINIONS: A MODULAR INTEGRATION OF TWO THEORIES Will Kalkhoff, Noah E. Friedkin and Eugene C. Johnsen ABSTRACT This chapter focuses on two theories in the landscape of research on social influence – status characteristics theory and social influence network theory – between which heretofore there has been little communication. We advance these two approaches by dovetailing them in a ‘‘modular integration’’ that retains the assumptions of each theory and extends their scope of application. Here, we concentrate on the extension of status characteristics theory to multiactor task-oriented groups and develop new insights on the effects of status characteristics in such groups. We address the implications for opinion changes of status differentiations in which some individuals are deemed more socially worthy and capable than others.

The study of persons’ opinions and how they form and change has been a major focus of research across fields within the social sciences. Topical investigations are diverse, spanning a broad range of social issues and problems, including gender equality (Bolzendahl & Myers, 2004), climate change (Brewer, 2005), public elections (Sciarini & Hanspeter, 2003), Advances in Group Processes, Volume 27, 1–38 Copyright r 2010 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0882-6145/doi:10.1108/S0882-6145 (2010)0000027004

1

2

WILL KALKHOFF ET AL.

interracial relations (Patchen, Davidson, Hofmann, & Brown, 1977), social security benefits (Groskind, 1988), and the death penalty (Bohm, 1992), to name a few. Basic research on social influence processes has contributed to our understanding of opinion formation and change, which is also one of the legacies of social psychology in both of its parent fields, psychology and sociology. In psychology, the early work on social influence tended to employ a ‘‘main effects’’ approach that emphasized characteristics of the influence source (e.g., Sherif, 1935, 1936; Asch, 1951, 1952, 1955, 1956; Hovland, Janis, & Kelley, 1953). Subsequent work has moved toward a ‘‘more multifactoral and interactive conceptualization of social influence’’ (Crano, 2000, p. 74). Contemporary psychological research on social influence, as it bears on processes underlying the formation and change of opinions, emphasizes the effects of various individual and group-level factors in the immediate context within which influence occurs. In sociology, by contrast, a defining feature of theories of social influence is their consideration of both the immediate microsocial frame in which influence processes occur and the larger macrostructural context. This is clearly evident in the two theories that are our present focus – status characteristics theory (SCT) and social influence network theory (SINT). SCT, part of the larger expectation states research program, is a multilevel theory of social inequalities that explains how macrostatus dimensions give rise to status microstructures in goal-oriented face-to-face encounters, such as committee meetings, jury deliberations, and focus group discussions (e.g., Berger, Hamit Fisek, Norman, & Zelditch, 1977). Through the process of status generalization, individuals in goal-oriented groups develop a shared set of performance expectations for one another based on their external status characteristics (e.g., their race, class, gender, and education). The underlying distribution of performance expectations in a group is evidenced in its behavioral power and prestige order (or status microstructure), such that higher status group members have greater influence over the group’s decisions. SINT, like SCT, is a multilevel treatment of social influence (Friedkin & Johnsen, 1990, 1999; Friedkin, 1998). At the microlevel, SINT presents a cognitive explanation of how people weigh and combine the opinions of influential others. At the macrolevel, it explains how social influence networks enter into and constrain the opinion formation process. The theory is unrestricted with respect to the basis of the direct interpersonal influences that define the influence network and, as such, it may be easily integrated with other theories that emphasize particular bases of social influence. ‘‘Differentiated lines of research [on related phenomena] set up pressures for intellectual integration of the resulting theories and findings’’

Status, Networks, and Opinions

3

(Fararo & Skvoretz, 1987, pp. 1183–1184). SINT and SCT have each achieved a sufficient degree of ‘‘theoretical cumulation’’ (Turner, 1989) that creates an opportunity for an integrative advancement. Our present work is a response to this opportunity: we seek to bring SINT and SCT into formal relationship with one another by integrating them. We concentrate on an advancement of one link of the tripartite integration of SCT, affect control theory, and SINT proposed by Friedkin and Johnsen (2003). Ridgeway and Smith-Lovin (1994) have concentrated on the advancement of the integration of status characteristic theory and affect control theory. Given two or more related theories, a greater theoretical unification may occur either through elimination, subsumption, or modular integration. The first two approaches are well known. The third (see Markovsky et al., 2008) treats two or more theories as integral modular components that can be used separately or jointly, as needed, much like different modularized electronic components can be used either alone or together in an integrated circuit for specific applications. We show below that SCT and SINT are well-suited for modular integration. Our integration brings the two theories together for a specific theoretical purpose: the explanation of opinion change in influence networks stratified by persons’ status characteristics. Each theory by itself does not explicitly focus on this situation. On the one hand, SCT describes the social structure of an influence situation in terms of a group’s power and prestige order (as determined by status differentiation), but lacks a model of the social process by which opinion disagreements and agreements evolve over time. On the other hand, SINT offers a model of the social process by which disagreements and agreements evolve, but does not specifically deal with situations in which status differentiation shapes the formation of influence networks that operate as power and prestige orders. The formal mathematical structure of each theory facilitates their modular integration in a manner that affords precise predictions about opinion change in stratified influence networks starting from only two ‘‘knowns’’: actors’ initial opinions and the distribution of status across the network. In what follows, we first provide separate overviews of SCT and SINT. We then explain our rationale for pursuing a modular integration of these two theories. Next we present the details of our integration and illustrate how it can be used to generate predictions for opinion change in stratified influence networks. We conclude with an assessment of our integration, including a discussion of how it deals with the agency of actors and the possibility of social change (Sewell, 1992).

4

WILL KALKHOFF ET AL.

STATUS CHARACTERISTICS THEORY In sociology, the most well-developed account of how sociodemographic status dimensions give rise to behavioral performance inequalities in small groups (e.g., patterns of social influence) is provided by SCT (Berger et al., 1977). Numerous experimental tests and replications have supported the basic claims of the theory (Wagner & Berger, 2002; Kalkhoff & Thye, 2006). In addition, researchers have drawn on SCT to help explain various naturally occurring phenomena, including interaction in school settings (Cohen, 1982; Natriello & Dornbusch, 1983; Cohen & Lotan, 1995, 1997), jury foreperson selection and decision-making (Strodtbeck & Lipinski, 1985; York & Cornwell, 2006), work group performance (Cohen & Zhou, 1991; Bunderson, 2003), wage disparities (Stewart & Moore, 1992), ability-test achievement (Lovaglia, Lucas, Houser, Thye, & Markovsky, 1998), interaction among police officers (Gerber, 2001), motherhood and workplace disadvantages (Ridgeway & Correll, 2004; Correll, Bernard, & Paik, 2007), doctor–patient interactions (Gallagher, Gregory, Bianchi, Hartung, & Harkness, 2005), and career-related outcomes (Correll, 2001; Kogan, McConnell, & Schoenfeld-Tacher, 2004; Correll, 2004). SCT consists of a set of concepts, formal assumptions, and scope conditions. Scope conditions are statements that set the limits of a theory’s domain of applicability; they are abstract statements (i.e., statements not bound to a particular place or time period) that specify the features of social settings in which a theory’s predictions are expected to hold.1 Two scope conditions are essential for the study of status processes: task orientation and collective orientation (Berger, 2007, p. 360).2 SCT applies to social settings where the interactants are primarily focused on a task that they are motivated to accomplish (task orientation), and where they consider it necessary and appropriate to take into account each other’s contributions (collective orientation). Examples of day-to-day tasks that fall within these conditions range from officers who conduct police investigations (Gerber, 2001) to committees that evaluate job applicants (Foschi, 1996). In these kinds of social settings, SCT postulates that status characteristics become ‘‘the central means by which groups of interactants organize their behavior and make collective decisions’’ (Ridgeway & Walker, 1995, p. 282). Status characteristics are a special type of social attribute whose categories (or levels) have unequal status value. If it is widely believed by the members of a culture that it is preferable to occupy one category of an attribute over another, then that attribute operates as a status characteristic in the culture.

Status, Networks, and Opinions

5

The most important insight of SCT is that macrobeliefs that imbue categories of a social attribute with different levels of status value generate corresponding ‘‘expectation states.’’ People who occupy the more preferable social categories are deemed more socially worthy and, thus, more competent. This attribution is an instance of the ‘‘halo effect’’ (Thorndike, 1920). The ‘‘reach’’ of the expectation states associated with a status characteristic identifies its type. Each level of a specific status characteristic corresponds to a similarly valued specific expectation state, often abbreviated SPE[x], which is associated with a limited range of tasks. Such characteristics include distinct areas of expertise, skill, or ability. For example, if the members of a culture tend to believe that it is generally better to be good at math rather than bad at it, and if they expect a math whiz to be competent at tasks requiring mathematical reasoning in particular, then ‘‘math ability’’ operates as a specific status characteristic in that culture. By contrast, each level of a diffuse status characteristic not only is culturally associated with valued specific skills (i.e., the SPE[x]) but also and more importantly corresponds to a valued general expectation state, abbreviated GES[x], which is relevant to a broad array of tasks. Such characteristics include sociodemographic characteristics such as race, class, gender, and educational attainment. For example, gender operates as a diffuse status characteristic when both of the following conditions are met: (i) stereotypes convey that men and women tend to exhibit different specific skills (e.g., men are better at fixing cars and women are better at nurturing); and (ii) those whose gender is culturally associated with greater status value are expected to be more competent at almost any task. Thus, diffuse status characteristics are especially powerful in their ability to maintain macro inequalities by ‘‘transsituationally’’ shaping status microstructures and encounters (Lawler, Ridgeway, & Markovsky, 1993, p. 281). In task-oriented and collectively oriented groups, status characteristics that differentiate group members become activated in the encounter and lead to the formation of performance expectations, which are situationally specific beliefs that group members develop for one another concerning their task-related abilities.3 The impact of an external status characteristic on the formation of performance expectations in a given microencounter depends on its degree of relevance to the task: the more relevant the characteristic, the stronger its effect on performance expectations. The strongest effects occur when group members are directly differentiated on the task ability itself (labeled C in the theory), such as being good or bad at solving math problems when that is the task. Weaker effects occur when the SPE[x] associated with levels of a specific or diffuse status

6

WILL KALKHOFF ET AL.

characteristic imply corresponding levels of task ability (C). The weakest effects occur when the SPE[x] associated with levels of a specific or diffuse status characteristic initially have no connection to levels of task ability. Indeed, one might question whether status characteristics would have any effect at all on performance expectations in such situations, but they do. In one of its most fascinating claims, SCT postulates that activated status characteristics that are initially task-irrelevant will become relevant through a ‘‘burden-of-proof’’ process. That is, group members presume the relevance of any noticeable status characteristic to their task, provided that they do not encounter a convincing reason to believe otherwise.4 However, compared to situations where the individuals’ a priori beliefs (i.e., the SPE[x]) establish a direct linkage between levels of a status characteristic and corresponding levels of task ability, status characteristics that become task-relevant through the burden-of-proof process will produce weaker performance expectations. Individuals have to perform more cognitive work in such situations to create the linkage. In sum, performance expectations are strongest when group members are differentiated on the ability instrumental to task success, weaker when they are differentiated on a specific or diffuse status characteristic that is explicitly relevant to the task through the SPE[x] and weakest when they are differentiated on a nonexplicitly relevant specific or diffuse status characteristic that becomes task-relevant through the burden-of-proof process.

Calculating Performance Expectations The ideas outlined earlier are formalized in a graph formulation of SCT that was introduced in the 1970s by Berger et al. (1977). This significant landmark in the development of SCT has been instrumental to its continued growth and empirical success (Berger, 2000). Most importantly, the graph formulation clarifies the underlying logical and mathematical structure of SCT in a manner that permits precise predictions about behavioral inequalities (e.g., differences in social influence) for a broad range of status situations. While central to the progress of SCT, the graph formulation can be daunting (see, e.g., Miller, 1977) and also cumbersome to use as the number of status characteristics grows large. Fortunately, in recent path-breaking theoretical work, Whitmeyer (2003) has shown that under rather general conditions SCT’s graph formulation can be represented more simply by an

7

Status, Networks, and Opinions

algebraic equation of the following form: " # " # X X l r gðrÞ  exp a hr gðrÞ Fi ¼ exp a r

(1)

r

where Fi is the predicted performance expectation for group member i who possesses any number of status elements, r is one of the following r¼1 r¼2 r¼3

specific task ability (C) explicitly relevant specific or diffuse status characteristic nonexplicitly relevant specific or diffuse status characteristic

lr is the number of status elements of level of relevance r on which actor i possesses the low/disadvantaged state, hr is the number of status elements of relevance r on which actor i possesses the high/advantaged state, a is a constant 8 > < :0865 Berger et al: ð1977Þ (1a) a ¼ :0671 Balkwell ð1991aÞ > : :0770 Fisek; Norman; and Nelson-Kilger ð1992Þ and g(r) is one of the following corresponding functions: 8 4r Berger et al: ð1977Þ >

: 4r 2:618 Fisek et al: ð1992Þ

(1b)5

In turn, the calculated performance expectations for all group members {F1, F2, y, FN} can be linearly transformed into a measure (Bij) of group member i’s resistance to influence from group member j; for example, B12 ¼ b0þb1(F1F2) in the case of a dyadic group. This equation says that group member 1’s resistance to influence from group member 2 is a direct function of group member 1’s expectation advantage over group member 2, which is captured by F1F2.

Multiactor Groups The traditional focus of SCT has been on the status rank of one group member vis-a`-vis a single other group member, even in multiactor

8

WILL KALKHOFF ET AL.

encounters (Ridgeway, 1984; Fisek, Berger, & Norman, 1991; Friedkin & Johnsen, 2003; Kalkhoff, 2005). Here, as indicated earlier, the measure of relative performance expectations is termed expectation advantage (FiFj), which represents the difference between a group member’s expectations for self and those for one other group member. This is calculated separately for each dyadic pair in a group. Fisek et al. (1991), however, argue that it is not appropriate to calculate expectation advantage in multiactor situations because the status of other group members may bias the status difference between any given pair of group members. The argument is Simmelian in flavor. For example, in a task-oriented group, the presence of the president of a company might create less of a status difference between a worker and a manager than might the presence of another worker. On the basis of such concerns, Fisek et al. (1991) present an alternative measure of relative performance expectations: si ¼

1 þ Fi N P ð1 þ Fj Þ

(2)

j¼1

where si is the expectation standing of group member i, Fi and Fj are the performance expectations for group members i and j (calculated using Eq. (1)), and N is the number of group members in the encounter. The result of Eq. (2), si, represents group member i’s proportional share of the total performance expectations that are available in the situation, where si is in the open interval between P0N and 1 (i.e., 0osio1), and the sum of all the si is equal to 1 (i.e., i si ¼ 1). A key feature of this measure, in comparison with the dyad-based expectation advantage, is that it captures the ‘‘biasing effect’’ of other group members’ performance expectations. The measure predicts rates of participation in status heterogeneous task-oriented and collectively oriented groups with a high degree of accuracy (Fisek et al., 1991), and it has also been shown to predict the ‘‘validation’’ of high status actors in multiactor task groups (Kalkhoff, 2005). And as we argue later in text, the measure of expectation standing, si, is the conceptual-mathematical nexus between SCT and SINT. When combined with SINT, SCT is extended to predict opinion change in status heterogeneous multiactor task groups in addition to its already successful use in predicting participation rates and collective validation.

9

Status, Networks, and Opinions

SOCIAL INFLUENCE NETWORK THEORY Like SCT, SINT (Friedkin & Johnsen, 1990, 1999; Friedkin, 1998) recognizes social influence as a fundamental social process, one that is vital to social organization insofar as it serves as an important foundation of individuals’ ‘‘socialization, identity, and decisions’’ (Friedkin, 1998, p. 4). Macrolevel applications of the theory emphasize the significance of shared understandings and agreements for social organization and social control (Durkheim, 1933, 1951) and explain how such understandings and agreements can emerge in complexly differentiated social systems. Microlevel applications of the theory extend the work of symbolic interactionists on the social construction of shared meaning (Mead, 1925, 1934) and provide a precise model of how individuals ‘‘mutually adjust’’ to one another’s attitudes and cognitively integrate conflicting viewpoints. In SINT social process and social structure are dually important. The theory provides an account of how individual and group outcomes emerge from a process unfolding in an influence structure. Over the years of its development, SINT has gained empirical support from both experimental and nonexperimental studies examining various influence-related phenomena, including opinion change (Friedkin, 1998; Friedkin & Johnsen, 1999), choice shift and group polarization (Friedkin, 1999), and norm formation (Friedkin, 2001).

The Formal Model SINT postulates a simple recursive definition for the influence process in groups consisting of N actors (i.e., groups of any size). The matrix algebraic equation is yðtþ1Þ ¼ AWyðtÞ þ ðI  AÞyð1Þ ¼ VðtÞ yð1Þ (1)

(3)

where y is an N by 1 column vector consisting of actors’ initial opinions on an issue; y(t) is an N by 1 column vector of actors’ opinions at time t; A is an N by N diagonal matrix of actors’ susceptibilities or ‘‘openness’’ to influence on the issue (with each value ranging from 0 to 1; i.e., 0raiir1); I is the N by N identity matrix; W is an N by N matrix of direct endogenous interpersonal influences, wij, each of which represents the direct unmediated weight that actor i assigns to the position held by actor j (including the selfweight that actor i assigns to her or his own position, wii); and   (4a) VðtÞ ¼ ðAWÞt þ ðAWÞt1 þ ðAWÞt2 þ    þ ðAWÞ þ I ðI  AÞ

10

WILL KALKHOFF ET AL.

for times t ¼ 1, 2, 3, 4, etc. Each influence weight is nonnegative, ranging from 0 to 1 (i.e., 0rwijr1, 0  vðtÞ ij  1), and the set of weights held by any given actor for all group members, including him- or herself, sums to 1 PN ðtÞ P w ¼ 1, v ¼ 1). (i.e., N ij j j ij Thus Eq. (3) says that actors modify their opinions on an issue by forming a weighted average of group members’ opinions as the microlevel influence process plays out over time. Furthermore, the outcome of this process depends on the macrostructure in which the process occurs. The structure consists of (i) the set of actors in the system; (ii) the direct influence network among them, represented by W; (iii) the openness of the actors to influence, represented by A; and (iv) the actors’ exogenously determined initial opinions on the issue under consideration, represented by y(1). A special case of SINT assumes that the influence structure of the group remains fixed as the process unfolds.6 It also assumes that given A, W, and y(1), the subsequent opinion changes in the group over time are determined. Given that the opinion formation process reaches an equilibrium state, where the process is no longer producing further changes in opinions (i.e., the group has developed either a consensus or a stable pattern of disagreement), Eq. (3) becomes yð1Þ ¼ AWyð1Þ þ ðI  AÞyð1Þ

(4b)

and actors’ settled or final opinions are given by yð1Þ ¼ Vyð1Þ

(5)

where     V ¼ lim VðtÞ ¼ lim ðAWÞt þ ðAWÞt1 þ ðAWÞt2 þ    þ ðAWÞ þ I ðI  AÞ t!1

t!1

(6) is a matrix of reduced-form coefficients describing the total (direct and indirect) interpersonal influence effects that transform initial opinions into final opinions, assuming this limit exists. The coefficients in V (i.e., the vij) are nonnegative, each ranging between 0 and 1 (i.e., 0rvijr1), and the set  N vij j¼1 held by any given P actor for all group members, including him- or herself, sums to 1 (i.e., N j vij ¼ 1). Importantly, the V matrix takes into account not only the direct effect of actor j’s influence over actor i, but also the indirect effects of j on i through the flows of interpersonal influence that arise in the influence structure. Hence, vij gives the overall relative weight of the initial opinion of actor j

Status, Networks, and Opinions

11

in determining the final opinion of actor i. Furthermore, if (IAW) is nonsingular,7 then V can be derived from V ¼ ðI  AWÞ1 ðI  AÞ

(7)

Otherwise, V can be obtained from Eq. (6) when the limit exists.

Operationalizing the Constructs Operationalizations of SINT’s constructs depend on the researcher’s substantive application, thus allowing application of the theory to a wide variety of influence settings and problems. For example, to apply the model to groups where a stable network of interpersonal communications has formed, values for the model’s key constructs, A, W, and even y(1) (if initial opinions are not otherwise known) may be obtained by analyzing the macrostructural features of the communication network. Friedkin (2001) illustrated this approach using Roethlisberger and Dickson’s (1939) classic observations on the Bank Wiring Observation Room (also see Friedkin, 1998). Friedkin used workers’ centrality in the communication network to operationalize workers’ susceptibilities to influence, A.8 He operationalized the direct influence network of the group, W, as a function of these susceptibilities and estimates of the probability of an interpersonal attachment between each pair of workers. Finally, Friedkin derived workers’ initial positions on job-related issues from their locations in the multidimensional social space, as defined by their structural equivalence in the influence network.9 Given these three pieces of information, N) ) are determined. Elsewhere, in a study of small equilibrium positions (y(N groups in experimental settings, Friedkin and Johnsen (1999) operationalized A and W in terms of subjects’ reports of accorded influence and initial positions on an issue. Here our goal is to explain opinion change in task oriented and collectively oriented status heterogeneous groups, with stratified influence networks, that are the theoretical focus of SCT. We argue that this can be achieved by formally integrating SCT and SINT. Specifically, we integrate SCT’s concept of individual expectation standing in SCT with the concept of interpersonal influence in SINT to operationalize the constructs that describe the influence structure of the group.

12

WILL KALKHOFF ET AL.

COMBINING THE THEORIES In presenting our integration of SCT and SINT, we address three central issues. We begin by describing our method of theoretical integration. Second, we establish that the domain of applicability of SCT (i.e., its scope) is compatible with SINT. Finally, we lay out the conceptual details of our integration and illustrate how it can be used to derive precise predictions for opinion change in both complete networks, where all members have status evaluations about all other members, and incomplete networks, where such is not the case.

Modular Integration Efforts to integrate different theoretical agendas in sociology promote the communal vitality of the discipline. In various ways that it can occur, ‘‘integrative metatheoretical and theoretical work undertaken in the spirit of unification y can create bridges between disparate theory enterprises so as to help break down particularistic barriers within sociological theory’’ (Fararo, 1989, p. 175). On the heels of integration, researchers find that they have much more in common and much more to talk about in the way of a common research agenda. Unification, then, is often desirable in sociology and may occur through elimination, subsumption, or modular integration. The first two approaches are well known. The third (Markovsky et al., 2008) treats two or more theories as integral modular components that can be used separately or jointly, as needed, much like different modularized electronic components can be used either alone or together in an integrated circuit for specific applications. In formal logic terms, modular integration can be thought of as theoretical ‘‘conjunction.’’ Markovsky et al. (2008) discuss the merits of a ‘‘modular’’ unification of existing theories. It is the approach that we take. There are other examples. In his status-value theory of power, Thye (2000) demonstrates how the concept of ‘‘value’’ provides a ‘‘natural bridge’’ between exchange theories of power and research on status hierarchies (p. 412). Barnum and Markovsky (2007) use the concept of a ‘‘behavioral interchange pattern’’ (Balkwell, 1991b; Fisek et al., 1991; Skvoretz & Fararo, 1996) to create a link between SCT and self-categorization theory (Turner, 1985). Fararo and Skvoretz (1987) provide a conceptual link between Blau’s (1977) social differentiation theory and Granovetter’s (1973) strength-of-weak-ties principle in a manner that ‘‘includes each of the two theories as [separate]

13

Status, Networks, and Opinions

components’’ (p. 1207). Friedkin and Johnsen (2003) establish theoretical links between SINT, SCT, and affect control theory. Although none of these authors used the term ‘‘modular integration,’’ they do it without saying it. However, not all theories are ideal candidates for modular integration. Those that are must share concepts and constructs by which they can be integrated operationally. As shown later in text, this condition is met in the case of SCT and SINT. Moreover, candidates for integration, in general, must be (i) logically coherent and clear; (ii) not culturally, historically, or geographically bound to particular objects, times, and places (abstract); (iii) empirically supported and explain a range of phenomena (general); and (iv) capable of representation in an unambiguous mathematical language from which precise predictions can be derived (Fararo & Skvoretz, 1987). It is clear that SCT and SINT satisfy these four criteria as well.

Scope Issues The scope conditions of SCT do not prevent it from being linked with SINT. SCT applies to task oriented and collectively oriented groups. We argue that collective orientation poses no obvious problems for the integration of SCT and SINT. Any theory of influence presupposes that actors are at least to some degree open to influence. Past SCT research has employed influence tasks that are presented as having objectively correct and incorrect solutions (see Berger et al., 1977, for a detailed description of the standardized experimental setting). Opinions, however, refer to personal views based on subjective personal judgment. One cannot as clearly regard an ‘‘opinion,’’ per se, as being correct or incorrect. In our broader interpretation of task orientation, though, what matters is not whether group members are motivated to successfully complete a task that has an objectively correct outcome, but whether they are motivated to produce some desirable outcome. Our interpretation of task orientation includes both objective cases (i.e., correct versus incorrect outcomes) as well as subjective ones (i.e., outcomes perceived by an actor to be right, fair, or reasonable). Thus our integration of SCT and SINT applies to settings where actors (i) consider it necessary and appropriate to take into account each other’s input (collective orientation) and (ii) are motivated to produce what each perceives as a right, fair, or reasonable outcome (task orientation). Thus, for example, our integration would not apply to encounters where group members pay no

14

WILL KALKHOFF ET AL.

heed to one another’s contributions or are apathetic about the potential outcome. Such would be the case in a jury consisting of disengaged, obstinate members (failed collective orientation) who could not care less about the result of their deliberations (failed task orientation).

Conceptual Details and Theoretical Illustrations In task-oriented and collectively oriented settings, we argue that higher status actors will exert more influence over group members’ opinions. To proceed with the modular integration of SCT and SINT, we begin by redefining expectation standing (Eq. (2)) in terms amenable to sociometric analysis: S ¼ R1 MP

(8)

where S is an N by N matrix of expectation standing values; M is an N by N sociomatrix representing the interpersonal ties and self (reflexive) ties among actors in a group (1 ¼ presence of a tie; 0 ¼ absence), where we assume that for M ¼ [mij], mii ¼ 1 for all i; P is an N by N diagonal matrix of scaled performance expectations Fi þ 1 (see Eq. (1)); and R is an N by N diagonal matrix with the row totals of MP. The elements in each row of S represent the expectation standing values held by a given actor for self and all other actors with whom she or he has direct ties; the implicit simplifying assumption being that actors only consider the status of those to whom they are directly connected. Recall that in SCT an actor’s expectation standing represents her or his share of the total performance expectations that are available in a group. The entire set of expectation standing values for actors in a group captures its influence structure, referred to in SCT as the underlying power and prestige order of the group, which is represented by S. In SINT, the influence structure of the group is represented by W (see Eq. (3)). Conceptually, then, S can be employed as W. And because the elements in S do not violate the restrictions for the elements in W (values are within the interval [0,1]; each row sums to unity), the expectation standing values in S are straightforwardly assimilated into SINT as the interpersonal influence weights in W. More precisely, the conceptual nexus between SCT and SINT is: S ¼ W. With this step, the two theories are brought into formal relationship with one another. The result is an integrated theory that includes SCT and SINT as modular components.

Status, Networks, and Opinions

15

Fully Connected Networks (M ¼ J, the All-One Matrix) To demonstrate how SCT and SINT interface and to illustrate how the two theories can be used together to predict opinion change in stratified influence networks, we begin by considering the three fully connected networks shown in Table 1. In each of these networks, the actors are differentiated by a single, nonexplicitly relevant diffuse status characteristic (D). In Network 3, the actors are differentiated by a diffuse status characteristic as well as a relevant specific status characteristic (C). Given the status information assumed to be salient in each network (i.e., the summary of actor attributes in Table 1), we can use Eq. (1) and the Berger et al. (1977), Balkwell (1991a), or Fisek, Norman, and Nelson-Kilger (1992) values for a and g(r) [given by Eqs. (1a) and (1b)] to compute a performance expectation for each actor within each network. We can then compute S (the expectation standings) for each network. Detailed calculations for Network 3, the most complex network in Table 1, are presented in the Appendix A. Next, for illustrative purposes, we chose values for the hypothetical initial opinion arrays in Table 1 to represent situations of interpersonal disagreement by levels of the diffuse status characteristic. That is, in both of the four-person networks in Table 1, the average initial opinion for the low diffuse status positions (B and C) is 17, and the average initial opinion for the high diffuse status positions (A and D) is 67. More concretely, if the only salient diffuse status characteristic happens to be gender, the two female members of a new city planning committee might believe, on average, that 17 percent of their traffic engineering budget should be allocated to a particular road improvement project, while the two male members of the committee might believe, on average, that roughly two-thirds of the budget should be allocated to the project. Given the arrays of hypothetical initial opinions and given that our theoretical integration specifies that the values in S (derived using SCT) can be input as the interpersonal influence weights in SINT (i.e., S ¼ W), we can calculate predicted final opinions using Eqs. (7) and (5). The simulated results in Table 1 suggest the general hypothesis that higher status actors exert more influence over group members’ opinions. In the dyad, for example, the higher status actor’s opinion is predicted to shift only 8 points toward the lower status actor’s initial opinion, but conversely, the lower status actor’s opinion is predicted to shift 27 points toward the higher status actor’s initial opinion. The same is true for Network 2 where the two lower status actors are predicted to shift their opinions more towards those of the higher status actors than vice versa, and the overall mean final

C

A

C

A

A

B

B

D

D

B

A ¼ Dþ, C B ¼ D, Cþ C ¼ D, Cþ D ¼ Dþ, C

A ¼ Dþ B ¼ D C ¼ D D ¼ Dþ

A ¼ Dþ B ¼ D :408 :592

:873 :127



:174 6 :174 6 6 4 :174 :174

2 :266 6 :075 6 6 4 :075 :092

32

:326 :326 :174 7 7 7 :326 :326 :174 5 :326 :326 :174

:326 :326 :174

3

:588 :262 :075 7 7 7 :262 :588 :075 5 :321 :321 :266

:321 :321 :092

:240 :111 :111 :538

2



:298 :202 :202 :298



3 :538 :111 :111 :240 6 :273 :328 :126 :273 7 7 6 7 6 4 :273 :126 :328 :273 5

:596 :404

:596 :404

Total Interpersonal Influences (V)

3 :298 :202 :202 :298 6 :298 :202 :202 :298 7 7 6 7 6 4 :298 :202 :202 :298 5 2



Expectation Standings (S ¼ W)

75

 67



3 36 6 22 7 6 7 6 7 4 27 5 33 x ¼ 29:7 59 x ¼ 42

2

54 x ¼ 50:1

36 x ¼ 51:5 2 3 58 6 43 7 6 7 6 7 4 46 5



3 75 6 9 7 6 7 6 7 4 25 5 2

59 x ¼ 42

9 x ¼ 42 2 3 75 6 9 7 6 7 6 7 4 25 5



27

8



3 39 6 13 7 7 6 7 6 4 2 5 26 2

5

3 17 6 34 7 7 6 7 6 4 21 5 2



Change Hypothetical Predicted Final (y(N)y(1)) Initial Opinions Opinions (y(N) ¼ Vy(1)) (y(1))

Notes: Dþ and D denote the high and low states of a nonrelevant diffuse status characteristic, respectively; Cþ and C denote the high and low states of a relevant specific status characteristic, respectively.

(3)

(2)

(1)

Summary of Actor Attributes

Theoretical Predictions for Three Hypothetical Fully Connected Stratified Influence Networks.

Network

Table 1.

16 WILL KALKHOFF ET AL.

Status, Networks, and Opinions

17

opinion of the group (50.1) more closely matches the mean initial opinion of the higher status actors (67) than that of the lower status actors (17). Here, the two actors with higher diffuse status are predicted to produce comparatively more opinion change among their lower status counterparts and to have more sway over settled opinions at the group level. But consider what happens in Network 3 where the two actors who are disadvantaged in terms of diffuse status are now given an advantage in terms of a relevant specific status characteristic. To return to the example used earlier, here it may be the case that the two female members of the city planning committee, though disadvantaged by their gender, have an offsetting status advantage by each possessing a task-relevant degree in budget and fiscal management. In this case, in contrast to what was predicted for Network 2, the actors in positions B and C, who now have a specific status advantage that is stronger (according to SCT) than their diffuse status disadvantage, are predicted to change their initial opinions the least. Note that the mean final opinion of the group (29.7) now much more closely matches the mean initial opinion of the actors with high specific status (17) than that of the actors with low specific status (67). Importantly, this prediction accords with past experimental and nonexperimental SCT research showing that interaction inequalities produced by diffuse status characteristics such as gender, race, and educational attainment can be reduced or even eliminated when high- and low-status group members have countervailing levels of a specific status characteristic (e.g., Cohen & Roper, 1972; Pugh & Wahrman, 1983; Markovsky, Smith, & Berger, 1984; Cohen & Lotan, 1995). Nonfully Connected Networks For the three networks in Table 1, the rows in each S are identical. Equivalence of the rows in S represents a situation where actors hold the same status-based beliefs and expectations about one another. Although past SCT research usually assumes such consensus as an initial condition, for various reasons situations may arise where there exists disagreement on expectations (see, e.g., Troyer & Younts, 1997). Because we assume that actors only consider the status of those to whom they are directly connected (see above), actors may have different sets of expectation standing values if they are not all directly connected to one another; thus, the rows in S may not be identical. Here we demonstrate how situations involving conflicting status beliefs are easily accommodated by our formulation. Consider the two nonfully connected networks in Table 2a. Each involves two four-person subnetworks connected by a ‘‘bridge’’ between positions D

B

D

C

D

C

A

B

A

G

H

F

H

F

E

G

E

0

:272 6 :272 6 6 6 :272 6 6 :214 6 6 6 0 6 6 0 6 6 4 0

2

2 A ¼ Dþ, C :206 B ¼ Dþ, C 6 :206 6 C ¼ Dþ, C 6 6 :206 D ¼ D-, Cþ 6 6 :170 E ¼ Dþ, C 6 6 0 F ¼ D, Cþ 6 6 G ¼ Dþ, C 6 6 0 H ¼ Dþ, C 6 4 0 0

A ¼ Dþ B ¼ Dþ C ¼ Dþ D ¼ D E ¼ Dþ F ¼ D G ¼ Dþ H ¼ Dþ

0

:206 :206 :382 0 0 0 0

0 0 0

0

0

0

0

0

0

0

0

0 0

0

:206 :382 :206 :206 :382 :206

:206 :382 :206

:170 :170 :320 :170

:170 :170 :320 :170

0

:206 :206 :382

0

0

0

3 2

3

:011 :011 :075 :084 :382 :111 :317

:325 :119 :358 :027 :038 :011 :011 7 7 7 :119 :325 :358 :027 :038 :011 :011 7 7 :088 :088 :598 :046 :063 :018 :018 7 7 7 :029 :029 :199 :225 :312 :091 :091 7 7 :008 :008 :058 :066 :679 :086 :086 7 7 7 :011 :011 :075 :084 :382 :317 :111 5

:325 :119 :119 :358 :027 :038 :011 :011

3

:466 :194 :072 :035 :007 :016 :016 7 7 7 :194 :466 :072 :035 :007 :016 :016 7 7 :186 :186 :218 :105 :022 :048 :048 7 7 7 :036 :036 :043 :399 :084 :183 :183 7 7 :018 :018 :021 :194 :286 :223 :223 7 7 7 :016 :016 :019 :173 :091 :471 :199 5

:194 :194 :072 :035 :007 :016 :016

Total Interpersonal Influences (V)

:016 :016 :016 :019 :173 :091 :199 :471

:466 6 :194 6 6 6 :194 6 6 :186 6 6 6 :036 6 6 :018 6 6 4 :016

3 2

6 0 7 7 6 :119 7 6 0 7 6 :119 7 6 6 0 7 7 6 :088 7 6 :170 7 6 :029 7 6 6 :206 7 7 6 :008 7 6 :206 5 4 :011 :011 :206

0

:272 :184 :272 :272

0

0

0 0

:155

0

0 0 0 0 7 7 7 0 0 0 0 7 7 :214 0 0 0 7 7 7 :230 :155 :230 :230 7 7 :272 :184 :272 :272 7 7 7 :272 :184 :272 :272 5

0

:206 :206 :382

0

0 0

0 0 0

0

0

:214 :214 :144

:272 :272 :184

:272 :272 :184

:272 :272 :184

Expectation Standings (S ¼ W)

Notes: Dþ and D denote the high and low states of a nonrelevant diffuse status characteristic, respectively; Cþ and C denote the high and low states of a relevant specific status characteristic, respectively.

(2)

(1)

Summary of Actor Attributes

Theoretical Predictions for Two Hypothetical Nonfully Connected Stratified Influence Networks.

Network

Table 2a.

19

Status, Networks, and Opinions

and E. In the first of these two networks, the actors are differentiated by a single, nonexplicitly relevant diffuse status characteristic (D). In the second network, the actors are differentiated by a diffuse status characteristic as well as a relevant specific status characteristic (C). Using the same procedures as mentioned earlier, we compute S for both the networks. Detailed calculations for the more complex Network 2 are presented in the Appendix B. The simulated results of the influence process are presented in Table 2b. Here again we chose values for the hypothetical initial opinion arrays to represent situations of relative disagreement by levels of the diffuse status characteristic. In both the eight-person networks in Table 2b, the average initial opinion for the low diffuse status positions (D and F) is 87.5, and the average initial opinion for the high diffuse status positions (A, B, C, E, G, and H) is 25.17. Given the arrays of hypothetical initial opinions and the values for W ¼ S (in Eq. (5)), we use the same procedures as above to calculate predicted final opinions. Table 2b.

The Simulated Results of the Influence Process.

Hypothetical Initial Opinions (y1) (1)

(2)

3 10 6 33 7 6 7 6 7 6 24 7 6 7 6 80 7 6 7 6 7 6 28 7 6 7 6 95 7 6 7 6 7 4 37 5 2

Predicted Final Opinions (y(N) ¼ Vy(1)) 3 24 6 30 7 6 7 6 7 6 28 7 6 7 6 38 7 6 7 6 7 6 35 7 6 7 6 48 7 6 7 6 7 4 37 5 2

19

32

x ¼ 40:75 2 3 10 6 33 7 6 7 6 7 6 24 7 6 7 6 80 7 6 7 6 7 6 28 7 6 7 6 95 7 6 7 6 7 4 37 5

x ¼ 34:07 2 3 43 6 48 7 6 7 6 7 6 46 7 6 7 6 62 7 6 7 6 7 6 58 7 6 7 6 76 7 6 7 6 7 4 59 5

19

55

x ¼ 40:75

x ¼ 55:82

Change (y(N)y(1)) 3 14 6 3 7 7 6 7 6 6 4 7 7 6 6 42 7 7 6 7 6 6 7 7 7 6 6 47 7 7 6 7 6 4 0 5 2

13

3 33 6 15 7 7 6 7 6 6 22 7 7 6 6 18 7 7 6 7 6 6 30 7 7 6 6 19 7 7 6 7 6 4 22 5 2

36

20

WILL KALKHOFF ET AL.

Consistent with what we found earlier in our analysis of the fully connected networks, the simulated results for the nonfully connected networks again suggest the general hypothesis that higher status actors exert more influence over group members’ emergent opinions. In Network 1 of Table 2b, the two lower status actors in positions D and F are predicted to shift their opinions more toward those of the higher status actors than vice versa. Positions D and F are predicted to shift their opinions an average of 44.5 points, whereas the higher status positions (A, B, C, E, G, and H) are predicted to shift their opinions an average of only 5.83 points. Furthermore, the overall mean final opinion of the group (34.07) much more closely matches the mean initial opinion of the higher status actors (25.17) than that of the lower status actors (87.5). Here again, then, the higher status actors are predicted to produce comparatively more opinion change among their lower status counterparts and to have more sway over settled opinions at the group level. Yet consider what is predicted to occur in Network 2 of Table 2b where the two low diffuse status actors (D and F) are now advantaged on a relevant specific status characteristic. Opposite what was predicted to occur in the first network, the average opinion change is less for the actors in positions D and F than for the other actors. Taken together, positions D and F are predicted to shift their opinions an average of 18.5 points, whereas the remaining higher status positions are predicted to shift their opinions an average of 26.33 points. Furthermore, even though the two actors in positions D and F are outnumbered, their ‘‘deviant’’ initial opinions carry more weight in determining settled opinions at the group level: the overall mean final opinion of the group (55.82) is about equally close to the mean initial opinion of the two low diffuse status actors (55.8287.5 ¼ 31.68) as it is to that of the six high diffuse status actors (55.8225.17 ¼ 30.65). The ‘‘interaction disability’’ faced by positions D and F in Network 1 is dampened in Network 2 where group members possess levels of a specific status characteristic opposite the levels of diffuse status.

DISCUSSION In this chapter, we have integrated SCT and SINT. The former describes the influence structure of status-differentiated groups, and the latter provides a general model of the social process by which opinion changes unfold in an influence network. When thought of as modular components and combined, the theories can be used to derive precise predictions about opinion change

Status, Networks, and Opinions

21

in both fully connected and nonfully connected networks starting from only two ‘‘knowns’’: actors’ initial opinions on some issue and the distribution of status across the network. The integration of these two theories highlights both the agency of actors and the occurrence of social change. For Sewell (1992), agency is the capacity to exert ‘‘some degree of control over the social relations in which one is enmeshed’’ (p. 20), and the ability to transform social relations (i.e., to effect social change) ‘‘arises from the actor’s control of resources’’ (p. 20; emphasis added). The illustrations of our integration discussed earlier and summarized in Tables 1 and 2 exemplify these ideas. Higher status actors in stratified influence networks do possess a valued resource: presumed knowledge or competence (Ridgeway, 1984). Actors may use this resource and produce social structural change, which occurs through modification of other actors’ social positions or opinions at both the individual and wider systemic levels. One especially promising prospect is that low diffuse status individuals (e.g., females and ethnic minorities) may not only be able to overcome their ‘‘interaction disability’’ in local groups (or subnetworks) through status interventions (see Webster & Whitmeyer, 2001), but through indirect ties may also gain more sway over social positions on a wider systemic scale over time. Our integration describes how such change might occur. Although the model we have presented earlier assumes that the network structure is fixed, this is not by necessity. SCT assumes that status orders, once formed, tend to persist. As such, a model for static structural conditions makes theoretical sense and is parsimonious. However, the model we employ is a special case of a more general SINT model in which actors’ susceptibilities to influence (the A matrix) and interpersonal influences (the W matrix) may change over time (Friedkin & Johnsen, 1990, 2003). Thus, our modular integration of SCT and SINT could be adapted to handle situations where one assumes a dynamic influence network in which the status structure is updated as the process unfolds over time. Another contribution of our integration concerns its treatment of actor attributes. In almost all existing network studies, the effect of attributes on relative influence is assumed to be based on homophily. But SCT addresses heterophily by assuming that influence increases as differentiation increases on status attributes. Thus, by combing SCT with SINT, our work models the relationship between relative influence and heterophily (i.e., differentiation on attributes that actors generally regard as status characteristics). At the same time, clear avenues exist for our formulation to address homophily. For instance, Kalkhoff and Barnum (2000) and Oldmeadow, Platow, Foddy, and Anderson (2003) show how status differentiation and

22

WILL KALKHOFF ET AL.

shared group membership operate jointly in producing influence. More recently, Barnum and Markovsky (2007) show how group membership can be modeled using the same formal mathematical framework employed by SCT. Inclusion of this line of work into our formulation is thus straightforward and would permit examination of relative influence in networks where actors are heterophilous on status attributes and homophilous on group memberships. Lastly, as emphasized earlier, perhaps one of the most important contributions of our formulation is that it deals with the relationship between status differentiation and opinion change in multiactor social systems. Notwithstanding a few past efforts (e.g., Ridgeway, 1984; Kalkhoff, 2005), most SCT-driven research on influence processes has focused on a single pair of interactants. However, as Simmel (1950) points out, ‘‘the sociological situation between the superordinate and the subordinate is completely changed as soon as a third element is added’’ (p. 141), and it may be changed with the addition of a fourth, a fifth, and so on. Finding ways to make sense of that ‘‘third element’’ and beyond has proven difficult for SCT researchers, both theoretically and empirically. Friedkin (1998) explains in general terms why this is problematic for a theory of social influence: A structural social psychology begins with the understanding that an episode of interpersonal influence is not an isolated event, but one that occurs among many other interpersonal influences. In the context of such a system of interpersonal effects, one cannot understand how actors come to hold particular opinions or behave in particular ways without taking the system of effects into account. (p. 34)

This limitation has long been recognized by SCT researchers (see Ridgeway, 1984). Our theoretical integration addresses this problem. The stage is now set for empirical investigation and theoretical evaluation and development of SCT in settings where agreements and disagreements evolve in complexly status-differentiated social systems.

NOTES 1. While not always specified, scope conditions play a vital role in the growth of sociological knowledge by providing the criteria that are necessary for conducting unambiguous tests of a theory. See Walker and Cohen (1985) for the details of this argument. See Foschi (1997) for additional information on the role of scope conditions in evaluating sociological theories. 2. See Berger et al. (1977) for more details on the scope conditions of SCT.

23

Status, Networks, and Opinions

3. Despite a long history of controversy, research suggests that nondifferentiating status characteristics can also affect performance expectations and resultant task behaviors, but only when they are ‘‘explicitly associated with success or failure at the job to be accomplished’’ (Balkwell, 1991a, p. 147). For an overview of empirical research on this topic, see Walker and Simpson (2000). For additional insights on how status processes might occur in status-homogenous groups, see Ridgeway (1988). 4. For tests of SCT’s burden-of-proof assumption, see Moore (1968) and Berger, Cohen, and Zelditch (1972). 5. The three sets of values for a and g(r) produce similar results; thus, the choice of which to use currently appears arbitrary. 6. For an extension of this model that allows susceptibilities to influence (A) and interpersonal influences (W) to change over time, see Friedkin and Johnsen’s (2003) application of the model to affect control theory and status characteristics theory. 7. Eq. (7) calls for the inverse (IAW)1. If (IAW) is not invertible, Eq. (6) is used. These and other computations required by SINT are readily performed in software programs such as Microsoft Excel, MATLAB, GAUSS, or Mathematica. 8. Centrality is a structural attribute of positions or ‘‘nodes’’ in a network, not individuals. In general it captures an actor’s potential for prominence, prestige, or influence based on occupancy of a particular network position. 9. Here structural equivalence occurs if a given pair of actors have the exact same profile of interpersonal weights in W (Friedkin & Johnsen, 1997). The logic of the relationship between structural equivalence and identical initial opinions is explained in Friedkin (2001; see also Friedkin, 1998, pp. 29–30). 10. In an empirical analysis, this theoretical assumption might be tested against a simpler assumption specifying that all actors are maximally open to influence (i.e., aii ¼ 1, or A ¼ I).

ACKNOWLEDGMENT This project was made possible with support from the National Science Foundation (award no. SES-0719310).

REFERENCES Asch, S. E. (1951). Effects of group pressure upon the modification and distortion of judgments. In: H. Guetzkow (Ed.), Groups, leadership, and men (pp. 177–190). Pittsburgh, PA: Carnegie Press. Asch, S. E. (1952). Social psychology. Englewood Cliffs, NJ: Prentice Hall. Asch, S. E. (1955). Opinions and social pressure. Scientific American, 193, 31–35. Asch, S. E. (1956). Studies of independence and conformity: I. A minority of one against a unanimous majority. Psychological Monographs, 70, 70.

24

WILL KALKHOFF ET AL.

Balkwell, J. W. (1991a). Status characteristics and social interaction. In: E. J. Lawler, B. Markovsky, C. L. Ridgeway & H. A. Walker (Eds), Advances in group processes (Vol. 8, pp. 135–176). Greenwich, CT: JAI Press. Balkwell, J. W. (1991b). From expectations to behavior: An improved postulate for expectation states theory. American Sociological Review, 56, 355–369. Barnum, C., & Markovsky, B. (2007). Group membership and social influence. Current Research in Social Psychology, 13, 22–38. Available at http://www.uiowa.edu/Bgrpproc/ crisp/crisp13_3.pdf. Retrieved on January 21, 2008. Berger, J. (2000). Theory and formalization: Some reflections on experience. Sociological Theory, 18, 482–489. Berger, J. (2007). The standardized experimental situation in expectation states research: Notes on history, uses, and special features. In: M. Webster, Jr. & J. Sell (Eds), Laboratory experiments in the social sciences (pp. 353–378). San Diego, CA: Academic Press. Berger, J., Cohen, B. P., & Zelditch, M., Jr. (1972). Status characteristics and social interaction. American Sociological Review, 37, 241–255. Berger, J., Hamit Fisek, M., Norman, R. Z., & Zelditch, M., Jr. (1977). Status characteristics and social interaction: An expectation states approach. New York: Elsevier. Blau, P. M. (1977). Inequality and heterogeneity: A primitive theory of social structure. New York: The Free Press. Bohm, R. M. (1992). Toward an understanding of death penalty opinion change in the United States: The pivotal years, 1966 and 1967. Humanity and Society, 16, 524–542. Bolzendahl, C. I., & Myers, D. J. (2004). Feminist attitudes and support for gender equality: Opinion change in women and men, 1974–1988. Social Forces, 83, 759–789. Brewer, T. L. (2005). US public opinion on climate change issues: Implications for consensusbuilding and policymaking. Climate Policy, 4, 359–376. Bunderson, J. S. (2003). Recognizing and utilizing expertise in work groups: A status characteristics perspective. Administrative Science Quarterly, 48, 557–591. Cohen, B. P., & Zhou, X. (1991). Status processes in enduring work groups. American Sociological Review, 56, 179–188. Cohen, E. (1982). Expectation states and interracial interaction in school settings. Annual Review of Sociology, 8, 209–235. Cohen, E. G., & Lotan, R. A. (1995). Producing equal-status interaction in the heterogeneous classroom. American Educational Research Journal, 32, 99–120. Cohen, E. G., & Lotan, R. A. (1997). Working for equity in heterogeneous classrooms: Sociological theory in practice. New York: Teachers College Press. Cohen, E. G., & Roper, S. S. (1972). Modification of interracial disability: An application of status characteristics theory. American Sociological Review, 37, 643–657. Correll, S. J. (2001). Gender and the career choice process: The role of biased self-assessments. American Journal of Sociology, 106, 1691–1730. Correll, S. J. (2004). Constraints into preferences: Gender, status and emerging career aspirations. American Sociological Review, 69, 93–113. Correll, S. J., Benard, S., & Paik, I. (2007). Getting a job: Is there a motherhood penalty? American Journal of Sociology, 112, 1297–1338. Crano, W. D. (2000). Milestones in the psychological analysis of social influence. Group Dynamics: Theory, Research, and Practice, 4, 68–80. Durkheim, E. (1933). The division of labor in society. G. Simpson (Trans.). New York: Free Press.

Status, Networks, and Opinions

25

Durkheim, E. (1951). Suicide. J. A. Spaulding & G. Simpson (Trans.). New York: Free Press. Fararo, T. J. (1989). The spirit of unification in sociological theory. Sociological Theory, 7, 175–190. Fararo, T. J., & Skvoretz, J. (1987). Unification research programs: Integrating two structural theories. American Journal of Sociology, 92, 1183–1209. Fisek, M. H., Berger, J., & Norman, R. Z. (1991). Participation in heterogeneous and homogeneous groups: A theoretical integration. American Journal of Sociology, 97, 114–142. Fisek, M. H., Norman, R. Z., & Nelson-Kilger, M. (1992). Status characteristics and expectation states theory: A priori model parameters and test. Journal of Mathematical Sociology, 16, 285–303. Foschi, M. (1996). Double standards in the evaluation of men and women. Social Psychology Quarterly, 59, 237–254. Foschi, M. (1997). On scope conditions. Small Group Research, 28, 535–555. Friedkin, N. E. (1998). A structural theory of social influence. New York: Cambridge University Press. Friedkin, N. E. (1999). Choice shift and group polarization. American Sociological Review, 64, 856–876. Friedkin, N. E. (2001). Norm formation in social influence networks. Social Networks, 23, 167–189. Friedkin, N. E., & Johnsen, E. C. (1990). Social influence and opinions. Journal of Mathematical Sociology, 15, 193–205. Friedkin, N. E., & Johnsen, E. C. (1997). Social positions in influence networks. Social Networks, 19, 209–222. Friedkin, N. E., & Johnsen, E. C. (1999). Social influence networks and opinion change. In: S. R. Thye, E. J. Lawler, M. W. Macy & H. A. Walker (Eds), Advances in group processes (Vol. 16, pp. 1–29). Stamford, CT: JAI Press. Friedkin, N. E., & Johnsen, E. C. (2003). Attitude change, affect control, and expectation states in the formation of influence networks. In: S. R. Thye & J. Skvoretz (Eds), Advances in group processes: Power and status (vol. 20, pp. 1–29). Oxford, England: Elsevier. Gallagher, T. J., Gregory, S. W., Jr., Bianchi, A., Hartung, P. J., & Harkness, S. (2005). Examining medical interview asymmetry using the expectation states approach. Social Psychology Quarterly, 68, 187–203. Gerber, G. L. (2001). Women and men police officers: Status, gender, and personality. Westport, CT: Praeger. Granovetter, M. (1973). The strength of weak ties. American Journal of Sociology, 78, 1360–1380. Groskind, F. (1988). A change in public opinion on social security benefits between 1954 and 1983. Social Service Review, 62, 171–175. Hovland, C. I., Janis, I. L., & Kelley, H. H. (1953). Communication and persuasion: Psychological studies of opinion change. New Haven, CT: Yale University Press. Kalkhoff, W. (2005). Collective validation in multi-actor task groups: The effects of status differentiation. Social Psychology Quarterly, 68, 57–74. Kalkhoff, W., & Barnum, C. (2000). The effects of status organizing and social identity processes on patterns of social influence. Social Psychology Quarterly, 63, 95–116. Kalkhoff, W., & Thye, S. R. (2006). Expectation states theory and research: New observations from meta-analysis. Sociological Methods and Research, 35, 219–249. Kogan, L. R., McConnell, S. L., & Schoenfeld-Tacher, R. (2004). Gender differences and the definition of success: Male and female veterinary students’ career and work performance expectations. Journal of Veterinary Medical Education, 31, 154–160.

26

WILL KALKHOFF ET AL.

Lawler, E. J., Ridgeway, C. L., & Markovsky, B. (1993). Structural social psychology: An approach to the micro-macro problem. Sociological Theory, 11, 268–290. Lovaglia, M., Lucas, J. W., Houser, J. A., Thye, S. R., & Markovsky, B. (1998). Status processes and mental ability test scores. American Journal of Sociology 104:195–228. Markovsky, B., Dilks, L., Koch, P., McDonough, S., Triplett, J., & Velasquez, L. (2008). Modularizing and integrating theories of justice. Advances in Group Processes, 25, 211–237. Markovsky, B., Smith, L. F., & Berger, J. (1984). Do status interventions persist? American Sociological Review, 49, 373–382. Mead, G. H. (1925). The genesis of self and self control. International Journal of Ethics, 25, 251–289. Mead, G. H. (1934). Mind, self and society. Chicago: University of Chicago Press. Miller, L., III. (1977). Review of Status characteristics and social interaction: An expectationstates approach, by J. Berger, M.H. Fisek, R.Z. Norman, and M. Zelditch, Jr. Social Forces 56:742–744. Moore, J. C. (1968). Status and influence in small group interactions. Sociometry, 31, 47–63. Natriello, G., & Dornbusch, S. M. (1983). Bringing behavior back in: The effects of student characteristics and behavior on the classroom behavior of teachers. American Educational Research Journal, 20, 29–43. Oldmeadow, J. A., Platow, M. J., Foddy, M., & Anderson, D. (2003). Self categorization, status and social influence. Social Psychology Quarterly, 67, 138–152. Patchen, M., Davidson, J. D., Hofmann, G., & Brown, W. R. (1977). Determinants of students’ interracial behavior and opinion change. Sociology of Education, 50, 55–75. Pugh, M. D., & Wahrman, R. (1983). Neutralizing sexism in mixed-sex groups: Do women have to be better than men? American Journal of Sociology, 88, 746–762. Ridgeway, C. L. (1984). Dominance, performance, and status in groups: A theoretical analysis. In: E. J. Lawler (Ed.), Advances in group processes (Vol. 1, pp. 59–93). Greenwich, CT: JAI Press. Ridgeway, C. L. (1988). Gender differences in task groups: A status and legitimacy account. In: M. Webster & M. Foschi (Eds), Status generalization: New theory and research (pp. 188–206). Stanford, CA: Stanford University Press. Ridgeway, C. L., & Correll, S. J. (2004). Motherhood as a status characteristic. Journal of Social Issues, 60, 683–700. Ridgeway, C. L., & Smith-Lovin, L. (1994). Structure, culture and interaction: Comparing two generative theories. In: B. Markovsky, K. Heimer & J. O’Brien (Eds), Advances in group processes (vol. 11, pp. 213–239). Greenwich, CT: JAI Press. Ridgeway, C. L., & Walker, H. A. (1995). Status structures. In: K. Cook, G. Fine & J. House (Eds), Sociological perspectives on social psychology (pp. 281–310). Boston: Allyn and Bacon. Roethlisberger, F. J., & Dickson, W. J. (1939). Management and the worker. Cambridge, MA: Harvard University Press. Sciarini, P., & Hanspeter, K. (2003). Opinion stability and change during an electoral campaign. International Journal of Public Opinion Research, 15, 431–453. Sewell, W. F. (1992). A theory of structure: Duality, agency, and transformation. American Journal of Sociology, 98, 1–29. Sherif, M. (1935). A study of some social factors in perception. Archives of Psychology, 187, 1–60. Sherif, M. (1936). The psychology of social norms. New York: Harper.

Status, Networks, and Opinions

27

Simmel, G. (1950). The sociology of Georg Simmel. K. Wolff (Ed. & Trans.). Glencoe, IL: Free Press. Skvoretz, J., & Fararo, T. J. (1996). Status and participation in task groups: A dynamic network model. American Journal of Sociology, 101, 1366–1414. Stewart, P. A., & Moore, J. C., Jr. (1992). Wage disparities and performance expectations. Social Psychology Quarterly, 55, 78–85. Strodtbeck, F. L., & Lipinski, R. M. (1985). Becoming first among equals: Moral considerations in jury foreman selection. Journal of Personality and Social Psychology, 49, 927–936. Thorndike, E. L. (1920). A constant error in psychological ratings. Journal of Applied Psychology, 4, 469–477. Thye, S. R. (2000). A status value theory of power in exchange networks. American Sociological Review, 65, 407–432. Troyer, L., & Younts, C. W. (1997). Whose expectations matter? The relative power of first and second order expectations in determining social influence. American Journal of Sociology, 2, 692–732. Turner, J. C. (1985). Social categorization and the self concept: A social cognitive theory of group behavior. In: E. Lawler (Ed.), Advances in group processes (vol. 2, pp. 77–121). Greenwich, CT: JAI Press. Turner, J. H. (1989). Theory building in sociology: Assessing theoretical cumulation. Newbury Park, CA: Sage. Wagner, D. G., & Berger, J. (2002). The evolution of expectation states theories. In: M. Zelditch, Jr. & J. Berger (Eds), Contemporary sociological theories (pp. 41–78). New York: Rowman and Littlefield. Walker, H. A., & Cohen, B. P. (1985). Scope statements: Imperatives for evaluating theory. American Sociological Review, 50, 288–301. Walker, H. A., & Simpson, B. (2000). Equating characteristics and status organizing processes. Social Psychology Quarterly, 63, 175–185. Webster, M., Jr., & Whitmeyer, J. M. (2001). Applications of theories of group processes. Sociological Theory, 19, 250–270. Whitmeyer, J. M. (2003). The mathematics of expectation states theory. Social Psychology Quarterly, 66, 238–253. York, E., & Cornwell, B. (2006). Status on trial: Social characteristics and influence in the jury room. Social Forces, 85, 455–477.

APPENDIX A. OBTAINING PREDICTED FINAL OPINIONS FOR NETWORK 3 IN TABLE 1 Step 1: Calculate a Predicted Performance Expectation for Each Actor Based on Salient Status Characteristics As indicated in column 2 of Table 1, actor A possesses the advantaged state of one initially nonrelevant diffuse status characteristic (Dþ) and the disadvantaged state of one relevant specific status characteristic (C).

28

WILL KALKHOFF ET AL.

Using Eq. (1) with the Balkwell (1991a) values for a and g(r), A’s predicted performance expectation is " FA ¼ exp a "

X

#

"

l r gðrÞ  exp a

r

¼ exp :067

X

# 1 ð3:192

42

X r

# hr gðrÞ "

Þ  exp :067

X

r

# 43

1 ð3:192

Þ

r

¼ exp½:067ð10:189Þ  exp½:067ð3:192Þ ¼ exp½:683  exp½:214 ¼  :302

Because actor D possesses the same status attributes (Dþ and C), actor D’s predicted performance expectation is also .302. Furthermore, in comparison with actors A and D, actors B and C possess the oppositely valued states of both salient status elements; therefore, the predicted performance expectation for actor B and actor C is simply the additive inverse (i.e., the negative) of the predicted performance expectation for actors A and D, or .302.

Step 2: Compute the Matrix of Expectation Standing Values, S, Using Eq. (8) Using the values obtained from Step 1, we first generate a 4 by 4 diagonal matrix, P, whose elements are the scaled performance expectations (Fi þ 1) for each of the four actors (A, B, C, and D) in the group: 2

:698 6 0 6 P¼6 4 0

0 0 1:302 0 0 1:302

0

0

0

0 0 0

3 7 7 7 5

:698

Next we generate a 4 by 4 matrix, M, whose elements represent the interpersonal ties and self (reflexive) ties among the actors in the group. Because Network 3 in Table 1 is fully connected (each actor can directly communicate with every other actor), each element in M is equal to 1. 3 1 1 1 1 61 1 1 17 7 6 M¼6 7 41 1 1 15 2

1 1 1 1

29

Status, Networks, and Opinions

Following Eq. (8), the matrix product of M and P is 2

:698 1:302 1:302 :698

3

6 :698 1:302 1:302 :698 7 7 6 MP ¼ 6 7 4 :698 1:302 1:302 :698 5 :698 1:302 1:302 :698

Matrix R in Eq. (8) is a diagonal matrix whose elements are the row totals from the matrix product of M and P: 4 60 6 R¼6 40

0 0 4 0 0 4

3 0 07 7 7 05

0

0 0

4

2

the inverse of which is 2

R1

:250 0 0 6 0 :250 0 6 ¼6 4 0 0 :250 0

0

0

0

3

0 7 7 7 0 5 :250

Finally, multiplying the inverse of R by the matrix product of M and P (calculated above) gives the matrix of expectation standing values: S ¼ R1 MP 2 :250 0 0 6 6 0 :250 0 6 ¼6 6 0 0 :250 4 2

0 0 0 :174 :326 :326

6 6 :174 :326 :326 6 ¼6 6 :174 :326 :326 4 :174 :326 :326

3 2

:698 7 6 7 6 0 7 6 :698 76 6 0 7 5 4 :698 :698 :250 3 :174 7 :174 7 7 7 :174 7 5 :174 0

1:302 1:302 :698

3

7 1:302 1:302 :698 7 7 7 1:302 1:302 :698 7 5 1:302 1:302 :698

Step 3: Given S, Compute the Matrix of Total Interpersonal Influences Using Eq. (7) Using S to operationalize the matrix of direct endogenous interpersonal influences, W (in SINT), and noting that IAW is nonsingular, we use Eq. (7) to obtain the matrix of total (direct and indirect) interpersonal influences, V. For the values in A in Eq. (7), one might assume, as we do here, that actors’ openness to influence is inversely related to the weight

30

WILL KALKHOFF ET AL.

that each actor assigns to her or his own opinion (i.e., aii ¼ 1–wii).10 With this assumption, and following Eq. (7), the matrix of total interpersonal influences is V ¼ ðI  AWÞ1 ðI  AÞ 02 311 3 2 3 2 :174 :326 :326 :174 :826 0 0 0 1 0 0 0 B6 7C 7 6 7 6 B6 7C 7 6 7 6 :674 0 0 7 6 :174 :326 :326 :174 7C B6 0 1 0 0 7 6 0 B6 7C 7 6 7 6 ¼ B6 7C 76 76 B6 0 0 1 0 7 6 0 7 6 :174 :326 :326 :174 7C 0 :674 0 B6 7C 7 6 7 6 @4 5A 5 4 5 4 :174 :326 :326 :174 0 0 0 :826 0 0 0 1 02 31 3 2 :826 0 0 0 1 0 0 0 B6 7C 7 6 B6 7C 7 6 :674 0 0 7C B6 0 1 0 0 7 6 0 B6 7C 7 6  B6 7C 76 B6 0 0 1 0 7 6 0 7C 0 :674 0 B6 7C 7 6 @4 5A 5 4 0 0 0 :826 0 0 0 1 02 311 02 3 2 31 :144 :269 :269 :144 1 0 0 0 :174 0 0 0 B6 7C 7 6 B6 7C B6 7C 7 6 B6 7C B6 0 1 0 0 7 6 :117 :220 :220 :117 7C B6 0 :326 0 0 7C C B6 7 7 6 B6 7C ¼ B6 7C  B6 76 7C B6 0 0 1 0 7 6 :117 :220 :220 :117 7C C B6 0 0 :326 0 7 B6 7C 7 6 B6 7C @4 5A 5 4 @4 5A :144 :269 :269 :144 0 0 0 1 0 0 0 :174 02 311 02 31 :856 :269 :269 :144 :174 0 0 0 B6 7C B6 7C B6 7C B6 7C B6 :117 :780 :220 :117 7C :326 0 0 7C B6 0 B6 7C B6 7C ¼ B6 7C  B6 7C B6 :117 :220 :780 :117 7C C B6 0 0 :326 0 7 B6 7C B6 7C @4 5A @4 5A :144 :269 :269 :856 0 0 0 :174 31 02 31 02 :174 0 0 0 1:526 :986 :986 :526 7C B6 7C B6 7C B6 7C B6 :326 0 0 7C B6 :429 1:805 :805 :429 7C B6 0 7C B6 7C B6 ¼ B6 7C 7C  B6 C B6 :429 :805 :1805 :429 7C B6 0 0 :326 0 7 7C B6 7C B6 @ 5A 4 @4 5A 0 0 0 :174 :526 :986 :986 1:526 3 2 :266 :321 :321 :092 7 6 7 6 6 :075 :588 :262 :075 7 7 6 ¼6 7 6 :075 :262 :588 :075 7 7 6 5 4 :092 :321 :321 :266

31

Status, Networks, and Opinions

Step 4: Given V and an Array of (Hypothetical) Initial Opinions, Compute Predicted Final Opinions Using Eq. (5) Using the array of hypothetical initial opinions for actors A, B, C, and D in Network 3 (Table 1), and following Eq. (5), predicted final opinions are given by yð1Þ ¼ Vyð1Þ 2 :266 6 6 :075 6 ¼6 6 6 :075 4

:321

:321

:588

:262

:262

:588

:092 :321 3 36 6 7 6 22 7 6 7 7 ¼6 6 7 6 27 7 4 5

:321

:092

3 2

75

3

7 6 7 6 7 :075 7 7 6 9 7 76 7 7 6 7 :075 7 6 25 7 5 4 5 :266

59

2

33

APPENDIX B. OBTAINING PREDICTED FINAL OPINIONS FOR NETWORK 2 IN TABLE 2 Step 1: Calculate a Predicted Performance Expectation for Each Actor Based on Salient Status Characteristics As indicated in column 2 of Table 2a, actor A possesses the advantaged state of one initially nonrelevant diffuse status characteristic (Dþ) and the disadvantaged state of one relevant specific status characteristic (C). Thus A’s predicted performance expectation in this network is equivalent to A’s predicted performance expectation in the preceding example (i.e., FA ¼ .302). Furthermore, because Actor A in Network 2 (Table 2) possesses the same status attributes (Dþ and C) as actors B, C, E, G, and H in this network, the predicted performance expectation for each is .302. Finally, in comparison with this subset of actors, actors D and F in Network 2 (Table 2) possess the oppositely valued states of both salient status

32

WILL KALKHOFF ET AL.

elements; therefore, the predicted performance expectation for actor D and actor F is simply the additive inverse (i.e., the negative) of the predicted performance expectation for actors B, C, E, G, and H, or .302.

Step 2: Compute the Matrix of Expectation Standing Values, S, Using Eq. (8) Generate an 8 by 8 diagonal matrix, P, containing the scaled performance expectations (Fi þ 1) for each of the eight actors in the group: 2 6 6 6 6 6 6 6 P¼6 6 6 6 6 6 4

3

:698

0

0

0

0

0

0

0

0

:698

0

0

0

0

0

0 0

0 0

:698 0

0 1:302

0 0

0 0

0 0

0 0

0 0

0 0

0 0

:698 0

0 1:302

0 0

0

0

0

0

0

0

:698

0 7 7 7 0 7 7 0 7 7 7 0 7 7 0 7 7 7 0 5

0

0

0

0

0

0

0

:698

Generate an 8 by 8 matrix, M, with elements representing the interpersonal ties and self (reflexive) ties among the actors in the group. Because Network 2 in Table 2 is not fully connected, some of the elements in M are zero, signifying the absence of a network tie (i.e., the two actors in question cannot directly communicate with one another). 1 61 6 6 61 6 61 6 M¼6 60 6 60 6 6 40

1 1

1 1

1 1

0 0 0 0

0 0

1

1

1

0 0

0

1 0

1 0

1 1

1 0 1 1

0 1

0 0

0 0

0 0

1 1 1 1

1 1

3 0 07 7 7 07 7 07 7 7 17 7 17 7 7 15

0

0

0

0

1 1

1

1

2

33

Status, Networks, and Opinions

And following Eq. (8), the matrix product of M and P is 2

:698 6 :698 6 6 6 :698 6 6 :698 6 MP ¼ 6 6 0 6 6 0 6 6 4 0

0 0

0 0

0 0

:698 :698

1:302 1:302

:698 :698

3 0 0 7 7 7 0 7 7 0 7 7 7 :698 7 7 :698 7 7 7 :698 5

0

0

0

:698

1:302

:698

:698

:698 :698 1:302 :698 :698 1:302

0 0

0 0

0 0

:698 :698 1:302

0

0

0

:698 :698 1:302 0 0 1:302

:698 :698

0 1:302

0 :698

0

As above, R is a diagonal matrix containing the row totals from the matrix product of M and P: 2

3:396

0

0

0

0

0

0

0

0 0

3:396 0

0 3:396

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

4:094 0

0 4:698

0 0

0 0

0 0

0

0

0

0

0

3:396

0

0

0 0

0 0

0 0

0 0

0 0

0 0

3:396 0 0 3:396

0

0

0

0

0

0 0

0 0

0 0

0 0

0 0

6 6 6 6 6 6 6 R¼6 6 6 6 6 6 4

the inverse of which is 2

R1

6 6 6 6 6 6 6 ¼6 6 6 6 6 6 4

:294 0 0

0

0

:294 0 0 :294

0

0

0

:244

0

0

0

0

0 0

0 0

0 0

0 0

:213 0

0 :294

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

:294 0

0 :294

3 7 7 7 7 7 7 7 7 7 7 7 7 7 5

3 7 7 7 7 7 7 7 7 7 7 7 7 7 5

34

WILL KALKHOFF ET AL.

And multiplying the inverse of R by the matrix product of M and P (calculated above) gives the matrix of expectation standing values: S ¼ R1 MP 2 :294 0 0 0 0 0 0 6 6 0 :294 0 0 0 0 0 6 6 6 6 0 0 :294 0 0 0 0 6 6 6 6 0 0 0 :244 0 0 0 6 ¼6 6 0 0 0 0 :213 0 0 6 6 6 6 0 0 0 0 0 :294 0 6 6 6 6 0 0 0 0 0 0 :294 4 0 2 :698 6 6 :698 6 6 6 6 :698 6 6 6 6 :698 6 6 6 0 6 6 6 6 0 6 6 6 6 0 4

0

0

0

0

0

:698 :698 1:302

0

:698 :698 1:302

0

:698 :698 1:302

0

:698 :698 1:302 :698 1:302 :698

0

0

0

0

0

:698

0

0

0

:698

0 0 2 :206 :206 6 6 :206 :206 6 6 6 6 :206 :206 6 6 6 6 :170 :170 6 ¼6 6 0 0 6 6 6 6 0 0 6 6 6 6 0 0 4

0

0

0

0

:206 :382 :206 :382 :206 :382 :170 :320 0

:277

0

0

0

0

0

0

0

0

3

7 0 7 7 7 7 0 7 7 7 7 0 7 7 7 0 7 7 7 7 0 7 7 7 7 0 7 5 :294

0 3 7 0 0 0 7 7 7 7 0 0 0 7 7 7 7 0 0 0 7 7 7 1:302 0:698 :698 7 7 7 7 1:302 0:698 :698 7 7 7 7 1:302 0:698 :698 7 5 0

0

:698 1:302 0:698 :698 3 0 0 0 0 7 0 0 0 0 7 7 7 7 0 0 0 0 7 7 7 7 :170 0 0 0 7 7 7 :149 :277 :149 :149 7 7 7 7 :206 :382 :206 :206 7 7 7 7 :206 :382 :206 :206 7 5 :206 :382 :206 :206

35

Status, Networks, and Opinions

Step 3: Given S, Compute the Matrix of Total Interpersonal Influences Using Eq. (7) Using S to operationalize the matrix of direct endogenous interpersonal influences, W (in SINT), and noting that IAW is nonsingular, we use Eq. (7) to obtain the matrix of total (direct and indirect) interpersonal influences, V (assuming, as in the preceding example, that aii ¼ 1–wii): V ¼ ðI  AWÞ1 ðI  AÞ 02 1 0 0 0 0 0 B6 B6 0 1 0 0 0 0 B6 B6 B6 0 0 1 0 0 0 B6 B6 B6 0 0 0 1 0 0 B6 ¼ B6 B6 0 0 0 0 1 0 B6 B6 B6 0 0 0 0 0 1 B6 B6 B6 0 0 0 0 0 0 @4

0 0 0 0 0 0 0 1

3

2

:794

0

0

:794

7 6 6 07 7 6 7 6 6 07 7 6 7 6 6 07 7 6 76 7 07 6 6 7 6 6 07 7 6 7 6 7 05 6 4

:206 6 6 :206 6 6 6 :206 6 6 6 :170 6 6 6 0 6 6 6 0 6 6 6 0 4 0 02 1 B6 B6 0 B6 B6 B6 0 B6 B6 B6 0 B6  B6 B6 0 B6 B6 B6 0 B6 B6 B6 0 @4 0

0 0 0 0 0

0 0 0 0 0 0 0 1 2

0

:206 :206 :382

0

0

:206 :206 :382

0

0

:206 :206 :382

0

:170 :170 :320 :170 0

0

0 0

:277 :149 :277

0

0

0

:206 :382

0

0

0

:206 :382

0 0 0 :206 :382 3 2 :794 0 0 0 0 0 0 0 7 6 6 0 1 0 0 0 0 0 07 7 6 7 6 6 0 1 0 0 0 0 07 7 6 0 7 6 6 0 0 1 0 0 0 07 7 6 0 76 6 0 0 0 1 0 0 07 7 6 0 7 6 7 0 0 0 0 1 0 07 6 6 0 7 6 6 0 0 0 0 0 1 07 5 4 0 0 0 0 0 0 0 0 1

0

0

0

0

0

0

3

7 0 7 7 7 0 :794 0 0 0 0 0 7 7 7 0 0 :680 0 0 0 0 7 7 7 0 0 0 :851 0 0 0 7 7 7 0 0 0 0 :618 0 0 7 7 7 0 0 0 0 0 :794 0 7 5 0 0 0 0 0 0 :794 311 0 0 7C C 0 0 7 7C 7C 7 0 0 7C C 7C C 0 0 7 7C 7C C :149 :149 7 7C 7C 7 :206 :206 7C C 7C C :206 :206 7 5A :206 :206 31 0 0 0 0 0 0 0 7C C :794 0 0 0 0 0 0 7 7C 7C 7 0 :794 0 0 0 0 0 7C C 7C C 0 0 :680 0 0 0 0 7 7C 7C C 0 0 0 :851 0 0 0 7 7C 7C 7 0 0 0 0 :618 0 0 7C C 7C C 0 0 0 0 0 :794 0 7 5A 0

0

0

0

0

0

0

0

0

0

0

:794

36

WILL KALKHOFF ET AL. 02

1 B6 B6 0 B6 B6 B6 0 B6 B6 B6 0 B6 ¼ B6 B6 0 B6 B6 B6 0 B6 B6 B6 0 @4

2 :164 7 6 6 07 7 6 :164 7 6 6 :164 07 7 6 7 6 6 07 7 6 :116 76 7 07 6 6 0 7 6 6 07 7 6 0 7 6 7 05 6 4 0

0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0

3

:164 :164 :304

0

0

0

:164 :164 :304

0

0

0

:164 :164 :304

0

0

0

0

0

:116 :116 :218 :116

:236 :127 :236 :127

0

0

0

0

0

0

0

:164 :303 :164

0 0 0 0 0 0 0 0 0 0 1 02 :206 0 0 0 0 0 0 B6 B6 0 :206 0 0 0 0 0 B6 B6 B6 0 0 :206 0 0 0 0 B6 B6 B6 0 0 0 :320 0 0 0 B6  B6 B6 0 0 0 0 :149 0 0 B6 B6 B6 0 0 0 0 0 :382 0 B6 B6 B6 0 0 0 0 0 0 :206 @4

0

:164 :303 :164 31

02

0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1

0

0

0

0

0

:836 :164 :164 :304

B6 B6 :164 B6 B6 B6 :164 B6 B6 B6 :116 B6 ¼ B6 B6 0 B6 B6 B6 0 B6 B6 B6 0 @4 0 02 :206 B6 B6 0 B6 B6 B6 0 B6 B6 B6 0 B6  B6 B6 0 B6 B6 B6 0 B6 B6 B6 0 @4 0

0

0

0

:127 :236 :127

7C C 0 7 7C 7C 7 0 7C C 7C C 0 7 7C 7C 7 0 7C C 7C C 0 7 7C 7C 7 0 5C A :206

0

0

0

0

:836 :164 :304

0

0

0

:164 :836 :304

0

0

0

0

0

:116 :116 :782 :116 0

:236 :873 :236 :127

0

0

0

0

:127 :764 :127

0

0

0

:164 :303 :836

0 0

0

0 0

:164 :303 :164 31 0 0 0 7C C :206 0 0 0 0 0 0 7 7C 7C C 0 :206 0 0 0 0 0 7 7C 7C 7 0 0 :320 0 0 0 0 7C C 7C C 0 0 0 :149 0 0 0 7 7C 7C 7 0 0 0 0 :382 0 0 7C C 7C C 0 0 0 0 0 :206 0 7 5A 0 0 0 0 0 0 :206 0

0

0

311

7C 7C 7C 7C C 0 7 7C 7C 7 0 7C C 7C C :127 7 7C 7C 7 :127 7C C 7C C :164 7 5A :836 0

0

311

7C C 0 7 7C 7C C 0 7 7C 7C C 0 7 7C 7C 7 :127 7C C 7C C :127 7 7C 7C 7 :164 5C A :164

37

Status, Networks, and Opinions 02

1:578 :578

:578 1:118 :183

:099

:053

:053

31

7C C :053 7 7C 7C C :578 :578 1:578 1:118 :183 :099 :053 :053 7 7C 7C 7 :427 :427 :427 1:869 :306 :165 :089 :089 7C C 7C C :142 :142 :142 :622 1:512 :816 :440 :440 7 7C 7C 7 :041 :041 :041 :181 :440 1:777 :419 :419 7C C 7C C :053 :053 :053 :233 :567 1:000 1:540 :540 7 5A :053 :053 :053 :233 :567 1:000 :540 1:540 02 31 :206 0 0 0 0 0 0 0 B6 7C C B6 0 :206 0 0 0 0 0 0 7 B6 7C B6 7C C B6 0 0 :206 0 0 0 0 0 7 B6 7C B6 7C B6 0 7 0 0 :320 0 0 0 0 7C C B6  B6 7C B6 0 7C 0 0 0 :149 0 0 0 B6 7C B6 7C C B6 0 0 0 0 0 :382 0 0 7 B6 7C B6 7C C B6 0 0 0 0 0 0 :206 0 7 @4 5A 0 0 0 0 0 0 0 :206 3 2 :325 :119 :119 :358 :027 :038 :011 :011 7 6 6 :119 :325 :119 :358 :027 :038 :011 :011 7 7 6 7 6 6 :119 :119 :325 :358 :027 :038 :011 :011 7 7 6 7 6 6 :088 :088 :088 :598 :046 :063 :018 :018 7 7 6 ¼6 7 6 :029 :029 :029 :199 :225 :312 :091 :091 7 7 6 7 6 6 :008 :008 :008 :058 :066 :679 :086 :086 7 7 6 7 6 6 :011 :011 :011 :075 :084 :382 :317 :111 7 5 4 :011 :011 :011 :075 :084 :382 :111 :317 B6 B6 B6 B6 B6 B6 B6 B6 B6 ¼ B6 B6 B6 B6 B6 B6 B6 B6 @4

:578 1:578 :578 1:118 :183

:099

:053

Step 4: Given V and an Array of (Hypothetical) Initial Opinions, Compute Predicted Final Opinions Using Eq. (5) Using the array of hypothetical initial opinions for actors A, B, C, D, E, F, G, and H in Network 2 (Table 2b), and following Eq. (5), predicted final

38

WILL KALKHOFF ET AL.

opinions are given by yð1Þ ¼ Vyð1Þ 2 :325 6 6 :119 6 6 6 :119 6 6 6 6 :088 6 ¼6 6 :029 6 6 6 6 :008 6 6 6 :011 4

:119

:119

:358

:027

:038

:325

:119

:358

:027

:038

:119

:325

:358

:027

:038

:088

:088

:598

:046

:063

:029

:029

:199

:225

:312

:008

:008

:058

:066

:679

:011

:011

:075

:084

:382

:011 :011 3 43 6 7 6 48 7 6 7 6 7 6 46 7 6 7 6 7 6 7 6 62 7 6 7 ¼6 7 6 58 7 6 7 6 7 6 7 6 76 7 6 7 6 7 6 59 7 4 5

:011

:075

:084

:382

2

55

:011 :011

3 2

10

3

7 6 7 6 7 :011 :011 7 7 6 33 7 7 6 7 6 7 :011 :011 7 7 6 24 7 7 6 7 7 6 7 80 7 :018 :018 7 6 7 6 7 76 7 6 7 :091 :091 7 7 6 28 7 7 6 7 7 6 7 :086 :086 7 6 95 7 7 6 7 7 6 7 6 7 :317 :111 7 5 4 37 5 :111 :317

19

COLLECTIVE EVENTS, RITUALS, AND EMOTIONS$ J. David Knottnerus ABSTRACT This chapter addresses the issue of how special collective ritual events operate and influence actors’ emotional states and commitment to a group. It is argued that in such events (e.g., rallies, holiday celebrations, and religious ceremonies) the greater the emotional intensity experienced by persons, the greater will be their commitment to and solidarity within the group. A model is proposed, which identifies several factors involved in such a process. The model builds on a body of theory and research, ‘‘structural ritualization theory (SRT),’’ which focuses on the role symbolic rituals play in social interaction and the generation and transformation of social structure. Four factors play a crucial role in the model: focus of attention, interactional pace, interdependence, and resources. Several of these factors also involve subcomponents that are identified and discussed. Attention is directed to how the formulation presented here is influenced by, and differs in certain ways from, classic and contemporary analysts including those working in the areas of social psychology and the sociology of emotions. Various examples are provided

$

An earlier version of this chapter was presented at the 101st annual meeting of the American Sociological Association, Montreal (2006).

Advances in Group Processes, Volume 27, 39–61 Copyright r 2010 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0882-6145/doi:10.1108/S0882-6145 (2010)0000027005

39

40

J. DAVID KNOTTNERUS

to illustrate the ability of the model to understand collective ritual events. Directions for further theory development and possible research investigating the arguments of the theory are also discussed.

This chapter is concerned with the question of how special collective ritual events operate and impact actors’ emotional states and commitment to a group. This is an important topic because collective ritual events are a type of social experience that can affect individuals, collective arrangements, and social developments. They, as have been pointed out by classic theorists such as Durkheim and more contemporary analysts, can play a pivotal role in social life profoundly affecting the nature and strength of social relations between the members of society. For instance, in attempting to develop a theory of public ritual, Etzioni (2000) and Etzioni and Bloom (2004) point out that holidays, especially recommitment holidays, can enforce commitments to shared beliefs. Holidays, as one type of collective ritual event, are important, ubiquitous social occurrences, which oftentimes strengthen peoples’ shared feelings, ties, and overall loyalty to a society or a particular group within that society. This is a topic deserving further theoretical attention and research because various questions are raised concerning how such events actually do this. For instance, what are the key components of these social activities that affect actors? What are the specific effects on actors of these collective experiences? And, do the components of collective ritual events vary and if so, in what way do they influence actors and the groups they belong to? In considering this issue it should be recognized that special collective ritual events are quite common occurring in societies around the world and through history. Examples of such collective ritual events include religious ceremonies, community or ethnic festivals, political (or other types of) rallies, military celebrations, commemorations of important historical events and developments, religious, civic, nationalistic, or military holidays, wedding celebrations, certain sporting events, pep rallies, etc. Furthermore, while such events often have macrosociological significance both reflecting and influencing conditions within an entire society, the events themselves take place in groups of varying sizes.1 Before addressing the question of how such events affect actors, it is necessary to more precisely identify what is meant by a special collective ritual event. Several features distinguish such ‘‘special’’ events. First, this type of social occurrence is clearly demarcated and separated from everyday

Collective Events, Rituals, and Emotions

41

social life and behaviors. The nature of the demarcation can be quite extensive and definitive involving, for instance, temporal, spatial, and definitional markers that specify when and where the event should take place and even the significance of the occurrence. In essence, it is considered to be a distinctive social activity set apart from daily social activities. Second, such events occur in a regularized manner. They are usually engaged in on a periodic basis whether that involves, for instance, a set time schedule or their staging being connected to other social developments, for example, a military celebration marking the completion of basic training or a pep rally preceding a ball game. Third, such special events involve to varying degrees stylized activities. Whether involving one activity or an array of practices such events involve people engaging in behaviors that are quite recognizable due to their definitive form or style, for example, marching, dancing, singing, praying, speech-making, and making vows or oaths. Finally, this type of event involves multiple actors. Indeed, there usually is a belief and expectation among people that such events are collectively engaged in. Two additional points should be made, however, about this feature. The number of people engaging in such events can greatly vary ranging from thousands (or more) individuals to just a handful of persons. Furthermore, it should be recognized that while falling outside the scope of the present discussion, it is possible that under certain circumstances a special case may exist in which a single actor may engage in practices related to the event. It is possible that an individual who is totally isolated from others (e.g., solitary confinement or a lone researcher in an arctic station) may celebrate the event and perform some facsimile of the ritual, albeit in a modified, private manner.2 With that said, a plausible answer to the question of how special collective occurrences impact people is that the greater the emotional state or feelings experienced by actors in ritual events, the greater will be their commitment to and, therefore, integration within the group. This leads in turn to the question of what are the key dimensions of collective occurrences that influence the feelings of participants and their commitment to a group. In what is to follow, I propose a model, which identifies several factors involved in such a process. I should first emphasize, however, that this formulation is influenced by and indebted to various scholars including Durkheim ([1915] 1965), Collins (2004), and Allan (1998, unpublished manuscript) (the latter two having focused to varying degrees on Durkheim’s notion of collective effervescence) and theory and research in social psychology. Particularly

42

J. DAVID KNOTTNERUS

important are the works of Lawler, Thye, and Yoon (2008, 2009) and Lawler and Yoon (1996) whose research has demonstrated how social interaction and exchange can generate positive emotions, that is, an emotional buzz, which heighten the strength of group ties and solidarity and Thibaut and Kelley (1959) whose theory and research (although focused on a different set of issues from those examined here and employing a different conceptual framework) emphasized how social interaction is influenced by the interdependence of the participants (see also Irwin, McGrimmon, & Simpson, 2008). Also relevant for the model presented here is theory and research dealing with legitimation and endorsement within groups and the contributions of various analysts within the sociology of emotions (to be discussed later).

STRUCTURAL RITUALIZATION THEORY The model builds on a distinct body of theory and research, ‘‘structural ritualization theory (SRT),’’ which focuses on the role symbolic rituals play in social interaction and the processes by which ritualization occurs and leads to the formation, reproduction, and transformation of social structure (Knottnerus, 1997, 2005, 2009, 2010). This theory is grounded in the basic assumption that daily life is normally characterized by an array of personal and social rituals. Such rituals help create stability to social life while expressing various symbolic meanings that give significance to our actions. Moreover, everyday rituals, whether occurring in small groups or organizations can lead to consequences unintended or unanticipated by group members while both being fed by and feeding into larger societal levels of interaction. Stated somewhat differently, the theory argues that ritualized practices help structure group dynamics. They contribute to the patterning of everyday behavior and interaction in various social milieus. Ritualized activities refer to the widespread form of social behavior in which people engage in regularized activities when interacting with others. Such practices are found throughout social life and can include ritualized forms of interaction within different institutions, subcultures, and groups of varying size, for example, patterns of behavior and communication in a youth group, family gatherings, religious occurrences, or sporting and recreational events. This perspective emphasizes that ritualized practices that comprise much of the taken-for-granted daily lives of people play a crucial role in the structuring of social behavior and group processes.

Collective Events, Rituals, and Emotions

43

The basic assumption that rituals are crucial to human behavior is consistent with the arguments of various scholars including Emile Durkheim ([1915] 1965), Erving Goffman (1967), Randall Collins (2004), Peter Berger and Thomas Luckmann (1966), Lloyd Warner (1959), Anthony Giddens (1984), Victor Turner (1967), and Mary Douglas (1970). The theory differs from nearly all these and other treatments, however, because it, among other things, provides more formal definitions of rituals. In this way it provides a more precise theory, allows for further theory development, and makes it easier to employ the perspective for the purposes of analysis and empirical research. It also focuses on ritualized interaction sequences and actions that are found in both secular and sacred settings. A number of empirical investigations have been (and are currently being) carried out providing tests, applications, and exemplifications of the theory. This work has progressed to the point that several lines of theory development and research employing multiple methodologies are currently under way each of which builds on and involves an extension of the original theory. At present this research focuses on deritualization, that is, disruptions to personal and social rituals, their consequences, and the ways people may cope with such experiences (Knottnerus, 2002; Thornburg, Knottnerus, & Webb, 2007, 2008; Sarabia & Knottnerus, 2009; Wu & Knottnerus, 2005, 2007); identity construction and ritual (Guan & Knottnerus, 1999; Minton & Knottnerus, 2008); the enactment of ritualized practices in organizations and communities (Knottnerus, Ulsperger, Cummins, & Osteen, 2006; Ulsperger & Knottnerus, 2007, 2008, 2009a, 2009b, 2010); reproduction of ritualized behaviors and social structure among groups (Sell, Knottnerus, Ellison, & Mundt, 2000; Knottnerus & Van de Poel-Knottnerus, 1999; Van de PoelKnottnerus & Knottnerus, 2002; Knottnerus, 1999, Knottnerus, Monk, & Jones, 1999; Knottnerus & Berry, 2002); strategic ritualization and the role of power (Knottnerus & LoConto, 2003; Edwards & Knottnerus, 2007, 2010; Guan & Knottnerus, 2006); ritual dynamics involving social inequality, distinction, and exclusion (Mitra & Knottnerus, 2004, 2008; Varner & Knottnerus, 2002; Minton & Knottnerus, 2008); and applied research involving social policy and interventions focusing, for example, on ritualized deviance in nursing homes (Ulsperger & Knottnerus, 2007, 2010). The present discussion focuses on a new topic: collective ritual events and the role of emotions. Before beginning, several definitions presented in the original theory (Knottnerus, 1997, pp. 260–261) need to be presented and a modification made in one of the concepts. ‘‘Socially standardized’’ refers to a regularly

44

J. DAVID KNOTTNERUS

engaged in social practice. This concept implies (to greater or lesser degrees) the continuance of a social usage or custom. ‘‘Action repertoire’’ is formally defined as a set of elements that are socially standardized practices. ‘‘Schema’’ refers to a cognitive structure. With these definitions ‘‘ritualized symbolic practice’’ is formally defined as an action repertoire that is schema-driven. Nomenclature for a ritualized symbolic practice is RSP. Given the issues addressed here, the original definition of an RSP is insufficient. Recognition also needs to be provided for the emotional component of a ritualized practice (see footnote 6 in the original formulation of the theory). For this reason, a revised definition of an RSP will be utilized in the present discussion. An RSP is an action repertoire that is schema-driven and emotion laden. This concept refers to that common form of social behavior in which individuals engage in regularized actions that are grounded in actors’ cognitive maps or symbolic frameworks and that possess emotional content (other terms that might be used would include sentiments, feelings, or affect). Finally, the original theory discusses the scope conditions or contexts to which the theory applies. The theory applies to scope conditions designated as ‘‘domains of interaction.’’ This is a theoretical concept that refers to a bounded social arena containing two or more actors. These actors are assumed to be at least part of the time engaged in face-to-face interaction. A domain of interaction is a delimited region or sphere of activity that has the power to produce effects, that is, it affects the probability of occurrences. Such occurrences involve actors’ cognitions and behaviors and, reflecting the revised definition of an RSP, emotions. This is an abstract concept that helps to clarify the nature of a social milieu and that can involve many different empirical cases. In the present discussion, attention is directed to domains of interactions or bounded social arenas in which actors collectively engage in special or distinctive ritual events (involving RSPs) which at least part of the time involve face-to-face interaction.

EMOTIONAL INTENSITY AND COMMITMENT IN RITUAL EVENTS In addressing the topic of collective ritual events and emotions, this formulation builds on certain ideas put forth by Randall Collins (2004) in his interaction ritual chain theory. Some of the factors identified as influencing collective emotions and commitment are similar to those

Collective Events, Rituals, and Emotions

45

emphasized by Collins. There are differences though between the two formulations. I do not for instance include, as he does, emotional mood or sentiment as one of the factors affecting emotional enthusiasm and commitment. By including emotion as one of the causal factors in his model, this approach exhibits a tautological quality in its argument. In other words, by placing an emotional condition in the explanatory (causal) model such a formulation assumes a particular outcome, that is, the generation of an emotional state among actors. I have tried to avoid this by formally identifying only social/social psychological conditions not explicitly involving emotions as the factors that lead to group commitment and integration. Furthermore, certain factors are discussed in the present model that are not addressed in Collin’s theory. Four factors play a crucial role in this model. In essence, they involve attention, interactional pace, interdependence, and resources. Each will be defined and discussed in greater detail. The shared focus of attention of actors in a collective event refers to the degree participants’ attention is directed to certain objects (see Collins, 2004; Allan, 1998, Unpublished Manuscript; and Durkheim [1915] 1965 who originally discussed this topic). Objects that are the focus of actors’ perceptions can vary including, for instance, physical items, persons, symbols, or logos within the collective event. The shared focus of attention of individuals in a collective episode can range from an absence of attention to an object (or minimal awareness of the object) to extremely high levels of focus, concentration, or awareness. The greater the shared focus of attention the more intense the collective emotions experienced by participants in the collective event. Various situational conditions and characteristics of actors can influence people’s levels of attention. Physical layout and arrangements, technology, and the way collective events are orchestrated can directly impact the extent to which group members are focused on the collective occasion and particular objects or aspects of the collective experience. Personal attributes such as the state of mind of participants and the degree to which they are capable of cognitively and perceptually apprehending their environment due to, for example, alcohol or drug use or physical activities they may be engaged in can also influence focus of attention. Consider, for instance, the difference between a political rally where, due to the physical layout and qualities of a speaker, all participants’ attention is directed to and captivated by a magnetic, powerful speaker and a wedding reception where people are distracted by friends, waiters, etc. and their attention to the host and newly married couple is more sporadic and perhaps even of a superficial nature.

46

J. DAVID KNOTTNERUS

The interactional pace of a ritual event refers to the degree to which actors are engaged in a sequence of (socially) interconnected acts and the nature of the recurring or repetitive acts. Interactional pace is a function of both the (a) rate of interaction and whether there is a (b) rhythmic motion to their physical movements in the interaction sequence (Allan, 1998, pp. 91–96). More precisely, rate of interaction refers to the frequency with which people interact. It deals with the speed or pace of the acts, which comprise the interaction sequence. While sometimes measured in terms of seconds and even microseconds relatively great differences can exist in the rate that acts occur between people in a ritualized collective event. Rhythmic motion refers to whether and to what degree physical movements in the social interaction recur in a uniform manner (see Wiltermuth & Heath, 2009; Collins, 2004; and Condon & Sander, 1974 for a revealing study that demonstrates the importance of rhythm for humans by documenting how infants in even the first day of life engage in rhythmic movements that are synchronous with the natural, rhythmic speech of adults). To what extent do physical movements in the interaction exhibit a measured, patterned, and regularized quality in which the physical movements or actions of different actors complement and coordinate with each other? Although some actions in ritual events may not exhibit such a rhythmic quality, others may to varying degrees recur in a regularized fashion. For instance, the rate of interaction in collective events such as a military celebration or a 4th of July ceremony may occur at a moderate or even somewhat slow pace – note the almost plodding nature of some civic ceremonies, with one uninspired speaker after another coming to the stage at a normal or measured rate. On the contrary, occasions such as protest rallies may be marked by a much quicker turn of events as speakers quickly replace each other, leaders constantly encourage the audience to chant and voice their feelings, and musical performances and other entertainment sustain the quickened tempo of the event as they fill the interludes between different speakers. So too, actors in some ritual occurrences such as the aforementioned civic ceremonies may exhibit little or no rhythmic quality in their physical movements. In contrast, other collective events may find people engaging in quite distinct, albeit different kinds of rhythmic behaviors, such as audience members swaying together and lifting their arms in unison during a church service or people chanting together at a sporting event or saluting and verbally responding in a coordinated manner in response to the admonitions of a political leader.

Collective Events, Rituals, and Emotions

47

Increases in the interactional pace as determined by either the rate of interaction or the rhythmic motion result in the strengthening of emotions shared by actors in the collective episode. Interdependence of actors deals with the distribution or relative occurrence of acts by different participants in the ritual event and how differentiated the actions are that are required to conduct the collective ritual. This component of the model, therefore, involves two key dimensions. First, interdependence of actors refers to the degree to which actors may or may not be equally (a) contributing to the ritual performance. At one end of the spectrum, marked inequalities may exist in the degree to which those who are present participate in the ritual enactment. At its most extreme, one or a limited number of persons may actually engage in the ritual act, whereas all the others assume a quite passive role merely observing the performance of the few. On the contrary, all those who are present may be equally involved in the collective episode, fully participating in the ritual performance. Interdependence of actors also involves the (b) level of complexity of different actions involved in the ritual event. It may be that there are simply one or a few activities, which group members need to participate in to conduct the ritual event. In contrast, a significantly larger number of activities may be required for the successful production of the ritualized collective event. The episode may exhibit to differing degrees a more differentiated and complex quality with individuals engaging in various kinds of specific practices as they collectively work together to accomplish the ritual event. In regard to actors unequally contributing to a ritual performance, a simple example would involve a religious gathering or political rally where the audience is completely passive and not actively involved in the event, merely observing and listening to a religious leader or politician. On the contrary, while the behaviors are quite simple, examples of actors equally contributing to the ritual event would be a military celebration where all those present are involved in drilling and marching activities or a religious service where everyone collectively prays and sings together. The theory argues that the more actors actually participate in the collective event, that is, the more equally they are involved in the collective episode, they greater the emotional impact of that event on those individuals. Concerning the level of complexity of actions involved in the ritual event, the cases just mentioned are examples of collective events that are simple involving only one or two types of behaviors, that is, drilling/marching and

48

J. DAVID KNOTTNERUS

praying/singing. On the contrary, a collective ritual episode may involve multiple practices, which are integral to the enactment of the event. A largescale, carefully designed political rally would provide one such example. In such an event participants may engage in a number of different activities, each of which is important to the overall performance whether they involve different actors marching, singing, carrying and presenting banners or other logos, speaking, playing music, driving vehicles, flying planes, operating lights and visual effects, performing dances or different types of physical movements and exercises, etc. The theory suggests that the greater the number of activities required for the production of the collective event, that is, the higher the level of complexity of different actions, the greater the sense of collective contribution to and mutual dependence among actors in the collective episode. This in turn enhances the impact of the event on actors’ emotional state. Concrete examples of such events can readily be found in recent history, for example, massive rallies such as the famous Nuremberg rallies conducted in Nazi Germany. Here, such events not only involved a more complex array of activities but often called upon all present to contribute to the collective episode. Both dimensions related to the interdependence of actors operated to create what were, by all accounts, quite moving and powerful collective events. Finally, the presence or absence of resources is quite important. As defined in the original presentation of SRT (Knottnerus, 1997) resources are materials needed to engage in RSPs, which are available to actors. This factor emphasizes the significance and necessity of resources for enacting ritualized behaviors including collective ritual occurrences. Moreover, an analytical distinction is made between two kinds of resources: human and nonhuman resources. Human resources are defined as the abilities and characteristics of actors perceived by group members to be of value (or have utility) for themselves or the group. Nonhuman resources refer to all that is not human that is perceived by group members to be of value (or have utility) for themselves or the group. Various kinds of human resources are vital to the production of a collective event such as the social, cognitive, and physical skills of group members, number of persons involved in the episode, and the arrangement or positioning of people in relation to each other. Probably the most important human resource, however, is the degree to which actors are copresent or visible to each other (see Collins, 2004, and Allan, 1998, unpublished manuscript for discussion of this topic). Ritual events can vary in the degree to which actors are aware or conscious of each other’s

Collective Events, Rituals, and Emotions

49

participation in the collective episode. The more likely that all individuals involved in the ritual event are seen by, visible to, and conscious of each other, the greater actors’ awareness that they are part of the collective ritual effort and the greater the impact the collective activity will have on them, for example, the feelings they have about the group experience. That this is so is fundamental to social psychology and involves to a large degree the idea of influence and legitimation and the perceived efficacy of the individual’s behavior. The more that actors are aware of each other’s participation in the ritual event the greater the consensus, or perception that a consensus exists, among the participants in the ritual episode. The greater the degree to which a consensus exists among participants, the higher the level of endorsement (as one type of legitimation) or support from one’s peers in the collective ritual performance (for discussions of legitimacy and endorsement, see Johansson & Sell, 2004; Dornbusch & Scott, 1975; Walker, Thomas, & Zelditch, 1986). The more that co-presence creates a perception of consensus or support from peers in the event, the greater the sense of social validation or confirmation among ritual participants. This in turn leads to enhanced feelings of satisfaction and confidence about the collective ritual experience. Of course, many kinds of nonhuman resources can also play a crucial role in the staging of a ritualized collective event. Various sorts of props and objects may be required to conduct a ritual performance including musical instruments, noise makers, special kinds of clothing or costumes, banners, pictures or paintings, body ornamentation, food, or weapons. So too, the physical layout in which the ritual event occurs can be quite important because the setting may be to varying degrees conducive to the actual enactment of the collective activity. For instance, contrast a large flat open field in which a large political gathering occurs to a stadium or carefully designed physical structure, which enables a larger number of people to have a clear view of the key speakers, and more effectively directs the attention of everyone to center stage. Furthermore, actors may use technology to facilitate certain features of the ritual. For example, the use of an audio system and large television screens may direct peoples’ attention to important persons or activities. A case in point involving the use of technology to facilitate human resources would involve large church services (such as those occurring in an auditorium in a mega-church) employing huge television screens to repeatedly show other members of the audience participating in the ritual event, thus, enhancing actors’ sense of co-presence. The discussion of resources and the role they play in ritualized collective events raises a number of questions that deserve further attention.

50

J. DAVID KNOTTNERUS

For instance, what are the generalized characteristics or properties of resources such as props or physical layouts? I would suggest, for example, that important properties of physical layouts where collective events occur would at the very least include the size and contour of the area, physical structures such as buildings and their arrangement, symbolic markers and logos placed in the setting, and the physical location of the site in relation to important features of the wider social environment such as population centers and transportation routes. Each of these could affect the staging of the collective event and the impact it has on participants in terms of both the number of actors involved and ultimately the degree of emotional intensity. Other questions concern how technology may vary; does technology increase or decrease the other three factors – focus of attention, pace, and interdependency – and influence other resources; and how does variation in technology impact emotional intensity? Such issues are undoubtedly complex. For instance in a recent study (Simpson, Knottnerus, & Stern, unpublished manuscript) of Raids, that is, special collective events involving online video games (massive multiplayer online role-playing games), questions addressed include whether the technological medium of online games enhances actors’ ability to focus on the computer task, allows people to interact at different frequency levels, and creates among individuals a sense of interdependency because of the contribution required of each player. At the same time this type of technology’s affect on co-presence may be complicated in the sense that such a medium may facilitate actors’ mutual awareness of their co participation in virtual games but after a certain point a very high degree of playing (i.e., actors who devote a large proportion of their waking hours to such games) could undermine a person’s sense of copresence and instead foster a more self-centered and obsessive mode of behavior focused only on the task at hand and the needs and wants of that individual. Issues such as these will be addressed by this study and future work that will build on the basic formulation outlined in this chapter. In essence, the theory argues that all four factors and their subcomponents can vary. Taken together the four factors and their separate parts influence the emotional intensity of persons in a ritual event and, therefore, the strength of their commitment to ritual activities, beliefs, and the group. Emotional intensity refers to the degree or strength of positive emotions generated by these four factors and their subcomponents. Although the specific nature of these positive affective states may vary depending on the particular type of ritual event under investigation (e.g., religious, political,

Collective Events, Rituals, and Emotions

51

civil, recreational) and the situational context in which the event occurs (which may be influenced by various historical, cultural, and structural conditions), a general characterization is possible. It is important to note that emotions have been conceptualized in different ways. For instance, various researchers in sociology and psychology have suggested that a small set of basic or primary emotions exists from which all other emotions are derived. Ekman and Friesen (1975), Kemper (1987), and Turner (2002) have identified groups of four or five basic emotions including such states as anger, fear, sadness, depression, and disgust. In all these taxonomies only one positive emotion is identified: satisfaction – happiness. Of course, many other more specific emotions exist which Turner (2002) and Kemper (1987) argue may involve variants of primary emotions (e.g., low, moderate, and high intensity) or first-order elaborations or combinations of primary emotions. A number of these more specific emotions could be produced by the social processes identified in the model. Depending on the particular nature of the ritual activity and situational conditions, the repertoire of more specific emotions that could be generated in a collective ritual event could include the states of: content, gratification, enjoyment, delight, friendly, joy, rapture, exhilaration, pride, reverence, triumphant, hope, and awe (see Turner, 2002, pp. 68–76, for a discussion of different kinds of emotions). Probably of greatest relevance to the present formulation is Russell’s circumplex model of affect (I am indebted to Shane R. Thye for pointing this out). Russell (1980) provides strong empirical support for the argument that emotions are best mapped by a circumplex model that consists of two orthogonal factors (i.e., the two-dimensional bipolar space of): pleasure/ satisfaction and degree of arousal (arousal/sleepiness). Rather than seeing affect as a set of independent dimensions, he argues that affective dimensions are interrelated in a systematic way and are best represented by a spatial model in which affective concepts are arranged in a circle in the following order: pleasure, excitement, arousal, distress, displeasure, depression, sleepiness, and relaxation. Employing this spatial model where the horizontal dimension involves pleasure-displeasure and the vertical dimension involves degrees of arousal, other affect terms define the quadrants of the space (e.g., anger is an emotion with high arousal and low satisfaction; depression is low on both factors; excitement is a combination of high pleasure and high arousal; contentment involves high pleasure and lower levels of arousal). The upshot (as it relates to the theory presented in this chapter) is that one or more emotional states, such as aroused, astonished, excited, delighted, happy, pleased, glad, serene, content, at ease, satisfied,

52

J. DAVID KNOTTNERUS

relaxed, or calm could be generated by the four factors and their subcomponents outlined in the model of ritual dynamics. In this regard, I would note that the analysis of emotional intensity put forth here acknowledges that ritual dynamics can generate different types of specific emotions. Such an analysis would go beyond how Durkheim and Collins discuss the more general ideas of collective effervescence and emotional energy. Although the present formulation does not address this issue in greater depth, future work will explore how differences in ritual processes and situational conditions can influence the kind of emotions generated in collective ritual events, especially as they pertain to Russell’s circumplex model of affect. Finally, it is argued that increases in these emotional states, that is, emotional intensity, result in increased commitment to the ritualized practices engaged in by actors in these collective events and the beliefs or symbolic themes expressed in these ritual practices. Positive emotions such as pleasure, arousal, excitement, delight, happiness, serenity, or contentment enhance people’s enthusiasm, devotion, and commitment to the ritualized activities and beliefs of the group during the ritual event. This condition then affects the dedication, loyalty, and allegiance of ritual participants to the group and ultimately social integration and bonds within the social body. Enhanced commitment to ritualized behaviors and beliefs during the ritual episode results in increased commitment to the group as a whole and strengthened social ties and relations among group members. Having discussed and defined the main concepts in this formulation, the basic argument can be more fully outlined. Essentially, the four factors and their subcomponents influence emotional intensity, which shapes commitment to shared ritualized practices and beliefs of the collective ritual event, which then influences commitment to the group and integration in the group, that is, social ties or bonds in the group. Stated in a more precise manner the theory argues that: (1) shared focus of attention þ (2) interactional pace [(a) rate of interaction and (b) rhythmic motion] þ (3) interdependence of actors [(a) differential/equal contribution of actors and (b) level of complexity of different actions] þ (4) resources (human resources/co-presence and nonhuman resources) influence the emotional intensity of actors, which impacts actors’ commitment to shared ritualized symbolic practices and beliefs/themes of the ritual event which then influences commitment to the group and integration/solidarity of the group. The following model graphically summarizes the key components of the formulation.

53

Collective Events, Rituals, and Emotions

Model 1

Rate of Interaction

Shared Focus of Attention

Rhythmic Motion

Interactional Pace

Differential /Equal Contribution of Actors

Interdependence of Actors

Level of Complexity of Different Actions

Human Resources / CoPresence

NonHuman Resources

Resources

Emotional Intensity of Actors

Actors’ Commitment to Shared Ritualized Symbolic Practices & Beliefs/Themes of the Ritual Event

Commitment to the Group and Integration/Solidarity of the Group

To restate, emotional intensity and the resulting commitment to ritualized practices, beliefs, and the group is the sum of shared focus of attention, interactional pace, interdependence of actors, and resources. The greater these factors, the higher the emotional intensity and commitment to group rituals and the group as a whole. The four factors are weighted equally. Although it is expedient to make this assumption at this point, subsequent work could alter the weightings. These are the key arguments of the theory. Future work will investigate the social dynamics operating in actual ritual events, thereby, allowing for an assessment of the formulation. Still, a brief example can be presented,

54

J. DAVID KNOTTNERUS

which indicates the potential ability of the model to understand this type of phenomena. One of the most well-known collective ritualized events of the twentieth century was the Reich Party Congresses conducted in Nazi Germany (Burden, 1967). From 1928 to 1938 these large mass rallies were held in Nuremberg at the beginning of September and lasted for a week. As various commentators have noted (e.g., Kertzer, 1988), these were monumental architectural structures for celebrating national socialism. A clearly defined set of ritualized practices occurred at all rallies (Knottnerus, Van Delinder, & Edwards, 2010). Rallies began with memorial services and large columns of uniformed and flag bearing members of groups such as the Hitler Youth and Nationalist Socialist Motor Corps. During the week, ceremonies were held, sports demonstrations were staged, objects such as banners were sanctified, and oaths of loyalty were taken. Culminating the week was a major speech by Hitler before huge numbers of people assembled on the grounds where spotlights were shone in an orchestrated manner. At the center of and above the massive columns Hitler stood alone on a stage where he delivered his speeches through a carefully prepared speaker system. The structuring of activities and the location of Hitler in this carefully designed site clearly directed participants’ focus of attention to their leader and other important activities and objects. Through the entire week collective activities and interaction proceeded in an uninterrupted, quick pace. Some of the collective actions such as saluting were also carried out in a synchronized manner. At the same time, a complex array of different activities contributed to the enactment of the ritual event with all actors present participating in the ritual event. Finally, a large amount of nonhuman and human resources were employed to stage the lengthy ritual event, which because of the way it was orchestrated enhanced peoples’ awareness of their mutual involvement or copresence. These events created an elevated degree of emotional intensity among those present which contributed to their commitment to the shared practices and beliefs associated with the ritual event and the nationalistic political group they belonged to. Social dynamics like these might also be found in other rallies, political or otherwise, that have occurred during the twentieth century. For instance, anecdotal evidence suggests that political rallies in the Soviet Union and subsequently Russia have changed over the past several decades exhibiting to varying degrees the factors identified in the present discussion. Indeed, similar analyses could be conducted of many special ritual activities such as holidays, religious gatherings, powwows, commemorative events, sports extravaganzas, civic celebrations, festivals, particular kinds of parading

Collective Events, Rituals, and Emotions

55

(e.g., the Orange Order parades in Northern Ireland) or large-scale community processions (Bhandari, Okada, & Knottnerus, unpublished manuscript).

FURTHER THEORY DEVELOPMENT This model could be integrated with the framework presented in the original formulation of SRT, which focuses on the factors determining the rank, that is, importance or dominance, of RSPs in social environments. The original theory emphasizes how ritual enactments are symboliccognitive (schema-driven) action repertoires. Four factors – salience, repetitiveness, homologousness, and RSP resources – are identified that determine the rank of RSPs. Subsequent work could address in a comprehensive manner how cognitive and emotional dimensions of ritualization processes operate. This discussion would include the four factors influencing emotional intensity, commitment to shared ritualized practices in a ritual event, and solidarity within the group (and might expand the scope of the model to apply to various kinds of group settings and not just special collective ritual events). These four components, it is suggested, parallel in significant ways and can be conceptually linked to the four factors of the original theory that influence rank. More precisely, focus of attention relates to salience, interactional pace parallels repetitiveness, interdependence of actors exhibits certain similarities to homologousness, and resources (such as co-presence in addition to other human and nonhuman resources) directly overlap with RSP resources. Further theory development could explore in greater depth this topic, explicate these interconnections, and provide a more synthetic formulation incorporating these different dimensions of rituals. These efforts would expand the analysis of ritualization in ways that reflect a commitment to theory integration, a goal discussed in the original presentation of SRT.

DISCUSSION AND CONCLUSION The formulation developed here is presented as a generalizable framework focused on the ways certain social dynamics influence emotions, commitment to ritualized practices, and group integration. It can be applied to various cases for the purposes of explanation, prediction of outcomes, and exemplification of collective ritual events. For this reason, the formulation could be investigated in several ways.

56

J. DAVID KNOTTNERUS

One possible project would involve the examination of several case studies, comparing and demonstrating how the arguments of the theory can explain the dynamics operating in these different settings and commitment to these ritual events and groups. This type of analysis would quite likely utilize historical and secondary forms of evidence. The model could also be used to investigate and analyze a particular type of collective ritual event (e.g., a religious ceremony, political rally, nationalistic celebration, or a certain kind of holiday event) occurring in different social contexts, for example, within different societies within the same time period, within the same society at different time periods, or within the same society or different societies at the same time period under different conditions. This sort of comparative investigation would focus on variations in the social dynamics of the same kind of ritual event and would allow for an assessment of the theory’s ability to explain and predict outcomes involving emotional intensity, commitment, and solidarity. Different kinds of evidence could be employed including both quantitative and qualitative sources. A different kind of analysis involving the experimental method could also be conducted in which all or certain of the four factors affecting emotional intensity are varied in a laboratory-generated collective ritual event. Measures could then be taken of actor’s emotional intensity and/or commitment to the group to determine whether such outcomes are consistent with the arguments of the theory. The kind of measures used could vary involving, for instance, questionnaires with scaled questions dealing with strength of emotions and group commitment or possibly physiological measures related to emotional reactions. Several other issues also deserve attention. As already noted, this analytical model deals with the social dynamics of ritual events influencing emotions and ritual/group commitment. The formulation does not address the specific nature of the beliefs associated with ritualized practices and the group. The nature of the beliefs or themes expressed through ritualized practices enacted by group members, I would suggest, is influenced by the empirical situation involving, for instance, historical, cultural, organizational, political, and economic conditions. Be that as it may it is possible that the nature of the beliefs expressed through ritual events could have a significant effect on the commitment of actors to a group. More precisely, the more the beliefs stress the group itself, the stronger the commitment toward the group. Contrast, for instance, a celebration directed toward a specific event, like a wedding, versus a celebration of the group itself, like a school pep rally (I am indebted to Jane Sell for this observation). Future research will address this topic.

Collective Events, Rituals, and Emotions

57

On a different note, depending on the situation the practices and beliefs involved in a ritual event may be similar to those found in the wider society which means that the ritual event acts to reaffirm or reproduce ritualized arrangements and beliefs in the larger social milieu, that is, the greater the emotional intensity and commitment to the ritual event and group, the more reaffirming the collective experience is. On the contrary, if circumstances should result in ritualized practices and beliefs in the ritual event differing from what is found in the wider social environment, such a situation could lead to conflict with and ultimately changes in the wider society, that is, the greater the emotional intensity and commitment to the ritual event and group, the greater the difference between the ritual event and the wider society and possible societal changes resulting from the collective experience. In other words, the formulation presented here could address social processes involved with either the reaffirmation of societal conditions or social change. These issues deserve further investigation. Finally, on a more speculative note, it is possible that intense collective ritual events may create changes in the beliefs held by group members. This is an idea that was suggested by Durkheim ([1915] 1965) in his classic study of religion and ritual (some contemporary theorists have noted Durkheim’s discussion of this issue although analysis of such occurrences has been limited. See for instance Young, 1994; Emirbayer, 1996; Olaveson, 2001). Perhaps through a synthesis of different beliefs or symbolic themes associated with different rituals or some other mechanism, certain ritual events involving high emotional intensity may serve as a conduit or catalyst for the creation of new RSPs and beliefs. This is a topic that also deserves further examination. In conclusion, it has been argued that several social mechanisms operating in collective ritual events play important roles in the development of emotional states and commitment to practices, beliefs, and groups. In addressing these issues such a formulation attests to the important role emotions along with cognitions play in rituals, particularly special collective ritual episodes. The model presented here provides the basis for research and further analytical efforts concerned with the dynamics of ritual enactments and group processes.

ACKNOWLEDGMENTS I thank Jane Sell and Shane R. Thye for their useful comments.

58

J. DAVID KNOTTNERUS

NOTES 1. Part of the reason they occur so often is that the groups enacting collective ritual events can range from a hundred thousand persons or more to just a handful of individuals. Of course, this assessment does not take into account the role of the media that enables a much larger number of people to be exposed to ritual events, albeit not as mutually interacting, co-participants which is the specific focus of the present discussion. 2. The phenomenon of private rituals is worthy of analysis in and of itself (see Allan, 1998). One such issue (I am indebted to Ken Allan for pointing this out) deals with how rituals may vary in terms of time decay. It seems likely that the longer the time from a group ritual, the less emotion the private use of similar rituals would be able to produce and vice versa. For instance, some people may regularly participate in group rituals and then practice them privately. In this case the group ritual would act as a sort of ‘‘charger’’ that reenergizes the emotional and symbolic content of the private ritual. It is also possible that people who practice private rituals do not lose their motivation to participate in public rituals. A related question also concerns why certain individuals who under extreme conditions are prevented from participating in public rituals that they had practiced in the past are still able to continue engaging in private versions of them, finding emotional fulfillment and meaning in doing so (see Knottnerus, 2002, for a description of persons who have done so after being interned in the highly restrictive and coercive environment of a concentration camp).

REFERENCES Allan, K. (1998). The meaning of culture: Moving the postmodern critique forward. Westport, CT: Praeger. Allan, K. Ritual and the ontological sources of human reality. Unpublished manuscript. Berger, P. L., & Luckmann, T. (1966). The social construction of reality. Garden City, NY: Doubleday. Bhandari, R. B., Okada, N., & Knottnerus, J. D. Urban ritual events and coping with disaster risk: A case study of Kathmandu. Unpublished manuscript. Burden, H. T. (1967). The Nuremberg party rallies: 1923–39. New York: Frederick A. Praeger. Collins, R. (2004). Interaction ritual chains. Princeton, NJ: Princeton University Press. Condon, W. S., & Sander, L. W. (1974). Neonate movement is synchronized with adult speech: Interactional participation and language acquisition. Science, 183(New Series), 99–101. Dornbusch, S., & Scott, W. R. (1975). Evaluation and the exercise of authority. San Francisco, CA: Jossey-Bass. Douglas, M. (1970). Natural symbols. New York: Vintage. Durkheim, E. ([1915]1965). The elementary forms of the religious life. New York: Free Press. Edwards, J., & Knottnerus, J. D. (2007). The orange order: Strategic ritualization and its organizational antecedents. International Journal of Contemporary Sociology, 44, 179–199. Edwards, J., & Knottnerus, J. D. (2010). The orange order: Parades, other rituals, and their outcomes. Sociological Focus, 43, 1–23.

Collective Events, Rituals, and Emotions

59

Ekman, P., & Friesen, W. V. (1975). Unmasking the face: A guide to recognizing emotions from facial clues. Oxford, England: Prentice-Hall. Emirbayer, M. (1996). Useful Durkheim. Sociological Theory, 14, 120–130. Etzioni, A. (2000). Toward a theory of public ritual. Sociological Theory, 18, 44–59. Etzioni, A., & Bloom, J. (2004). We are what we celebrate: Understanding holidays and rituals. New York: New York University Press. Giddens, A. (1984). The constitution of society: Outline of the theory of structuration. Berkeley, CA: University of California Press. Goffman, E. (1967). Interaction ritual: Essays on face-to-face behavior. Garden City, NY: Anchor Books. Guan, J., & Knottnerus, J. D. (1999). A structural ritualization analysis of the process of acculturation and marginalization of Chinese Americans. Humboldt Journal of Social Relations, 25, 43–95. Guan, J., & Knottnerus, J. D. (2006). Chinatown under siege: Community protest and structural ritualization theory. Humboldt Journal of Social Relations, 30, 5–52. Irwin, K., McGrimmon, T., & Simpson, B. (2008). Sympathy and social order. Social Psychology Quarterly, 71, 379–397. Johansson, A. C., & Sell, J. (2004). Sources of legitimation and their effects on group routines: A theoretical analysis. In: C. Johnson (Ed.), Research in the Sociology of Organization, Volume 22. Legitimacy processes in organizations (pp. 89–116). Amsterdam: Elsevier. Kemper, T. D. (1987). How many emotions are there? Wedding the social and autonomic components. American Journal of Sociology, 93, 263–289. Kertzer, D. I. (1988). Ritual, politics, and power. New Haven, CO: Yale University Press. Knottnerus, J. D. (1997). The theory of structural ritualization. In: B. Markovsky, M. J. Lovaglia & L. Troyer (Eds), Advances in Group Processes (Vol. 14, pp. 257–279). Greenwich, CT: JAI Press. Knottnerus, J. D. (1999). Status structures and ritualized relations in the slave plantation system. In: T. J. Durant,, Jr. & J. D. Knottnerus (Eds), Plantation society and race relations: The origins of inequality (pp. 137–147). Westport, CT: Praeger. Knottnerus, J. D. (2002). Agency, structure and deritualization: A comparative investigation of extreme disruptions of social order. In: S. C. Chew & J. D. Knottnerus (Eds), Structure, culture and history: Recent issues in social theory (pp. 85–106). Lanham, MD: Rowman & Littlefield. Knottnerus, J. D. (2005). The need for theory and the value of cooperation: Disruption and deritualization. (Presidential Address, Mid-South Sociological Association, Baton Rouge, 2003). Sociological Spectrum, 25, 5–19. Knottnerus, J. D. (2009). Structural ritualization theory: Application and change. In: J. D. Knottnerus & B. Phillips (Eds), Bureaucratic culture and escalating world problems: Advancing the sociological imagination (pp. 70–84). Boulder, CO: Paradigm Publishers. Knottnerus, J. D. (2010). Ritual as a missing link: Sociology, structural ritualization theory and research (November). Boulder, CO: Paradigm Publishers. Knottnerus, J. D., & Berry, P. E. (2002). Spartan society: Structural ritualization in an ancient social system. Humboldt Journal of Social Relations, 27, 1–42. Knottnerus, J. D., Van Delinder, J. L., & Edwards, J. (2010). Strategic ritualization and power: Nazi Germany, The orange order, and native Americans. In: Ritual as a missing link: Sociology, structural ritualization theory and research (November). Boulder, CO: Paradigm Publishers.

60

J. DAVID KNOTTNERUS

Knottnerus, J. D., & LoConto, D. G. (2003). Strategic ritualization and ethnicity: A typology and analysis of ritual enactments in an Italian American community. Sociological Spectrum, 23, 425–461. Knottnerus, J. D., Monk, D. L., & Jones, E. (1999). The slave plantation system from a total institution perspective. In: T. J. Durant, Jr. & J. D. Knottnerus (Eds), Plantation society and race relations: The origins of inequality (pp. 17–27). Westport, CT: Praeger. Knottnerus, J. D., & Van de Poel-Knottnerus, F. (1999). The social worlds of male and female children in the nineteenth century French educational system: Youth, rituals and elites. Lewiston, NY: Edwin Mellen Press. Knottnerus, J. D., Ulsperger, J. S., Cummins, S., & Osteen, E. (2006). Exposing Enron: Media representations of ritualized deviance in corporate culture. Crime, Media, Culture, 2, 177–195. Lawler, E. J., Thye, S. R., & Yoon, J. (2008). Social exchange and micro social order. American Sociological Review, 73, 519–542. Lawler, E. J., Thye, S. R., & Yoon, J. (2009). Social commitments in a depersonalized world. New York: Russell Sage Foundation. Lawler, E. J., & Yoon, J. (1996). Commitment in exchange relations: Test of a theory of relational cohesion. American Sociological Review, 61, 89–108. Minton, C., & Knottnerus, J. D. (2008). Ritualized duties: The social construction of gender inequality in Malawi. International Review of Modern Sociology, 34, 181–210. Mitra, A., & Knottnerus, J. D. (2004). Royal women in ancient India: The ritualization of inequality in a patriarchal social order. International Journal of Contemporary Sociology, 41, 215–231. Mitra, A., & Knottnerus, J. D. (2008). Sacrificing women: A study of ritualized practices among women volunteers in India. Voluntas: International Journal of Voluntary and Nonprofit Organizations, 19, 242–267. Olaveson, T. (2001). Collective effervescence and communitas: Processual models of ritual and society in Emile Durkheim and Victor Turner. Dialectical Anthropology, 26, 89–124. Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology, 39, 1161–1178. Sarabia, D., & Knottnerus, J. D. (2009). Ecological stress and deritualization in East Asia: Ritual practices during dark age phases. International Journal of Sociology and Anthropology (Open Access Online Journal), 1(1), 012–025, May. Available at www.academicjournals.org/IJSA/contents/2009cont/May.htm Sell, J., Knottnerus, J. D., Ellison, C., & Mundt, H. (2000). Reproducing social structure in task groups: The role of structural ritualization. Social Forces, 79, 453–475. Simpson, J., Knottnerus, J. D., & Stern, M. Virtual rituals: Structural ritualization theory and massive multiplayer on-line role-playing games. Unpublished paper. Thibaut, J. W., & Kelley, H. H. (1959). The social psychology of groups. New York: Wiley. Thornburg, P. A., Knottnerus, J. D., & Webb, G. R. (2007). Disaster and deritualization: A re-interpretation of findings from early disaster research. The Social Science Journal, 44, 161–166. Thornburg, P. A., Knottnerus, J. D., & Webb, G. R. (2008). Ritual and disruption: Insights from early disaster research. International Journal of Sociological Research, 1, 91–109. Turner, J. (2002). Face-to-Face: Towards a sociological theory of interpersonal behavior. Stanford: Stanford University Press.

Collective Events, Rituals, and Emotions

61

Turner, V. (1967). The forest of symbols: Aspects of Ndembu ritual. Ithaca, NY: Cornell University Press. Ulsperger, J. S., & Knottnerus, J. D. (2007). Long-term care workers and bureaucracy: The occupational ritualization of maltreatment in nursing homes and recommended policies. Journal of Applied Social Science, 1, 52–70. Ulsperger, J. S., & Knottnerus, J. D. (2008). The social dynamics of elder care: Rituals of bureaucracy and physical neglect in nursing homes. Sociological Spectrum, 28, 357–388. (2008 Sociological Spectrum Article of the Year Award). Ulsperger, J. S., & Knottnerus, J. D. (2009a). Illusions of affection: Bureaucracy and emotional neglect in nursing homes. Humanity and Society, 33, 238–259. Ulsperger, J. S., & Knottnerus, J. D. (2009b). Institutionalized elder abuse: Bureaucratic ritualization and transformation of physical neglect in nursing homes. In: J. D. Knottnerus & B. Phillips (Eds), Bureaucratic culture and escalating world problems: Advancing the sociological imagination (pp. 134–155). Boulder, CO: Paradigm Publishers. Ulsperger, J. S., & Knottnerus, J. D. (2010). Elder care catastrophe: Rituals of abuse in nursing homes – and what you can do about it. (November). Boulder, CO: Paradigm Publishers. Van de Poel-Knottnerus, F., & Knottnerus, J. D. (2002). Literary narratives on the nineteenth and early twentieth-century French elite educational system: Rituals and total institutions. Lewiston, NY: Edwin Mellen Press. Varner, M. K., & Knottnerus, J. D. (2002). Civility, rituals and exclusion: The emergence of American golf during the late nineteenth and early twentieth centuries. Sociological Inquiry, 72, 426–441. Walker, H. A., Thomas, G. M., & Zelditch, M., Jr. (1986). Legitimation, endorsement and stability. Social Forces, 64, 620–643. Warner, W. L. (1959). The living and the dead: A study of the symbolic life of Americans (Yankee City Series). New Haven, CT: Yale University Press. Wiltermuth, S. S., & Heath, C. (2009). Synchrony and cooperation. Psychological Science, 20, 1–5. Wu, Y., & Knottnerus, J. D. (2005). Ritualized daily practices: A study of Chinese ‘educated youth’. Shehui (Society) (6), 167–185 (Chinese academic journal). Wu, Y., & Knottnerus, J. D. (2007). The origins of ritualized daily practices: From Lei Feng’s diary to educated youth’s diaries. Shehui (Society) (1), 98–119 (Chinese academic journal). Young, F. W. (1994). Durkheim and development theory. Sociological Theory, 12, 73–82.

GROUP REFLEXIVITY AND PERFORMANCE Richard L. Moreland and Jamie G. McMinn ABSTRACT Many papers have been written about group reflexivity. Testimonials by practitioners often contain strong claims about its performance benefits. Research papers, by scientists, seem to support such claims at first glance, but a closer look reveals methodological problems and weak results, even in the studies that show performance benefits, and there are several studies that show no performance benefits. We have begun our own program of research on group reflexivity, and so far, we have found no performance benefits either. All of this suggests that enthusiasm for group reflexivity should be tempered, until more and better research has been done.

Practice makes perfect, so they say, and although few of us are perfect at performing any task, most of us perform tasks better as we gain experience with them. These improvements are large at first and then become progressively smaller, producing ‘‘learning curves’’ (see Argote, 1999, for a comparative analysis of learning curves in individuals, groups, and organizations). The performance benefits of task experience may be due to several factors, but an especially important one is reflection (see Boud, Keogh, & Walker, 1994; Kolb, 1984; Lee & Hutchison, 1998). People often think back about their performance after completing a task. Several issues could be

Advances in Group Processes, Volume 27, 63–95 Copyright r 2010 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0882-6145/doi:10.1108/S0882-6145 (2010)0000027006

63

64

RICHARD L. MORELAND AND JAMIE G. MCMINN

considered at that time, such as what happened, why it happened, how things might have turned out differently, and whether task performance could be improved. The resolution of these issues sometimes yields a deeper understanding of the task, helping people to perform it better in the future. Reflection is normally a private activity, carried out by individuals, but groups certainly perform many tasks, particularly in today’s complex society. Can groups also reflect about task performance, discussing what happened, why it happened, and so on? An interesting trend among group researchers in recent years has been the ‘‘discovery’’ of new group phenomena by analyzing individual characteristics or behaviors and then extrapolating them to the group level. The individual, in other words, has often served as a model for the group (consider the assumed parallels between self-efficacy in individuals and collective efficacy in groups, or between decision making by individuals and group decision making). This trend has advantages and disadvantages, but its power to generate research on groups cannot be denied. What about reflection? Can groups, like individuals, reflect on their task performance, and if so, then do they also perform better as a result?

CLAIMS ABOUT GROUP REFLEXIVITY BY PRACTITIONERS There is consensus in the literature that groups can indeed reflect, although there is also evidence that groups do not enjoy reflection, seldom reflect on their own (without being led to do so by others), and may not reflect especially well (Daudelin, 1996; Gurtner, Tschan, Semmer, & Nagele, 2007; West, 1996, 2000; Widmer, Schippers, & West, 2009). In fact, many papers have been written about group reflection. These papers can be divided into two distinct sets. One set of papers involves efforts by practitioners from various professions to help groups perform better by encouraging people to reflect more often on their performance and guiding them to reflect in better ways. These professions include the military, where ‘‘after action reviews’’ are conducted after many unit missions (see ‘‘A Leader’s Guide to After-Action Reviews,’’ 1993); business (Baird, Holland, & Deacon, 1999; Clark & Fujimoto, 1991; Daudelin, 1996; Pacanowski, 1995); education (Kolodner et al., 2004; Schank & Cleary, 1995); medicine (Salas et al., 2008; Vashdi, Bamberger, Erez, & Weiss-Meilik, 2007); and even sports (Goldsmith, 2009; O’Donoghue, 2006; Petruzzi, 1999). Papers on group reflection written by practitioners are nearly always positive; they often resemble ‘‘testimonials’’

Group Reflexivity and Performance

65

that contain strong claims about the value of group reflection and encourage readers to use such reflection more often. Many of the papers include tips about how group reflection should be done (see Baird et al., 1999; Goldsmith, 2009; Salas et al., 2008; Smith-Jentsch, Zeisig, Acton, & McPherson, 1998; Tannenbaum, Smith-Jentsch, & Behson, 1998). Baird and his colleagues, for example, suggest that reflection should (a) occur very soon after a task is performed; (b) be structured (maybe by an external facilitator), rather than informal; (c) focus on a few issues, rather than every issue; (d) involve every group member, rather than just a few; and (e) be followed, as quickly as possible, by further task performance. Tips such as these represent a kind of ‘‘lore’’ about group reflection, presumably generated through experience with many groups that have reflected in different ways, under various conditions. Although the tips certainly seem plausible, no one (to our knowledge) has yet done research to assess their (absolute or relative) merits. After reading this first set of papers, we were intrigued, but also rather skeptical, because no data or statistical tests were ever presented, nor were any citations made to scientific research. We wondered whether group reflection was akin to group brainstorming (see Mullen, Johnson, & Salas, 1991), namely something that seems appealing to practitioners, but has no proven value. Our skepticism led us to search for scientific research on how group reflection actually affects group performance. We found such research in a second set of papers.

WEST’S (1996) ANALYSIS OF GROUP REFLEXIVITY Scientific research on group reflection includes case studies, correlational studies, and experiments. Much of this work can be traced back to an influential paper by West (1996) (see also Swift & West, 1998; West, 2000). In his paper, West (1996) defined group reflection as ‘‘the extent to which group members overtly reflect upon the group’s objectives, strategies, and processes, and adapt them to current or anticipated endogeneous or environmental circumstances’’ (p. 559). He argued that reflection can occur during task performance, as well as after it (see also Cunliffe & EasterlySmith, 2004; Daudelin, 1996; Rasker, Post, & Schraagen, 2000)1 and that reflection can be performed by individual members of a group, as well as by the group as a whole. The latter claim raises several interesting issues, some of which we shall revisit later on. Can individual and group reflection occur simultaneously, for example, and if so, then do they facilitate or inhibit one

66

RICHARD L. MORELAND AND JAMIE G. MCMINN

another? Does one form of reflection have more value than the other (see Brav, Andersson, & Lantz, 2009; Sutton & Dalley, 2008), and if so, then which one is more valuable? Does the value of group reflection depend on the number of members who participate in it, and if so, then why? Could reflection by a single member, such as the leader of a group, be enough to help the group perform better, even if no one else reflects? West’s (1996) definition of group reflection seemed broad to us, and it was broadened even further by his description of the activities that are likely to occur in groups as they reflect. According to West, ‘‘Reflection involves behaviors such as questioning, planning, exploratory learning, analysis, diversive exploration, making use of knowledge explicitly, planfulness, learning at a meta level, reviewing past events with self-awareness, digestion, and coming to terms over time with a new awareness’’ (p. 560). Group reflection is apparently a complex process with many components, each of which might (on its own) influence performance. This complexity raises several other interesting issues. For example, are all the components of group reflection equally likely to occur, or do some occur more often than others, and if so, then which ones are most common? And are some components of group reflexivity more valuable than others? If so, then which ones have the greatest value? Does performance depend on the order in which the components of reflection occur? If so, then which components ought to occur earlier and which ones later? Finally, can some components interact with others to affect group performance, and if so, then what forms do such interactions take? West (1996) also discussed other aspects of group reflexivity, such as when it will occur, where it can focus (e.g., task vs. socioemotional aspects of the group), and possible boundary conditions for its performance benefits. Regarding those conditions, West’s analysis focused on complex decisionmaking groups, which are relatively autonomous; work in uncertain (and thus unpredictable) environments, using complicated technology that can itself be unpredictable; and perform complex tasks that require considerable coordination among group members, and which can also change over time. Moreover, several criteria can often be used to evaluate performance on these tasks, making it ambiguous at times whether a group is performing well or poorly. Any of these boundary conditions can be re-conceptualized as a moderator variable that might shape the effects of group reflexivity on performance, either strengthening or weakening such effects. Several of these variables were later studied, producing results that generally supported West’s analysis. And many other moderator variables were later proposed. These variables include the depth of the group’s reflection process (Schippers,

Group Reflexivity and Performance

67

Den Hartog, & Koopman, 2007; Gurtner et al., 2007); the strength of its ‘‘safety climate’’ (Edmondson, 1999, 2002; Fay, Borrill, Amir, Haward, & West, 2006), how much group members already know about their task, before they reflect (Hackman & Wageman, 2005), and how motivated group members are to learn more about their task (Bunderson & Sutcliffe, 2003). For us, however, the most interesting portion of West’s (1996) paper was his analysis of whether group reflexivity improves group performance. There was little research on this issue at the time his paper was written, so West focused on the performance benefits of various processes that he viewed as components of group reflexivity. These included task communication (which enables the growth of shared mental models among group members about their task and about one another), planning (including the development of better strategies), constructive controversy among group members about their task, and problem identification by group members. There is much research related to each of these group processes, and that work generally shows that they are beneficial. Thus, a group performs better when its members have shared and accurate mental models about their task and about one another (DeChurch & Mesmer-Magnus, 2009), engage in planning and develop effective task performance strategies as a result (Wittenbaum, Vaughan, & Stasser, 1998); experience conflict regarding the task (Nemeth & Staw, 1989), and try to identify problems associated with their task performance (Moreland & Levine, 1992). All this research led West to conclude that group reflexivity itself is beneficial. West’s conclusion surprised us, in part because it seemed premature. Why? Although shared mental models, planning, constructive controversy, and problem identification may all be associated with group reflexivity, we would not equate them with group reflexivity. Imagine a researcher who wanted to perform an experiment on the performance effects of group reflexivity. Would that researcher compare the performance of groups from a planning versus a no planning condition? Probably not, because experiments of that kind have already been done. Instead, the researcher would compare the performance of groups from a reflection versus a no reflection condition, and maybe (as West suggested in his 1996 paper) try to record the behavior of group members while they reflect, to see whether they actually do much planning, and how the amount of planning that is done relates to the group’s performance. So, rather than simply equating reflection in groups with shared mental models, planning, constructive controversy, and problem identification, it seems to us that a better approach is to view group reflection as an opportunity for such processes to occur. Put another way, why not view shared mental models, planning,

68

RICHARD L. MORELAND AND JAMIE G. MCMINN

constructive controversy, and problem identification as possible mediators of any effects that group reflexivity has on performance? One advantage of this approach is that other plausible mediators can be added to the list. Examples include goal setting (O’Leary-Kelly, Martocchio, & Frank, 1994) and the strengthening of group cohesion (Mullen & Copper, 1994). Another advantage is the realization that group reflection can also provide opportunities for negative processes to occur, thereby harming group performance. Consider, for example, the potential for destructive conflict among group members when they reflect (De Dreu & Weingart, 2003), or for a weakening of the group’s collective efficacy (Lindsley, Brass, & Thomas, 1995) when reflection proves to be unhelpful.

SCIENTIFIC RESEARCH ON GROUP REFLEXIVITY AND PERFORMANCE Let us return now to the second set of papers that was mentioned earlier, namely papers that describe scientific research on group reflexivity and performance. Most of that research can be traced, directly or indirectly, to West’s (1996) analysis. These papers could be summarized in several ways (see Widmer et al., 2009), but we have chosen (rather irreverently) a ‘‘good news/ bad news’’ approach. We will thus begin with papers that seem to show better group performance after reflection, and then consider papers that seem to show no benefits of reflection for group performance.

THE GOOD NEWS At first glance, there seems to be a lot of ‘‘good news’’ regarding the performance benefits of group reflection. This research includes case studies (Barry, Britten, Barber, Bradley, & Stevenson, 1999; Edmondson, 2002; Pacanowski, 1995; Pisano, Bohmer, & Edmondson, 2001; Stiltanen, Willis, & Scobie, 2008; Sutton & Dalley, 2008; Vashdi et al., 2007; West, 2000); correlational studies (Carter & West, 1998; Gevers, van Eerde, & Rutte, 2001; Hirst & Mann, 2004; Hoegl & Parboteeah, 2006; Lee, 2008; Somech, 2006); and experiments (Lewis, Belliveau, Herndon, & Keller, 2007; Muller, Herbig, & Petrovic, 2009; see also Blickensderfer, Cannon-Bowers, & Salas, 1997b; Salas, Nichols, & Driskell, 2007; Smith-Jentsch, Cannon-Bowers, Tannenbaum, & Salas, 2008; Tannenbaum et al., 1998). But after reading these

Group Reflexivity and Performance

69

papers, we were largely unimpressed by the evidence that they contained. Much of the research suffered from methodological problems, which raised doubts about the conclusions drawn by the researchers. In our opinion, this research does not really offer convincing evidence that groups perform tasks better after reflecting on them. The methodological problems with case studies, which report largely qualitative data on just one or a few groups, are well known, and some of these problems were apparent in the case studies that we read on group reflexivity and performance. One problem was whether the groups that researchers chose to study were representative, making the results generalizable to other groups that were not studied. Case studies involving more than one group were less problematic in this regard, but such studies seldom involved more than a handful of groups, so the problem was not eliminated. It would also have been helpful if researchers explained how they chose the groups that they studied, but they rarely did so, and it seemed to us that the groups described in some reflexivity case studies were probably chosen because their performance had benefited from reflection. Another methodological problem with these case studies was that some of the data that researchers collected about group reflection and/or performance may have been inaccurate. For example, none of these researchers actually observed group reflection, despite strong advocacy for that practice by West (1996). Instead, they simply asked group members to recall how much they reflected, how they reflected, and so on. Memories about group reflection might well be inaccurate, especially when people are trying to remember experiences that occurred long ago.2 And such memories could be biased by implicit theories about group performance (see Guzzo, Wagner, Maguire, Herr, & Hawley, 1986; Martell & Guzzo, 1991; Staw, 1975). Many people have formulated their own personal ‘‘theories’’ about why groups succeed or fail. Such theories often include group process variables, such as reflection. In fact, reflection (aka strategizing) is viewed by many people as a key factor in group performance (see Guzzo et al., 1986; Peterson, 1992). Someone who views his or her group as successful may thus ‘‘remember’’ more reflection in that group than actually took place. The bias can work in the other direction too – someone who remembers a lot of reflection in his or her group might view that group as more successful than it actually is. One way in which this bias (working in either direction) could be weakened is to collect data on group reflection or performance from independent sources, and indeed, some of the studies that we read boasted this feature. But notice that just collecting data from multiple sources does not eliminate the problem. Insofar as other people have developed implicit theories about

70

RICHARD L. MORELAND AND JAMIE G. MCMINN

group performance that resemble the theories held by a group’s members, their descriptions of that group will be biased in similar ways. Causality is yet another methodological problem associated with case studies. In case studies of group reflection, researchers typically ask group members to describe their reflection practices and performance over a single, specified period of time. But all such data can show is that there is a correlation between group reflection and performance, not that reflection causes groups to perform better. A different, yet still acceptable interpretation of the results from such studies is that both group reflection and performance were caused by some third (unmeasured) variable, creating an apparent effect of reflection on performance. If we consider the reasons why some groups reflect more than others, then many candidates for that third variable come to mind. For example, West (1996) speculated that group reflection arises from especially good or poor group performance, conflict among group members (associated perhaps with diversity in membership), efforts by members to cope with problems of time allocation or synchronization, and a positive affective tone in the group (e.g., group cohesion or a ‘‘safety climate’’). Research has now confirmed that several of these factors are indeed associated with group reflection, and other factors have recently been identified as well, including group feedback seeking (West, 2000), cooperative interdependence among group members (De Dreu, 2007; Schippers, Den Hartog, Koopman, & Wienk, 2003; Tjosvold, Hui, & Yu, 2003; Tjosvold, Tang, & West, 2004), a sense of collective efficacy (Edmondson, 1999; Gevers et al., 2001), motivated members who prefer an ‘‘active’’ approach to the group’s tasks (Brav et al., 2009; Lantz, In press; Lee, 2008; Schippers et al., 2007), and leadership carried out in a transformational or ‘‘coaching’’ style (Edmondson, 1999; Hirst, Mann, Bain, Pirola-Merlo, & Richter, 2004; Schippers et al., 2008; Somech, 2006). Many of these factors, and some of those identified by West, have been shown to have positive effects on performance, as well as on group reflection. That might explain why positive correlations between group reflection and performance arise and are sometimes (mis)interpreted as evidence that reflection has ‘‘caused’’ improvements in group performance. Finally, even when researchers who do case studies of group reflexivity and performance correctly interpret the correlation between those variables, the exact strength of that correlation is usually unclear. Qualitative data do not allow an actual correlation to be calculated, so its size is unknown, which can create further difficulty later on (e.g., the inability to predict performance from group reflexivity with any precision). This particular problem, at least, can be avoided by moving away from case studies and performing

Group Reflexivity and Performance

71

correlational studies of group reflexivity and performance instead. Correlational studies can also solve the problem of generalizability, if a large (and thus more representative) sample of groups is studied. Unfortunately, many of the other methodological problems mentioned earlier regarding case studies are still worrisome. The data on group reflexivity and/or performance in correlational studies may still be inaccurate, due to possible biases associated with respondents’ implicit theories about group performance. And correlations between group reflexivity and performance may still reflect the influence of other factors that have affected both variables, rather than any causal effects of group reflexivity on performance. Other problems can arise in correlational studies of group reflexivity and performance as well. We saw evidence of such problems in several of those studies. For example, in nearly all the studies (except for Gevers et al., 2001), the proportion of group members who chose to participate was disappointing, raising issues of self-selection effects on the results. Yet none of the researchers investigated whether people who chose to participate in their study differed from those who did not, which made it impossible to see whether and how self-selection might have influenced their research results. And in many of the studies, there were measurement problems involving either group reflection or performance.3 Carter and West (1998), for example, used a group reflexivity scale that actually embodied two different factors (social reflexivity and task reflexivity) rather than one, which complicated the interpretation of their results. When group reflexivity and/ or performance scales are administered to individual group members, researchers should also check whether responses from people who belong to the same groups are similar enough to warrant the aggregation of those responses into group scores. Yet some researchers (e.g., Hirst & Mann, 2004) did not do this. In fact, Lee (2008) appeared to use the individual (rather than the group) as the unit of analysis, violating the criterion of response interdependence (given that all the respondents were group members) that must be met for most statistical tests to be performed. Researchers must also consider how strongly group reflexivity correlates with other relevant predictors of group performance. In all these studies, reflexivity was correlated strongly with several other aspects of group process, making it unclear how much those variables contributed to the correlation between group reflexivity and performance. Even when all these problems are avoided, correlational studies can produce results that are disappointing. In several of the studies we read, for example, the correlation between group reflexivity and performance was positive, but small (Carter & West, 1998; Hirst & Mann, 2004; Hoegl &

72

RICHARD L. MORELAND AND JAMIE G. MCMINN

Parboteeah, 2006). And in all the studies, group reflexivity was positively correlated with some measures of group performance, but not with others. Performance is a complex construct, of course, with several different aspects, so it may be naı¨ ve to expect that group reflexivity will affect every aspect of group performance in the same way. Before moving on to discuss experiments on group reflexivity and performance, there is another set of correlational studies that should probably be considered, because their results can also be interpreted as ‘‘good news.’’ Group reflexivity was treated in these studies as a moderator or a mediator variable, something that could strengthen the helpful effects of other factors on performance (moderation)4 or that is necessary for such effects to occur at all (mediation). The moderating effects of group reflexivity were studied by Borrill et al. (2000), De Dreu (2002, 2007), and Fay et al. (2006). De Dreu (2002) found that the performance benefits of minority dissent were stronger in groups with greater reflexivity. In all the other studies, the performance benefits of member diversity were found to be stronger in groups with greater reflexivity. The mediating effects of group reflexivity were studied by Drach-Zahavy and Somech (2001), Edmondson (1999), Hirst et al. (2004); Schippers et al. (2008), Schippers et al. (2003), Somech (2006), Tjosvold et al. (2003), and Tjosvold et al. (2004). Drach-Zahavy and Somech found that reflexivity mediated the performance benefits of both member diversity and frequent group meetings; Edmondson found that reflexivity mediated the performance benefits of group safety climates; Hirst and his colleagues found that reflexivity mediated the performance benefits of both experiened leadership and ‘‘coaching’’ leadership; Schippers and her colleagues (2008) found that reflexivity mediated the performance benefits of transformational leadership; and Tjosvold and his colleagues found (in both their studies) that reflexivity mediated the performance benefits of cooperation among group members. In the other studies, more complicated forms of mediation were found. Schippers and her colleagues (2003), for example, found that reflexivity mediated the performance benefits of two different interactions; one of these involved member diversity and group longevity and the other involved member diversity and outcome interdependence. Somech found that reflexivity mediated the performance benefits of an interaction involving member diversity and participative leadership. Although these moderation and mediation studies suggest that group reflexivity can be valuable, they sometimes suffered from the same methodological problems that were described earlier. And in a few of the papers, we found direct correlations between group reflexivity and

Group Reflexivity and Performance

73

performance that were disturbing. For example, there were negative correlations between group reflexivity and performance in both of the De Dreu studies. Moreover, group reflexivity was uncorrelated with at least some performance measures in the studies by Hirst et al. (2004), Somech (2006), and Tjosvold et al. (2004). The opposite ‘‘problem’’ arose in the Edmondson (1999) and the Schippers et al. (2003) studies, where reflexivity correlated very strongly with some performance measures, so strongly that we doubted whether the researchers had actually distinguished (psychometrically) between those constructs. Many of the methodological problems that plague case studies and correlational studies can be avoided by performing experiments on group reflexivity and performance. In particular, a well-designed experiment can provide convincing evidence that reflection (independently of any other factors) actually causes group performance to improve.5 Several experiments on group reflexivity and performance have indeed been done, but most of them involve a special training technique called ‘‘guided team selfcorrection’’ (Smith-Jentsch et al., 1998), and their results are mixed. We will thus set those experiments aside for a moment and turn instead to the other two experiments that appear to show performance benefits of group reflexivity, namely Lewis et al. (2007) and Muller et al. (2009). The experiment by Lewis et al. is impressive, which seems ironic because their research was not primarily about reflexivity at all, but rather about transactive memory. Transactive memory (Wegner, 1987) is a shared awareness among group members of who is good at what. That awareness can arise by several means, but experience is often an important factor – as group members spend more time together, engaging in a wider variety of activities, they have more opportunities to learn about one another’s abilities. Many laboratory and field studies (see DeChurch & MesmerMagnus, 2009) have now shown that groups perform better when their transactive memory systems are stronger, perhaps because of improvements in their planning, coordination, and problem solving (see Moreland, 1999). However, groups that rely too heavily on transactive memory systems are vulnerable to certain performance threats. In particular, turnover among members can weaken a transactive memory system, even converting it from an asset to a liability. When people leave (or enter) a group, they often take (or bring) with them valuable expertise. As a result, those who remain in the group may have difficulty tracking exactly who knows what. The main purpose of the research by Lewis et al. (2007) was to assess how serious this problem is and learn more about why it occurs. In their primary experiment, the dynamics and performance of groups in various turnover

74

RICHARD L. MORELAND AND JAMIE G. MCMINN

conditions were compared. In one condition, groups experienced no turnover at all. In another condition, partial turnover occurred – one member left, to be replaced by a newcomer. In the last condition, total turnover occurred. Not surprisingly, performance (on an electronics assembly task) by the groups that experienced no turnover was better than that of groups in the other conditions, especially the partial turnover condition, where group performance was worst. And these performance differences were indeed related to various aspects of the transactive memory systems within those groups. Groups in the partial turnover condition were especially interesting in this regard. The researchers wondered whether the ‘‘oldtimers’’ in those groups would try to adapt their transactive memory systems in ways that reflected the expertise of their new members, or try instead to maintain the transactive memory systems that they had developed before those new members arrived. And what would the new members themselves do? Would they simply maintain their current expertise, or try to adapt that expertise to better ‘‘fit’’ into the transactive memory systems of the groups that they entered? As it turned out, the oldtimers maintained and the newcomers adapted, and insofar as these things occurred, the performance of these groups suffered. Lewis et al. thought that group reflection might help to solve this problem, so they performed an auxiliary experiment involving a replication (with new participants) of the partial turnover condition from the original experiment. But this time, the oldtimers from each group were led to reflect (privately, soon after the newcomers arrived) on the breadth and depth of their task expertise. The researchers suspected that this might help them to understand their transactive memory systems better and maybe to realize that they should adapt those systems to incorporate the expertise of the new members. That is exactly what happened, and to the extent that it occurred, the performance of these groups improved, reaching levels comparable to those of the no turnover groups from the original experiment. So, group reflexivity had performance benefits. The experiment by Muller et al. (2009) was less impressive. Those researchers asked groups of engineering students to design a new product. Experts later evaluated their designs on several dimensions, producing performance measures. In a control condition, groups completed their task without interruption. But in two other conditions, groups were interrupted soon after they began to work, to carry out (with some guidance) a rather complex ‘‘explication’’ procedure. In one condition, that procedure was meant to help individual group members (separated from one another) discover any tacit knowledge that they possessed about the task. The researchers expected this to help those groups perform better, in part because

Group Reflexivity and Performance

75

individual members could share with each other what they knew, making more knowledge available to the group as a whole. A comparable procedure was used in the other condition, but this time group members discussed the task with each other, to discover any tacit knowledge that they possessed about relevant group processes. After they completed the explication process, groups in both of these conditions went back to the product design task, working under the same circumstances as groups in the control condition. The results showed that groups in the individual and the group explication conditions outperformed groups in the control condition. And there was evidence that the performance benefits of group explication were greater than those of individual explication. Those benefits appeared to arise from the time and energy that groups saved (while working on the task) in discussing their goals. All these results, however, must be interpreted with caution. The differences in performance across conditions were seldom significant, due perhaps to the fact that the total sample was small in size. This led the researchers to analyze the data in unusual ways (‘‘exact test procedures’’) that focused on effect sizes rather than significance levels. And many of the effects observed in the experiment were restricted to a subset of the performance measures. As we noted earlier, there is another set of experiments related to group reflexivity that involves guided team self-correction, a special training technique originally developed for the United States Navy. According to Smith-Jentsch et al. (2008), ‘‘Guided team self-correction is a team debriefing strategy in which members are given the responsibility for diagnosing and solving their team’s performance problems, with guidance as to what topics they should discuss and how to do so constructively’’ (p. 304). A more detailed description of the many activities associated with this technique can be found in Smith-Jentsch et al. (1998). Guided team selfcorrection is complex and requires considerable instruction and practice for everyone involved (trainers and trainees). The experiments on guided team self-correction include Blickensderfer et al. (1997b), Tannenbaum et al. (1998), and Smith-Jentsch et al. (2008). A meta-analysis by Salas et al. (2007) is also relevant. That analysis evaluated the effects of several different team training strategies (including guided team self-correction) and then explored why they are helpful. We debated where all this work should be located in our chapter. Guided team self-correction certainly involves group reflexivity, and many analysts believe that such training can improve group performance. But as we shall see, research on guided team self-correction has produced unimpressive results, so maybe we should have moved it to the ‘‘bad news’’ portion of the chapter.

76

RICHARD L. MORELAND AND JAMIE G. MCMINN

Blickensderfer et al. (1997b; see also Blickensderfer, Cannon-Bowers, & Salas, 1997a) carried out one of the first experiments on guided team selfcorrection. In that experiment, teams of college students worked on a computer simulation of tactical naval decision making. The monitor served as a ‘‘radar screen’’ that displayed information about each team’s own ship and several unidentified contacts. Every team’s task was to identify those contacts by type (air, surface, or subsurface), status (civilian or military), and intention (hostile or peaceful). This was accomplished by accessing information about the contacts on the computer, passing such information among team members, and then making decisions based on the information. The experiment began with an hour of practice to familiarize teams with the simulation. After a brief break, the teams actually worked on the simulation for awhile and discussed their performance. Next, half of the teams received self-correction training, while the rest carried out a control activity. The self-correction training included (a) an audio-taped lecture that explained why team self-correction is useful and how it should be done, and presented brief vignettes that illustrated effective and ineffective team selfcorrection behaviors; and (b) two practice sessions in which the teams first completed an exercise, then discussed their performance, and finally received feedback from a trainer about how those discussions could be improved. Afterwards, all the teams returned to the computer simulation and worked on it twice more, discussing their performance after each session. All team discussions were recorded and later evaluated by trained observers for several behaviors. These evaluations yielded measures of how much team self-correction occurred and how well team members communicated with one another. And questionnaire responses by team members yielded measures of shared task expectations and team cohesion. Finally, the computer simulation generated scores that reflected how well teams performed. The results showed that self-correction training produced several benefits, including more team self-correction behavior, more shared task expectations, better communication among team members, and greater team cohesion. However, self-correction training did not affect team performance. Smith-Jentsch et al. (1998) developed and implemented, over several years, a one-day workshop that the Navy used to instruct officers in a process called ‘‘team dimensional training,’’ a version of guided team selfcorrection that involved various activities, including pre-briefings and debriefings, modeling of appropriate task behaviors, and coaching. The goal of these workshops was to help officers learn how to train groups of sailors to perform tasks more effectively. The researchers began by evaluating the task mental models (using a card-sorting task) of officers who participated

Group Reflexivity and Performance

77

in the workshop, then compared those models with an ‘‘expert’’ model derived from various sources (e.g., comparing the mental models of people already known to be successful versus unsuccessful at the task). The results showed that the officers’ mental models were more similar to the expert model after their training than before. When the officers later began to apply what they had learned by actually training groups, records were kept on those groups, including the task behaviors and mental models of their members and levels of group performance. Comparable records were also kept on groups whose members were trained in more traditional ways. In their paper, Smith-Jentsch et al. focused on a single group, one that had recently participated in the workshop. This group was asked to perform a series of progressively more difficult tasks. Task performance was evaluated by experienced observers, who rated the group on several dimensions. The performance of this group was then compared to that of five comparable groups (from the database) that had not participated in the workshops. Although the performance of those groups had declined as the tasks became more difficult, the performance of the target group actually improved. This was interpreted as evidence for the effectiveness of guided team self-correction. Tannenbaum et al. (1998) performed an experiment in which naval officers received either training in briefing skills or control training. Training in briefing skills included (a) a lecture about learning cycles in groups, effective briefing behaviors, and how briefing behaviors relate to group learning; (b) trainee evaluations of a videotape that showed someone actually debriefing a group; (c) a role-playing exercise in which trainees practiced debriefing a fictional group (portrayed on another videotape); and (d) self-critiques by the trainees of their own briefing behaviors, along with a critique of those behaviors by the trainer. The control training, which was equally long, involved practice in how to use a computer console that was part of a larger combat information center. The officers were trained separately from the sailors whose groups had to be trained (the sailors received the control training too). Each group of sailors was asked to perform several tasks (over the course of a day), with briefings and debriefings before and after each task. The tasks became more difficult over time. All the tasks involved simulated crises (presented using computers) that required group members to cooperate with each other and work in coordinated ways. Experienced observers evaluated the behaviors of the officers and the sailors, including several aspects of teamwork. Evaluations were also made of group performance (how quickly the groups reacted to each crisis and how appropriate their reactions were).

78

RICHARD L. MORELAND AND JAMIE G. MCMINN

The results showed that (a) officers trained in briefing skills did a better job of briefing their groups (more so after each crisis than before) than did officers who received the control training; (b) groups led by officers trained in briefing skills were more likely than other groups to display self-correction behaviors (reflexivity), especially after each crisis; and (c) many (but not all) kinds of teamwork behavior occurred more often in groups led by officers trained in briefing skills, when compared with groups led by officers who received the control training. Perhaps as a result (though no mediation tests were done), the latter groups also performed their tasks less well. Two other experiments were later performed by Smith-Jentsch et al. (2008), who were disappointed that guided team self-correction had failed to produce any performance benefits in the original experiment by Blickensderfer et al. (1997b). That failure was attributed to several factors, including (a) facilitators may have stressed the importance of shared task expectations among group members, without ensuring that such expectations were accurate; (b) members of the teams had little experience with the task they had to perform, making it difficult for them (even with a facilitator) to develop effective strategies; and (c) facilitators were similar to group members in status (based on work expertise), limiting how much influence could be exerted by the facilitators within the groups. SmithJentsch et al. decided that new research was thus needed, research designed to overcome these limitations. In their first study, which lasted for two years, Smith-Jentsch et al. analyzed naval submarine attack center groups. During the first year, one set of groups (forming a kind of control condition) was briefed before and after each of its exercises, using the Navy’s standard methods. Meanwhile, a cadre of officer instructors was trained (at great length) in the briefing methods associated with guided team self-correction. During the second year, another set of groups was given briefings before and after its exercises by these instructors. All the groups were later asked to participate in two special battle exercises. Before each exercise, groups in the control condition received a briefing that simply described their mission and equipment, as well as the data that would be available to them. Afterwards, their debriefing was simply a summary of what occurred during the exercise. Groups in the experimental condition were given a similar briefing before each exercise, but they were also told that an important goal of their training was to learn an expert model of how to act during the exercise. Members of these groups expected to evaluate themselves later on, during their debriefing, for compliance with that model. After each exercise, groups in the experimental condition were given a quick summary of what occurred, followed by an

Group Reflexivity and Performance

79

analysis of the group’s behavior, focusing on ways in which it did and did not match the expert model. Finally, keeping the expert model in mind, group members were asked to generate specific goals for improvement on future exercises. The main dependent measures in this experiment were the accuracy of group members’ mental models about teamwork, and the degree to which group members shared those models. The results showed that groups in the experimental condition had more accurate mental models about teamwork than did groups in the control condition, but there was no difference between conditions in the degree to which those models were shared. In their second experiment, Smith-Jentsch et al. assigned officers to receive either task training related to the submarine attack center, or training in how to guide teams using self-correction practices. A set of groups was then asked to perform a series of exercises (similar to those just described), led by someone trained in one of those two ways. During the last of these exercises, each group was videotaped while it worked, allowing trained observers to later evaluate the quality of group members’ teamwork. The same observers made evaluations of each group’s performance as well. The results showed that the groups trained in guided self-correction displayed better teamwork behaviors (on most, but not all the dimensions evaluated) than did the groups that were trained more traditionally. Guided self-correction training also led to better performance during the exercise. Finally, the effects of guided team self-correction were also explored in an unusual meta-analysis by Salas et al. (2007). Salas and his colleagues first identified a set of seven studies in which researchers had assessed the effects of some training intervention on group performance. Two judges then evaluated every intervention to assess its reliance on each of three broad strategies, namely (a) cross-training, (b) coordination/adaptation, and (c) guided self-correction. Overall, the results of the meta-analysis showed that all training interventions improved group performance, although this effect was not large in size. But when the data were examined more closely, in terms of how much the various training interventions relied on each of the three strategies, different results emerged. Training interventions that relied on cross-training had little impact on group performance. Stronger effects were found for interventions that relied on guided self-correction; interventions that relied on coordination/adaptation had the strongest effects. However, several correlations were also found among the reliance scores for the three training strategies. For example, interventions that relied on crosstraining did not tend to rely on coordination/adaptation or guided selfcorrection; interventions that relied on coordination/adaptation also tended to rely on guided self-correction. These correlations led the researchers to

80

RICHARD L. MORELAND AND JAMIE G. MCMINN

estimate (statistically) the independent effects of each training strategy on group performance. The results showed that only the coordination/ adaptation strategy had such effects. The cross-training and guided selfcorrection strategies had no effects of their own on group performance, once their correlations with other strategies were considered. So, guided team selfcorrection did not (on its own) appear to be an effective way to train groups, raising further doubts about how much group reflexivity contributes to performance.

THE BAD NEWS Some of the ‘‘good’’ news about the performance benefits of group reflexivity was not all that good. But there is also some actual bad news to report – studies that show no performance benefits after groups reflect on a task. This research again includes case studies (Goodman & Wilson, 2000), correlational studies (Brav et al., 2009; Edmondson, Bohmer, & Pisano, 2001; Savelsbergh, van der Heijden, & Poell, 2009), and experiments (Daudelin, 1996; Gurtner et al., 2007; Rasker et al., 2000). There are fewer studies in this section of the chapter than in the preceding section, but does that really mean that research generally favors group reflexivity? Not necessarily, because there may be some researchers who investigated the performance benefits of group reflexivity, but failed to find any benefits, and thus did not even try to publish their work, or tried to publish it, but failed.6 Goodman and Wilson (2000) conducted a case study of computer emergency response teams, which are responsible for dealing with Internet attacks. After one of these attacks, the researchers observed a particular team struggle with media requests for information, requests that interfered with the team’s operation. During a subsequent review of this incident, team members discussed what had happened and decided that they would try to avoid such problems in the future by holding press conferences early in the process of responding to new attacks. In other words, team members decided to be more proactive than reactive in their media relations. Yet when later attacks indeed occurred, the team did not actually hold any press conferences, and so problems associated with media attention continued to arise. Group reflection thus failed to improve this team’s performance. As for correlational studies, Brav et al. (2009) investigated work groups in large manufacturing companies. One of the researchers’ goals was to discover why some groups reflected more often than others. Reflexivity was expected to depend on several characteristics of job design and the levels of

Group Reflexivity and Performance

81

social support and cooperation within the groups. The researchers were also interested in the effects of reflexivity on two work group outcomes, namely initiative and self-organizational activities. Both of those outcomes were expected to depend on social support and cooperation, as well as on reflexivity. The results showed that group reflexivity was indeed affected by job design, but not by social support or cooperation. Both social support and cooperation affected group initiative, but reflexivity did not. The only thing that group reflexivity affected was self-organizational activities. Edmondson et al. (2001) studied the implementation in several hospitals of a new procedure for cardiac surgery, one that is less invasive than the traditional procedure. At issue was why some surgical teams adjusted to the new procedure more quickly than did others. Several variables that the researchers expected to be important predictors of adjustment proved to be unimportant. Among those variables were debriefings, project audits, and after-action reports, all forms of reflexivity. Surgical teams seemed to do most of their learning during actual operations, by doing things in different ways and evaluating how such changes were affecting the patients’ outcomes. Finally, Savelsbergh et al. (2009) studied customer service teams working in a bank. Workers completed a survey that measured several team learning behaviors, some of which involved reflection (on team processes and outcomes). Evaluations of team performance were obtained from team members and from their supervisors. Although some team learning behaviors (e.g., exploring different perspectives, creating shared meanings) were indeed related to team performance, behaviors involving reflection were not. As for experiments, Daudelin (1996) studied the effects of reflection on managers in a large company. Each person was assigned to either a control condition (no reflection), or to one of several reflection conditions (reflect alone, reflect with a partner/coach, reflect with a group of peers). Participants were provided with guidelines by the researchers about how to reflect. All the reflection sessions (with lasted for one hour) were videotaped for later analysis. Reflection focused on a ‘‘challenging work experience’’ selected by each participant. After reflection, the participants completed a questionnaire that measured their satisfaction with the experience and asked them to describe anything they had learned from it. Several other variables that might have affected reflection (e.g., the severity of the work problem) were also measured. Control participants were asked, over the telephone, to identify a work problem themselves, and then to describe whatever they had learned while solving it. The results revealed no differences in satisfaction across conditions. However, learning differences were found – participants who

82

RICHARD L. MORELAND AND JAMIE G. MCMINN

reflected alone or with a partner/coach learned more than did participants who reflected with a group of peers or did not reflect at all. And participants who reflected with a group of peers did not learn more than those who did not reflect at all. Why wasn’t group reflection more helpful? A review of the videotapes suggested three reasons, namely that (a) group members emphasized personal similarities, rather than differences, when discussing problems; (b) group discussions were rather ‘‘shallow,’’ moving quickly from one topic to the next, without exploring anything in much detail; and (c) groups were less likely than individuals to follow the reflection guidelines provided by the researchers. Rasker et al. (2000) performed two experiments, both involving small groups (dyads) of college students working on a computer simulation that allowed them to ‘‘fight fires’’ in a virtual ‘‘city.’’ One group member was supposed to observe what was happening in the city and share that information with his or her partner. The other member was supposed to allocate available resources (e.g., personnel, equipment) to help extinguish any fires that were observed. At the beginning of each experiment, both group members received an instruction manual describing their roles and explaining the simulation’s operation. Each group then took part in two training sessions, both involving more than a dozen fire-fighting scenarios. Each member interacted with a simulated ‘‘partner’’ during this period, someone who played the complementary role perfectly. Afterwards, groups were assigned to treatment conditions and the experimental session (which also involved more than a dozen scenarios) was held. Verbal behavior (by the real participants) during that session was recorded, so that later evaluations of discussion content could be made by trained observers. And the computer simulation generated scores reflecting the performance of the groups. In the first experiment, half the groups were assigned to an unrestricted communication condition in which group members sat together and could talk with each other at any time (both during and between scenarios) about what was happening. The other groups were assigned to a restricted communication condition in which group members were separated and could only communicate (during, but not between scenarios) by e-mail, using standard messages provided by the researchers. The results showed that groups in the unrestricted communication condition performed better than those in the restricted communication condition. Content analyses of the group discussions in the unrestricted communication condition showed that people were more likely to make ‘‘activity-based’’ comments (e.g., information exchange, performance monitoring) than ‘‘task-related’’ comments (e.g., evaluation, strategy determination, sharing knowledge about

Group Reflexivity and Performance

83

the task or the group). Evaluation comments, which included analyses by group members of things they had done well or poorly, seemed to embody reflection, suggesting that group reflexivity had few performance benefits for these groups. However, no internal analyses were carried out by the researchers to see whether there was any correlation between the number of evaluation comments that group members made and the overall performance of their group. In the second experiment, group members always sat together and could thus speak with each other freely about what was happening. Half of the groups, however, were only allowed to communicate during (but not between) scenarios. The remaining groups were only allowed to communicate between (but not during) scenarios. Group reflection, in the form of evaluations, could have occurred in either of these conditions, but evaluations made between scenarios seem closer to the traditional notion of group reflection. Yet the results showed that groups performed better when their members could communicate only during the scenarios, rather than between them. Moreover, within the groups that communicated only during the scenarios, members exchanged more activity-based than task-related comments (including evaluations), whereas the opposite pattern arose in groups that communicated only between the scenarios. Reflection thus appeared to be harmful, rather than helpful, for group performance. Once again, however, the researchers carried out no internal analyses to see whether there was any correlation between the number of evaluation comments that group members made and the overall performance of their group. Finally, an impressive experiment by Gurtner et al. (2007) also showed no performance benefits of group reflection. In that experiment, small groups of college students participated in a simulated military air surveillance task. Each group contained two ‘‘specialists’’ of lower rank and a ‘‘commander.’’ Group members worked at the task using computers and communicated with one another by e-mail. On the computer monitors, ‘‘airplanes’’ could be seen moving in a military environment. The group’s goal was to classify these airplanes correctly, according to their threat levels. To achieve this goal, information was needed about several characteristics of the airplanes, including their size, altitude, airspeed, and flight direction. That information was distributed to the specialists in such a way that neither person had all the information needed. So, specialists had to either share information with one another or send it to the commander. The commander’s job was to process and integrate the information (relying on a mathematical formula unknown to the specialists) and make a final decision about whether each airplane was friendly or not. That decision was then sent to the specialists.

84

RICHARD L. MORELAND AND JAMIE G. MCMINN

Everything that group members did (including the messages they sent to one another) was recorded on the computers, and thus available for later analysis. Each group was first taught how to operate the simulation. Then it had to complete seven work ‘‘shifts,’’ spread across two days. Three of these shifts occurred during the first day, and then one week later, the four remaining shifts occurred. After the final shift, group members completed a questionnaire that assessed their mental models regarding the task. Performance was measured by computing the proportion of airplanes correctly classified by the group during its final work shift. Each group was assigned to one of three reflection conditions. In the group reflection condition, group members discussed (through e-mail) three sets of questions, which led them to (a) review how they had carried out their tasks, (b) imagine how they might perform those tasks in other ways, and (c) consider what they could do to improve their task performance. In the individual reflection condition, group members answered those same questions, but they did so privately, writing their answers on paper. Finally, in the control condition, no reflection occurred at all – group members instead discussed (through e-mail) why some people are more successful than others in their careers. All of this occurred just before the first work shift on the second day. The results showed that the performance of groups in the individual reflection condition, but not that of groups in the group reflection condition, exceeded the performance of groups in the control condition. However, group performance did not differ between the individual and group reflection conditions. Individual reflection thus improved performance, whereas group reflection did not. Comparable results were found for several group process measures, including how often the commander sent strategy suggestions to the specialists, how often the specialists actually implemented strategies that were suggested, and similarities in the mental models of group members. Causal analyses revealed that the influence of reflection on group performance was mediated by differences among the groups in commander strategy suggestions, specialist strategy implementation, and similarity in group members’ mental models. This experiment is thus a rare example of research involving an attempt to understand why group reflection did (or did not) improve performance. The researchers explored that issue further by examining more closely the messages that group members exchanged. They found that commanders in the group reflection condition were less active than those in the other two conditions, in the sense that commanders in the other conditions initiated more communication with the specialists (rather

Group Reflexivity and Performance

85

than just responding to communication from the specialists). And compared with commanders in the other two reflection conditions, commanders in the group reflection condition suggested more general strategies to the specialists, which produced more shallow (and thus less useful) reflection in their groups.

OUR OWN RESEARCH ON GROUP REFLEXIVITY The unsettling combination of (a) bold claims about the performance benefits of group reflexivity, and (b) unconvincing scientific evidence for those benefits, led us to do research of our own on group reflexivity and performance. We would like to describe at least some of that work here. All our research is experimental in nature and done in a laboratory setting where three-person groups of college students work at a computer simulation (commercially available) called Casino Empire. In that simulation, each group controls a Las Vegas-style gambling casino that it must operate for two ‘‘years.’’ The general goal is to make the casino as profitable as possible by making wise decisions about its operation. Cash prizes are given to the groups that perform best. The simulation is very engaging, but also rather complex. Literally dozens of decisions can be made by each group, covering issues related to customers (e.g., what kinds of people are most desirable?; how can such persons be led to enter the casino, stay for a long time, and spend lots of money?); employees (e.g., how many and what kinds of people should be hired?; how much should employees who perform different jobs be paid?); casino facilities (e.g., how many and what kinds of gambling machines should be installed?; should bars and restaurants be available, and if so, then where they should be located and how much should be charged for the amenities they provide?); and the casino itself (e.g., how should it be decorated, inside and out?; what forms of security are needed?). To help with making these and other decisions, a group can call up on the computer many special screens that summarize what is happening in the casino (e.g., each customer’s personal background, how much he or she has spent, and how satisfied he or she is; how popular each game is and how much profit it has been generating; how hard each employee has been working). Group decisions must be entered into the computer software every ‘‘month.’’ That software then uses an internal algorithm to compute the impact of those decisions on the casino’s success. Several performance measures are made available to the group. These include the total number of customers who

86

RICHARD L. MORELAND AND JAMIE G. MCMINN

have visited the casino; a rating that reflects the popularity of the casino with its customers; the casino’s ranking, when compared to other casinos in the city; and how much cash there is in the casino’s bank account. To clarify the simulation’s operation for participants, we provide each group with a ‘‘tip sheet’’ before the simulation begins and allow it to keep that sheet thereafter. The tip sheet offers no suggestions about strategy or tactics, just information about the operation of the simulation.7 Participants in our experiments are told that we are studying ‘‘top management teams’’ to learn why some teams do better than others at managing their companies. Then we randomly assign participants to groups, which are randomly assigned in turn to our treatment conditions. Most of our experiments are simple, involving just a reflection condition and a no reflection (control) condition. Each group is first seated in front of the computer, told that it will be operating a simulated gambling casino, and then given the tip sheet to consult throughout the simulation. Next, the simulation is activated, and the group starts to make its decisions and learn how those decisions affect the casino. During that process, the group works by itself, isolated from the experimenter. It is possible for a group to ask the experimenter for help, but barring unusual circumstances, such as a computer failure, he or she provides no assistance (e.g., no advice about strategy or tactics). When the first ‘‘year’’ of casino operation is over, the experimenter returns, records how well the group’s casino performed, blanks out the computer screen (temporarily), and activates a video recorder that is mounted on the wall, aimed at the group. The group is then told that it will have a 10-minute ‘‘break.’’ Groups in the reflection condition are told to use that time to discuss ‘‘what went right and what went wrong’’ with their casino. Groups in the control condition are told to play the game Solitaire on their computer. We have found that playing Solitaire indeed prevents groups from reflecting on the operation of their casinos. The experimenter stays out of the computer room during the break. After the break, the experimenter returns to the room and reactivates the simulation. Each group then begins to operate its casino again, for another ‘‘year.’’ At the end of that time, the experimenter returns and debriefs participants about the true purpose of the research. They are then thanked for their help and dismissed. Our first experiment involved 315 students in 105 groups, which were assigned to either a reflection or a control condition. To evaluate the performance of these groups, we focused on popularity ratings, although any of the other performance measures would have served as well, because

Group Reflexivity and Performance

87

they are all strongly correlated. We performed a multiple regression analysis in which each group’s popularity rating at the end of Year 2 was predicted from (a) its popularity rating at the end of Year 1, and (b) a binary variable representing whether the group was in the reflection or the control condition. (We have also added other predictors to this analysis at times, such as the gender composition of the group, how many group members seemed to know one another before the experiment began, and how many group members had prior experience with computer simulations.) The results showed a strong positive effect of popularity in Year 1 on popularity in Year 2. However, popularity in Year 2 was completely unaffected by reflexivity condition. And the other predictors we have tried (see above) have proven to be unimportant as well. Why wasn’t reflexivity more helpful to the groups in our experiment? We examined the videotapes of their reflection sessions to find some clues. Two (related) aspects of those tapes were immediately apparent. First, many of the groups seemed uncomfortable about reflecting (e.g., awkward conversation, with long silences; avoidance of eye contact; fidgeting). Several analysts, writing about various groups, have also noted that reflection indeed seems to make groups uncomfortable (see Barry et al., 1999; Edmondson, 2002; West, 1996; Widmer et al., 2009). Second, few groups reflected throughout the time period made available for that activity. Reflection typically occurred in ‘‘bursts,’’ which became progressively shorter as time passed, and eventually ended altogether, usually before the break was over. Several groups even asked the experimenter whether they could start operating the simulation again sooner than planned (such requests were always denied). Participants spent a surprising amount of time during the break discussing topics unrelated to the simulation (e.g., where they lived, what courses they were taking, which bars they most enjoyed), rather than reflecting. Participants’ discomfort could, perhaps, be attributed to fear that group reflection would be a bad experience, with members arguing about the casino, blaming each other for the group’s failures, trying to take undeserved credit for the group’s successes, and so on. Note that most group members were strangers to one another, so they had no established relationships that might have clarified for them what to expect from a group discussion, or prevented a nasty discussion from occurring. If this interpretation of our videotapes is correct, then maybe group reflection would improve performance if we studied groups of friends, rather than strangers, or used some team-building activity to strengthen cohesion in groups of strangers, or assigned to each group a trained facilitator who could keep its discussion civil.

88

RICHARD L. MORELAND AND JAMIE G. MCMINN

Participants’ failure to use all the time that we made available for reflection could be attributed to the discomfort associated with reflection, or perhaps to the belief that reflection is useless. In fact, one of the participants in our first experiment exclaimed (to his group, while it was supposed to be reflecting) that ‘‘If they really want us to get better at this, then they ought to just let us keep playing – talking about it won’t help.’’ If this interpretation of our videotapes is correct, then maybe group reflection would improve performance if we attested to the value of reflection when we asked groups to reflect, or composed groups of people with higher levels of self-efficacy, or assigned to each group a trained facilitator who could keep discussion flowing, and focused on the casino. Perhaps what matters, when it comes to performance, is not how much a group reflects, but rather the quality of that reflection, or when it occurs. To assess the quality of reflection by our groups, we have coded transcriptions of their discussions made from the videotapes. On the basis of both theory (what social processes could occur during reflection that might affect group performance?) and pragmatics (what did the groups actually talk about as they reflected?), we have developed a content coding scheme and applied it (reliably) to our videotapes. In that scheme, every utterance is first classified as relevant or irrelevant to the operation of the casino. Relevant utterances are then classified further as examples of (a) identifying some problem; (b) noting some accomplishment by the group; (c) making plans for the future; (d) setting specific goals for the group; (e) theorizing about casinos or their customers; (f) agreeing with something another person said; (g) expressing confusion about the simulation; or (h) reading aloud from the tip sheet. We have performed many data analyses in which the frequencies with which various aspects of discussion content (e.g., the total number of relevant utterances; the balance between relevant and irrelevant utterances; the number of utterances from particular categories, or from subsets of the categories) were used as additional predictors of Year 2 popularity ratings in the regression analysis described earlier (analyses involving only groups in the reflection condition). Sadly, none of these analyses has produced very interesting results, in the sense that discussion content seems unrelated to changes in group performance. One option that we may pursue soon is a new kind of content analysis (see Ballard, Tschan, & Waller, 2008), carried out using special software that searches for sequences of utterances that are correlated with group performance. Two other possible explanations have also occurred to us regarding the apparent inability of group reflection to improve performance in our

Group Reflexivity and Performance

89

research. One of them involves the realization that group reflection occurs not only after tasks but also during tasks. Maybe the latter type of reflection has more impact on group performance (see Rasker et al., 2000). We did not videotape the groups while they operated their casinos, so we have no measures of ‘‘in-process’’ reflection, but such measures could be obtained in any new research. Another possible explanation is that individual and group reflection inhibit each other. As noted earlier, there is considerable evidence that individual reflection can improve performance. But by insisting that participants in our group reflection condition discuss the casino’s operation with one another, we may have prevented them from engaging in much individual reflection, and if it is the latter kind of reflection that really benefits performance, then that may explain why we did not find any performance benefits for group reflection. Interesting experiments could be done in which the effects of various combinations/sequences of individual and group reflection on performance are studied. A recent experiment by Baruah and Paulus (2008) on brainstorming by individuals and groups provides an illustrative example, albeit in a different topic area.

CONCLUSIONS We have several experiments on group reflexivity and performance underway and plans for more experiments in the future. All of this effort may be surprising, given our review of past research on group reflexivity (research that offers limited evidence for its performance benefits), and the results of our own research on group reflexivity, which has yet to reveal any performance benefits from that practice. Why not simply surrender and conclude that group reflexivity has no performance benefits? That conclusion is justifiable, we think, but maybe too extreme. Another possible conclusion that also seems justifiable, but is much less extreme, is that group reflexivity can have performance benefits, but only under very special conditions, some of which are hard to create (e.g., group members who are knowledgeable about the task and highly motivated, a task that is complex and yet susceptible to group analysis; a carefully trained facilitator who can spend time with the group, guiding its reflection in productive ways). If that is the right conclusion, then using reflexivity to help groups perform better may be possible, but not often feasible. Our real conclusion, judging from our behavior (see Bem, 1972), is that there is still hope for group reflexivity and that further research may show how it can be used to reliably enhance performance.

90

RICHARD L. MORELAND AND JAMIE G. MCMINN

NOTES 1. In a recent paper by Van Ginkel, Tindale, and van Knippenberg (2009), reflexivity was even conceptualized as something that can occur before a task is performed, but this makes little sense to us and seems potentially dangerous because it further blurs how reflection differs from other group processes, such as planning and goal setting. 2. And some analysts (e.g., Daudelin, 1996; Stiltanen et al., 2008) have argued that reflection is a natural, almost unconscious aspect of life, something that occurs continuously over time, rather than a special activity that only occurs occasionally, when group members decide to reflect or are led to do so by others. If these analysts are correct, then group members’ memories of reflection may not be very clear. 3. To be fair, measurement problems would probably just have weakened the results of these studies, without necessarily biasing them in a particular direction. 4. Group reflexivity could also help groups by weakening the negative effects of other group factors on performance, but no one seems to have studied moderation of that type. 5. Of course, experiments can suffer from methodological problems of their own, such as poor external validity. 6. To be fair, some of the measurement problems mentioned earlier may have also affected the research in this section of the chapter, and such problems could have weakened the results of that research, making the positive effects of group reflexivity on performance undetectable. 7. A copy of that tip sheet can be obtained by contacting the first author of this chapter.

ACKNOWLEDGMENTS We want to thank Brigitte Armon, Tara Fertelmes, Joshua Fetterman, Jeffrey Flagg, Steve Gonzalez, Michael Hansen, Katie Hayes, Joshua Hirshenhorn, Stephanie Lamison, and Vincent Losasso for their help in collecting, transcribing, and analyzing the data from our group reflexivity experiments. Our thanks also go to Kin Andersson, Elizabeth Blickensderfer, Agneta Brav, Stuart Bunderson, Andrea Gurtner, Britta Herbig, Annika Lantz, Norbert Semmer, Franziska Tschan, Michael West, and Pascale Widmer for their help with this chapter.

REFERENCES A Leader’s Guide to After-Action Reviews. (1993). Available at http://www.au.af.mil/.au/awc/ awcgate/army/tc_25-20/tc25-20.pdf. Retrieved on June 1, 2010.

Group Reflexivity and Performance

91

Argote, L. (1999). Organizational learning: Creating, retaining, and transferring knowledge. Norwell, MA: Kluwer Academic Publishers. Baird, L., Holland, P., & Deacon, S. (1999). Learning from action: Embedding more learning into the performance fast enough to make a difference. Organizational Dynamics, 27, 19–32. Ballard, D. I., Tschan, F., & Waller, M. J. (2008). All in the timing. Small Group Research, 39, 328–351. Barry, C. A., Britten, N., Barber, N., Bradley, C., & Stevenson, F. (1999). Using reflexivity to optimize teamwork in qualitative research. Qualitative Health Research, 9, 26–44. Baruah, J., & Paulus, P. B. (2008). Effects of training on idea generation in groups. Small Group Research, 39, 523–541. Bem, D. J. (1972). Self-perception theory. In: L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 6, pp. 1–62). New York: Academic Press. Blickensderfer, E., Cannon-Bowers, J. A., & Salas, E. (1997a). Theoretical bases for team self-correction: Fostering shared mental models. In: M. M. Beyerlein, D. A. Jackson & S. T. Beyerlein (Eds), Advances in interdisciplinary studies of work teams (pp. 249–279). Greenwich, CT: JAI Press. Blickensderfer, E., Cannon-Bowers, J. A., & Salas, E. (April, 1997b). Training teams to selfcorrect: An empirical investigation. Paper presented at the annual meeting of the Society for Industrial and Organizational Psychology, St. Louis. Borrill, C. S., Carletta, J., Carter, A. J., Dawson, J. F., Garrod, S., Rees, A., Richards, A., Shapiro, D., & West, M. A. (2000). The effectiveness of health care teams in the National Health Service. Birmingham, UK: Action Centre for Health Service Organization Research. Boud, D., Keogh, R., & Walker, D. (Eds). (1994). Reflection: Turning experience into learning. New York City: Nichols. Brav, A., Andersson, K., & Lantz, A. (2009). Group initiative and self-organizational activities in industrial work groups. European Journal of Work and Organizational Psychology, 18, 347–377. Bunderson, J. S., & Sutcliffe, K. M. (2003). Management team learning orientation and business unit performance. Journal of Applied Psychology, 88, 552–560. Carter, S. M., & West, M. A. (1998). Reflexivity, effectiveness, and mental health in BBC TV production teams. Small Group Research, 29, 583–601. Clark, K., & Fujimoto, T. (1991). Product development performance. Boston: Harvard Business School Press. Cunliffe, A. L., & Easterly-Smith, M. (2004). From reflection to practical reflexivity: Experiential learning as lived experience. In: M. Reynolds & R. Vince (Eds), Organizing reflection (pp. 30–46). Aldershot, UK: Ashgate Publishing. Daudelin, W. W. (1996). Learning from experience through reflection. Organizational Dynamics, 24, 36–48. DeChurch, L. A., & Mesmer-Magnus, J. R. (2009). The cognitive underpinnings of effective teamwork: A meta-analysis. Journal of Applied Psychology, 95, 32–53. De Dreu, C. (2002). Team innovation and team effectiveness: The importance of minority dissent and reflexivity. European Journal of Work and Organizational Psychology, 11, 285–298. De Dreu, C. (2007). Cooperative outcome interdependence, task reflexivity, and team effectiveness: A motivated information processing perspective. Journal of Applied Psychology, 92, 628–638.

92

RICHARD L. MORELAND AND JAMIE G. MCMINN

De Dreu, C., & Weingart, L. R. (2003). Task versus relationship conflict, team performance, and team member satisfaction: A meta-analysis. Journal of Applied Psychology, 88, 741–749. Drach-Zahavy, A., & Somech, A. (2001). Understanding team innovation: The role of team processes and structures. Group Dynamics: Theory, Research, and Practice, 5, 111–123. Edmondson, A. C. (1999). Psychological safety and learning behavior in work teams. Administrative Science Quarterly, 44, 350–383. Edmondson, A. C. (2002). The local and variegated nature of learning in organizations: A group-level perspective. Organization Science, 13, 128–146. Edmondson, A. C., Bohmer, R., & Pisano, G. (October, 2001). Speeding up team learning. Harvard Business Review, 79, 125–134. Fay, D., Borrill, C. S., Amir, Z., Haward, R., & West, M. A. (2006). Getting the most out of multidisciplinary teams: A multi-sample study of team innovation in health care. Journal of Occupational and Organizational Psychology, 79, 553–567. Gevers, J. M. P., van Eerde, W., & Rutte, C. G. (2001). Time pressure, potency, and progress in project groups. European Journal of Work and Organizational Psychology, 10, 205–221. Goldsmith, W. (2009). Ten tips to make sure your end-of-season re-view is a pre-view for success for next year. Available at http://www.sportscoachingbrain.com. Retrieved on June 1, 2010. Goodman, P. S., & Wilson, J. M. (2000). Substitutes for socialization in exocentric teams. In: M. Neale, B. Mannix & T. Griffith (Eds), Research in groups and teams (Vol. 3, pp. 53–77). Stamford, CT: JAI Press. Gurtner, A., Tschan, F., Semmer, N. K., & Nagele, C. (2007). Getting groups to develop good strategies: Effects of reflexivity interventions on team process, team performance, and shared mental models. Organizational Behavior and Human Decision Processes, 102, 127–142. Guzzo, R. A., Wagner, D. B., Maguire, E., Herr, B., & Hawley, C. (1986). Implicit theories and the evaluation of group process and performance. Organizational Behavior and Human Decision Processes, 37, 279–295. Hackman, J. R., & Wageman, R. (2005). A theory of team coaching. Academy of Management Review, 30, 269–287. Hirst, G., & Mann, L. (2004). A model of R & D leadership and team communication: The relationship with project performance. R&D Management, 34, 147–160. Hirst, G., Mann, L., Bain, P., Pirola-Merlo, A., & Richter, A. (2004). Learning to lead: The development and testing of a model of leadership learning. Leadership Quarterly, 15, 311–327. Hoegl, M., & Parboteeah, K. P. (2006). Team reflexivity in innovative projects. R&D Management, 36, 115–125. Kolb, D. A. (1984). Experiential learning: Experience as the source of learning and development. Englewood Cliffs, NJ: Prentice-Hall. Kolodner, J. L., Camp, P. J., Crismond, D., Fasse, B., Gray, J., Holbrook, J., & Ryan, M. (2004). Promoting deep science learning through case-based reasoning: Rituals and practices in ‘‘Learning by Design’’ classrooms. In: N. M. Seel & S. Dijkstra (Eds), Curriculum, plans, and processes in instructional design (pp. 89–113). Mahwah, NJ: Erlbaum. Lantz, A. (in press). Teamwork on the line can pay off down the line. Journal of Workplace Learning.

Group Reflexivity and Performance

93

Lee, A. Y., & Hutchison, L. (1998). Improving learning from examples through reflection. Journal of Experimental Psychology: Applied, 4, 187–210. Lee, L. T. S. (2008). The effects of team reflexivity and innovativeness on new product development performance. Industrial Management and Data Systems, 108, 548–569. Lewis, K., Belliveau, M., Herndon, B. J., & Keller, J. (2007). Group cognition, membership change, and performance: Investigating the benefits and detriments of collective knowledge. Organizational Behavior and Human Decision Processes, 103, 159–178. Lindsley, D. H., Brass, D. J., & Thomas, J. B. (1995). Efficacy-performance spirals: A multilevel perspective. Academy of Management Review, 20, 645–678. Martell, R. F., & Guzzo, R. A. (1991). The dynamics of implicit theories of group behavior: When and how do they operate. Organizational Behavior and Human Decision Processes, 50, 51–74. Moreland, R. L. (1999). Learning who knows what in work groups and organizations. In: L. Thompson, D. Messick & J. Levine (Eds), Shared cognition in organizations: The management of knowledge (pp. 3–31). Mahwah, NJ: Erlbaum. Moreland, R. L., & Levine, J. M. (1992). Problem identification in groups. In: S. Worchel, W. Wood & J. A. Simpson (Eds), Group process and productivity (pp. 17–47). Newbury Park, CA: Sage. Mullen, B., & Copper, C. (1994). The relation between group cohesiveness and performance: An integration. Psychological Bulletin, 115, 210–227. Mullen, B., Johnson, C., & Salas, E. (1991). Productivity loss in brainstorming groups: A metaanalytic integration. Basic & Applied Social Psychology, 12, 3–23. Muller, A., Herbig, B., & Petrovic, K. (2009). The explication of implicit team knowledge and its supporting effect on team processes and technical innovations. Small Group Research, 40, 28–56. Nemeth, C. J., & Staw, B. M. (1989). The tradeoffs of social control and innovation within groups and organizations. In: L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 22, pp. 175–210). New York: Academic Press. O’Donoghue, P. (2006). The use of feedback videos in sport. International Journal of Performance Analysis in Sport, 6, 1–14. O’Leary-Kelly, A. M., Martocchio, J. J., & Frank, D. D. (1994). A review of the influence of group goals on group performance. Academy of Management Journal, 37, 1285–1301. Pacanowski, M. (1995). Team tools for wicked problems. Organizational Dynamics, 23, 36–51. Peterson, S. E. (1992). College students’ attributions for performance on cooperative tasks. Contemporary Educational Psychology, 17, 114–124. Petruzzi, J. (1999). Football coaching – half time psychology. Available at http://www.pponline. co.uk/encyc/football-coaching-half-time-psychology. Retrieved on June 1, 2010. Pisano, G. P., Bohmer, R. M. J., & Edmondson, A. (2001). Organizational differences in rates of learning: Evidence from the adoption of minimally invasive cardiac surgery. Management Science, 47, 752–768. Rasker, P. C., Post, W. M., & Schraagen, J. M. C. (2000). Effects of two types of intra-team feedback on developing a shared mental model in command and control teams. Ergonomics, 43, 1167–1189. Salas, E., Klein, C., King, H., Salisbury, M., Augenstein, J. S., Birnbach, D. J., Robinson, D. W., & Upshaw, C. (2008). Debriefing medical teams: 12 evidence-based best practices and tips. Joint Commission Journal on Quality and Patient Safety, 34, 518–527.

94

RICHARD L. MORELAND AND JAMIE G. MCMINN

Salas, E., Nichols, D. R., & Driskell, J. E. (2007). Testing three team training strategies in intact teams: A meta-analysis. Small Group Research, 38, 471–488. Savelsbergh, C. M. J. H., van der Heijden, I. J. M., & Poell, R. F. (2009). The development and empirical validation of a multi-dimensional measurement instrument for team learning behaviors. Small Group Research, 40, 578–607. Schank, R. C., & Cleary, C. (1995). Engines for education. Mahwah, NJ: Erlbaum. Schippers, M. C., Den Hartog, D. N., & Koopman, P. L. (2007). Reflexivity in teams: A measure and correlates. Applied Psychology: An International Review, 16, 189–211. Schippers, M. C., Den Hartog, D. N., Koopman, P. L., & Van Knippenberg, D. (2008). The role of transformational leadership in enhancing team reflexivity. Human Relations, 61, 1593–1616. Schippers, M. C., Den Hartog, D. N., Koopman, P. L., & Wienk, J. A. (2003). Diversity and team outcomes: The moderating effects of outcome interdependence and group longevity and the mediating effect of reflexivity. Journal of Organizational Behavior, 24, 779–802. Smith-Jentsch, K. A., Cannon-Bowers, J. A., Tannenbaum, S. L., & Salas, E. (2008). Guided team self-correction: Impacts on team mental models, processes, and effectiveness. Small Group Research, 39, 303–327. Smith-Jentsch, K. A., Zeisig, R. L., Acton, B., & McPherson, J. A. (1998). Team dimensional training: A strategy for guided team self-correction. In: J. A. Cannon-Bowers & E. Salas (Eds), Making decisions under stress: Implications for individual and team training (pp. 272–297). Washington, DC: APA Press. Somech, A. (2006). The effects of leadership style and team process on performance and innovation in functionally heterogeneous teams. Journal of Management, 32, 132–157. Staw, B. M. (1975). Attribution of the ‘‘causes’’ of performance: A general alternative interpretation of cross-sectional research on organizations. Organizational Behavior and Human Performance, 13, 414–431. Stiltanen, J., Willis, A., & Scobie, W. (2008). Separately together: Working reflexively as a team. International Journal of Social Research Methodology, 11, 45–61. Sutton, L., & Dalley, J. (2008). Reflection in an intermediate care team. Physiotherapy, 94, 63–70. Swift, T. A., & West, M. A. (1998). Reflexivity and group processes: Research and practice. Sheffield, UK: ESRC Centre for Organization and Innovation. Tannenbaum, S. I., Smith-Jentsch, K. A., & Behson, S. J. (1998). Training team leaders to facilitate team learning and performance. In: J. A. Cannon-Bowers & E. Salas (Eds), Making decisions under stress: Implications for individual and team training (pp. 247–270). Washington, DC: APA Press. Tjosvold, D., Hui, C., & Yu, Z. (2003). Conflict management and task reflexivity for team in-role and extra-role performance in China. The International Journal of Conflict Management, 14, 141–163. Tjosvold, D., Tang, M. M. L., & West, M. A. (2004). Reflexivity for team innovation in China: The contribution of goal interdependence. Group and Organizational Management, 29, 540–559. Van Ginkel, W., Tindale, R. S., & Van Knippenberg, D. (2009). Team reflexivity, development of shared task representations, and the use of distributed information in group decision making. Group Dynamics: Theory, Research, and Practice, 13, 265–280. Vashdi, D. R., Bamberger, P. A., Erez, M., & Weiss-Meilik, A. (2007). Briefing-debriefing: Using a reflexive organizational learning model from the military to enhance the performance of surgical teams. Human Resource Management, 46, 115–142.

Group Reflexivity and Performance

95

Wegner, D. M. (1987). Transactive memory: A contemporary analysis of the group mind. In: B. Mullen & G. R. Goethals (Eds), Theories of group behavior (pp. 185–208). New York: Springer-Verlag. West, M. A. (1996). Reflexivity and work group effectiveness: A conceptual integration. In: M. A. West (Ed.), Handbook of work group psychology (pp. 555–579). Chichester, UK: Wiley. West, M. A. (2000). Reflexivity, revolution, and innovation in work teams. In: D. A. Johnson & S. T. Beyerlein (Eds), Product development teams (Vol. 5). Advances in interdisciplinary studies of work teams (pp. 1–29). Stamford, CT: JAI Press. Widmer, P. S., Schippers, M. C., & West, M. A. (2009). Recent developments in reflexivity research: A review. Journal of Everyday Activity, 2, 4–11. Wittenbaum, G. M., Vaughan, S. I., & Stasser, G. (1998). Coordination in task-performing groups. In: R. S. Tindale, J. Edwards, E. J. Posavac, F. B. Bryant, Y. Suarez-Balcazar, E. Henderson-King & J. Myers (Eds), Theory and research on small groups (pp. 177–204). New York City: Plenum Press.

TRUST AS AN EXPRESSIVE RATHER THAN AN INSTRUMENTAL ACT David Dunning and Detlef Fetchenhauer ABSTRACT Trust involves making oneself vulnerable to another person with the prospect of receiving some benefit in return. Contemporary theoretical accounts of trust among strangers emphasize its instrumental nature. People are assumed to trust to the extent that they can tolerate the risk and are sufficiently optimistic that their trust will be reciprocated. We describe evidence from laboratory economic games showing that this account empirically fails. Participants often trust even though their risk tolerance and social expectations suggest they should not. We propose, instead, that trust is largely an expressive act. People trust because of dynamics that surround the act itself rather than its potential outcomes. Evidence for the expressive nature of trust comes in two forms. First, studies of the emotions surrounding trust indicate that it is significantly predicted by how people feel about the act itself, not how they feel about its potential outcomes. Second, trust rates rise significantly if people are placed in a relationship with another person, no matter how anonymous, fleeting, or minimal that relationship is – presumably because being placed in a relationship evokes social norms that promote trust. We end our discussion by explaining a curious fact that participants grossly

Advances in Group Processes, Volume 27, 97–127 Copyright r 2010 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0882-6145/doi:10.1108/S0882-6145 (2010)0000027007

97

98

DAVID DUNNING AND DETLEF FETCHENHAUER

underestimate the trustworthiness of others. We also discuss possible motives for reciprocating trust and questions for future research.

You must trust and believe in people or life becomes impossible. – Anton Chekhov It is impossible to go through life without trust; that is to be imprisoned in the worst cell of all, oneself. – Graham Greene Trusting too much to others’ care is the ruin of many. – Benjamin Franklin There’s no trust, no faith, no honesty in men; all perjured, all forsworn, all naught, all dissemblers. – William Shakespeare

These four quotations exemplify the vexing paradox presented by trust. Thoughtful scholars across the history of Western scholarship have given diligent attention to the issue of trust and have written what would appear to be good advice about whether to trust others. The problem is that so much of this reasonable wisdom is contradictory. As Chekhov and Greene both passionately note, trust provides the raw material for a valuable life. So much of what individuals need – money, food, love – involves dealing with other people, and it is difficult to imagine any long-term or worthwhile relationship that does not involve trust. To the extent that individuals have strong social networks, forged by trust, they are in a stronger position to act both individually and collectively. Indeed, data show just how important trust can be to promote good social outcomes within a society. Trust is an essential component of stable interpersonal relationships, such as love and marriage (Deutsch, 1958; Fehr, 1988; for a review, see Simpson, 2007). It is crucial for organizations to thrive (Kramer, 1998; Kreps, 1990) and for democratic governments to flourish (Fukuyama, 1995; Sullivan & Transue, 1999). To the extent that people within a nation display trust in business contracts, and honor that trust in return, their national economy achieves greater economic growth (Putnam, 1993; Knack & Keefer, 1997). However, as Franklin and Shakespeare reasonably and compellingly counter, placing trust in another person is inevitably the act of a fool, an argument that many other Western scholars and thinkers have endorsed through the years. Trust is the pavement on the road to ruin. This road starts with the very definition of what it means to trust. Several definitions of trust

Trust as an Expressive Act

99

have been offered in philosophy, psychology, economics, and sociology (McKnight & Chervany, 2001), but for our purposes we adopt a definition, with one alteration, that has been articulated by Rousseau, Sitkin, Burt, and Camerer (1998) and that contains many of the elements present in other definitions of trust. Our definition is that people trust another when they make themselves vulnerable to another person’s exploitation with the prospect of receiving some benefit if that other person honors the trust given to them. For example, by sharing a secret, one makes oneself vulnerable to public embarrassment, but may receive important social support in return. By handing money over to a bank, one makes oneself open to the risk that the bank may fail or steal the money, but one also receives the potential benefit of having a safe place to keep the money and the possibility of gaining interest. The issue for scholars like Franklin and Shakespeare is that other people are often rational actors who behave in their own material self-interest. Because of this, people will inevitably, and frequently, betray the trust placed in them because it is in their interest to do so. Important thinkers such as Machiavelli (2003 [1515]) and Hobbes (1997 [1660]) have all agreed with Franklin and Shakespeare that people are inherently selfish, and so will inescapably violate trust whenever honoring that trust stands in the way their own material advantage. Modern neoclassical economics reaches a similar conclusion and suggests instead that people should never make their prospects vulnerable to others unless the actions of those other people can be constrained by a compulsion to honor trust or by onerous sanctions that makes violating trust unprofitable (Bolle, 1998; Berg, Dickhaut, & McCabe, 1995).

TRUST AMONG STRANGERS Given this fundamental contradiction, how does one approach trust? Both trusting others and avoiding trust appear to be equally wise courses of action. How does one thread the needle between these two diametrically opposing pieces of advice? The literature contains some answers, at least when it comes to behavior in long-lasting relationships. Trust in the shadow of uncertain exploitation can be built up over several interactions by gradual and repeated demonstrations of trust and cooperation. Two parties that start out distrusting each other can come to a more mutually cooperative stance after each has repeatedly signaled that he or she can be trusted, in a process traditionally described as graduated reciprocation in tension reduction (GRIT; Osgood, 1962; Lindskold, 1978). One party makes small, unilateral

100

DAVID DUNNING AND DETLEF FETCHENHAUER

gestures of trust to the other person and communicates that he or she wishes the other party to match it. If that other party does, they are rewarded with an even bigger display of trust. Over time, mutual displays of trust that are rewarded can lead to a relationship in which trust is inculcated. Life, however, often precludes the incremental, long-term establishment of trusting relations. One is faced with an immediate, and one-off, decision about whether to trust the other person. One buys housing materials at a local hardware store trusting that the material is of adequate quality. Or, people must judge whether to ask the individual sitting next to them on the train to keep an eye on their personal belongings versus taking those belongings with them. Getting the answers to these questions is important. Over the past few decades, this type of potential trust among strangers has been at the heart of treatments of social capital within a certain society or a collective group of actors. In this line of research trust (and trustworthiness) toward strangers is regarded as one prerequisite for the functioning of modern and complex societies, as it enables mutual cooperation and lowers transaction costs. As supportive evidence, the World Value Survey contains a one-item measure of trust (i.e., ‘‘Generally speaking would you say that most people in your country can be trusted or that one cannot be too careful?’’). It has been shown that answers to this question substantially differ between countries. For example, in Scandinavian countries about 60% percent of all respondents say that most others can be trusted; this is the case for only 20% in France. Although one might doubt whether it is possible to measure trust with such a crude and dichotomous one-item scale, it has been shown that this measure correlates with a number of different other variables like financial honesty or economic growth (Fetchenhauer & Van der Vegt, 2001; Knack & Keefer, 1997). Thus, in this chapter, we focus on trust among strangers, noting that answering the question of whether people should trust requires, first, addressing what causes people to trust – and to reciprocate trust – in social interaction. What are the social or psychological dynamics that cause people to be vulnerable to others, and which cause them to honor those acts of trust?

THE TRUST GAME In our work, we try to capture this decision whether or not to trust a stranger by asking our participants to play an economic game in the laboratory. Imagine that participants find themselves in an experiment

Trust as an Expressive Act

Fig. 1.

101

Schematic Depiction of the Possible Decisions and Outcomes for the Players in the Typical Trust Game Used in This Research.

surrounded by a large number of other people, none of whom know each other. The experimenter hands participants $5 each and gives them a decision to go along with it. Participants, now playing the role of chooser, must decide to keep the $5 or to hand it to another person in the room. That other person will be randomly chosen, and neither the chooser nor that other person, playing the role of the responder, will never know the identity of each other. If the chooser decided to give the money, $5 is inflated to $20, and the responder has two options about what to do with the new amount. Responders can either keep the entire $20 or give $10 back to the chooser. Both choosers and responders have the game completely explained to them. Thus, responders know that their choice is dependent on whether the chooser gives their money to them. A schematic representation of this game is given in Fig. 1. This game makes it possible to study trust in the laboratory and under controlled conditions. Another advantage of that game is that it not only measures trust as an attitude or a cognitive appraisal, but also as a specific and costly behavior, in that all our participants had to invest actual money. Most important, we would argue that the trust game nicely captures the essence of trust: to make oneself vulnerable and dependent on the moral character of another person, a complete stranger (see also Berg et al., 1995; Camerer, 2003; Snijders & Keren, 2001). Some researchers have combined laboratory trust games or some of its variants with large-scale survey research. Results show that, at least under some conditions, participants’ answers to attitudinal questions (e.g., the trust measure from the World Value Survey) and their behavior in the trust game

102

DAVID DUNNING AND DETLEF FETCHENHAUER

correspond to a certain degree (Holm & Danielson, 2005; Naef & Schupp, 2009). However, in this realm much more research appears to be worthwhile and necessary, in that the ways that survey and behavioral measures of trust mirror or diverge from each other may prove quite informative.

GOALS OF THE CHAPTER In this chapter, we review data garnered using this game to argue that contemporary accounts of trust largely fail to explain why or when people trust strangers. According to current theoretical accounts, trusting strangers is largely or exclusively an instrumental action in which people are primarily focused on the consequences they expect to unfurl as a consequence of their actions. People make themselves vulnerable to others via trust not as an act in itself, but rather as a means to some benefit they are interested in. Trust is a mere instrument used to gain that benefit, and it is the size and the probably of receiving that benefit that encourages or discourages trust. Here, however, we argue that trust is in large part neither instrumental nor consequentialist. People do not trust because of any calculus involved about its outcomes. Instead, trust is predominantly an expressive act. People trust because of direct rewards the behavior itself provides or because the performing the act itself fulfills some goal. We argue for the expressive character of trust by showing how data collected in our laboratory, and elsewhere, have forced us to this position. In the sections that follow, we examine how well instrumental accounts of trust explain when and how often people trust strangers when playing economic games in the laboratory and show how instrumental accounts of trust largely fail to anticipate what happens in the laboratory. We then describe how studies on the emotions surrounding trust indicate that it is not the consequences that people are most concerned about as they decide to trust, but rather the behavior itself. Next, we discuss how decisions to trust are ‘‘switched on’’ or ‘‘off’’ depending on whether people are placed in an actual social relationship with the person they are playing the game with – no matter how superficial or transient that relationship is. Finally, we discuss the choice to reciprocate trust, to show that it may be produced by the same types of expressive dynamics governing decisions to trust. People appear to be following the dictates of being placed in a social relationship, although they are also quite sensitive to how much respect the other person is displaying to them.

Trust as an Expressive Act

103

INSTRUMENTAL ACCOUNTS OF TRUST Virtually all theories of trust in economics and psychology construe trust among strangers as instrumental, in that people are largely concerned with the potential consequences that will be brought about by a decision to trust. Central to this analysis are two different factors that should heavily influence trust decisions. The first is tolerance for risk. People should gamble on another person to the extent that they are willing to tolerate any sort of risk for potential gain. The second is expectation. To the extent that people are optimistic that their trust will be reciprocated, they should be more willing to make themselves more vulnerable, and if their expectation lies within their risk tolerance, they should trust the other person.

Does Risk Tolerance Influence Trust? The problem with this analysis is that it largely fails to account for the rate of trust observed in laboratory experiments like ours. Take risk tolerance, for example. Several studies have shown that people’s trust behavior is not very sensitive to tolerance for risk. Kanagaretnam, Mestelman, Nainar, and Shehata (2009), for example, asked participants about how attractive they found certain risky gambles. People who found risky gambles to be attractive were no more likely to take their chances with another person in a trust game than those who found the gambles less attractive (see also Ashraf, Bohnet, & Piankov, 2006). Ben-Nur and Halldorsson (2010) measured inclinations toward risk through both survey questions and reactions to risky lotteries. Regardless of their measure, they found no relationship between risk attitudes and behavior in the trust game. Finally, Eckel and Wilson (2004) conducted the most extensive study of risk and trust. They measured risk tolerance in any number of ways – through survey questions about risk, decisions with risky gambles – even with measures of thrill-seeking, desire for novelty, and disinhibition. None of these measures predicted the degree to which people would take a risk in the trust game.

Do Social Expectations Influence Trust? Even more troubling for the instrumental account is the role played by expectation in decisions to trust. In psychological and economic treatments of trust, there is no variable that has played such a central role in theories of

104

DAVID DUNNING AND DETLEF FETCHENHAUER

trust. According to theorists in psychology, people will trust others if they think others will reciprocate that trust; they will not trust if they believe that their trust is likely to be abused (Bacharach & Gambetta, 2001; Barber, 1983; Deutsch, 1958; Kramer & Carnevale, 2001; Pruitt & Rubin, 1986). The Rotter (1967, 1971) personality measure of trust centrally focuses on expectations about other people. Respondents are asked, for example, if ‘‘other people have a vicious streak’’ and are ‘‘primarily interested in their own welfare’’ despite their claims to the contrary. That trust is so infused with the notion of expectation is best exemplified by the change we have made from Rousseau et al.’s (1998) definition of trust, which is to substitute the phrase ‘‘prospect of future benefit’’ instead of what Rousseau et al. used, which was the expectation of future benefit. Economic treatments of trust similarly highlight the importance of expectation. According to the neoclassical analysis of trust, the reason why no one should trust a stranger in a one-off exchange is that there is no expectation that a stranger would honor trust. This is best exemplified by the neoclassical analysis of the $5/$20 economic game described earlier. According to the neoclassical analysis, no chooser should give up the $5 because responders are guaranteed to keep the $20, being rational actors focused exclusively on their own material self-interest. Responders take no profit from handing money back and face no sanction or loss of reputation by keeping the entire $20. Thus, their choice is a certainty, and choosers should not make themselves vulnerable to assured exploitation (e.g., Berg et al., 1995). Behavior in the laboratory, however, fails to adhere to this analysis. Choosers decide to give money to the responder, whether the format of the game is like ours, in which must give up their entire endowment (Fetchenhauer & Dunning, 2009) or can choose instead to give up only some portion of it (Berg et al., 1995). Why might this be so? The economic analysis may dismally forecast that no one would reciprocate trust, but the psychological analysis at least suggests that there is some expectation that Responders will actually honor trust that has been placed in them. And, in fact, people do reciprocate trust at some nonzero rate (Berg et al., 1995; Camerer, 2003; Fetchenhauer & Dunning, 2009). Thus, people might harbor enough expectation to allow them to take a risk on the trustworthiness of other people. That is, if people believe the likelihood that their trust would be honored is high enough to be acceptable given their tolerance level of risk, then they would be correct to trust a stranger. In this way, decisions to trust may still be instrumental and consequentialist.

Trust as an Expressive Act

105

We have conducted several studies demonstrating that trust does not follow this logic, at least in our economic game in the laboratory (Fetchenhauer & Dunning, 2009). In these studies, we asked participants if they would give up the $5 as a chooser (although the amounts of money involved vary somewhat depending on the geographic location of the experiment). While doing so, we also asked questions designed to see if participants’ choices followed instrumental principles. First, we asked participants to estimate the likelihood that responders will return $10 versus keep the entire $20. Second, we asked participants to consider a lottery in which they could gamble $5 to win $10, and then to tell us the minimum chance of winning they would require to gamble the $5. Comparing the responses to these two queries allows us to gauge the proportion of choosers who should rationally trust the other person in the trust game. Specifically, choosers should trust if they believe the chance the responder will give $10 back is equal to or greater than the minimum chance of winning they would demand in the lottery. In essence, if they are willing to gamble $5 on a lottery at an equal or lower probability of success, then they are making a decision to trust that is consistent with their expectations and their tolerance for risk. However, if they believe that the chance of the responder reciprocating their trust is lower than the chance of winning they would demand in the lottery scenario, then they should refuse to trust the responder. Across the first three studies we conducted with this paradigm (Fetchenhauer & Dunning, 2009, Studies 1a–c), we found that choosers were rather pessimistic about their peers, in that they thought that only 45% of their peers would return the $10 in the role of the responder. And, given this pessimism, it is not a surprise that our analysis suggested that only 30% of choosers should choose to hand over their $5 – if the decision to trust was purely instrumental. However, when we examined real behavior, our data presented two clear surprises. First, although choosers on average thought that a majority of responders would choose to keep the $20, in reality a full 80% of responders chose to reciprocate trust. Second, although a clear majority of choosers displayed expectations and risk attitudes suggesting they should keep their $5, a full 65% decided to trust the anonymous responder they had been paired with. A full 30–35% of participants trusted ‘‘irrationally’’ in that they made decisions that contradicted their expectations and risk tolerance. They trusted the other person even though they stated that the chance they would receive money back was lower than the chance they demanded to gamble the same amount in the lottery. In many of these cases, they decided to trust the other person

106

DAVID DUNNING AND DETLEF FETCHENHAUER

even though they thought the odds were greater that the Responder would keep all the money than would give $10 back. To be sure, participants’ expectations mattered – somewhat. Roughly 9% of the variance in their decisions could be explained by the level of optimism they expressed about the chances that their trust would be reciprocated – but, at the end of the day, we were impressed by how little expectation influenced trust decisions rather than how much it did.

Alternative Explanations Before moving to how these data suggest that trust decisions are neither instrumental nor consequential, we need to address two possible responses. Each response argues that participants’ trust decisions in our experiments were instrumental, despite the data just described. Where People’s Expectations Accurately Reported? The first response is that participants were instrumental in their trust decision; they just were not honest in reporting their expectations about their peers. Perhaps, in reality, they were more optimistic that their peers – after all, that is what their behavior indicated – but provided pessimistic estimates to us merely to soften any blow to their esteem or reputation if the other person failed to honor their trust. Given the self-report nature of the data above, this is not an account that we could dismiss immediately. However, we have collected data suggesting that expectations largely fail to drive decisions to trust even when imposed by an external agent (Fetchenhauer & Dunning, 2010a). In one study, we began by asking a number of people to play a responder in the $5/$20 game, and from their responses formed two groups. In one group, 80% of participants indicated that they would honor trust – a rate equal to what we had observed in previous studies. In the second group, only 46% of participants stated they would honor trust – a rate close to the subjective expectations participants had previously reported. We then approached a second wave of participants and asked them, for real, to play the role of chooser, after being paired with a responder who had already indicated his or her decision. We truthfully told some of our choosers that there was an 80% chance of gaining $10 if they trusted the other person, and truthfully told the rest that the chance of return was 46%. Of key importance, we also asked the first group if they would gamble $5 if

Trust as an Expressive Act

107

they had an 80% chance to win $10 in return in a lottery. The second group was asked if they would gamble the $5 to win $10 if the chance in the lottery were 46% instead. Participants’ responses indicated that the chance of return mattered a great deal for decisions in the lottery, but not for decisions to trust. In the lottery, only 29% of participants chose to gamble with only a 46% chance of winning, but 78% chose to gamble when the chance of winning was 80%. These chances of winning, however, did not similarly influence decisions to trust. In the 46% condition, 54% decided to trust – a rate comparable to previous studies and one that was significantly higher than those choosing to gamble in the lottery. In the 80% condition, only 70% decided to trust – a rate that was not significantly higher than the other group and actually less than the proportion willing to gamble on an 80% lottery. In short, when expectations of gaining through trust were handed to participants, ‘‘too many’’ people chose to trust when the probability of return was low, relative to the number willing to gamble on a chance event. In addition, decisions to trust were insensitive to the rate of return, whereas lottery gambles were quite sensitive to the changing the rate of return (Fetchenhauer & Dunning, 2010a).

Are People Concerned About the Common Good? The second response to our data involves a potential concern for the common good. We have asserted that decisions to trust are instrumental and consequential if those decisions are consistent with the decisions participants make on a lottery offering similar odds and payoffs for the self. That is, if a person is willing to bet $5 on a lottery at some chance of winning $10, then they should trust another person if they think the chance of having their trust honored is equal or superior to the chance of winning in the lottery. However, in the lotteries described earlier, there were no consequences for other people, only the self. This differs significantly from the trust game, where another person might gain from the decisions of the chooser, and perhaps choosers were responsive to that. Specifically, trusting another person creates $20 of wealth in total, whereas declining to trust creates only $5. Perhaps people are sensitive to the overall wealth or utility their decision can create, even if though their decisions may deprive them of some of that increased wealth. In short, people might be motivated to benefit the common good, just not the self.

108

DAVID DUNNING AND DETLEF FETCHENHAUER

This is a plausible account that explains our data and yet renders trust decisions as instrumental in nature. However, we have tested whether giving people a chance to benefit the common good leads them toward more risktaking behavior. In that study, participants considered three different decisions, counterbalanced in order across participants. One decision was whether to bet $5 on a flip of a coin with a chance to win $10. Another was whether to trust another person in the $5/$20 game in which there was a 50% chance of having one’s trust reciprocated. The third decision was the telling one. It returned to the format of a coin flip, but with ‘‘extended’’ consequences. If a person decided to gamble their $5 and won on the coin flip, not only would they win $10 but some random person in the experiment would also win $10. If the person gambled and lost, than some random person would win $20. Note that this ‘‘extended coin flip’’ is not the trust game, but presents the same odds and outcomes for self and other as the trust game (Dunning & Fetchenhauer, 2010). Did adding possible benefits for another person prompt people to gamble on a coin flip more frequently? The answer appears to be no. Adding benefits to another person caused the percentage of participants choosing to gamble to move from 43% to 42%, a most trivial difference, even though 64% of choosers decided to gamble on another person when presented with the trust game (Dunning & Fetchenhauer, 2010). These results have been replicated in a between-subject format (Mensching, Schlo¨sser, Dunning, & Fetchenhauer, 2010), which again showed that adding potential benefits to some other random individual failed to entice people to gamble more on a coin flip, whereas people were more likely to gamble on another person when presented with a trust game. Other research shows more directly that trust is not instrumental in nature, in that trust rates are altered from what they are naturally if people are forced to think in instrumental or consequentialist terms. For example, consider a trust game, slightly different from ours, in which people are given $20 and can give $0, $10, or $20 to a responder, with whatever amount sent being tripled. Kugler, Connolly, and Kausel (2009) discovered that forcing participants to consider the consequences of their actions, by making them consciously estimate how much money they expected to receive in return, caused them to lower the amount they sent by 40%, with the proportion sending all $20 dropping from 56% to a mere 19%. In a follow-up study, asking participants to consider how regretful they would feel if they ‘‘over-’’ or ‘‘under-trusted’’ another individual – again, an emphasis on the consequences of their actions – caused the percentage of choosers sending all of their money to drop from 30% to 0%.

Trust as an Expressive Act

109

EMOTIONAL CORRELATES OF TRUST: EXPRESSIVE VS. INSTRUMENTAL Work we have done in our labs on the emotional correlates of trust also guides us toward thinking the decision is related more to expressive than to instrumental dynamics. Besides the money involved, it is clear that people can derive utility from the emotions they feel that are attached to trust. One motivator for trust may be the joy and gratitude people anticipate feeling if the other person reciprocates their trust; a motivator against trust might be the anger, embarrassment, and disappointment they expect if the other person violates their trust. Some theorists, for example, have suggested that a major determinant of trust behavior is a worry over betrayal (Bohnet & Zeckhauser, 2004). Other theorists have recognized the importance of emotions in determining people’s economic decisions under risk. Economic theorists, for example, have outlined the potentially important role played by regret (Bell, 1982; Loomes & Sugden, 1982) and disappointment (Bell, 1985; Loomes & Sugden, 1986) in decision making. When making decisions, people consider how regretful they will feel if the decision turns out badly, and so steer away from those decisions, or at least make them with more caution. Other theorists have discussed how the pleasure or displeasure of an event can be accentuated if it is a surprising one (Mellers, Schwartz, Ho, & Ritov, 1997; Mellers, Schwartz, & Ritov, 1999). Thus, with colleague Thomas Schlo¨sser, we examined the potential relationship between emotion and decisions to trust. We presented participants with an opportunity to trust another person in our typical economic game, telling them there was a 50% chance that their trust would be reciprocated, but importantly asked them how they would feel in various circumstances. Of key note, we realized there were two layers of emotions to query participants about. If one looks at previous treatments of emotion in decision making, one finds that they tend to be consequentialist in nature. People consider their decisions and take into account how much pleasure, regret, disappointment, and surprise they will feel based on the outcomes their decisions might produce. But this is not the only layer of emotion in play. People can experience emotions in the immediate here-and-now, emotions that attach to the specific action they choose to take. One can, for example, feel fear or thrill about the act of flipping a coin, double or nothing, regardless of how one feels about the consequences of that action. One can feel relieved to keep one’s money in the trust game. Thus, an

110

DAVID DUNNING AND DETLEF FETCHENHAUER

emerging scholarship has importantly made a distinction between anticipated emotions, which are those emotions people forecast they will feel once the outcomes of decisions are known, and immediate emotions, which are the feelings people experience about choosing one course of action over another at the time they moment the choice (Loewenstein, Weber, Hsee, & Welch, 2001; Loewenstein & Lerner, 2003). One’s immediate emotions can be quite distinct from the emotions people anticipate they will feel once the consequences of an action become clear. Thus, we took pains to ask participants about their anticipated and immediate emotions as they contemplated trusting another person. In terms of anticipated emotions, participants were asked to report how they would feel in all four potential scenarios that could unfold in the trust game. They were asked how they would feel if they trusted the other person and received money back, if they trusted and received nothing back, if they decided not to trust and learned later that the person would have returned money, and decided to forgo trust and learned that the other person would have failed to reciprocate that trust. We queried participants by asking them how positive or negative they would feel in each situation, how aroused they would feel, and how much in control they would think themselves to be – three common dimensions that researchers have used to investigate people’s emotional lives (Bradley & Lang, 1994). Participants also reported on the immediate emotions they attached to the two decisions they could make, trusting or avoiding trust. Participants reported how positive, aroused, and in control they felt about handing over their money to the responder, and also how they would feel along these dimensions if they decided to keep the money. Note that these emotions are more expressive in nature, in that they indicate that the mere choice made in the trust game could be a direct source of feeling or utility for the participant, independent of any utility participants anticipating gaining in the future once the trust transaction was complete. In total, participants provided emotional reports for a total of six scenarios. Two focused on immediate emotions (i.e., how they felt about giving their money, keeping their money). The remaining four focused on potential outcomes of the game (i.e., giving versus keeping the money crossed with the responder deciding to give money back versus keep it all). For each of these six scenarios, participants sorted themselves neatly into two statistical groups after we conducted a cluster analysis, one that reported more positive feelings about the scenario than the other group did. For each scenario, we then examined how well falling into the first group rather than the second predicted decisions to trust.

Trust as an Expressive Act

111

The results of this analysis were clear, and perhaps just less surprising to us as they would be to an economist. Anticipated emotions failed to predict decisions to trust. Reports of immediate emotion, however, predicted decisions to trust rather strongly – in fact, very strongly relative to other variables that economists would consider more important. Recall that expectations that the responder would honor trust across all our studies above was a significant, but rather weak, predictor of who would trust (accounting for 9% of variance). Here, immediate emotions proved to be a strong predictor. Of those who felt more positively about keeping the money than they did about giving it, only 35% decided to trust the responder. Of those who felt the opposite, a full 91% gave their money. This explained 21% of the variance in decisions to trust – much more than the nonsignificant 10% explained by anticipated emotions (Schlo¨sser et al., 2010). Two notes must be made about these findings. At least where emotion is concerned, it appears that the emotions that are most heavily related to decisions to trust are those that are expressive in nature, in that they attach to the decision itself rather than its outcomes. It is how people feel about the act of trusting versus its opposite that predicts who trusts. Emotions attached to more consequentialist facets, for example, how people anticipate how they would feel as a consequence of their exchange with the responder, predicted little if anything. This pattern arose even though participants turned out to be accurate predictors of how they would feel once the trust game was played out. Participants’ predictions of their anticipated emotions were remarkably accurate forecasts of how they really felt after the trust game was executed and outcomes known. Second, it was not the case that immediate emotions served as a ‘‘backdoor’’ for anticipated emotions to predict trust decisions. Immediate and anticipated emotions were largely independent, and so it was not the case that anticipated emotions predicted trust decisions by first influencing immediate emotions (Schlo¨sser et al., 2010).

EVIDENCE FOR EXPRESSIVENESS: THE RELATIONAL DYNAMICS OF TRUST If decisions to trust are not instrumental, what are they? What prompts participants to hand money over to a stranger even in situations in which we are pessimistic about receiving any money back? We do not feel we have a completely crystallized answer to this question, but we believe we have some

112

DAVID DUNNING AND DETLEF FETCHENHAUER

insight into the dynamic involved. Consider the following scenario, first presented by Kahneman, Knetsch, and Thaler (1986): A small photocopying shop has one employee who has worked in the shop for six months and earns $9 per hour. Business continues to be satisfactory, but a factory in the area has closed and unemployment has increased. Other small shops have now hired reliable workers at $7 an hour to perform jobs similar to those done by the photocopy shop employee. The current employee leaves, and the owner decided to pay a replacement $7 an hour.

Of 125 respondents asked, 73% thought it to be acceptable to pay a new employee $7 if the current one left. However, 83% of 98 respondents thought it unfair if the shop owner reduced the current employee’s wages to $7 if that employee instead stayed on. Similar responses were found when participants considered relationships between landlords and potential tenants. The range of rent increases landlords could acceptably ask of new tenants was much larger than those which could be imposed on continuing tenants – although any new landlord buying the building would not be held to the same constraint. Our interpretation of these responses is that the relationship status between the shop owner and employee matters. The ethical code people use in their dealings with others depends on whether one is placed in some sort of relationship with that other. For example, if a random person walked into a room of several people, muttering that he or she needed to borrow a cell phone, our intuition is that people would feel some small but not very compelling urge to share their own cell phone. But if that person were a business associate who had come expressly arrived to work with them, people would feel much more of a compulsion to share their phone. Relationships come with different norms about the other person should be treated, with people constrained to consider the outcomes of the other person when in a relationship. Other work shows how relations can change the way people treat others, with decisions moving from more instrumental to more expressive. For example, in their work on affective approaches to social exchange, Lawler and colleagues (Lawler, 2001; Lawler, Thye, & Yoon, 2008) have shown that repeated positive exchanges with another person leads individuals to value that relationship. Exchanges with that individual may start with only consequentialist concerns, with people staying in the relationship only if it is materially rewarding. But after a while, the relationship becomes a source of reward in itself, with people continuing exchanges with that person even when there are other more materially advantageous alternatives available.

Trust as an Expressive Act

113

In essence, actions in a relationship begin as consequentialist becomes expressive.

The Minimal Relationship Effect We argue that people are sensitive to expressive or relational concerns not only after a period of time within a relationship, but that actions within a relationship carry important expressive concerns at the outset. The mere circumstance of being placed in a relationship is enough to evoke expressive concerns and to set aside consequentialist ones. This had led us to assert the existence of a minimal relationship effect, in which people act to potentially benefit others for which they are in a relationship, no matter how minimal, anonymous, or fleeting that relationship is. People hold a norm to be ‘‘nice’’ to other people they are in relationship, even if that relationship is not a deep or long-lasting one. However, absent the presence of that relationship, people will not give weight to the outcomes of others. In a sense, a minimal relationship effect would echo a long-documented phenomenon in social psychology, namely the minimal group effect (Tajfel, 1970). This effect refers to the fact that people begin to favor groups they have been assigned to over other groups, even if the basis for sorting people into groups is arbitrary, uninformative, or unimportant. For example, if people are sorted into groups based on whether they like the paintings of abstract artists Klee or Kandinsky more, or whether they tend to over- or underestimate the number of dots on a projection screen, they begin to describe their own group in more favorable terms than the other group, make more self-serving attributions for their own group’s behavior, and even pay their own group more (for a review, see Brewer, 1979). Interestingly, much of these effects are due to treating in-group members better rather than by treating out-group members more negatively (Brewer, 1979). There is something about the binding into some sort of social unit that alters how favorably people treat one another, no matter how minimal or free of meaning that binding is. A mere relationship effect might be responsible for one phenomenon already extant in the literature – the victim identifiability effect in altruistic behavior in others. In demonstrations of this phenomenon, participants are asked in the laboratory whether they are willing to give up some of their money to another participant who has lost theirs through no fault of their own. For some participants, no specific ‘‘victim’’ to be helped is identified, although participants know that an anonymous person who has been

114

DAVID DUNNING AND DETLEF FETCHENHAUER

victimized will be identified by random method should they choose to offer their money. For other participants, the person they can help has already been designated at random, although they are told nothing that would reveal the personal identity of the individual. Participants provide much more money to the victim in the second circumstance in which that individual has already been designated. A similar effect is shown if participants are approached to contribute to Habitat for Humanity. If the recipient of their donation has already been designated, people are much more willing to give than if the recipient is yet to be assigned (Small & Loewenstein, 2003; see also Small, Loewenstein, & Slovic, 2007). The identifiable victim effect is usually explained by noting that people may be more willing to help a single individual they know they can help with certainty rather than those small proportion of an indistinct aggregate of other individuals. We, however, can assert that the identifiable victim effect arises because assigning a ‘‘victim’’ to the participant places that participant in a relationship with that victim. Now in a relationship, the obligations and norms of behavior change, with participants now impelled to be more sensitive to the outcomes of that other individual. The same assertion can be applied to similar effects in the Dictator Game. People assigned to another person choose to share whatever money the experimenter has given them with that other person. However, if instead given a choice about whether they want to take part in a Dictator Game, far fewer participants choose to share money voluntarily. In short, being placed in a relationship with another person appears to impel sharing. People will not share because of some intrinsic taste for altruism. If they first have a choice about entering into that relationship, many decline the opportunity (Lazear, Malmendier, & Weber, 2009). Why would people adopt a stance toward treating strangers nicely once they have been placed in a minimal relationship with them? One recent analysis is that a norm toward concern for the outcomes of others may arise as societies become more complex in the social relations that people rely on. As societies move from being small ones, in which most dealings are with kin or with long-term relationships, to more complex ones, which involve people interacting more commonly on one-shot occasions with strangers, norms may develop that allow those ephemeral interactions to flourish. In those more complex societies, people become more sensitive to fairness, opportunities to cooperate, and the benefit of others even among relatively short-term relationships. They also become more prone to punish people who fail to live up to these norms, even if those others are of relatively little future relevance (Henrich et al., 2010).

Trust as an Expressive Act

115

This analysis has been empirically supported by comparing the decisions made by people who live in more complex societies, such as the United States, with those in which fewer essential interactions are made with relative strangers, such as the Hazda from Tanzania. Across 15 different societies, Henrich et al. (2010) found dramatic differences in the willingness to give money to a stranger in a dictator or ultimatum game – related to how much members of those societies depended on the market – that is, strangers – for the calories in their diet. Among Americans, for example, who buy 100% of their calories from strangers, respondents on average gave nearly 50% of their money to a stranger in a dictator game. Among the Hazda, who obtain 0% of their calories from strangers, the figure was closer to 25%. Henrich et al. (2010) also found that people from more complex societies were more willing incur costs to punish strangers who did not live up to norms of fairness in their dealings with others.

Evidence for the Minimal Relationship Effect Our own work suggests a minimal relationship effect in trust behavior, in that people need to be assigned to an interaction partner for ‘‘undue’’ or ‘‘inexplicable’’ levels of trust to arise. For example, when we present choosers with a mere hypothetical decision to trust, trust rates collapse from over 60% to closer to 30% (Fetchenhauer & Dunning, 2009). A similar pattern has been observed by Holm and Nystedt (2008), who found trust rates of 69% when participants considered games with real financial consequences but trust rates of only 27% when the decision involved mere hypothetical payoffs. This collapse is surprising because, if anything, respondents should be more eager to claim they would trust if the decision carries no real material consequence, in that making the claim allows them to portray themselves as generous and ethical people. And much other work shows that people tend to overclaim being generous and ethical, relative to their actual behavior, when asked hypothetical questions about how they would act in situations with moral overtones (Balcetis & Dunning, 2008; Balcetis, Dunning, & Miller, 2008; Epley & Dunning, 2000, 2006). When it is ‘‘cheap talk,’’ people should claim the moral high ground. Here, they do not. This seemingly contradictory finding with the trust game can be explained, however, through a minimal relationship effect. When posed as a hypothetical situation, people are not in a relationship with a specific other as they reach their choice. Thus, they are freer to decline to be nice to some

116

DAVID DUNNING AND DETLEF FETCHENHAUER

other random person. However, when the trust game is presented as a real exchange, there is another person involved and the chooser is placed in a relationship, albeit minimal. Thus placed, the mandate of being nice to that other person is evoked. We have conducted work showing that putting a person explicitly in a relationship is a necessary condition for the high rates of trust seen in previous experiments (Fetchenhauer & Dunning, 2010b). In one such experiment, we asked participants in one condition to play our typical $5/ $20 trust game after being told they had been assigned to a specific responder, with roughly 55% of participants choosing to trust in this circumstance. In a second condition, however, we removed the presence of the relationship to the responder, stating to participants that a responder had not been assigned yet – and would not if they chose to keep the $5. In this group, only 35% of choosers decided to trust – close to what we observed when the trust game was described as hypothetical. Of key interest, though, was a third condition, in which participants were told that a responder had been assigned to them, but that the responder did not know about the game and would not know if choosers decided to keep the $5. In this circumstance, trust rates rose to over 60%. In essence, putting the chooser in a relationship was sufficient to prompt trust rates to rise significantly and well over what one would anticipate as a purely instrumental act. Note that it was not necessary for the responder to know about the chooser’s choice. Even if keeping the $5 was to be a completely private act, people still chose to trust at rates far higher than if the responder did not exist at all. But this study left one important confound. Placing people in a relationship moved their decision from being about some small portion of a social aggregate to being about the totality of one single individual. People may be prompted to act in a pro-social manner when they know that their action, with absolute certainty, determines the fate of a single individual. Thus, it is not the presence of a relationship that matters as much as the presence of ‘‘fate control’’ over one individual. Thus, we conducted a followup study to pit the power of the relationship against the presence of a single individual. This study, again, involved three conditions. In the first, participants played the trust game as typically described, with the proviso that the responder would not know of the game unless the chooser decided to give away the $5. In this circumstance, over 60% of participants decided to give the $5. In the last two conditions, the trust game was instead described as a ‘‘buyin’’ into a Dictator Game. Participants could pay the experimenter $5, and

117

Trust as an Expressive Act

the experimenter would then approach an anonymous participant who would be given $20 and asked if he or she would share $10 with some other participant – who, the experimenter assured, would be the chooser. In one condition, the other participant who would be approached had yet to be designated. In the other, the participant to be approached had been already designated and assigned to the chooser. Note that this buy-in to the Dictator Game has the same payoff structure as the trust game, in that the chooser must decide whether to pay $5 with the prospect to receive $10 in return. However, psychologically, it is very different. The exchange – and relationship – does not take place between chooser and responder. Instead, the exchange, and any relationship it represents, takes place between chooser and the experimenter. Thus, according to the minimal relationship account, chooser should be reticent to give up their $5 regardless of whether their decisions, with certainty, control the fate of whether a responder gets to participate. And, as expected, we observed that only 15% and 14% of choosers decided to give up their $5 in the two conditions as described, respectively. We should note, however, that one follow-up study to this research found only a nonsignificant trend in the same direction (Mensching et al., 2010).

OTHER ISSUES Other than calling into question the instrumental nature of trust decisions, our work has presented other puzzles, some of which are explicable and some that are less so.

Explaining Undue Cynicism Our studies suggest that people are rather unduly cynical about the likelihood that responders will reciprocate trust. Recall that, in our studies, participants believe that only 45% of participants will return money in the $5/$20 trust game, but the actual rate is 80% (Fetchenhauer & Dunning, 2009). This is a stunning failure of intuition, made even more stunning because there is good deal of evidence suggesting that people make fairly accurate predictions about the pro-social behavior of others. People estimate with remarkable accuracy the rate at which other people will donate to charity, vote, walk away from a person asking others to sign a petition, remain in their romantic relationships, and volunteer for a less

118

DAVID DUNNING AND DETLEF FETCHENHAUER

desirable experiment so that a child does not have to take part in it (e.g., Balcetis & Dunning, 2008; Balcetis et al., 2008; Epley & Dunning, 2000, 2006; Nisbett & Kunda, 1985). What prompts this cynicism? We believe we have identified one important dynamic that explains why participants would have so drastically underestimated their peers (as well as explain why other studies have found accuracy rather than cynicism). We believe the key ingredient is the feedback about others that people have gained throughout their life. Feedback about the trustworthiness of others is incomplete and ‘‘asymmetric’’ and presses toward a cynical direction. Consider the following scenario in which Stanley decides to trust another person. Two possible outcomes of that trust may arise. First, the other person may reciprocate that trust, thus rewarding Stanley for his action. But, second, the other person may violate that trust, punishing Stanley and also alerting him that he has been somewhat too optimistic about human nature. Thus, he revises his theory about the trustworthiness of others in a cynical direction. But, suppose, instead, that Stanley decides not to trust another individual. Stanley in this circumstance will fail to receive any feedback about the potential wisdom of this action. Because he fails to give the other person a chance to confirm or contradict the wisdom of his decision, Stanley will never learn when he has been too cynical – when his trust would have been honored even though he was initially skeptical that it would. Thus, there will be no countervailing pressure on Stanley’s theories of human nature to alert him when he should hold a more optimistic view of his peers. In essence, this is an experiencing sampling account (Denrell, 2005; Eiser, Fazio, Stafford, & Prescott, 2003; Smith & Collins, 2009) of the undue cynicism we observed. The experiences people create for themselves do not provide all the data they need to form an accurate view of trustworthiness in human nature. This feature of trust made it different from the other types of social situations in which people show accuracy in prediction. In these situations (e.g., voting, romantic relationships), people learn much about other people’s behavior by being passive observers (Epley & Dunning, 2000, 2006). They do not engage in social interaction, which might produce incomplete and asymmetric feedback, and thus are left with an unbiased set of experiences leaving them with accurate impressions of human nature and social action. To test this account, we conducted a trust game study in which we altered the nature of the feedback that participants encountered – to see if eliminating the asymmetry in the feedback people received also removed their undue cynicism (Fetchenhauer & Dunning, 2010c). Participants viewed

Trust as an Expressive Act

119

quick 10-second videos of 56 different individuals. For each, they were asked whether they would trust that person in a trust game, with one game being played for real after participants had viewed all the videotapes. Participants completed this task under three different feedback conditions. In the first, participants received no feedback about whether their decisions to trust were correct or incorrect. In the second condition, involving asymmetric feedback, participants received feedback much in the same way we assert people do in their everyday lives. That is, whenever participants chose to trust an individual, we informed them of whether the other person would honor or violate that trust. However, if they declined to trust, we provided no feedback. Finally, in the third condition, involving full, complete, and symmetric feedback, participants were given feedback on their decisions regardless of whether they chose trust or its opposite. Much like in our other studies, roughly 80% our responders indicated that they would reciprocate trust. However, choosers in the no feedback or asymmetric feedback tended to think, at the conclusion of the study, that roughly 58% and 63%, respectively, of the people they had witnessed would reciprocate trust – estimates that were far too low. Choosers in the complete feedback condition, however, quickly learned just how trustworthy their peers were, thinking at the end that 79% were trustworthy individuals, a far more accurate estimate. Relative to participants in the other conditions, the complete feedback group was more likely to trust (70% versus 59% and 57% for asymmetric and no feedback conditions, respectively) and earned substantially more money when their trust decision was played (Fetchenhauer & Dunning, 2010c).

Accounting for Trustworthiness Our work also presents a mystery about the behavior of responders. So far, we have focused our discussion exclusively on why choosers decide to make themselves vulnerable to other individuals for only an uncertain chance of gaining from it. We have mentioned that a neoclassical analysis of trust suggests that choosers should transfer nothing to the sender, and that is it something of a puzzle to economic analysis why people might choose to do so. We should note, however, that it is an equally vexing puzzle to neoclassical economics why responders send any money back to the chooser. They are under no compulsion to do so. Because the decision can be rendered anonymous, they have no reputational concerns at stake. Thus, what leads them to act in a pro-social way?

120

DAVID DUNNING AND DETLEF FETCHENHAUER

At first blush, the answer to why they send money back would seem to be rather straightforward. The norm of reciprocity suggests that if someone does something nice for you, you must reciprocate that kindness at some point in the future. Much work in the influence and compliance literature notes the central role that reciprocity can play in shaping social behavior (for a review, see Cialdini, 2001). Thus, is the norm of reciprocity the impetus for responder’s actions in the trust game? Despite it being straightforward explanation, data suggest that reciprocity is not necessarily the engine behind the actions of responders. For example, Mensching et al. (2010) asked participants either to play a trust game, or to play a Dictator Game using the same amounts of money. Responders in the trust game had to decide whether to split h20 with a chooser who had given up h5, and 72% did so, which can be taken as prima facie evidence of reciprocation. However, when Mensching et al. removed the chooser from the interaction, and merely asked participants whether they would split h20 with another person as a dictator in the dictator game, a full 70% decided to split the money, even though there was no action on the part of the other person to reciprocate. An opportunity to reciprocate was not a necessary condition for people to generously give up part of the money given to them by the experimenter (see Kiyonari, Yamagishi, Cook, & Cheshire, 2006, for similar conclusions). Other data support the notion that reciprocity plays a far smaller role in trustworthiness than one might think. Ben-Nur and Halldorsson (2010) examined how much responders gave back to choosers, looking to see how much two separate variables predicted trustworthiness. One variable was reciprocity, which they measured by asking how much each participant would give to another person in a hypothetical dictator game after receiving money from that person in a previous dictator game. The other variable was ‘‘unconditional kindness,’’ which they measured by asking participants how much they would give to another person in a dictator game in which they had had no prior dealings with the other person. They found that both variables failed to predict trustworthiness. Thus, data suggest that responders are not necessarily acting out of a need to honor a reciprocity norm. They send back to the chooser roughly the same amount of money whether or not the chooser has had a chance to first benefit the responder with their own act of kindness. That said, there are data suggesting that if responders fail to be sensitive to whether the chooser has acted first with a kindness, they are sensitive to whether the chooser has been grudging about it. Responders send back a great deal of money if the chooser has been generous with

Trust as an Expressive Act

121

trust, but are not so charitable if the chooser has shown any sign of incomplete trust. This has been demonstrated in two different ways. The first involves studies in which choosers can send only a portion of their money rather than the entire endowment. If choosers decide to send a large portion or all of their endowment, responders are generous in sending back a good portion of the money they now have in their disposal. However, as choosers send less, responders react by sending a smaller proportion of the money they have (Pilutla, Malhotra, & Murnighan, 2003). In addition, if choosers decide to ensure that they will receive sufficient money back by imposing monetary sanctions on responders who give too little back, responders react by giving just enough back to avoid the sanction, and very little more (Fehr & Rockenbach, 2002; Malhotra & Murnighan, 2002).

Accounting for Expressive Influences Finally, although our data suggest that trust decisions are expressive rather instrumental in nature, there is still much work to be done to specify what it is about the action of trust that causes people to do it. Earlier, we suggested that trusting another person, even a stranger, may be a norm that develops in societies that successfully make the transition from those that involve mostly transactions among kin to those that involve a wide range of social exchanges taking place in a constant flux of emerging and disappearing relationships (Henrich et al., 2010). Conducting successful exchanges among strangers would require norms of behavior promoting harmony over disharmony, even among strangers who have only the most fleeting of relevance for each other. Thus, the trust behavior we observe in our laboratory experiments might be the product of social norms that have evolved to support the complex societies our participants lived in. But even if that explanation bears out to be correct, we still have work to do to specify what exact norm people are following. Are people following a norm to be pro-social? If so, then we can expect people to be quite nice whenever dealing with other people. Or, are people just trying to avoid being antisocial? This second norm would lead to a very different type of behavior if we, for example, let people send just a little of their endowment to the responder rather than the whole amount. Suffice it to say that our participants, in post-experiment interviews, seem not to be able to articulate the specific norm they are following – an observation that has been made by others (Zak, 2008).

122

DAVID DUNNING AND DETLEF FETCHENHAUER

There may be other expressive facets of the trust decision beyond norms, however. People may choose to trust because it allows them to establish a self-identity they wish to bolster or maintain. That is, people want to think of themselves as nice, pro-social people, and deciding to trust even when the odds of profiting from it are low is one sure way to establish that selfimage. Many theorists in economics and psychology have discussed how people shape their decisions in the service of bolstering favored self-images (Akerlof & Kranton, 2000; Bodner & Prelec, 2002; Dunning, 2007), and, as such, trust might be decision that is made with a wary eye for what the decision says about its maker. In addition, trust decisions might be more directly driven by emotion than recognized. As people live their lives, favorable feelings might become more attached to trust and unfavorable feelings attached to the decision not to trust. In short, people may develop somatic markers, visceral or physiological reactions, that lead them to act in a more pro-social way than their risk tolerance and social expectations. These markers may contain the sum of knowledge they have of rewards and punishments they have encountered in the past (Damasio, 1994) in situations that match the trust decision they face. In sum, the work we have conducted so far suggest that it is factors surrounding the behavior of trust itself, and not its potential consequences, that largely determines the choice that people make. Thus, trust should be considered more of an expressive behavior than an instrumental one. That said, much more research must be conducted to better specify what it is in the expressive realm that determines what people choose to do.

CONCLUDING REMARKS We, like others, have defined trust as making oneself vulnerable to another person’s exploitation with the prospect of receiving some benefit if that other person honors the trust given to them. Although we feel this definition best captures the essence of trust, it has one shortcoming that can be found in essentially all definitions of trust: it does not distinguish between trust as a behavior, trust as a cognition, and trust as a feeling. Most researchers, at least implicitly, have assumed that people behave trustful and experience feelings of trust because they believe that the other person will turn out to be trustworthy. Therefore, if one measures which choices participants make in the trust game, one has also measured their cognitions and emotions. For methodological reasons, many researchers, especially those in the realm

123

Trust as an Expressive Act

of experimental economics and rational choice theory, would argue that one should not spend too much time measuring some fuzzy concepts like cognitions and emotions, but should rather focus on participants’ actual behavior as that behavior also reveals participants’ cognitions and emotions. However, the research reviewed in this chapter has shown that such an approach may miss out on revealing the essence of trust. People do not only trust merely because they anticipate positive gain when doing so. Instead, they commonly trust a stranger even when they anticipate that the other person will not reciprocate their trust – and they feel better about handing over the money than they would about keeping it. Thus, we hope that in the future others will join us in distinguishing behavior, cognitions, and emotions with regard to trust decisions. It makes the whole issue of trust more complicated than is often assumed – and potentially reveal it to be an expressive rather than an instrumental act – but it would also allow us to understanding the dynamics behind this most important type of decision for social life.

REFERENCES Akerlof, G., & Kranton, R. E. (2000). Economics and identity. Quarterly Journal of Economics, 115, 715–753. Ashraf, N., Bohnet, I., & Piankov, N. (2006). Decomposing trust and trustworthiness. Experimental Economics, 9, 193–208. Bacharach, M., & Gambetta, D. (2001). Trust in signs. In: K. Cook (Ed.), Trust in society (Vol. 2, pp. 148–184). New York: Russell Sage Foundation. Balcetis, E., & Dunning, D. (2008). A mile in moccasins: How situational experience reduces dispositionism in social judgment. Personality and Social Psychology Bulletin, 34, 102–114. Balcetis, E., Dunning, D., & Miller, R. L. (2008). Do collectivists ‘‘know themselves’’ better than individualists?: Cross-cultural investigations of the ‘‘holier than thou’’ phenomenon. Journal of Personality and Social Psychology, 95, 1252–1267. Barber, B. (1983). The logic and limits of trust. New Brunswick, NJ: Rutgers University Press. Bell, D. E. (1982). Regret in decision making under uncertainty. Operations Research, 30, 961–981. Bell, D. E. (1985). Disappointment in decision making under uncertainty. Operations Research, 33, 1–27. Ben-Nur, A., & Halldorsson, F. (2010). Trusting and trustworthiness: What are they, how to measure them, and what affects them. Journal of Economic Psychology, 31, 64–79. Berg, J., Dickhaut, J., & McCabe, K. (1995). Trust, reciprocity, and social history. Games and Economic Behavior, 10, 122–142.

124

DAVID DUNNING AND DETLEF FETCHENHAUER

Bodner, R., & Prelec, D. (2002). Self-signaling and diagnostic utility in everyday decision making. In: I. Brocas & J. Carillo (Eds), Collected essays in psychology and economics (pp. 1–22). Oxford University Press. Bohnet, I., & Zeckhauser, R. (2004). Trust, risk and betrayal. Journal of Economic Behavior and Organization, 55, 467–484. Bolle, F. (1998). Rewarding trust: An experimental study. Theory and Decision, 45, 83–98. Bradley, M. M., & Lang, P. J. (1994). Measuring emotion: The self-assessment manikin and the semantic differential. Journal of Behavioral Therapy and Experimental Psychiatry, 25, 49–59. Brewer, M. B. (1979). In-group bias in the minimal intergroup situation: A cognitivemotivational analysis. Psychological Bulletin, 85, 307–324. Camerer, C. F. (2003). Behavioral game theory: Experiments on strategic interaction. Princeton, NJ: Princeton University Press. Cialdini, R. B. (2001). Influence: Science and practice (4th ed.). Boston: Allyn & Bacon. Damasio, A. (1994). Descartes error: Emotion, reason, and the human brain. New York: Putnam. Denrell, J. (2005). Why most people disapprove of me: Experience sampling in impression formation. Psychological Review, 112, 951–978. Deutsch, M. (1958). Trust and suspicion. Conflict Resolution, 2, 265–279. Dunning, D. (2007). Self-image motives and consumer behavior: How sacrosanct self-beliefs sway preferences in the marketplace. Journal of Consumer Psychology, 17, 237–249. Dunning, D., & Fetchenhauer, D. (2010). Understanding the psychology of trust. In: D. Dunning (Ed.), Social motivation (pp. 147–170). New York: Psychology Press. Eckel, C., & Wilson, R. (2004). Is trust a risky decision? Journal of Economic Behavior and Organization, 55, 447–465. Eiser, R. J., Fazio, R. H., Stafford, T., & Prescott, T. (2003). Connectionist simulation of attitude learning: Asymmetries in the acquisition of positive and negative evaluations. Personality and Social Psychology Bulletin, 29, 1221–1235. Epley, N., & Dunning, D. (2000). Feeling ‘‘holier than thou’’: Are self-serving assessments produced by errors in self or social prediction? Journal of Personality and Social Psychology, 79, 861–875. Epley, N., & Dunning, D. (2006). The mixed blessings of self-knowledge in behavioral prediction: Enhanced discrimination but exacerbated bias. Personality and Social Psychology Bulletin, 32, 641–655. Fehr, B. (1988). Prototype analysis of the concepts of love and commitment. Journal of Personality and Social Psychology, 4, 557–579. Fehr, E., & Rockenbach, B. (2002). Detrimental effects of sanctions on human altruism. Nature, 422, 137–140. Fetchenhauer, D., & Dunning, D. (2009). Do people trust too much or too little? Journal of Economic Psychology, 30, 263–276. Fetchenhauer, D., & Dunning, D. (2010a). Betrayal aversion versus principled trustfulness: How to explain risk avoidance and risky choices in trust games. Manuscript under review. University of Cologne. Fetchenhauer, D., & Dunning, D. (2010b). The minimal connection effect in decisions to trust. Unpublished manuscript. University of Cologne. Fetchenhauer, D., & Dunning, D. (2010c). Why so cynical? Asymmetric feedback underlies misguided skepticism in the trustworthiness of others. Psychological Science, 21, 189–193.

Trust as an Expressive Act

125

Fetchenhauer, D., & Van der Vegt, G. (2001). Honesty, trust and economic growth. A crosscultural comparison of Western industrialized countries. Zeitschrift fu¨r Sozialpsychologie, 32, 189–200. Fukuyama, F. (1995). Trust: The social virtues and the creation of prosperity. New York: Free Press. Henrich, J., et al. (2010). Markets, religion, community size, and the evolution and punishment. Science, 327, 1480–1484. Hobbes, T. (1660/1997). The leviathan. New York: Touchstone Press. Holm, H., & Danielson, A. (2005). Tropic trust versus Nordic trust. Experimental evidence from Tanzania and Sweden. The Economic Journal, 115, 505–532. Holm, H., & Nystedt, P. (2008). Trust in surveys and games – A methodological contribution on the influence of money and location. Journal of Economic Psychology, 29, 522–542. Kahneman, D., Knetsch, J. L., & Thaler, R. H. (1986). Fairness as a constraint on profit seeking: Entitlements in the market. American Economic Review, 76, 728–741. Kanagaretnam, K., Mestelman, S., Nainar, K., & Shehata, M. (2009). The impact of social value orientation and risk attitudes on trust and reciprocity. Journal of Economic Psychology, 30, 368–380. Kiyonari, T., Yamagishi, T., Cook, K. S., & Cheshire, C. (2006). Does trust beget trustworthiness? Trust and trustworthiness in two games and two cultures: A research note. Social Psychology Quarterly, 69, 270–283. Knack, S., & Keefer, P. (1997). Does social capital have an economic payoff? A cross-country investigation. The Quarterly Journal of Economics, 112, 1251–1288. Kramer, R. M. (1998). Paranoid cognition in social systems. Thinking and acting in the shadow of doubt. Personality and Social Psychology Review, 2, 251–275. Kramer, R. M., & Carnevale, P. J. (2001). Trust and intergroup negotiation. In: R. Brown & S. Gaertner (Eds), Intergroup relations (Vol. 4). Blackwell Handbook of Social Psychology (pp. 431–450). Oxford, UK: Blackwell Publishers. Kreps, D. M. (1990). Corporate culture and economic theory. In: J. Alt & K. Shepsle (Eds), Perspectives on positive political economy (pp. 90–143). Cambridge, UK: Cambridge University Press. Kugler, T., Connolly, T., & Kausel, E. E. (2009). The effect of consequential thinking on trust game behavior. Journal of Behavioral Decision Making, 22, 101–119. Lawler, E. J. (2001). An affect theory of social exchange. American Journal of Sociology, 107, 321–352. Lawler, E. J., Thye, S. R., & Yoon, J. (2008). Social exchange and micro social order. American Sociological Review, 73, 519–542. Lazear, E. P., Malmendier, U., & Weber, R. A. (2009). Sorting and social preference. Unpublished manuscript. Stanford University. Lindskold, S. (1978). Trust development, the GRIT proposal, and the effects of conciliatory acts on conflict and cooperation. Psychological Bulletin, 85, 772–793. Loewenstein, G., & Lerner, J. S. (2003). The role of affect in decision making. In: R. J. Davidson, K. R. Scherer & H. H. Goldsmith (Eds), Handbook of affective sciences (pp. 619–642). New York: Oxford University Press. Loewenstein, G. F., Weber, E. U., Hsee, C. K., & Welch, N. (2001). Risk as feelings. Psychological Bulletin, 127, 267–286. Loomes, G., & Sugden, R. (1982). Regret theory: An alternative theory of rational choice under uncertainty. Economic Journal, 92, 805–824.

126

DAVID DUNNING AND DETLEF FETCHENHAUER

Loomes, G., & Sugden, R. (1986). Disappointment and dynamic consistency in choice under uncertainty. Review of Economic Studies, 53, 271–282. Machiavelli, N. (1515/2003). The prince [R. Goodwin (Trans.)]. Wellesley, MA: Dante University Press. Malhotra, D., & Murnighan, J. K. (2002). The effects of contracts on interpersonal trust. Administrative Science Quarterly, 47, 534–559. McKnight, D., & Chervany, N. (2001). Trust and distrust definitions: One bite at a time. In: R. Falcone, M. Singh & Y.-H. Tan (Eds), Trust in cyber-societies: Integrating the human and artificial perspectives (pp. 27–54). Berlin: Springer Verlag. Mellers, B. A., Schwartz, A., Ho, K., & Ritov, I. (1997). Decision affect theory: How we feel about risky options. Psychological Science, 8, 423–429. Mellers, B. A., Schwartz, A., & Ritov, I. (1999). Emotion-based choice. Journal of Experimental Psychology: General, 128, 1–14. Mensching, O., Schlo¨sser, T., Dunning, D., & Fetchenhauer, D. (2010). Why do people trust? A multi-perspective approach. Unpublished manuscript. University of Cologne. Naef, M., & Schupp, J. (2009) Measuring trust: Experiments and surveys in contrast and combination. DIW Berlin: SOEP papers on Multidisciplinary Panel Data Research. Nisbett, R. E., & Kunda, Z. (1985). Perception of social distributions. Journal of Personality and Social Psychology, 48, 297–311. Osgood, C. E. (1962). An alternative to war or surrender. Urbana, IL: University of Illinois Press. Pilutla, M. M., Malhotra, D., & Murnighan, J. K. (2003). Attributions of trust and the calculus of reciprocity. Journal of Experimental Social Psychology, 39, 448–455. Pruitt, D. G., & Rubin, J. Z. (1986). Social conflict: Escalation, stalemate, and settlement. New York: McGraw-Hill. Putnam, R. D. (1993). Making democracy work-civic traditions in modern Italy. Princeton, NJ: Princeton University Press. Rotter, J. B. (1967). A new scale for the measurement of interpersonal trust. Journal of Personality, 35, 651–665. Rotter, J. B. (1971). Generalized expectancies of interpersonal trust. American Psychologist, 26, 443–452. Rousseau, D. M., Sitkin, S. B., Burt, R., & Camerer, C. (1998). Not so different after all: A cross-discipline view of trust. Academy of Management Review, 23, 393–404. Schlo¨sser, T., Fetchenhauer, D., & Dunning, D. (2010). Emotional dimensions of trust behavior: Immediate versus anticipated emotions. Unpublished manuscript. University of Cologne. Simpson, J. A. (2007). Foundations of interpersonal trust. In: A. W. Kruglanski & E. T. Higgins (Eds), Social psychology: Handbook of basic principles (2nd ed, pp. 587–607). New York: Guilford. Small, D. A., & Loewenstein, G. (2003). Helping ‘‘A’’ victim or helping ‘‘THE’’ victim: Altruism and identifiability. Journal of Risk and Uncertainty, 26(1), 5–16. Small, D. A., Loewenstein, G., & Slovic, P. (2007). Sympathy and callousness: The impact of deliberative thought on donations to identifiable and statistical victims. Organizational Behavior and Human Decision Processes, 102, 143–153. Smith, E. R., & Collins, E. C. (2009). Contextualizing person perception: Distributed social cognition. Psychological Review, 116, 343–364.

Trust as an Expressive Act

127

Snijders, C., & Keren, G. (2001). Do you trust? Whom do you trust? When do you trust? Advances in Group Processes, 18, 129–160. Sullivan, J. L., & Transue, J. E. (1999). The psychological underpinnings of democracy: A selective review of research on political tolerance, interpersonal trust, and social capital. Annual Review of Psychology, 50, 625–650. Tajfel, H. (1970). Experiments in intergroup discrimination. Scientific American, 223, 96–102. Zak, P. J. (2008). The neurobiology of trust. Scientific American (June), 88–95.

BEING DIFFERENT OR BEING BETTER?: DISENTANGLING THE EFFECTS OF INDEPENDENCE AND COMPETITION ON GROUP CREATIVITY Jack A. Goncalo and Verena Krause ABSTRACT Accumulating evidence suggests that individualism provides an atmosphere conducive to creative idea generation. However, research in both cross-cultural and social psychology suggests that individualism may reflect either independence or competition; a distinction that has been overlooked in research on group creativity. In this chapter, we highlight the distinction between these two constructs and develop a series of testable propositions that help distinguish their unique effects on the creative process. In doing so, we uncover several theoretical insights, including the possibility that independence and competition (a) are theoretically and empirically distinct, (b) have differential effects on idea generation, (c) have similar effects on idea selection but through different mechanisms, and (d) may interact to stimulate group creativity. We conclude by suggesting methodological approaches to disentangling these constructs in future research.

Advances in Group Processes, Volume 27, 129–157 Copyright r 2010 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0882-6145/doi:10.1108/S0882-6145 (2010)0000027008

129

130

JACK A. GONCALO AND VERENA KRAUSE

Keywords: Individualism-collectivism; independence; competition; group creativity. Given that most groups are in an environment that becomes increasingly competitive over time (Barnett & Hansen, 1996), there is tremendous pressure to generate ideas that may lead in a profitable new direction (Amabile, 1996). In response to this pressure, work organizations, scientists, artists, and political decision makers (Paulus & Nijstad, 2003) employ various strategies, including the formation of brainstorming groups (Paulus & Yang, 2000) to promote creative idea generation. A creative idea is most often defined as one that is both novel and useful (Amabile, 1983). It is novel because it diverges from existing solutions and useful in that it presents a potentially viable solution to a problem. In organizations such ideas may relate to a wide variety of domains such as organizational products, practices, services or procedures (Shalley & Gilson, 2004). Creativity can be distinguished from innovation because whereas creativity is focused on the development and generation of new and useful ideas, innovation refers to the process through which they are successfully implemented at the organizational level (Amabile, 1996). Ideally, people can collaborate to generate more creative ideas than any one individual could come up with alone because they have the opportunity to build on, combine, and improve on the ideas suggested by others (Diehl & Stroebe, 1987). In collaboration, the whole might be more creative than the sum of its parts. This logic prompted Osborn (1957) to predict that a well functioning brainstorming group has the potential to generate more than twice the number of ideas produced by the same number of individuals working alone. Unfortunately, face-to-face brainstorming groups may suffer from a number of problems that make them less effective than a nominal group of individuals who work alone and then combine their ideas (Taylor, Berry, & Block, 1958; Diehl & Stroebe, 1987; Girotra, Terwiesch, & Ulrich, 2010). Process losses stemming from production blocking, evaluation apprehension or free-riding can cause individuals to withhold ideas during brainstorming sessions (Diehl & Stroebe, 1987). Far from being a hotbed of creative thought, the typical brainstorming group is either a cacophony of people talking over each other or a timid group; afraid their ideas will be rejected and privately hoping that no one will notice their lack of participation. A critical question, therefore, is what are the group processes that facilitate the expression of creative ideas (Rietzschel, Nijstad, & Stroebe, 2006; Girotra et al., 2010)?

Being Different or Being Better?

131

A growing body of research suggests that one way to mitigate processes loss and stimulate group ideation is to promote a culture of individualism, defined broadly as a culture in which the needs of the individual are prioritized over the needs of the group (Markus & Kitayama, 1991). Groups that attribute their success to individual achievement (Goncalo, 2004; Goncalo & Duguid, 2008), endorse individualistic rather than collectivistic values (Goncalo & Staw, 2006), maximize their own outcomes with little or no regard for the outcomes of others (Beersma & De Dreu, 2005) and are composed of independent as opposed to interdependent selves (Wiekens & Stapel, 2008; Goncalo & Kim, 2010) generate a wide range of novel ideas. We are struck by the consistency of the findings that have accumulated over the past 5 years. To our knowledge, no study has yet documented any advantage of collectivism for group creativity. We suspect that if collectivism does contribute to group creativity, that it probably does so infrequently and only under narrow conditions. It is also becoming apparent, however, that within this stream of research individualism has been defined and operationalized in a number of different ways. On the one hand, an advantage of this theoretical and methodological diversity is that converging findings provide evidence for the robustness and generalizability of the effect. The results attest to the strength of the phenomenon and not the idiosyncratic effects of a particular manipulation. On the other hand, the lack of precision over the basic definition of individualism may make accumulating results difficult to interpret. This ambiguity may lead to confusion over the underlying processes that explain why individualistic groups outperform collectivistic groups on tasks that demand creative solutions. In this chapter, we distinguish between two important facets of individualism, independence, and competition, which may each play a unique role in facilitating the expression of creative ideas. A long tradition of research on individualism from the perspective of culture and social cognition (Triandis, Bontempo, Villareal, Asai, & Lucca, 1988; Chen & West, 2008) and bargaining and negotiations (De Dreu & Boles, 1998; De Dreu, Weingart, & Kwon, 2000) has distinguished between these two forms of individualism both theoretically and empirically. However, our review suggests that research on group creativity has glossed over this important distinction and has treated these two underlying constructs as largely interchangeable. The purpose of this chapter is to apply this distinction to research on group creativity and to generate a series of testable propositions that emerge when independence and competition are considered separately and in interaction with each other. By highlighting this

132

JACK A. GONCALO AND VERENA KRAUSE

distinction, we hope to clarify seemingly conflicting findings and suggest new avenues for future research.

INDIVIDUALISM AND GROUP CREATIVITY There is growing evidence that individualism stimulates the expression of creative ideas. In this section, we review the evidence in detail and highlight key differences in the theoretical and methodological approaches that have been brought to bear on the issue. Initial evidence for the link between individualism and group creativity comes from research on the group serving bias, which describes the tendency to attribute success to factors that are internal to the group and failure to factors that are external to the group (Forsyth & Schlenker, 1977; Schlenker & Miller, 1977; Taylor & Doria, 1981). According to Goncalo (2004), the group-serving tendency to attribute success internally (e.g., we cooperated, we communicated, we worked well together) could highlight how the group behaved before a successful outcome (Cialdini, Reno, & Kallgren, 1990), thus creating pressure to conform to their point of view on subsequent tasks. The group-serving bias might be corrected by highlighting the contributions made by each individual group member (e.g., Sam is political, Ed is energetic, Bill is cooperative) and thereby promoting a sense of uniqueness (Goncalo, 2004; Goncalo & Duguid, 2008). In one study, Goncalo (2004) gave groups false-positive feedback on an initial group task and then asked groups to brainstorm new business ideas. Groups either attributed their success on the first task collectivistically (to the group as a whole), or individualistically (to the unique contributions made by individual group members). The results showed that individualistic attributions for past performance caused groups to generate more ideas and those ideas were more divergent. Divergent thinking is defined as thinking that moves outward from a problem in many possible directions whereas convergent thinking moves from a number of different alternatives to a single correct solution (Mayer, 1992). In other words, divergent thinking requires the group to break with a common theme to explore ideas that are very different from each other. These findings were replicated in a subsequent study in which success attributed to the individual caused groups to consider a wider range of decision alternatives before reaching consensus than success attributed to the group (Goncalo & Duguid, 2008). In addition, attributions to the individual also facilitated the sharing of unique information in a hidden

Being Different or Being Better?

133

profile task which, in turn, increased decision accuracy (Goncalo & Duguid, 2008). Analyses of the groups’ interaction provided some support for the role of conformity pressure; individualistic attributions liberated groups to express more disagreements. Further evidence comes from Beersma and De Dreu (2005) who investigated the consequences of social motives; one’s preferences for the distribution of outcomes between oneself and an interdependent other (Tjosvold, 1984). Motives can be either pro-self in which negotiators try to maximize their own outcomes, with no (or negative) regard for the outcomes obtained by others or motives can be pro-social in which the negotiator tries to maximize their own and others’ outcomes (De Dreu & Boles, 1998). Beersma and De Dreu (2005) hypothesized that pro-self social motives encourage competition, which should facilitate the expression of ideas (Dugosh & Paulus, 2005) but interfere with the group ability to reach consensus. In two studies, a pro-self motive was manipulated by rewarding people either individually or collectively in an initial negotiating task. That task was then followed by either a creative idea generation task in which groups were asked to come up with advertising slogans for a new marketplace or a planning task in which groups were asked to solve a specific problem. Competition was not measured directly, but the results did support the hypothesis that pro-self motives facilitated performance on the divergent task, while pro-social motives facilitated performance on the convergent task. In another study, Goncalo and Staw (2006) primed groups to think of themselves as either individualistic or collectivistic and then measured their creativity on a subsequent group brainstorming task. They predicted that reduced levels of conformity pressure characteristic of individualistic cultures (Bond & Smith, 1996) would facilitate group creativity by permitting greater independence from the group. People in individualistic cultures have an independent sense of self and therefore strive to express the attributes that make them unique, while people in collectivistic cultures have an interdependent sense of self and therefore strive to maintain harmonious relationships with other in-group members (Markus & Kitayama, 1991). Because individualistic cultures stress being ‘‘true’’ to one’s self and one’s unique set of needs and desires (Fiske, Kitayama, Markus, & Nisbett, 1998), people with an independent self concept may be encouraged to resist social pressure if it contradicts his/her personal opinions (Bond & Smith, 1996). In the Goncalo and Staw (2006) study, individualism was primed by asking participants to write three statements (a) describing yourself, (b) why you think you are not like most other people, (c) why you think it might be

134

JACK A. GONCALO AND VERENA KRAUSE

advantageous to stand out from other people. Collectivism was primed by asking participants to write three statements (a) describing the groups to which you belong, (b) why you think you are like most other people, (c) why it might be advantageous to blend in with other people. The results showed that groups primed to think individualistically generated more ideas that were rated as more novel and more divergent than groups primed to think collectivistically. Unlike the results obtained by Beersma and De Dreu (2005), however, individualistic groups also outperformed collectivistic groups when they were asked to select their most creative idea (convergent stage). Goncalo and Staw (2006) found that individualistic groups selected ideas that were more novel and reflected an original combination of more than one idea from their list. There is also more recent evidence suggesting that an individualistic sense of self may be useful, even on tasks completed alone. Construing the self as independent (‘‘I’’) induces the motivation to be alone and different, whereas construing the self as interdependent (‘‘we’’) induces the motivation to be accepted and to conform (Wiekens & Stapel, 2008). Consistent with research at the group level, participants with a salient ‘‘I’’ self-construal outperformed participants with a salient ‘‘we’’ self-construal on a task that called for divergent thinking. These results suggest that the group-level findings may have a cognitive component. In other words, individualism enhances the ability of group members to bring creative solutions to mind. At the group level, however, simply thinking of creative ideas might not be sufficient if those ideas are not voiced. Indeed, Goncalo and Kim (2010) investigated the effects of self-construal on a face-to-face group brainstorming task and found no main effects. In other words, although independent selves may be better able to think of more novel ideas, they did not necessarily express them. The results showed that idea expression was highest in groups that were not only primed to think independently but also endorsed a reward allocation rule that incited competition (equity) (Adams, 1963, 1965). It is worth noting that none of the aforementioned experiments manipulated individualism by making individual payoffs explicitly contingent on individual success. In other words, none of the participants in these studies were paid in exchange for the number of ideas they expressed during the brainstorming session (Toubias, 2006). Rather, in each study individualism was primed on one task which, in turn, shifted groups’ orientation on a subsequent task. For instance, Goncalo and Kim (2010) manipulated the equity rule not by paying each individual for their contributions but by asking the group to endorse the value of the equity rule

Being Different or Being Better?

135

during a discussion. Therefore, individualism in these studies is best understood as a psychological frame that guides how people view themselves in relation to others rather than a compensation system. We discuss this priming approach in more detail in a subsequent section. As a whole, this stream of research provides strong evidence that individualism, in various forms, stimulates group creativity. However, there are two important limitations in the current research. First, individualism is defined in many different ways, which makes it somewhat challenging to map the conceptual terrain. The danger is that the term itself may become underspecified and over-applied. Second, individualism is also operationalized in very different ways, which may make conflicting results difficult to interpret. For instance, why does individualism interfere with convergent outcomes in some studies but not others? We turn to these two important issues in the next section.

DEFINITIONS OF INDIVIDUALISM Individualism is a concept that has been extremely influential and the subject of considerable research in both cross-cultural psychology and social psychology. Although the concepts of individualism in these two literatures arose in different contexts and utilize different methodological approaches, they converge on the idea that individualism may reflect either the desire to remain independent from the group or the motivation to win in competitions. This distinction has not yet been made in research on group creativity, but bringing that distinction to the foreground may lead to a finer grained understanding of why individualism promotes creative expression.

Culture and Social Cognition The modern research on individualism in cross-cultural psychology has been shaped by Hofstede (1980), who initially described individualism as a cultural value that is bipolar and one-dimensional. In other words, collectivism, for Hofstede (1980), was simply low individualism. Since then, researchers have come to agree that individualism is orthogonal to rather than on a continuum with collectivism (Bontempo, 1993; Rhee, Uleman, & Lee, 1996; Singelis, 1994; Triandis et al., 1988; Oyserman, Coon, & Kemmelmeier, 2002), hence the two constructs can be studied separately.

136

JACK A. GONCALO AND VERENA KRAUSE

Subsequent research has refined these constructs further and elucidated a number of different forms that individualism-collectivism may take (Brewer & Chen, 2007). For instance, Singelis, Triandis, Bhawuk, and Gelfand (1995) introduced the vertical–horizontal distinction that refers to the extent to which a culture emphasizes equality or hierarchical differentiation. The vertical–horizontal dimension interacts with individualism to produce two distinct forms. Horizontal individualism is associated with the desire to be unique, self-reliant, and distinct from groups. In other words, the emphasis is on maintaining one’s independence; all people are equal but each person is unique. Vertical individualism, on the other hand, is associated with the desire to distinguish one’s self from others and to acquire status through competition. For example, college students in the United States are upset when they are labeled ‘‘average’’ (Weldon, 1984) because they strive to be and see themselves as superior to others (Triandis, 2000). Winning in competitions is a way to assert one’s uniqueness to the group. The distinction between independence and competition weaves its way through much of the research on individualism over the past two decades. In cultures that emphasize independence, there is a heightened concern for the self, personal autonomy and self-fulfillment, emotional independence, individual initiative, and the right to privacy (Hofstede, 1980). Such cultures also place a strong emphasis on personal responsibility and freedom of choice (Waterman, 1984). At the individual level, independence has been investigated in terms of how people construe themselves in relation to other people (Markus & Kitayama, 1991; Brewer & Gardner, 1996). Individuals with an independent self-construal behave in accordance with their personal cognition, emotions, and motivations, and they prioritize their own needs over those of the groups to which they belong (Markus & Kitayama, 1991; Hsu, 1985; Triandis et al., 1988). People with an independent self-construal are less attentive and sensitive to the needs of others (Markus & Kitayama, 1991; Singelis, 1994). This distinction has a number of implications for how people behave in social settings. For instance, an independent self-construal is also associated with the open expression of emotion, even in public. Such expression tends to focus on ego-focused emotions such as anger, pride, and frustration that may serve to assert one’s independence to others (Markus & Kitayama, 1991; Eid & Diener, 2001). When the definition of individualism has focused on competition, it is typically associated with striving for individual achievement and the desire to get ahead of others (Chen & West, 2008). Numerous other studies have used a definition of individualism that is equated with competition (Triandis, Leung, Villareal, & Clack, 1985; Triandis et al., 1988;

Being Different or Being Better?

137

Cox, Lobel, & McLeod, 1991; Diaz-Guerrero, 1984; De Dreu, Nijstad, & van Knippenberg, 2008). Triandis et al. (1985) found that individualists in the United States emphasize competition, social recognition, and personal pleasure. Cox et al. (1991) found that groups consisting of individualists behave competitively, whereas groups consisting of collectivists behaved more cooperatively. Diaz-Guerrero (1984) also found an emphasis on competitive behavior in individualists in comparison with Latin American collectivists. In fact, Triandis et al. (1988) determined in a factor analysis that the largest contribution to individualism in the United States, approximately 35% of the variance, stems from a single factor that was labeled ‘‘self-reliance with competition.’’ Interestingly, some have argued that competition is a different element of culture than individualism and should be considered separately (Schimmack, Oishi, & Diener, 2005; Brewer & Chen, 2007), but they are more often treated interchangeably, particularly in the research on group creativity.

Bargaining and Negotiation In social psychology, individualism has also been an extremely important construct but it arose in the context of group dynamics and has considerable influence in research on bargaining and negotiation (De Dreu & Boles, 1998; De Dreu et al., 2000). According to the Theory of Cooperation (Deutsch, 1949, 1973; Deutsch, Krauss, & Rosenau, 1962), social interaction can be understood in terms of how people perceive their goals to be related to others. In cooperation, people perceive their goals as positively related; the attainment of one’s goal facilitates the attainment of another’s goal. In competition, people perceive their goals as negatively linked; the attainment of one’s goal precludes the attainment of another’s goal. In Deutsch’s (1949) seminal study, he manipulated these social motives in two sections of an introductory psychology course. In one section, students were told that all group members would receive the same grade and that their grade depended on how well they performed relative to similar groups. In another section, students were told that their grade would be based on their individual contribution; the person who contributed the most would get the highest grade. There were striking differences between the two sections in terms of how they related to each other. The cooperative sections reported having friendly discussions, feeling more satisfied, being more attentive to others and feeling more personally secure. The competitive

138

JACK A. GONCALO AND VERENA KRAUSE

sections reported feeling aggressive, not listened to and not well understood by others. De Dreu et al. (2000) distinguish between egoistic and competitive social motives. Whereas someone with an egoistic social motive tries to maximize their own outcome without regard for the other parties involved, someone with a competitive social motive tries to maximize their own outcome with a negative regard for the other parties involved. Thus, either zero or negative weight respectively is being put on the opposing party’s outcome. De Dreu and Boles (1998) found that competitive negotiators mainly consult competitive heuristics, whereas individualistic (egoistic) negotiators consult both competitive and cooperative heuristics. But, both competitive and egoistic negotiators are willing to use competitive heuristics if they are more effective. De Dreu et al.’s (2000) meta-analysis found largely the same consequences for these two social motives: engagement in less problemsolving behavior, more contentious behavior, and achievement of lower joint outcomes in comparison with pro-social negotiators. Competitive negotiators engage in more contentious behavior, in less problem-solving behavior, and overall achieve sub-optimal solution relative to co-operative negotiators (De Dreu et al., 2000). Although existing research has shown that pro-self social motives stimulate greater creativity than pro-social motives, we do not yet know whether, within the pro-self category, competitive and egoistic motives might have similar effects. In sum, the distinction between competitive and egoistic motives is strikingly similar to the distinction between competition and independence in cross-cultural psychology. At a general level, the research from two distinct literatures converges insofar as both individualism and pro-self social motives facilitate creative performance more so than collectivism and pro-social motives. However, within these very broad categories, independence and competition may have differential effects on group creativity, but existing research has not yet disentangled the role of each motive in the creative process.

DIFFERENTIAL EFFECTS OF INDEPENDENCE AND COMPETITION In this section, we develop a series of propositions that distinguish between the consequences of independence and competition for group creativity. The first set of propositions focuses on idea generation, and the second set

139

Being Different or Being Better?

of propositions focuses on the process of idea selection. To date, most studies of group brainstorming have focused on productivity in terms of how many ideas a group generates in a fixed amount of time and the extent to which those ideas are novel (Rietzschel et al., 2006; Girotra et al., 2010). This longstanding emphasis on productivity has been criticized in part because there are many outcomes that indicate an effective brainstorming session other than the sheer number of ideas generated (Sutton & Hargadon, 1996). In an intriguing development, recent research has broadened to include the idea selection stage because the creative process does not involve merely generating ideas, but selecting one that might be implemented (Rietzschel et al., 2006; Faure, 2004; Putman & Paulus, 2009). Interestingly, idea generation and idea selection may only be loosely coupled in the sense that coming up with a large number of novel ideas does not necessarily guarantee that the group will select a creative idea (Rietzschel et al., 2006). Surprisingly, a recent study suggests that groups actually avoid novel ideas unless specifically instructed to do so (Rietzschel, Nijstad, & Stroebe, 2010). Here we examine the unique effects of independence and competition at both stages of the creative process.

Idea Generation There is considerable research suggesting that competition facilitates productivity in brainstorming groups in which the goal is to generate as many ideas as possible (Osborn, 1957; Simonton, 1999). Competition has been shown to facilitate idea generation in both electronic and face-to-face groups by motivating individuals to match their performance with a more productive member of the group (Paulus, Larey, Putman, Leggett, & Roland, 1996; Munkes & Diehl, 2003; Dugosh & Paulus, 2005). Groups are more productive (e.g., they generate more ideas) when each member of the group is trying to generate more ideas than everyone else. The expression of a large number of solutions (productivity) may, in turn, lead to more creative solutions. According to evolutionary theories of creativity (Campbell, 1960; Staw, 1991; Simonton, 1999, 2003) creativity is a probabilistic consequence of quantity that explains why, for instance, the most creative people in many fields are also the most prolific (Simonton, 2003). Therefore, competition may facilitate group creativity by increasing productivity because the more ideas that are generated, the more likely those ideas will be novel departures from existing solutions.

140

JACK A. GONCALO AND VERENA KRAUSE

Unlike competition, independence may not necessarily increase the number of ideas generated during group brainstorming sessions. The motive to remain independent may cause people to focus on their own ideas and be unmotivated to carefully consider the ideas shared by others. In other words, the preference for being alone and for working autonomously may make independent selves simply uninterested in collaborating to reach an optimal group outcome (Wiekens & Stapel, 2008). Indeed, Goncalo and Kim (2010) primed individual group members with either an independent or interdependent self-construal and did not observe any main effects of selfconstrual on the number of ideas or the novelty of ideas expressed in face-to-face brainstorming sessions. In other words, although an independent self-construal may stimulate divergent thinking among individual group members, their ideas may remain unexpressed. Therefore, unlike competition, independence alone may not have an impact on either the productivity or creativity of group ideation. Proposition 1. The positive relationship between competition and group creativity is mediated by productivity, whereas independence is unrelated to both productivity and group creativity.

Idea Selection Although competition may promote idea expression, it may become a liability at the idea selection stage. A competitive norm may cause group members to derogate each other’s ideas, promote their own ideas even if they are not optimal and refuse to compromise (De Dreu et al., 2008). These problems may be exacerbated in naturalistic settings where receiving credit for a highly creative idea may be extremely profitable (Audia & Goncalo, 2007). There is intriguing evidence to support this prediction from research on the effect of rivalry on knowledge valuation (Menon, Thompson, & Choi, 2006). The results of three studies showed that people ignore good ideas suggested by members of one’s own group because endorsing their good ideas would make one look like a follower and cause a loss of status. The potential threat to the self causes people to look outside the group for inspiration and even to endorse ideas suggested by external rivals (Menon et al., 2006). This research suggests that the degree of competition within the group should be inversely related to their ability to select creative ideas.

Being Different or Being Better?

141

Proposition 2. Competition causes the group to derogate each other’s ideas and reduces the likelihood of selecting the most creative idea from their pool of available ideas. Independence may also have negative consequences for idea selection but may do so through different underlying mechanisms. According to the motivated information processing in groups (MIP-G) model, the quality of group decision making is higher when groups are willing to ‘‘expend effort to achieve a thorough, rich and accurate understanding of the world, including the group task and decision problem at hand’’ an orientation called epistemic motivation (De Dreu et al., 2008, p. 23). Whereas, independent selves might be highly motivated to reach correct or creative solutions on individual tasks on which they can work alone and receive credit for their own work, they may be withdrawn, unmotivated, and unwilling to expend effort on group tasks that demand social interaction. There is indirect evidence to support this prediction from research on narcissism showing that narcissists generate more ideas in brainstorming groups when their individual contributions were identifiable than when individual contributions were anonymous (Wallace & Baumeister, 2002). Although narcissism and independence are clearly not identical, they do share a pre-occupation with the self; narcissists are more likely to use singular first person pronouns in speech (e.g., I and me) and the use of such pronouns has been shown to prime a independent self-construal (Raskin & Shaw, 1988; Brewer & Gardner, 1996). It is possible that independence may cause groups to reach premature closure on the first minimally acceptable idea simply to end the discussion and perhaps turn their attention to solo tasks where there is a greater opportunity to receive personal recognition. Therefore, we predict the following: Proposition 3. Independence reduces epistemic motivation on group tasks and reduces the group’s ability to select creative ideas.

INTERACTION BETWEEN INDEPENDENCE AND COMPETITION In the last section, we distinguished between independence and competition as important facets of individualism that may either exert different effects on group creativity or produce similar effects through different mechanisms. The possibility that independence and competition are conceptually distinct

142

JACK A. GONCALO AND VERENA KRAUSE

raises the intriguing possibility that they may interact to influence group processes and performance. In other words, one might think of independence and competition as two bi-polar continuums that, when considered together, may result in a 2  2 factorial (see Fig. 1 for a summary). Independence is on a continuum with conformity to the group (Asch, 1956; Allen, 1965) and competition is on a continuum with cooperation (Deutsch, 1949; Messick & McClintock, 1968). In this section, we consider the consequences for group creativity that may result from this interaction. Cell A in Fig. 1 represents an interaction between independence and competition and might be similar to vertical individualism that indicates a desire to assert one’s uniqueness by outcompeting others (Singelis et al., 1995). Indeed, these two separate characteristics of individualism have been shown to co-exist in the United States (Triandis et al., 1988; Chen & West, 2008; Oyserman et al., 2002). Interestingly, there is recent evidence to suggest that independence and competition may interact to stimulate group ideation. Goncalo and Kim (2010) theorized that groups composed of people with a salient independent self-construal might express novel ideas, but only if they also endorse an equity reward allocation rule that permits people to compete for a larger share of the group reward (Deutsch, 1985). Interdependent selves may not be as motivated by an equity rule since they do not seek opportunities to stand out but rather prefer to blend in and to maintain harmony with other group members (Markus & Kitayama, 1991). In contrast, independent selves might be more motivated to compete since the equity rule allows them the opportunity to stand out by expressing more ideas than others (Triandis & Gelfand, 1998). In an experiment, Goncalo and Kim (2010) crossed an independent versus interdependent self-construal with a competitive or cooperative reward allocation rule in a factorial design. The results did not reveal any main effect of either self-construal or of reward allocation rule. Instead, they showed that groups generate more ideas and more novel ideas when they are simultaneously primed to think independently and to endorse a competitive reward allocation rule. Participants in this condition also reported being more vertical individualistic (Triandis, Chen, & Chan, 1998), and video tape data of the groups’ brainstorming sessions also revealed that people in this condition showed less regard for other peoples’ ideas (and cut into each others’ turn during the brainstorming session). Each mechanism fully mediated the productivity gain observed in the groups that were both independent and competitive. Future research might investigate the possibility that independence causes creative ideas to come to mind (Wiekens & Stapel, 2008) and competition

Conformity

Independence

The Interactive Effects of Independence and Competition on Group Creativity.

D (1) Avoid conflict (2) Withold novel ideas that may trigger controversy (3) Suggest ideas that converge with those suggested by others

C (1) Compete not to be unique, but to imitate others (2) Triggers feelings of envy and counter-productive behavior (3) Causes ambivalence between the group and its members

Fig. 1.

B (1) Independence that is group oriented and open minded. (2) Willingness to compromise before task conflict escalates (3) Motivates both idea expression and idea selection

Cooperation

A (1) Assert one's unique qualities by "winning" in competitions (2) Motives similar to those underlying minority dissent (3) Facilitates idea expression but interferes with idea selection

Competition

Being Different or Being Better? 143

144

JACK A. GONCALO AND VERENA KRAUSE

provides the motivation to express ideas to others (Dugosh & Paulus, 2005). Group ideation suffers when either ingredient is missing. Subsequent work might also investigate whether this combination has any impact on idea selection. Apropos of our earlier point, the competitive behavior that results may interfere with the process of choosing an idea to pursue. Proposition 4. Independence interacts with competition to facilitate the expression of creative ideas such that group creativity is highest (and creative idea selection lowest) when individuals are motivated to demonstrate their independence by competing with other group members. Cell D in Fig. 1 represents what is, in our view, the least creative combination of elements: The desire to both cooperate and to conform to the group. Such groups might avoid productive conflict, withhold their most novel ideas for fear of causing controversy and suggest ideas that converge with those suggested by others. In other words, ‘‘You have a great idea! I have one just like it!’’ These processes might promote harmony but may be inimical to the task of generating creative solutions. Some might argue that conforming and cooperative groups might actually be in an excellent position to generate creative ideas if they are simply instructed to do so. After all, their willingness to galvanize their efforts to complete assigned goals should make them productive and efficient. Goncalo and Staw (2006) tested this possibility in an experiment in which they instructed individualistic and collectivistic groups to generate either creative or practical solutions. The results showed that collectivistic groups underperformed individualistic groups even when specifically instructed to be creative, thus casting doubt on this rather optimistic proposition. One suspects that the collectivistic groups were probably very happy to work together but blissfully unaware that their ideas were few and mundane. Cells A and D in Fig. 1 are the two cells that are typically contrasted in most studies of individualism-collectivism. In our view, however, it is the diagonal from cell B to C that is the most intriguing and unique in the context of existing research. It is theoretically possible that independence may co-exist with cooperation and conversely that competition can co-exist with conformity to the group. We explore these quadrants in more detail below.

Independence Motivated by the Desire to Cooperate The idea that independence can be motivated by cooperation (cell B) is surprisingly consistent with Asch’s (1956) original conception of

Being Different or Being Better?

145

independence from the group majority. Asch (1956) found that when confronted with an unanimous majority, people can sometimes ignore the evidence of their own senses and adopt the majority point of view. They do so either because they want to be liked and accepted or because they assume the majority must be privy to some information that the lone individual is unaware of (Deutsch & Gerard, 1955). As Levine (1999) reminded us, Asch (1956) assumed that individuals maintain an independent position because they want the group to detect a correct solution, and they are willing to surrender their position if the evidence suggests they are wrong. According to Levine (1999, p. 361), this kind of independence is, ‘‘group oriented, cooperative and open minded.’’ Much like the proverbial whistleblower, individuals may risk ostracism because they care about the group and want to see it prosper (Morrison & Milliken, 2000). Asch (1956) also correctly surmised that independence is critical to group functioning because exposure to dissenting opinion may arouse doubt and make the group more open to alternatives. This notion is consistent with subsequent research on minority influence showing that dissent stimulates divergent thinking in the majority (Nemeth, 1986; Nemeth & Goncalo, 2005; Nemeth & Goncalo, 2011). However, the assumed motives behind the expression of dissenting opinions in the minority influence literature (Maass & Clark, 1984) are quite different from those assumed by Asch (1956). According to Moscovici (1976), dissenting points of view are maintained when people are convinced they are right and they want their point of view to prevail (Levine, 1999). When dissent is maintained over time with consistency and confidence (Nemeth & Wachtler, 1983), then they are more likely to convert the majority, at least in private (Moscovici, 1980). To ‘‘win’’ dissenting opinions cannot falter in the face of majority pressure. However, unlike Asch (1956) the motives underlying minority influence are ‘‘self-oriented, competitive and close-minded’’ (Levine, 1999, p. 361). In that sense, this perspective is most consistent with the conditions present in cell A: independence combined with a sense of competition. Thus we would expect similar consequences: Independence motivated by competition may stimulate creative ideation but may interfere with the process of reaching consensus on an idea to pursue to the implementation stage. Independence motivated by cooperation may be the most advantageous for group creativity because it may stimulate both ideation and permit idea selection. Independence provides a mindset conducive to creative thought (Wiekens & Stapel, 2008) and the desire to cooperate with the group to reach a high-quality solution may motivate idea expression since building on, combining, and improving the ideas suggested by others will help the

146

JACK A. GONCALO AND VERENA KRAUSE

group reach its goal. At the idea selection stage, independence may provide the confidence to stand alone, and if necessary to advocate for unpopular ideas that the rest of the group may initially dismiss. Dissent at the idea selection stage may cause the group to consider more ideas and to deliberate more carefully before making a decision (Nemeth, Brown, & Rogers, 2001). However, independence motivated by the desire to cooperate may cause the dissenter to compromise if they are satisfied that the group has made an optimal selection. A more competitive orientation may cause conflict to escalate to the point of being destructive and personal as opposed to task focused (Jehn, 1995; De Dreu & Weingart, 2003) because the desire to prevail may override the willingness to eventually compromise. Research showing that individually focused attributions raise the quality of group decision making (a convergent task) might be reinterpreted in light of this interaction (Goncalo & Duguid, 2008). Attributions to the contributions made by each individual member may have promoted a sense of independence from the group (e.g., I make a unique contribution to the group), but a recent experience of shared success may imbue the group with positive affect which, in turn, increases cooperation and commitment (Lawler, 2001). An increased willingness to cooperate may explain why individualistic attributions facilitated the exchange of knowledge and raised the quality of group decision making following success but not following failure (Goncalo & Duguid, 2008). Therefore, we predict the following: Proposition 5. Independence interacts with cooperation to facilitate both the expression and selection of creative ideas such that group creativity and creative idea selection are highest when independence is motivated by the desire to cooperate with the group.

Competition Motivated by the Desire to Conform Our model also suggests the counterintuitive possibility that competitive behavior can be motivated by the desire to conform to the group (cell C). However, competition in this context will not be a driver of novelty but a motivator of counterproductive behavior. Like the catchphrase, ‘‘Keeping Up With the Joneses,’’ people might compete, not by diverging from their competitors, but by attempting to imitate them. Competition motivated by the desire to conform could trigger envy; a negative emotion felt ‘‘when a person lacks another’s superior quality, achievement, or possession and either desires it or wishes that the other

Being Different or Being Better?

147

lacked it’’ (Parrott & Smith, 1993, p. 906; Cohen-Charash & Mueller, 2007). Envy only occurs when one perceives the envied person to be similar to oneself and when one lacks something important to the self-concept. Harming the envied person can be seen as an affect-regulation technique (Baumeister, Smart, & Boden, 1996; Bushman, Baumeister, & Phillips, 2001) because frustration is reduced (Fox & Spector, 1999; Kulik & Brown, 1979; Smith, 1991; Spector, 1975, 1978). And, harming the envied person can also feel empowering and increase one’s self-esteem (Fein & Spencer, 1997). Unfortunately, feelings of envy might not be conducive to group creativity. If one member of the group suggests an idea that is greeted with excitement and hailed as a creative solution by others, envious team mates might respond by either sabotaging the idea or by attempting to come up with an idea that is very similar so as to bask in reflected glory. Instead of motivating divergence from the status quo, envy is more likely to constrain the group to solutions that are at best, poor imitations of more creative ideas. Proposition 6. Competition should interact with conformity to trigger envy and envy should, in turn, stifle creative expression. The desire to simultaneously compete with and be similar to others may also create a sense of ambivalence between the group and its members. On the one hand, one is attracted to the group enough to want to be liked and accepted by them. On the other hand, one is willing to derogate and harm the group to ensure that they remain similar to you. This is the dilemma often faced by minorities who try to succeed in school and advance in their career only to be told they are ‘‘acting white’’ by members of their own group (Fordham & Ogbu, 1986; Fryer & Torelli, 2006). The competitive drive to succeed is experienced at the same time as the fear of standing out from the group and garnering their disapproval. Ambivalence is an emotion in which people experience coexisting, opposing feelings toward a person, object, or idea (Fong, 2003). There are two types of ambivalence, potential (Kaplan, 1972) and felt (Jamieson, 1993; Priester & Petty, 1996) ambivalence. Potential ambivalence refers to two opposing beliefs held by a person, who is unaware of this opposition presumably because she has never thought about it. Felt ambivalence, on the contrary, is in a person’s awareness and creates internal conflict (van Harreveld, van der Pligt, & de Liver, 2009). One way to reduce ambivalence is to make a choice (e.g., reject the group and compete to get ahead). But, choices are associated with uncertainty about their

148

JACK A. GONCALO AND VERENA KRAUSE

consequences, which causes discomfort. Negative consequences are particularly prevalent in one’s thoughts and carry considerable weight (Skrowonski & Carlston, 1989; Ito, Larsen, Smith, & Cacioppo, 1998; Eyal, Liberman, Trope, & Walther, 2004). Emotions such as disappointment, fear, guilt, and especially regret (Loomes & Sugden, 1982) are anticipated to follow if the decision turns out to be wrong. Ambivalence is particularly intriguing because there is evidence that felt ambivalence can actually stimulate creative thought. For instance, holding two contradictory thoughts in mind at the same time increases the likelihood that these opposing thoughts will be integrated into a novel idea (Rothenberg, 1990). Fong (2003) theorized that ambivalent emotions are unusual; they signal an unusual environment and thus make people sensitive to novel associations. The results of two experiments showed participants who were asked to either recall a time they felt ambivalent or to read a proverb that conveyed ambivalent emotions performed better on test of creative problem solving (Fong, 2003). Amabile, Barsade, Mueller, and Staw (2005) suggested that ambivalent emotions might facilitate creativity by increasing the breadth of cognitive material available for recombination. Positive and negative emotional experiences are stored in different memory nodes and ambivalent emotions may trigger both networks and allow people to draw on a wider range of experiences (Bower, 1981; Blaney, 1986). This research suggests that ambivalent emotions about one’s group may actually stimulate creative thought. However, since this research has not yet been conducted in a group setting, it is unclear whether or how such emotions could influence the expression of creative ideas. Nevertheless, the proposition below would be intriguing to pursue in future research. Proposition 7. Conformity and competition interact to trigger felt ambivalence, and felt ambivalence, in turn, stimulates creative problem solving at the individual level.

METHODOLOGICAL QUESTIONS We have argued that individualism can refer either to independence or competition and each motive might exert different effects on group creativity. Future research may gain considerable leverage from making this distinction clear. However, separating these two constructs may create some empirical challenges that we address in this section.

Being Different or Being Better?

149

Interestingly, although research in cross-cultural psychology has largely relied on cross-national comparisons, relating individualism to group creativity has been strictly experimental. This approach is part of a trend that has been gathering force in recent years. Most of the research on the individual-collectivism dimension has found that national membership alters the person’s self concept (Markus & Kitayama, 1991) and that such differences in self-construal, much like a personality trait, can be used to predict behavior across situations. Typically, this research has compared differences between East Asians and European North Americans (Lehman, Chiu, & Schaller, 2004) using nationality as a proxy for a person’s underlying cultural values of individualism versus collectivism (Brockner, 2003). Despite such between-country differences, however, there is also substantial within-country variation (Oyserman et al., 2002). In other words, cultural values might also be subject to more immediate influences in the social situation such as those present in laboratory settings (Oyserman & Lee, 2008). Unfortunately, we do not know whether manipulations used in the research on group creativity emphasize independence, competition, or a combination of the two. The manipulation of pro-self social motives employed by Beersma and De Dreu (2005) seems to clearly tap competition, given the emphasis on acquiring a greater share of a reward than others. Conversely, the selfconstrual manipulation developed by Brewer and Gardner (1996) used by Wiekens and Stapel (2008) and Goncalo and Kim (2010) probably triggers independence, given the data showing that this priming technique prompts people to think of themselves as independent, alone, and different. The attribution manipulation used by Goncalo (2004) and Goncalo and Duguid (2008) probably also makes people feel independent from the group but may also trigger the desire to cooperate if the attribution follows success as opposed to failure (Lawler, 2001). Likewise, the priming manipulation used by Goncalo and Staw (2006) probably also promotes a sense of oneself as independent given that people were prompted to describe their individual attributes and think of reasons why they were unique. These assumptions are consistent with the theoretical framework we proposed in this chapter as well as the results of existing research. Competition facilitated creative idea generation but interfered with the group’s ability to reach consensus (Beersma & De Dreu, 2005). In contrast, independence has not been shown to have a main effect on creative idea generation without specific instructions to be creative (Goncalo & Staw, 2006), positive feedback on a prior task (Goncalo, 2004; Goncalo & Duguid, 2008) or a salient equity norm that promotes idea expression

150

JACK A. GONCALO AND VERENA KRAUSE

(Goncalo & Kim, 2010). Moreover, independence does not necessarily interfere with the convergent stage of the creative process since individualistic (and perhaps independent) groups in these studies selected more creative ideas when instructed to be creative (Goncalo & Staw, 2006) and made more accurate decisions after receiving positive feedback (Goncalo & Duguid, 2008). Future research, however, might include measures of independence and competition to verify which aspect of individualism is made salient to the participants. Oyserman et al. (2002) conducted two meta-analyses on the existing scales of individualism-collectivism. They found that 83% of the scales included in the meta-analyses focused on personal independence; more than any other underlying construct. Hui (1988), for example, developed a scale that distinguished collectivists from individualists on the basis of their interdependence with parents, spouses, relatives, coworkers, friends, and neighbors. Participants rated statements such as ‘‘It is desirable that a husband and a wife have their own sets of friends, instead of having only a common set of friends’’ or ‘‘When I am among my colleagues/ classmates, I do my own thing without minding about them.’’ Gudykunst et al. (1996) measured independence with statements such as ‘‘Being able to take care of myself is a primary concern for me,’’ ‘‘I prefer to be self-reliant rather than depend on others,’’ and ‘‘I should decide my future on my own.’’ Other scales include ones by Singelis (1994), who used Markus and Kitayama’s review to create a 24-item Self-Construal Scale that measured independence and interdependence directly. Similarly, Triandis, McCusker, and Hui (1990) and Singelis et al. (1995) developed scales including measurements for independence. Hence, various scales exist that measure independence, and they may provide items that are useful manipulation checks. There are also a number of promising measures of competition. For example, in their factor analysis, Triandis et al. (1988) used a scale that included items such as ‘‘I feel winning is important in both work and games,’’ and ‘‘Doing your best isn’t enough; it is important to win.’’ Simmons, Wehner, Tucker, and King (1988) measured competition through their cooperative/competitive strategy scale, which included items such as ‘‘It is important to me to do better than others,’’ ‘‘To succeed, one must compete against others,’’ and ‘‘Success can be best defined as a situation in which there are both winners and losers.’’ Another scale measuring competition among other constructs was developed by Cassidy and Lynn (1989). They included items such as ‘‘I try harder when I am in competition with other people,’’ ‘‘I judge my performance on whether I do better than

151

Being Different or Being Better?

others rather than on just getting a good result,’’ and ‘‘It annoys me when other people perform better than I do.’’ In contrast to independence, however, items related to competition are not as frequently included in measures of individualism (Oyserman et al., 2002). Approximately 15% of scales included measures of competition, which was determined by items such as ‘‘It is important to me that I perform better than others on a task.’’

SUMMARY Existing research clearly suggests that individualism provides an atmosphere conducive to creative idea generation. However, we have argued that individualism may reflect either independence or competition; a distinction that has been overlooked in research on group creativity. Highlighting the distinction between these two constructs uncovered several theoretical insights, including the possibility that independence and competition (a) are theoretically and empirically distinct, (b) have differential effects on idea generation but similar effects on idea selection through different mechanisms and (c) that they may interact to stimulate group creativity. Our review also underscored an important point: Creativity is most likely to flourish when individuals retain their sense of independence from the group, not merely for the sake of being different, or to impose their ideas on others, but with some sense of responsibility to the group with whom they have cast their lot. Therefore, we conclude with considerable homage to Solomon Asch (1956) who initially asserted that independence and cooperation might co-exist in this way.

REFERENCES Adams, J. S. (1963). Toward an understanding of inequity. Journal of Abnormal and Social Psychology, 67, 422–436. Adams, J. S. (1965). Inequity in social exchange. In: L. Berkowitz (Ed.), Advances in experimental social psychology (vol. 2, pp. 267–299). New York: Academic Press. Allen, V. L. (1965). Situational factors in conformity. In: L. Berkowitz (Ed.), Advances in experimental social psychology (vol. 2). New York: Academic Press. Amabile, T. (1983). The social psychology of creativity: A componential conceptualization. Journal of Personality and Social Psychology, 45(2), 357–376. Amabile, T. M. (1996). Creativity in context. Boulder, CO: Westview. Amabile, T. M., Barsade, S. G., Mueller, J. S., & Staw, B. M. (2005). Affect and creativity at work. Administrative Science Quarterly, 50(3), 367–403.

152

JACK A. GONCALO AND VERENA KRAUSE

Asch, S. E. (1956). Studies of independence and submission to group pressure: I. A minority of one against a unanimous majority. Psychological Monographs, 70(9, Whole no. 417). Audia, P. G., & Goncalo, J. A. (2007). Success and creativity over time: A study of inventors in the hard-disk drive industry. Management Science, 53, 1–15. Barnett, W. P., & Hansen, M. T. (1996). The red queen in organizational evolution. Strategic Management Journal, 17, 139–157. Baumeister, R. F., Smart, L., & Boden, J. M. (1996). Relation of threatened egotism to violence and aggression: The dark side of high self-esteem. Psychological Review, 103(1), 5–33. Beersma, B., & De Dreu, C. K. W. (2005). Conflict’s consequences: Effects of social motives on post-negotiation creative and convergent group functioning and performance. Journal of Personality and Social Psychology, 89, 358–374. Blaney, P. H. (1986). Affect and memory – A review. Psychological Bulletin, 99(2), 229–246. Bond, R., & Smith, P. B. (1996). Culture and conformity: A meta-analysis of studies using Asch’s (1952b, 1956) line judgment task. Psychological Bulletin, 119(1), 111–137. Bontempo, R. (1993). Translation fidelity of psychological scales: An item response theory analysis of an individualism-collectivism scale. Journal of Cross-Cultural Psychology, 24, 149–166. Bower, G. H. (1981). Mood and memory. American Psychologist, 36, 129–148. Brewer, M. B., & Chen, Y. R. (2007). Where (who) are collectives in collectivism? Toward conceptual clarification of individualism and collectivism. Psychological Review, 114(1), 133–151. Brewer, M. B., & Gardner, W. (1996). Who is this ‘‘we’’? Levels of collective identity and self representations. Journal of Personality and Social Psychology, 71, 83–93. Brockner, J. (2003). Unpacking country effects: On the need to operationalize the psychological determinants of cross-national differences. Research in Organizational Behavior, 25, 333–367. Bushman, B. J., Baumeister, R. F., & Phillips, C. M. (2001). Do people aggress to improve their mood? Catharsis beliefs, affect regulation opportunity, and aggressive responding. Journal of Personality and Social Psychology, 81, 17–32. Campbell, D. T. (1960). Blind variation and selective retention in creative thought as in other knowledge processes. Psychological Review, 67, 380–400. Cassidy, T., & Lynn, R. (1989). A multifactorial approach to achievement motivation: The development of a comprehensive measure. Journal of Occupational Psychology, 62, 301–312. Chen, F. F., & West, S. G. (2008). Measuring individualism and collectivism: The importance of considering differential components, reference groups, and measurement invariance. Journal of Research in Personality, 42, 259–294. Cialdini, R. B., Reno, R. R., & Kallgren, C. A. (1990). A focus theory of normative conduct: Recycling the concept of norms to reduce littering in public places. Journal of Personality and Social Psychology, 58, 1015–1026. Cohen-Charash, Y., & Mueller, J. S. (2007). Does perceived unfairness exacerbate or mitigate interpersonal counterproductive work behaviors related to envy? Journal of Applied Psychology, 92(3), 666–678. Cox, T. H., Lobel, S. A., & McLeod, P. L. (1991). Effects of ethnic group cultural differences on cooperative and competitive behavior on a group task. The Academy of Management Journal, 34(4), 827–847. Deutsch, M. (1949). A theory of co-operation and competition. Human Relations, 2, 129–152.

Being Different or Being Better?

153

Deutsch, M. (1973). The resolution of conflict: Constructive and destructive processes. New Haven, CT: Yale University Press. Deutsch, M. (1985). Distributive justice, a social psychological perspective. New Haven, CT: Yale University Press. Deutsch, M., & Gerard, H. B. (1955). A study of normative and informational social influence upon individual judgment. Journal of Abnormal and Social Psychology, 195, 629–636. Deutsch, M., Krauss, R. M., & Rosenau, N. (1962). Dissonance of defensiveness. Journal of Personality, 30(1), 16–28. De Dreu, C. K. W., & Boles, T. L. (1998). Share and share alike or winner take all?: The influence of social value orientation upon choice and recall of negotiation heuristics. Organizational Behavior and Human Decision Processes, 76(3), 253–276. De Dreu, C. K. W., Nijstad, B. A., & van Knippenberg, D. (2008). Motivated information processing in group judgement and decision making. Personality and Social Psychology Review, 12, 22–49. De Dreu, C. K. W., & Weingart, L. R. (2003). Task versus relationship conflict, team performance, and team member satisfaction: A meta-analysis. Journal of Applied Psychology, 88(4), 741–749. De Dreu, C. K. W., Weingart, L. R., & Kwon, S. (2000). Influence of social motives on integrative negotiation: A meta-analytic review and test of two theories. Journal of Personality and Social Psychology, 78(5), 889–905. Diaz-Guerrero, R. (1984). La psicologia de los Mexicanos: Un paradigma. Revista Mexicana de Psicologia, 1(2), 95–104. Diehl, M., & Stroebe, W. (1987). Productivity loss in brainstorming groups: Toward the solution of a riddle. Journal of Personality and Social Psychology, 79, 722–735. Dugosh, K. L., & Paulus, P. B. (2005). Cognitive and social comparison processes in brainstorming. Journal of Experimental Social Psychology, 41, 313–320. Eid, M., & Diener, E. (2001). Norms of experiencing emotions in different cultures: Inter- and intranational differences. Journal of Personality and Social Psychology, 81(5), 869–885. Eyal, T., Liberman, N., Trope, Y., & Walther, E. (2004). The pros and cons of temporally near and distant action. Journal of Personality and Social Psychology, 86, 781–795. Faure, C. (2004). Beyond brainstorming: Effects of different group procedures on selection of ideas and satisfaction with the process. Journal of Creative Behavior, 38, 13–34. Fein, S., & Spencer, S. J. (1997). Prejudice as self-image maintenance: Affirming the self through derogating others. Journal of Personality and Social Psychology, 73(1), 31–44. Fiske, A. P., Kitayama, S., Markus, H. R., & Nisbett, R. E. (1998). The cultural matrix of social psychology. In: D. T. Gilbert, S. T. Fiske & G. Lindzey (Eds), The handbook of social psychology (4th ed, pp. 915–981). New York: McGraw-Hill. Fong, C. T. (2003). The effects of emotional ambivalence on creativity. Academy of Management Journal, 49(5), 1016–1030. Fordham, S., & Ogbu, J. (1986). Black students’ school success: Coping with the ‘‘burden of ‘acting white’ ’’. Urban Review, 18(3), 176–206. Forsyth, D. R., & Schlenker, B. R. (1977). Attributing the causes of group performance: Effects of performance quality, task importance, and future testing. Journal of Personality, 45, 220–236. Fox, S., & Spector, P. E. (1999). A model of work frustration–aggression. Journal of Organizational Behavior, 20, 915–931.

154

JACK A. GONCALO AND VERENA KRAUSE

Fryer, R. G., & Torelli, P. (2006). Acting white: The social price paid by the best and brightest minority students. Education Next, 6, 1. Girotra, K., Terwiesch, C., & Ulrich, K. T. (2010). Idea generation and the quality of the best idea. Management Science, 56(4), 591–605. Goncalo, J. A. (2004). Past success and convergent thinking in groups: The role of groupfocused attributions. European Journal of Social Psychology, 34, 385–395. Goncalo, J. A., & Duguid, M. M. (2008). Hidden consequences of the group serving bias: Causal attributions and the quality of group decision making. Organizational Behavior and Human Decision Processes, 107, 219–233. Goncalo, J. A. & Kim, S. H. (2010). Distributive justice beliefs and group idea generation: Does a belief in equity facilitate productivity? Journal of Experimental Social Psychology, 46, 836–840. Goncalo, J. A., & Staw, B. M. (2006). Individualism-collectivism and group creativity. Organizational Behavior and Human Decision Processes, 100, 96–109. Gudykunst, W. B., Matsumoto, Y., Ting-Toomey, S., Nishida, T., Kim, K., & Heyman, S. (1996). The influence of cultural individualism-collectivism, self-construals, and individual values on communication styles across cultures. Human Communication Research, 22, 510–543. Hofstede, G. (1980). Culture’s consequences: International differences in work related values. Newbury Park, CA: Sage Publications. Hsu, F. L. K. (1985). The self in cross-cultural perspective. In: A. J. Marsella, G. De Vos & F. L. K. Hsu (Eds), Culture and self (pp. 24–55). London: Tavistock. Hui, C. H. (1988). Measurement of individualism-collectivism. Journal of Research in Personality, 22, 17–36. Ito, T. A., Larsen, J. T., Smith, N. K., & Cacioppo, J. T. (1998). Negative information weighs more heavily on the brain: The negativity bias in evaluative categorizations. Journal of Personality and Social Psychology, 75(4), 887–900. Jamieson, D. W. (1993, August). The attitude ambivalence construct: Validity, utility, and measurement. Paper presented at the annual meeting of the American Psychological Association, Toronto, Ontario, Canada. Jehn, K. (1995). A multi-method examination of the benefits and detriments of intragroup conflict. Administrative Science Quarterly, 40, 256–282. Kaplan, K. J. (1972). On the ambivalence–indifference problem in attitude theory and measurement: A suggested modification of the semantic differential technique. Psychological Bulletin, 77, 361–372. Kulik, J. A., & Brown, R. (1979). Frustration, attribution of blame, and aggression. Journal of Experimental Social Psychology, 15, 183–194. Lawler, E. J. (2001). An affect theory of social exchange. American Journal of Sociology, 107(2), 321–352. Lehman, D. R., Chiu, C., & Schaller, M. (2004). Psychology and culture. Annual Review of Psychology, 55, 689–714. Levine, J. M. (1999). Solomon Asch’s legacy for group research. Personality and Social Psychology Review, 3(4), 358–364. Loomes, G., & Sugden, R. (1982). Regret theory: An alternative theory of rational choice under uncertainty. The Economic Journal, 92, 805–824. Markus, H., & Kitayama, S. (1991). Culture and self: Implications for cognition, emotion, and motivation. Psychological Review, 98, 224–253.

Being Different or Being Better?

155

Maass, A., & Clark, R. D. (1984). Hidden impact of minorities: 15 years of minority influence research. Psychological Bulletin, 95(3), 428–450. Mayer, R. E. (1992). Thinking, problem solving, cognition. New York: Worth Publishers. Menon, T., Thompson, L., & Choi, H. S. (2006). Tainted knowledge vs. tempting knowledge: People avoid knowledge from internal rivals and seek knowledge from external rivals. Management Science, 52(8), 1129–1144. Messick, D. M., & McClintock, C. G. (1968). Motivational bases of choice in experimental games. Journal of Experimental Social Psychology, 4(1), 1–25. Morrison, E. W., & Milliken, F. J. (2000). Organizational silence: A barrier to change and development in a pluralistic world. Academy of Management Review, 25(4), 706–725. Moscovici, S. (1980). Towards a theory of conversion behaviour. In: L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 13). New York: Academic Press. Moscovici, S. (1976). Social influence and social change. London, UK: Academic Press. Munkes, J., & Diehl, M. (2003). Matching or competition? Performance comparison processes in an idea generation task. Group Processes and Intergroup Relations, 6, 305–320. Nemeth, C., Brown, K., & Rogers, J. (2001). Devil’s advocate versus authentic dissent: Stimulating quantity and quality. European Journal of Social Psychology, 31(6), 707–720. Nemeth, C. J. (1986). Differential contributions of majority and minority influence. Psychological Review, 93, 23–32. Nemeth, C. J., & Goncalo, J. A. (2005). Creative collaborations from afar: The benefits of independent authors. Creativity Research Journal, 17, 1–8. Nemeth, C. J., & Goncalo, J. A. (2011). Rogues and heroes: Finding value in dissent. In: J. Jetten & M. Hornsey (Eds), Rebels in groups: Dissent, deviance, difference and defiance. Hoboken, NJ: Wiley-Blackwell. Nemeth, C. J., & Wachtler, J. (1983). Creative problem solving as a result of majority vs. minority influence. European Journal of Social Psychology, 13, 45–55. Osborn, A. F. (1957). Applied imagination. New York: Scribner. Oyserman, D., Coon, H. M., & Kemmelmeier, M. (2002). Rethinking individualism and collectivism: Evaluation of theoretical assumptions and meta-analyses. Psychological Bulletin, 128(1), 3–72. Oyserman, D., & Lee, S. W. S. (2008). Does culture influence what and how we think? Effects of priming individualism and collectivism. Psychological Bulletin, 134(2), 311–342. Parrott, W. G., & Smith, R. H. (1993). Distinguishing the experiences of envy and jealousy. Journal of Personality and Social Psychology, 64, 906–920. Paulus, P. B., Larey, T. S., Putman, V. L., Leggett, K. L., & Roland, E. J. (1996). Social influence process in computer brainstorming. Basic and Applied Social Psychology, 18, 3–14. Paulus, P. B., & Nijstad, B. A. (2003). Group creativity: Innovation through collaboration. New York: Oxford University Press. Paulus, P. B., & Yang, H. C. (2000). Idea generation in groups: A basis for creativity in organizations. Organizational Behavior and Human Decision Processes, 82, 76–87. Priester, J. R., & Petty, R. E. (1996). The gradual threshold model of ambivalence: Relating the positive and negative bases of attitude to subjective ambivalence. Journal of Personality and Social Psychology, 71, 431–449. Putman, V. L., & Paulus, P. B. (2009). Brainstorming, brainstorming rules and decision making. Journal of Creative Behavior, 43(1), 23–39. Raskin, R. N., & Shaw, R. (1988). Narcissism and the use of personal pronouns. Journal of Personality, 56, 393–404.

156

JACK A. GONCALO AND VERENA KRAUSE

Rhee, E., Uleman, J. S., & Lee, H. K. (1996). Variations in collectivism and individualism by in-group and culture: Confirmatory facto analyses. Journal of Personality and Social Psychology, 71, 1037–1053. Rietzschel, E. F., Nijstad, B. A., & Stroebe, W. (2006). Productivity is not enough: A comparison of interactive and nominal brainstorming groups on idea generation and selection. Journal of Experimental Social Psychology, 42, 244–251. Rietzschel, E. F., Nijstad, B. A., & Stroebe, W. (2010). The selection of creative ideas after individual idea generation: Choosing between creativity and impact. British Journal of Psychology, 101, 47–68. Rothenberg, A. (1990). Creativity and madness: New findings and old stereotypes. Baltimore: Johns Hopkins University Press. Schimmack, U., Oishi, S., & Diener, E. (2005). Individualism: A valid and important dimension of cultural differences between nations. Personality and Social Psychology Review, 9(1), 17–31. Schlenker, B. R., & Miller, R. S. (1977). Egocentrism in groups: Self-serving biases or logical information processing? Journal of Personality and Social Psychology, 35, 755–764. Simmons, C. H., Wehner, E. A., Tucker, S. S., & King, C. S. (1988). The cooperative/ competitive strategy scale: A measure of motivation to use cooperative strategies for success. The Journal of Social Psychology, 128(2), 199–205. Shalley, C. E., & Gilson, L. L. (2004). What leaders need to know: A review of social and contextual factors that can foster or hinder creativity. Leadership Quarterly, 15, 33–53. Simonton, D. K. (1999). Creativity as blind variation and selective retention: Is the creative process Darwinian? Psychological Inquiry, 10, 309–328. Simonton, D. K. (2003). Scientific creativity as constrained stochastic behavior: The integration of product, person and process perspectives. Psychology Bulletin, 129(4), 475–494. Singelis, T. M. (1994). The measurement of independent and interdependent self-construals. Personality and Social Psychology Bulletin, 20, 580–591. Singelis, T. M., Triandis, H. C., Bhawuk, D. P. S., & Gelfand, M. J. (1995). Horizontal and vertical dimensions of individualism and collectivism: a theoretical and measurement refinement. Cross-Cultural Research, 29, 240–275. Skrowonski, J. J., & Carlston, D. E. (1989). Negativity and extremity biases in impression formation: A review of explanations. Psychological Bulletin, 105, 131–142. Smith, R. H. (1991). Envy and the sense of injustice. In: P. Salovey (Ed.), The psychology of jealousy and envy (pp. 79–99). New York: Guildford Press. Spector, P. E. (1975). Relationships of organizational frustration with reported behavioral reactions of employees. Journal of Applied Psychology, 60, 635–637. Spector, P. E. (1978). Organizational frustration: A model and review of the literature. Personnel Psychology, 31, 815–829. Staw, B. M. (1991). Dressing up like an organization – When psychological theories can explain organizational action. Journal of Management, 17(4), 805–819. Sutton, R. I., & Hargadon, A. (1996). Brainstorming groups in context: Effectiveness in a product design firm. Administrative Science Quarterly, 41, 685–718. Taylor, D. M., & Doria, J. R. (1981). Self-serving and group-serving bias in attribution. Journal of Social Psychology, 113, 201–211. Taylor, D. W., Berry, P. C., & Block, C. H. (1958). Does group participation when using brainstorming facilitate or inhibit creative thinking. Administrative Science Quarterly, 3(1), 23–47. Tjosvold, D. (1984). Cooperation theory and organizations. Human Relations, 37, 743–767.

Being Different or Being Better?

157

Toubias, O. (2006). Idea generation, creativity and incentives. Marketing Science, 25(5), 411–425. Triandis, H. C. (2000). Individualism and collectivism: Past, present and future. In: D. Matsumoto (Ed.), The handbook of culture and psychology. New York: Oxford University Press. Triandis, H. C., Bontempo, R., Villareal, M. J., Asai, M., & Lucca, N. (1988). Individualism and collectivism: Cross-cultural perspectives on self-ingroup relationships. Journal of Personality and Social Psychology, 54, 323–338. Triandis, H. C., Chen, X. P., & Chan, D. K. (1998). Scenarios for the measurement of collectivism and individualism. Journal of Cross-Cultural Psychology, 29, 275–289. Triandis, H. C., & Gelfand, M. J. (1998). Converging measurement of horizontal and vertical individualism and collectivism. Journal of Personality and Social Psychology, 74(1), 118–128. Triandis, H. C., Leung, K., Villareal, M. J., & Clack, F. L. (1985). Allocentric versus idiocentric tendencies: Convergent and discriminant validation. Journal of Research in Personality, 19, 395–415. Triandis, H. C., McCusker, C., & Hui, C. H. (1990). Multimethod probes of individualism and collectivism. Journal of Personality and Social Psychology, 59, 1006–1020. van Harreveld, F., van der Pligt, J., & de Liver, Y. (2009). The agony of ambivalence and ways to resolve it: Introducing the MAID model. Personality and Social Psychology Review, 13, 45–61. Wallace, H. M., & Baumeister, R. F. (2002). The performance of narcissists rises and falls with perceived opportunity for glory. Journal of Personality and Social Psychology, 82(5), 819–834. Waterman, A. S. (1984). The psychology of individualism. New York: Praeger. Weldon, E. (1984). Deindividualization, interpersonal affect and productivity in laboratory task groups. Journal of Applied Social Psychology, 14(5), 469–485. Wiekens, C. J., & Stapel, D. A. (2008). I versus we: The effects of self-construal level on diversity. Social Cognition, 26(3), 368–377.

APPLYING A STATUS PERSPECTIVE TO RACIAL/ETHNIC MISCLASSIFICATION: IMPLICATIONS FOR HEALTH Irena Stepanikova ABSTRACT This study applies a new taxonomy of racial/ethnic misclassification that considers shifts in racial/ethnic status to investigate physical and emotional responses to racial treatment among different misclassification types. It finds that the odds of reporting physical and emotional symptoms increase 3.3 and 2.9 times, respectively, among individuals who experience racial/ ethnic status loss (i.e., are misclassified into a racial/ethnic category with lower status compared to their self-reported category) compared to their correctly classified counterparts. In contrast, individuals who experience racial/ethnic status gain (i.e., are misclassified into a racial/ethnic category with higher status compared to their self-reported category) are no more likely to suffer from symptoms compared to correctly classified individuals. The results suggest that being misclassified per se does not necessarily harm well-being, but the loss of social status inherent in some types of misclassification does.

Advances in Group Processes, Volume 27, 159–183 Copyright r 2010 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0882-6145/doi:10.1108/S0882-6145 (2010)0000027009

159

160

IRENA STEPANIKOVA

According to Bobo and Fox (2003, p. 319) the insight that ‘‘the basic social processes invoked by the terms race, racism, and discrimination are quintessentially social psychological phenomena’’ has been long downplayed in mainstream sociology. This study joins the emerging efforts to apply a social psychological perspective to these social processes and makes several important contributions. It proposes a new taxonomy of racial/ethnic misclassification that considers how the observed racial/ethnic category of misclassified individuals compares to their self-reported racial/ethnic category in terms of social status. Using survey data, I demonstrate that this taxonomy helps to distinguish types of misclassification that are harmful to physical and mental health outcomes from those that pose no increased risk. Racial/ethnic misclassification is a theoretically interesting phenomenon for social psychologists because it is a notable example of a situation in which a shared understanding of social categories is lacking. Such a shared understanding is a sine qua non, a cornerstone of smooth social interaction. This is because social meanings attached to social categories help organize social interactions by anchoring a complex system of expectations about interaction partners’ roles, attributes, and appropriate behavior in a particular social situation (Bourg, 2002). When such a shared understanding is lacking because the interaction partners’ perception of social categories varies, expectation for behavior are likely to vary as well. Differences in expectations may make social interaction uncomfortable, difficult, or even impossible (Campbell & Troyer, 2007). Relating turns highly problematic, and social interaction becomes a source of psychological distress. Recent data reveal that racial and ethnic misclassification is a prevalent phenomenon in some racial and ethnic groups. In a recent national survey of young adults aged 18–26 years, interviewers misclassified 37 percent of nonHispanic Native Americans and 18 percent of Hispanic blacks (Campbell & Troyer, 2007). In another study relying on self-reports, the rates of misclassification were even higher. Forty-three percent of Hispanics and 63 percent of American Indian/Alaska Natives were misclassified (Saperstein, 2006). In this study, at least 5 percent of Americans overall, regardless of their race, experienced misclassification. Given the prevalence of misclassification in some groups, it is important to understand its consequences for the lives of those who experience it. Scholars have only recently started to explore this important problem. A study conducted by Campbell and Troyer (2007) found that racial misclassification among Native Americans was linked to an increased likelihood of considering or attempting suicide and to believing that one will die before the age of 35. The authors argue that since misclassification invalidates self-image and

Applying a Status Perspective to Racial/Ethnic Misclassification

161

identity, de-legitimizes claims for membership in own community, and threatens social status, it can increase stress levels and lead to negative mental health outcomes. Native Americans appear especially vulnerable to such negative outcomes, since as a group, they experience misclassification more commonly than other minorities. One important problem with previous scholarship is that it treats misclassification as a homogenous phenomenon, failing to recognize its various subtypes. In this chapter, I turn to a status perspective to address this shortcoming. I develop concepts of racial/ethnic status loss and racial/ ethnic status gain that consider the status implications of self-claimed and other-claimed racial/ethnic identity. I argue that racial/ethnic status loss and racial/ethnic status gain differ sharply in their consequences for a misclassified person’s health and well-being. Racial/ethnic status loss, which involves an incorrect placement of a person into a racial/ethnic category with lower status compared to the status of the self-claimed category, is stressful. I expect it to lead to negative emotional reactions and to higher physical symptom load. In contrast, racial/ethnic status gain that involves classifying a person as belonging to a racial/ethnic category with higher status compared to the self-proclaimed category may not compromise health at all. In fact, well-being may benefit from this type of misclassification since it helps to alleviate interpersonal stress commonly experienced by individuals from disadvantaged racial/ethnic backgrounds. Such a prediction stands in a sharp contrast to previous research that assumed that negative effects of misclassification are distributed uniformly across cases. In addition to the theoretical elaboration of misclassification types and empirical testing of their predictive power regarding health outcomes, I add to the scope and robustness of previous scholarship in several other ways. First, this is the first study to focus specifically on routine misclassification, that is, on misclassification that happens repeatedly, not just on isolated occasions. Since race is a social construct shaped in multiple repeated social interactions (Ferber, 1995; Harris & Sim, 2002; Rockquemore & Brunsma, 2002), any single instance of racial/ethnic misclassification may be of little consequence for most persons experiencing it (though some may find even an isolated instance of misclassification stressful). In contrast, repeated instances of misclassification may have profound impact, ultimately leading to stress overload and to health problems. Second, the measure of misclassification used here is respondent-reported, not interviewer-reported as in previous research. This is an important improvement since the actor’s interpretation of misclassification is likely to play a role in her emotional and physical responses. For a person to

162

IRENA STEPANIKOVA

experience misclassification as stressful, she must first subjectively understand that she has been misclassified – a process we can ascertain only from the misclassified person’s subjective responses. Thus, respondent-reported measure taps into cognitive processes that may link misclassification to health and well-being outcomes. Third, I focus specifically on emotional and physical responses to racial treatment. This is an important advance over previous scholarship that limited its focus on mental health, such as suicidal ideation, depression, use of psychological services, and fatalism (Campbell & Troyer, 2007). I argue that symptoms, both emotional and physical, that a respondent subjectively identifies as responses to racial treatment are important in understanding the lived racial experience and its implications for health. Finally, this chapter focuses on adults aged 18–99, extending Campbell and Troyer’s (2007) focus on young adults aged 18–26. This focus on a broader age range is important for evaluating the overall robustness of racial/ethnic misclassification effects on health-related outcomes.

RACE, ETHNICITY, AND SOCIAL STATUS Scholars have long recognized the considerable influence of social status on health. Social status, generally understood as the placement in the hierarchy of esteem, influence, and prestige, has been shown to play a key role in mortality rates and other health outcomes (Marmot, Shipley, & Rose, 1984; Marmot et al., 1991). Its notable effects persist even when factors such as material deprivation, access to health care, and job characteristics are held constant. These observations suggest that health is eroded by subjectively experienced ‘‘hierarchy stress’’ associated with the low placement in the status hierarchy, rather than solely by the material correlates of such a placement (Marmot et al., 1984, 1991). Racial/ethnic status is a form of social status. It is not surprising that its health implications are analogous to other forms of status: People at the top of the racial/ethnic status hierarchy enjoy better physical and emotional health than those at the bottom. Lower status attached to the membership in minority groups is linked to poorer health through threats to self-worth, everyday experiences of disrespectful treatment, and subtle or more overt forms of discrimination, among other factors. In fact, evidence shows that experience of discrimination, which represents a highly stressful aspect of daily lives of many minority persons, threatens their physical and mental health (Bird, Bogart, & Delahanty, 2004; Borrell, Kiefe, Williams,

Applying a Status Perspective to Racial/Ethnic Misclassification

163

Diez-Roux, & Gordon-Larsen, 2006; Kessler, Mickelson, & Williams, 1999). The impact of discrimination on mental health is considerable, with one study comparing it in its magnitude to major life events such as death of a loved one, divorce, or job loss (Kessler et al., 1999). Discrimination is an extreme form of esteem denial. Such esteem denial is a part of daily experience for many minority individuals as they interact in the social world. This is because race and ethnicity imply widely held consensual beliefs about social worth of individuals (Berger, Cohen, & Zelditch, 1972; Berger, Fisek, Norman, & Zelditch, 1977; Berger & Fisek, 2006). These cultural beliefs arise from countless social interactions between and among people with unequal material resources (Ridgeway & Balkwell, 1997; Ridgeway, Boyle, Kuipers, & Robinson, 1998), but they eventually spread to affect distribution of esteem and deference in a wide variety of situations, especially in task-oriented groups (Berger, Ridgeway, Fisek, & Norman, 1998; Ridgeway & Berger, 1986) but also in social situations more generally (Podolny, 1993). Since these cultural beliefs generally disadvantage minority individuals by portraying them as less competent and less worthy, social interaction guided by these beliefs generates more stress for minority individuals and may over time erode their health.

RACIAL/ETHNIC MISCLASSIFICATION FROM A STATUS PERSPECTIVE In most cases, misclassification means a vertical shift in social status.1 In principle, two types of such a status shift are possible. First, the observer may place the observed person into a higher status category compared to the category to which the observed person places herself. For example, the observer misclassifies a non-white person as white. Since whites are generally ranked higher in the status hierarchy than non-whites, the misclassified individual experiences status gain. We would expect that the misclassified person receives more respect and more positive attention compared to a situation in which her race was classified correctly. The observer in this scenario, for instance, may be more likely to engage in behaviors that express esteem, grant respect, and give access to resources. The misclassified individual’s health may benefit from such a more positive social interaction. The alternative is that the observer places the observed person into a lower status category compared to the category into which the observed person places herself, as in case of misclassifying a white person as black.

164

IRENA STEPANIKOVA

Arguably, the consequences of this type of misclassification for the interaction between the observer and the observed person are precisely the opposite of the consequences described in the first scenario. The misclassified individual experiences status loss. She may be treated with less respect, esteem, and deference and may even experience more hostility or neglect compared to what she could expect if her observer classified her racial/ethnic status correctly.

Hypotheses Since the loss of esteem inherent in racial/ethnic status loss is stressful, a higher emotional and physical symptom load can be expected among individuals experiencing racial/ethnic status loss compared to correctly classified individuals and those experiencing racial/ethnic status gain. In contrast, racial/ethnic status gain may help alleviate stress, as it places an individual into a category consensually held in higher esteem. Therefore, a lower symptom load can be expected among people experiencing status gain compared to correctly classified individuals. H1. (correct classification vs. status loss): Ceteris paribus, correctly classified individuals are less likely to suffer from physical and mental symptoms compared to individuals who experience racial/ethnic status loss (i.e., are misclassified as members of a lower racial/ethnic status category compared to the racial/ethnic category claimed for self). H2. (status gain vs. status loss): Ceteris paribus, individuals who experience racial/ethnic status gain (i.e., are misclassified as members of a higher racial/ethnic status category compared to the racial/ethnic category claimed for self) are less likely to suffer from physical and mental symptoms compared to individuals who experience racial/ethnic status loss (i.e., are misclassified as members of a lower racial/ethnic status category compared to the racial/ethnic category claimed for self). H3. (status gain vs. correct classification): Ceteris paribus, individuals who experience racial/ethnic status gain (i.e., are misclassified as members of a higher racial/ethnic status category compared to the racial/ethnic category claimed for self) are less likely to suffer from physical and mental symptoms compared to correctly classified individuals.

Applying a Status Perspective to Racial/Ethnic Misclassification

165

SPECIFYING THE RACIAL/ETHNIC STATUS HIERARCHY To know whether a misclassified individual experiences a racial/ethnic status loss or a racial/ethnic status gain, we must first outline a racial/ethnic status hierarchy in which different racial/ethnic categories can be placed. Some groups are easier to place than others. Non-Hispanics whites, in particular, are by most scholars assumed to occupy a higher rank in the racial/ethnic status hierarchy than any other racial/ethnic group. Several studies in the expectation states tradition, for instance, used this assumption when comparing non-Hispanic whites to minority individuals, including African Americans (Cohen & Roper, 1972; Riordan & Ruggiero, 1980; Webster & Driskell, 1978), East Indians (Foschi & Buchan, 1990), and Mexican Americans (Rozenholtz & Cohen, 1985). Non-white groups are more difficult to place. Studies of status, for instance, are mostly silent regarding how different non-white groups place in the racial/ethnic hierarchy relative to each other. For instance, how do blacks rank when compared to Hispanics? How do Native Americans compare to Asians? Some clues about the relative ranking of non-white groups can be harvested from studies of racial attitudes. Attitudes are linked to social status such that people tend to express more negative attitudes toward lower status groups and more positive attitudes toward higher status groups. The examination of survey evidence reveals that the most negative attitudes exist toward blacks (Devine & Elliot, 1995; Dovidio, Evans, & Tyler, 1986; Madon et al., 2001; Bobo, 2001). Bobo and Zubrinsky (1996, p. 891) poignantly summarize that ‘‘blacks receive the most negative overall ratings and whites, predictably, receive the most favorable ratings. Asians and Hispanics tend to fall in between y [T]his pattern y points to the presence of an American racial rank order, with whites consensually regarded as occupying the most preferred social position.’’ Importantly, such attitudes are not reported only by white Americans, but are prevalent also among individuals from various minority backgrounds (even though positive in-group bias may temper negative attitudes toward own minority group). Thus, minority Americans, just like their white counterparts, appear to be aware of the existence of a general racial/ethnic hierarchy and express this hierarchy in attitudinal surveys. I adopt for the purposes of this chapter a model or racial/ethnic status hierarchy in which whites place at the top, blacks place at the bottom and non-black racial/ethnic minorities stand in an intermediary position. More specifically, non-black minority groups (i.e., Native Americans,

166

IRENA STEPANIKOVA

Native Hawaiians/Pacific Islanders, Hispanics, Asians, and those that claim ‘‘other’’ race) occupy a joint rank in my model ‘‘sandwiched’’ between whites and blacks. By using a three-tier hierarchy, this model improves upon previous scholarship on race in the status tradition, which limited its focus on dichotomous comparisons between whites and a single minority group. At the same time, this model uses a potentially simplifying assumption of a joint rank of non-black minorities. In reality, a hierarchy of status is likely to exist among these groups, but the paucity of prior work makes it difficult to determine their relative placement. It is difficult to determine, for instance, whether Native Americans enjoy a higher, lower, or similar status compared to Native Hawaiians, or how their status compares to that of Hispanics. The simplified model used here can be seen as an initial step toward specifying a more comprehensive hierarchy. More detail on how this model is used for the coding of racial/ethnic misclassification variables appears in the Methods section.

METHODS Data Data come from the Behavioral Risk Factors Surveillance System (BRFSS), annual cross-sectional survey administered over telephone to a representative sample of the U.S. population 18 years of age or older living in households. The survey is conducted by Centers for Disease Control and Prevention. The goal of BRFSS is to measure health perceptions and behaviors linked to diseases and injuries in the adult population. Each year, states can add optional modules. A module on race is especially relevant for my study. It asks respondents how they classify their race/ethnicity and how their race/ ethnicity is routinely classified by others, which makes it possible to identify cases of racial/ethnic misclassification. It further asks about emotional and physical reactions to race, yielding measures used in this study as dependent variables. Several other measures of racial experience (specifically racial discrimination and racial cognitions) are used in this study as control variables. The module on race was added by Arkansas, Colorado, Delaware, Mississippi, Rhode Island, South Carolina, Wisconsin, and Washington, DC in 2004; by Rhode Island in 2007; and by Nebraska and Virginia in 2008.2 Random-digit-dial, disproportionate stratified sampling design was used to select telephone numbers. Numbers were initially divided into two strata based on the density of household numbers in each group consisting

Applying a Status Perspective to Racial/Ethnic Misclassification

167

of 100 numbers with the same area code, prefix, and first two digits of the suffix and with all possible combinations of the last two digits. Numbers from each stratum were sampled separately. The coverage ranged from 87 to 98 percent across states and was lower among residents of the south, minorities, and the poor, who are known to have lower telephone coverage. Therefore, all analyses reported here use post-stratification weights to adjust for demographic distortions due to non-response, differences in telephone coverage, and the probability of selection. Trained interviewers used computer-assisted telephone interviewing to collect data from one randomly selected adult per household. The core portion of the interview took about 10 minutes; state-added questions took additional 5–10 minutes. Median response rate ranged from 39 percent in Rhode Island to 63 percent in Colorado. Beside the availability of measures that help identify misclassification cases, a key feature of this dataset, is its large sample size (n ¼ 27,998), which ensures sizable numbers of respondents in racial/ethnic misclassification categories considered in this study. This is important, since racial/ethnic misclassification affects a small percentage of population, and the taxonomy proposed in this study further divides the misclassified individuals into several categories. A large sample is needed to ensure statistical power for conducting hypotheses tests regarding these categories. I limit my analyses to individuals who identified with a single racial category. Since individuals who identified with more than one racial category have two or more statuses within the racial/ethnic hierarchy, they could not be used for my purposes. Thus, the analytical sample consists of single-racial Hispanic and non-Hispanic whites, blacks, Asians, and American Indians/Alaska Natives, Native Hawaiians/Pacific Islanders, and those reporting ‘‘other’’ race.

Measurement I utilize two dependent variables to cover both the emotional and the physical reactions to race. For emotional symptoms in response to racial treatment, respondents were asked ‘‘During the past 30 days, have you felt emotionally upset, for example angry, sad, or frustrated, as a result of how you were treated based on your race?’’ For physical symptoms in response to racial treatment, respondents were asked, ‘‘Within the past 30 days, have you experienced any physical symptoms, for example headache, an upset stomach, tensing of your muscles, or a pounding heart, as a result of how

168

IRENA STEPANIKOVA

you were treated based on your race?’’ Response options for both questions were ‘‘yes’’ and ‘‘no.’’ Since these dependent variables express aspects of health that the respondents explicitly interpret as resulting from racial treatment, it is important to control for measures of racial experience to make sure I isolate the effects of racial/ethnic misclassification from the effects of such racial experiences. I measure perceptions of racial discrimination at work and in health care settings during the past year (‘‘Within the past 12 months at work, do you feel you were treated worse than, the same as, or better than people of other races?’’ and ‘‘Within the past 12 months when seeking healthcare work, do you feel your experiences were worse than, the same as, or better than for people of other races?’’) Respondents who selected ‘‘worse’’ were coded as 1, whereas all others were coded as 0. I also included a measure of frequency of racial cognition. Respondents were asked: ‘‘How often do you think about your race?’’ Responses were measured on a 1–7 scale ranging from ‘‘never’’ to ‘‘constantly.’’ This measure was entered as continuous in multivariate models. I also used this variable to conduct a subsample analyses separately with those who never think about their race and those who think about their race once a year or more often. The main explanatory factor is racial/ethnic misclassification. To determine whether respondents experienced it, I compared self-reported race/ethnicity to the reports of how they were routinely classified by others (henceforth referred to as observed race/ethnicity). For observed race/ ethnicity, a single question was used, ‘‘How do other people usually classify you in this country? Would you say White, Black or African American, Hispanic or Latino, Asian, Native Hawaiian or Other Pacific Islander, American Indian or Alaska Native, multiracial or some other group?’’ For self-reported race/ethnicity, separate questions were asked about ethnicity and about race. Respondents first reported their ethnicity (‘‘Are you Hispanic or Latino?’’). Next they were asked, ‘‘Which one or more of the following would you say is your race?’’ Response options included ‘‘White,’’ ‘‘Black or African American,’’ ‘‘Asian,’’ ‘‘Native Hawaiian or Other Pacific Islander,’’ ‘‘American Indian or Alaska Native,’’ and ‘‘some other group.’’ It is of note that in this survey, respondents could choose whether they were Hispanic or non-Hispanic independently of choosing their racial background. In contrast, Hispanic ethnicity was one of the categories within the observed race/ethnicity variable. This means that respondents could report that they were usually classified as Hispanic or that they were usually classified as one of the major racial groups (e.g., black, white, Native American). They could not, however, report that they were usually classified

169

Applying a Status Perspective to Racial/Ethnic Misclassification

Table 1.

Coding of Racial/Ethnic Misclassification Variables. Observed Race/Ethnicity White, Asian, AI/AN, NH/PI, Other, Hispanic Black, nonnonnonnonnonnonHispanic Hispanic Hispanic Hispanic Hispanic Hispanic

Self-reported race/ethnicity White, non-Hispanic CC Asian, non-Hispanic SG AI/AN, Non-Hispanic SG NH/PI, non-Hispanic SG Other, non-Hispanic SG Hispanic SG Black, non-Hispanic SG

SL CC OM OM OM OM SG

SL OM CC OM OM OM SG

SL OM OM CC OM OM SG

SL OM OM OM CC OM SG

SL OM OM OM OM CC SG

SL SL SL SL SL SL CC

Notes: CC ¼ correct classification; SL ¼ status loss; SG ¼ status gain; OM ¼ other misclassification; AI/AN ¼ American Indian/Alaska Native. NH/PI ¼ Native Hawaiian/Pacific Islander.

as Hispanic black, Hispanic white, Hispanic Native American, etc. Given this important difference between the self-reported and observed race indicators, it was necessary to create a single variable for self-reported race/ ethnicity with similar categories to those in the observed race/ethnicity variable, since identification of misclassification cases required comparison of self-reported and observed race/ethnicity. This new variable grouped all individuals who self-identified as Hispanic into one category but separated non-Hispanic individuals into unique categories based on their self-reported race. Thus, the Hispanic category in the self-reported race/ethnicity variable includes individuals of any race. Table 1 summarizes the coding of racial/ethnic misidentification variables. Individuals whose self-reported race/ethnicity matched their observed race/ ethnicity were coded as correctly classified. The remaining cases were assigned into categories representing status loss, status gain, and other type of misclassification according to the model of racial/ethnic hierarchy outlined earlier in this chapter. The status gain category contained individuals whose observed racial/ethnic status was higher compared to their self-reported racial/ethnic status, including Asians, American Indians/Alaska Natives, Native Hawaiians/Pacific Islanders, Hispanics, Blacks, and ‘‘others’’ misclassified as whites. The status loss category contained individuals whose observed racial/ethnic status was lower compared to their self-reported racial/ethnic status. It included whites misclassified as Asians, American Indians/Alaska Natives, Native Hawaiians/Pacific Islanders, Hispanics, blacks, or others; and Asians, American Indian/Alaska Natives, Native

170

IRENA STEPANIKOVA

Hawaiians/Pacific Islanders, Hispanics, and ‘‘others’’ misclassified as blacks. The remaining cases of misclassification were coded as ‘‘other misclassification.’’ They included persons who self-identified as American Indian/Alaska Natives, Native Hawaiians/Pacific Islanders, Hispanics, and ‘‘others’’ but who were misclassified as another non-black minority. It remained undetermined whether these individuals gained status, lost status, or whether their status remained the same. In addition to factors representing racial experiences that were described earlier (specifically, discrimination and racial cognitions), the models controlled for factors influencing health in general. I included several sociodemographic factors known to play a role in various health outcomes, including gender, age (in years), and the highest school grade completed (‘‘less then high school,’’ ‘‘high school graduate,’’ ‘‘some college,’’ and ‘‘college graduate or higher’’). Annual household income from all sources was measured in eight categories. Since the categories differed each year, it was not feasible to recode income into dollars. Instead, the variable was treated as an eightpoint scale. Healthcare coverage was included to account for the possibility that poor access to health care might lead to untreated health conditions, such as high blood pressure or depression, which may exacerbate subjective experience of symptoms in response to racial treatment, such as headaches or anxiety. It was measured by a yes/no question, ‘‘Do you have any kind of health care coverage, including health insurance, prepaid plans such as HMOs, or government plans such as Medicare?’’ Region as defined by the U.S. Census Bureau had four categories: midwest, northeast, south, and west. I also controlled for the year of the interview.

Analytic Strategy After estimating univariate and bivariate statistics, multivariate logistic regression models of physical and emotional symptoms were estimated. The models corrected for complex sampling design to ensure representativeness. Treating correct classification category as a reference group facilitated the evaluation of Hypotheses 1 and 3, which compare status loss and status gain categories to correct classification. Further tests comparing status loss to status gain were conducted to evaluate Hypothesis 2. For completeness, all models included other misclassification category, which collected cases of misclassification without clear status implications. The models were estimated first for the full sample. Next, to explore the possibility that racial cognitions, measured by frequency of thinking about race, play a role

Applying a Status Perspective to Racial/Ethnic Misclassification

171

in responses to misclassification, models were estimated separately for the first subsample containing respondents who reported that they never thought about race and for the second subsample including individuals who reported thinking about race anywhere from ‘‘at least once a year’’ to ‘‘constantly.’’ One-tailed tests were applied to test directional hypotheses; two-tailed tests we used for control variables.

RESULTS Table 2 reports characteristics of the sample. The sample consisted primarily of whites, males, and residents of southern states. Blacks and Hispanics constituted two larger minority groups, accounting for 11 and 6 percent of respondents, respectively. Given that the sample was predominately white, the reports of emotional and physical symptoms in response to racial treatment were surprisingly common: 6 percent of respondents reported emotional symptoms and 3 percent reported physical symptoms. Racial/ ethnic misclassification was rare in the general sample (about 4%), mainly because of low misclassification rates among whites, but it varied widely by racial/ethnic groups. American Indians/Alaska Natives and Hispanics commonly experienced status gain (47% and 26%, respectively). Native Hawaiians/Pacific Islanders had high rates of both status gain (32%) and status loss (38%). By contrast, status gain and status loss was rarely observed among blacks and whites. Bivariate comparisons (not shown) suggest increased vulnerability of misclassified individuals to physical and emotional symptoms. These symptoms were about twice as common in the group that experienced misclassification compared to the correctly classified individuals. Status loss was especially closely associated with symptoms, with as many as 29 percent of respondents in this category reporting emotional symptoms and 11 percent of reporting physical symptoms. Results of multivariate logistic regression models appear in Tables 3 and 4. Table 3 shows models of emotional symptoms. The first column, which reports results for the whole sample, reveals that individuals in the status loss category, who are routinely misclassified as belonging to a racial/ethnic group with lower status than the group they claim for themselves, have higher likelihood of reporting emotional symptoms compared to correctly classified individuals. When coefficients are expressed as odds ratios, we observe that the odds of reporting emotional symptoms increase by 2.8 times in the status loss group. This result yields initial support for Hypothesis 1, but further

172

IRENA STEPANIKOVA

Table 2.

Characteristics of the Sample.

Physical symptoms resulting from racial treatment (0,1) Emotional symptoms resulting from racial treatment (0,1) Racial/ethnic classification by observers (0,1) Correct classification Status loss Status gain Other misclassification Race/ethnicity (0,1) Black, non-Hispanic NH/PI, non-Hispanic Asian, non-Hispanic AI/AN, non-Hispanic White, non-Hispanic Other, non-Hispanic Hispanic Female (0,1) Annual household income (1–8) Highest grade completed (0,1) Less than high school High school graduate Some college College graduate Healthcare coverage (0,1) Age in years (18–99) Region (0,1) Northeast Midwest West South Survey year (0,1) 2004 2007 2008 Racial discrimination at work (0,1) Racial discrimination in health care (0,1) Racial cognition: Frequency (1–7)

Mean

Standard Error

.030 .064

.002 .004

.959 .009 .027 .005

.003 .002 .002 .001

.107 .003 .021 .009 .799 .006 .056 .458 6.308

.003 .001 .006 .001 .007 .001 .003 .006 .021

.058 .276 .264 .402 .873 4.551

.003 .005 .006 .006 .004 .167

.084 .243 .147 .527

.002 .005 .004 .006

.661 .030 .309 .057 .029 2.158

.007 .001 .008 .003 .002 .020

Sources: Behavioral Risk Factor Surveillance System 2004, 2007, and 2008. Notes: N ¼ 27,998. Means are weighted and corrected for survey design. AI/AN ¼ American Indian/Alaska Native. NH/PI ¼ Native Hawaiian/Pacific Islander.

.472 1.545 .396 .284 3.664 .019 .401 .178 .378 .161 .465 .089 .001

.351 .298 .632 .164 .657 .628 .478 .911 .222 .134 .037 .260 .156 .172 .170 .005

.442 2.190 1.573 .561 .511 .488 .450 .050 .238 .180 .021 .128 .013

.827 .709 1.658

Unstandardized coefficient

1.049www .308 .533

Standard error

.389 .272 .275 .275 .008

.298 .994 .978 .608 1.266 .327 .194 .054

.799 .468 .886

Standard error

(N ¼ 16,065)

(N ¼ 27,998)

Unstandardized coefficient

Never Thinks about Race

Full Sample

.295 .143 .033 .136 .019

.484 2.361 1.639 .610 .943 .577 .419 .020

1.101www .608w .792

Unstandardized coefficient

.300 .184 .173 .198 .006

.179 .665 .643 .526 .998 .233 .157 .042

.378 .323 .703

Standard error

(N ¼ 11,933)

Thinks about Race at Least Once a Year

Logistic Regression Models of Emotional Symptoms Resulting from Racial Treatment, Comparing Status Loss and Status Gain to Correct Classification.

Racial/ethnic misclassificationa Status loss Status gain Other misclassification Race/ethnicityb Black, non-Hispanic Asian, non-Hispanic NH/PI, non-Hispanic AI/AN, non-Hispanic Other, non-Hispanic Hispanic Female Annual household income Highest grade completedc Less than high school High school graduate Some college Healthcare coverage Age in years

Table 3.

Applying a Status Perspective to Racial/Ethnic Misclassification 173

.288 .195 3.270 2.028

.268 .177 .129 .173 .030 .346

.184 .093 2.772 1.869

.367 3.743 – 3.762

.281 .647 .188

.166 .179 .181

.228 .155 .211

Unstandardized coefficient

– .558

.479 .300 .228 .310

.276 .342 .317

Standard error

.289 3.294

.343 .021 2.552 1.735

.173 .037 .167

Unstandardized coefficient

.038 .412

.315 .207 .146 .200

.191 .207 .206

Standard error

(N ¼ 11,933)

Thinks about Race at Least Once a Year

Sources: Behavioral Risk Factor Surveillance System 2004, 2007, 2008. Notes: All estimates are weighted and corrected for survey design. Taylor linearized variance estimation is used. AI/AN ¼ American Indian/ Alaska Native. NH/PI ¼ Native Hawaiian/Pacific Islander. w po.05; wwwpo.001 (one-tailed test). po.05; po.01; po.001 (two-tailed test). a Reference category is correct classification. b Reference category is white, non-Hispanic. c Reference category is college graduate. d Reference category is south. e Reference category is 2004.

Regiond Northeast Midwest West Survey yeare 2007 2008 Racial discrimination at work Racial discrimination in health care Racial cognition: Frequency Constant

Standard error

(N ¼ 16,065)

(N ¼ 27,998)

Unstandardized coefficient

Never Thinks about Race

Full Sample

Table 3. (Continued )

174 IRENA STEPANIKOVA

.300 .179 .185 .175 .006 .203 .237 .223

.363 .099 .270

.183 .694 1.338 .384 .708 .277 .161 .045

.445 1.971 4.543 .631 1.637 .455 .623 .164 .413 .121 .211 .140 .005

.397 .326 .702

1.199www .344 .047

Standard Error

.314 .615 .416

.416 .310 .367 .336 .012

.826 .191 .383 .257 .001 .273 .154 .282

.335 1.281 1.414 .812 1.355 .381 .248 .078

.743 .661 1.338

Standard Error

.735 1.673 1.845 .490 2.488 .434 .266 .239

1.293w .131 1.023

Unstandardized Coefficient

(N ¼ 16,065)

(N ¼ 27,998)

Unstandardized Coefficient

Never Thinks about Race

Full Sample

.392 .078 .281

.307 .178 .273 .244 .006

.598 2.465 .506 .709 .148

f

.364 2.222

1.213www .411 .181

Unstandardized Coefficient

.256 .236 .255

.358 .217 .200 .196 .006

.414 .887 .293 .183 .052

.211 .713

.419 .346 .742

Standard Error

(N ¼ 11,933)

Thinks About Race at Least Once a Year

Logistic Regression Models of Physical Symptoms Resulting from Racial Treatment, Comparing Status Loss and Status Gain to Correct Classification.

Racial/ethnic misclassificationa Status loss Status gain Other misclassification Race/ethnicityb Black, non-Hispanic Asian, non-Hispanic NH/PI, non-Hispanic AI/AN, non-Hispanic Other, non-Hispanic Hispanic Female Annual household income Highest grade completedc Less than high school High school graduate Some college Healthcare coverage Age in years Regiond Northeast Midwest West

Table 4. Applying a Status Perspective to Racial/Ethnic Misclassification 175

.270 .214 .211 .031 .403 .346

.626 .118 1.216 .259

4.414 3.743

4.216 3.762

1.035 .198 1.437 –

Unstandardized Coefficient

.665 .558

.442 .339 .372 –

Standard Error

4.203 3.294

.431 .243 1.149 .220

Unstandardized Coefficient

.465 .412

.340 .264 .242 .044

Standard Error

(N ¼ 11,933)

Thinks About Race at Least Once a Year

Sources: Behavioral Risk Factor Surveillance System 2004, 2007, 2008. Notes: All estimates are weighted and corrected for survey design. Taylor linearized variance estimation is used. AI/AN ¼ American Indian/ Alaska Native. NH/PI ¼ Native Hawaiian/Pacific Islander. w po.05; wwwpo.001 (one-tailed test). po.05; po.01; po.001 (two-tailed test). a Reference category is correct classification b Reference category is white, non-Hispanic. c Reference category is college graduate. d Reference category is south. e Reference category is 2004. f No symptoms were reported by any NH/PI who think about their race at least once a year. Therefore, it was not possible to estimate a model with NH/PI. NH/PI are collapsed with ‘‘other’’ for the purposes of this model.

Survey Yeare 2007 2008 Racial discrimination at work Racial discrimination in health care Racial cognition: Frequency Constant

Standard Error

(N ¼ 16,065)

(N ¼ 27,998)

Unstandardized Coefficient

Never Thinks about Race

Full Sample

Table 4. (Continued )

176 IRENA STEPANIKOVA

177

Applying a Status Perspective to Racial/Ethnic Misclassification

analysis revels that the effect of status loss on emotional symptoms is limited to individuals who think about race once a year or more often. Among those who never think about their race, status loss is not related to emotional symptoms. In contrast, in Table 4, physical symptoms are linked to status loss both when the racial cognitions are present and when they are absent. In a model estimated with the full sample, the odds of reporting physical symptoms are 3.3 times higher in the status loss group compared to the correctly classified group. Results of tests evaluating Hypothesis 2 are displayed in Table 5. The coefficients come from models with an identical set of independent variables as models shown in Tables 3 and 4, except that status loss serves as a reference category to allow a comparison with status gain. This means that the coefficients shown in Table 5 can be interpreted as the differences in the likelihood of emotional and physical symptoms associated with status gain versus status loss. In models estimated for the full sample, individuals experiencing status loss had 3.8 times higher odds of emotional symptoms and 2.3 times higher odds of physical symptoms compared to individuals experiencing status gain. The analyses by subsample yielded significant effects of status gain only for those individuals who think about race at least once a year, limiting the support for Hypothesis 2. Not surprisingly, the sizes of these Table 5. Logistic Regression Models of Emotional and Physical Symptoms Resulting from Racial Treatment, Comparing Status Gain to Status Loss. Emotional Symptoms

Full sample (N ¼ 27,998)

Physical Symptoms

Unstandardized Coefficient

Standard Error

Unstandardized Coefficient

Standard Error

1.346www

.351

.844w

.460

Never thinks about race (N ¼ 16,065) .086

.861

Thinks about race at least once a year (N ¼ 11,933) 1.697www

.467

1.418 .819w

.944 .495

Sources: Behavioral Risk Factor Surveillance System 2004, 2007, and 2008. Notes: Except for the reference category for misclassification variables, the models were identical to those shown in Tables 3 and 4. All estimates are weighted and corrected for survey design. Taylor linearized variance estimation is used. w po.05; wwwpo.001 (one-tailed test).

178

IRENA STEPANIKOVA

effects were somewhat larger compared to the effects evident in the full sample (OR for emotional symptoms ¼ 5.5; OR for physical symptoms ¼ 2.7). Finally, Hypothesis 3 compared status gain to correct classification. Table 3 shows that individuals experiencing status gain have lower likelihood of emotional symptoms if they think about their race at least once a year. Support for Hypothesis 3 is not obtained for physical symptoms (Table 4). The examination of the effects of control variables reveals several interesting patterns. The most consistent effects across the models are found for the variables that measure aspects of racial experiences such as racial discrimination and racial cognitions. As we might expect, individuals who think about their race more often are more likely to suffer from physical and emotional symptoms in response to racial treatment. Similarly, the likelihood of suffering from emotional and physical symptoms increases among those who report having experienced discrimination at work or when seeking health care. The effects of racial discrimination extend to all respondents, regardless of their level of racial cognitions. This finding is interesting, as it suggests that racial discrimination is linked to compromised well-being even among those who do not think about their race. In contrast, the effects of several socio-demographic factors are evident only when racial cognitions are present. For instance, younger people and several minority groups, most notably non-Hispanics Blacks, Asians, Native Hawaiians/Pacific Islanders, and Hispanics, show an increase in the likelihood of emotional symptoms in response to racial treatment only if they think about their race at least once a year.

DISCUSSION The status perspective used in this study helped to outline a taxonomy of racial/ethnic misclassification, which was then applied to understand variation in emotional and physical symptoms in response to racial treatment. The results revealed that the type of misclassification clearly matters when we consider these symptoms. Individuals who experienced a loss of status as a result of their misclassification (and who thought about their race) were more vulnerable to physical and emotional symptoms, while such an increased vulnerability was not observed among those experiencing a status gain. These findings suggest that misclassification per se does not necessarily harm aspects of well-being studied here, but the loss of social status inherent in some types of misclassification does, especially for respondents who think about their race.

Applying a Status Perspective to Racial/Ethnic Misclassification

179

It is important to consider alternative explanations for these findings. One such alternative is offered by the identity perspective. From this perspective, being misclassified by others means a threat to self-identity. Misclassification may be connected to increased stress levels by preventing self-verification (Cast & Burke, 2002) and disrupting shared understanding of social meanings, especially as they pertain to role expectations (Troyer & Younts, 1997; Troyer, Younts, & Kalkhoff, 2001). As explained by Burke and Stets (1999), we constantly strive to verify our identities in interpersonal interaction. This process, known as self-verification, has important implications for emotional health. If efforts to self-verify are frustrated, we try to modify our behavior or cognitions to achieve a better match, but if self-verification is still denied, lower self-esteem and negative emotions invariably ensue. An actor who has been misclassified in terms of her race, for instance, may experience negative emotions when her interaction partners treat her in an unexpected manner or when she notices that her partners find her behavior unexpected. If, as identity perspective suggests, all cases of misclassification pose a threat to self-identity, and if threatened self-identity poses a risk to health, then we might expect that status gain harms health just as much as status loss does. The results presented in this study fail to support such a conclusion. On the contrary, this study found negative effects of status loss and no effect or positive effect of status gain. These results are more consistent with the arguments stressing status implications of race/ethnicity. Such arguments have divergent implications for individuals experiencing status loss and status gain, since misclassification involving status loss leads to increased stress and more physical and mental symptoms, while misclassification involving status gain leads to decreased stress and fewer physical and mental symptoms. Thus, it does not appear that identity processes can be a sole explanation for the findings reported here. However, it is possible that both status and identity processes play a role in misclassification. In fact, such a simultaneous influence of status and identity–based mechanisms may explain why limited support was obtained for the hypothesis that status gain is linked to better health outcomes. Threat to self-identity may indeed lead to an increased symptom load in cases of status gain, but this effect may be cancelled out by status processes that pull in the opposite direction. Being granted higher esteem and influence leads to positive emotions, decreases stress, and ultimately has positive effects on health and well-being. At the same time, being misclassified can independently generate stress, as arguments stressing self-verification predict. If such countervailing processes are in place, they could produce results similar to those shown in this study;

180

IRENA STEPANIKOVA

specifically, little difference between correctly classified individuals and those who experience status gain. In case of status loss, identity and status processes work in concert, reinforcing each other. The stress levels in the status loss category should increase because of self-identity threats and because of lost esteem and influence. Since both the lack of self-verification and the loss of esteem are stressful, being misclassified as belonging to a lower status racial/ ethnic category implies a considerable threat to well-being, as seen in the results reported earlier. Further research is needed to disentangle statusbased and identity-based processes in misclassification and to elucidate their relationship. The results concerning racial cognition are of note. Racial cognitions play a role in linking misclassification to emotional symptoms. Individuals who cognitively engage their racial/ethnic status may be especially vulnerable to negative emotions if they experience a status loss. Those who do not think about their race, on the contrary, may be more resistant to such emotions even when exposed to this potentially harmful type of misclassification. In case of physical symptoms, thinking about race matters less. Status loss is linked to higher likelihood of physical symptoms regardless of the presence of racial cognition. This suggests that status loss may put people at risk physically (though not emotionally) even when it is not cognitively processed. This study has several limitations. The data represented the populations of selected states from the southern, northeastern, Midwestern, and western region. Southern and northeastern states were in majority, while western and midwestern regions were each represented by a single state each, specifically Colorado and Wisconsin. It would have been preferable to have fuller representation of states, especially from the west and midwest. Even though there appears no theoretical reason to believe that the processes observed in this study are specific only to the states included, we cannot conclude that they apply to other states or to the general American population without data representing it. The treatment of Hispanic ethnicity in the observed race/ethnicity measure deserves some discussion. Hispanic ethnicity was not treated as orthogonal to race but constituted only one of the categories of the observed race/ethnicity measure. Therefore, it was not possible to fully separate out the effects of ethnic vs. racial misclassification, especially for Hispanics. Since Hispanics are an ethnic group composed of multiple races, this measurement limitation may have lead to a higher degree of error among Hispanics than among non-Hispanic groups. If information on self-reported and observed race and ethnicity (orthogonally treated) becomes available in future, it will be possible

Applying a Status Perspective to Racial/Ethnic Misclassification

181

to evaluate the relative contribution of racial vs. ethnic misclassification to health-related outcomes for Hispanics individuals more fully than it was possible in this study. Such evaluation is important (see Campbell & Rogalin, 2006), since Hispanics have recently outgrown all non-Hispanic minority groups in the United States and their growth is expected to continue in the future. In conclusion, it is important to note that the landscape of racial and ethnic relations in the United States has been changing rapidly in recent years. Racial/ethnic identities, especially those of non-blacks, are becoming more multidimensional and fluid than ever before and vary across contexts and observers (Harris & Sim, 2002; Lee & Bean, 2004). The numbers of interracial marriages have surged, leading demographers to expect a continuous growth of bi-racial and multiracial population (Gullickson, 2006; Lee & Bean, 2004). It would not be surprising to see racial/ethnic misclassification become more prevalent as these trends continue. Future research on its implications may prove an important component of more accurate understanding of evolving patterns of racial/ethnic relations.

NOTES 1. Misclassification cases that imply a horizontal shift between categories with a similar placement in the status hierarchy also exist, but are not theoretically considered here. In my analysis, these cases are identified as ‘‘other misclassification.’’ 2. 2005 and 2006 surveys were not used since they did not contain all the questions needed for the analysis.

ACKNOWLEDGMENTS I thank Brent Simpson, Shane Thye, Michael Lovaglia, Mary Campbell, and Brian Powell for their help in developing this manuscript.

REFERENCES Berger, J., Cohen, B. P., & Zelditch, M., Jr. (1972). Status characteristics and social interaction. American Sociological Review, 37, 241–255. Berger, J., & Fisek, M. H. (2006). Diffuse status characteristics and the spread of status value: A formal theory. The American Journal of Sociology, 111, 1038–1079.

182

IRENA STEPANIKOVA

Berger, J., Fisek, M. H., Norman, R. Z., & Zelditch, M., Jr. (1977). Status characteristics and social interaction. New York: Elsevier. Berger, J., Ridgeway, C. L., Fisek, M. H., & Norman, R. Z. (1998). The legitimation and delegitimation of power and prestige orders. American Sociological Review, 63, 379–405. Bird, S. T., Bogart, L. M., & Delahanty, D. L. (2004). Health-related correlates of perceived discrimination in HIV care. AIDS Patient Care STDS, 18, 19–26. Bobo, L. (2001). Racial attitudes and relations at the close of the twentieth century. In: N. Smelser, W. J. Wilson & F. Mitchell (Eds), America becoming: Racial trends and their consequences (pp. 262–299). Washington, DC: National Academy Press. Bobo, L., & Zubrinsky, C. L. (1996). Attitudes on residential integration: Perceived status differences, mere in-group preference, or racial prejudice? Social Forces, 74, 883–909. Bobo, L. D., & Fox, C. (2003). Race, racism, and discrimination: Bridging problems, methods, and theory in social psychological research. Social Psychology Quarterly, 66(4), 319–332. Borrell, L. N., Kiefe, C. I., Williams, D. R., Diez-Roux, A. V., & Gordon-Larsen, P. (2006). Self-reported health perceived racial discrimination, and skin color in African Americans in the CARDIA study. Social Science and Medicine, 63, 1415–1427. Bourg, M. C. (2002). Gender mistakes and inequality. Ph.D. dissertation, Department of Sociology, Stanford University, Stanford, CA. Burke, P. J., & Stets, J. E. (1999). Trust and commitment through self-verification. Social Psychology Quarterly, 62, 347–366. Campbell, M. E., & Rogalin, C. L. (2006). Categorical imperatives: The interaction of Latino and racial identification. Social Science Quarterly, 87, 1030–1052. Campbell, M. E., & Troyer, L. (2007). The implications of racial misclassification by observers. American Sociological Review, 72, 750–765. Cast, A. D., & Burke, P. J. (2002). A theory of self-esteem. Social Forces, 80, 1041–1068. Cohen, E. G., & Roper, S. S. (1972). Modification of interracial interaction disability: An application of status characteristic theory. American Sociological Review, 37, 643–657. Devine, P. G., & Elliot, A. J. (1995). Are racial stereotypes really fading? The Princeton trilogy revisited. Personality and Social Psychology Bulletin, 21, 1139–1150. Dovidio, J. F., Evans, N., & Tyler, R. B. (1986). Racial stereotypes: The contents of their cognitive representations. Journal of Experimental Social Psychology, 22, 22–37. Ferber, A. L. (1995). Exploring the social construction of race. In: N. Zack (Ed.), American mixed race: The culture of microdiversity (pp. 155–167). Lanham, MD: Rowman and Littlefield. Foschi, M., & Buchan, S. (1990). Ethnicity, gender, and perceptions of task competence. Canadian Journal of Sociology/Cahiers canadiens de sociologie, 15, 1–18. Gullickson, A. (2006). Black/White interracial marriage trends, 1850–2000. Journal of Family History, 31(3), 289–312. Harris, D. R., & Sim, J. J. (2002). Who is multiracial? Assessing the complexity of lived race. American Sociological Review, 67, 614–627. Kessler, R. C., Mickelson, K. D., & Williams, D. R. (1999). The prevalence, distribution, and mental health correlates of perceived discrimination in the United States. Journal of Health and Social Behavior, 40, 208–230. Lee, J., & Bean, F. D. (2004). America’s changing color lines: Immigration, race/ethnicity, and multiracial identification. Annual Review of Sociology, 30, 221–242. Madon, S., Guyll, M., Aboufadel, K., Montiel, E., Smith, A., Palumbo, P., & Jussim, L. (2001). Ethnic and national stereotypes: The Princeton trilogy revisited and revised. Personality and Social Psychology Bulletin, 27, 996–1010.

Applying a Status Perspective to Racial/Ethnic Misclassification

183

Marmot, M. G., Shipley, M. J., & Rose, G. (1984). Inequalities in death-specific explanations of a general pattern? Lancet, 1, 1003–1006. Marmot, M. G., Smith, G. D., Stansfeld, S., Patel, C., North, F., Head, J., White, I., Brunner, E., & Feeney, A. (1991). Health inequalities among British civil servants: The Whitehall II study. Lancet, 337, 1387–1393. Podolny, J. M. (1993). A status-based model of market competition. The American Journal of Sociology, 98, 829–872. Ridgeway, C. L., & Balkwell, J. W. (1997). Group processes and the diffusion of status beliefs. Social Psychology Quarterly, 60, 14–31. Ridgeway, C. L., & Berger, J. (1986). Expectations, legitimation, and dominance behavior in task groups. American Sociological Review, 51, 603–617. Ridgeway, C. L., Boyle, E. H., Kuipers, K. J., & Robinson, D. T. (1998). How do status beliefs develop? The role of resources and interactional experience. American Sociological Review, 63, 331–350. Riordan, C., & Ruggiero, J. (1980). Producing equal-status interracial interaction: A replication. Social Psychology Quarterly, 43, 131–136. Rockquemore, K. A., & Brunsma, D. L. (2002). Beyond black: Biracial identity in America. Thousand Oaks, CA: Sage. Rozenholtz, S. J., & Cohen, E. G. (1985). Activating ethnic status. In: J. Berger & M. Z. Josey-Bass, Jr. (Eds), Status, rewards, and influence: How expectations organize behavior (pp. 430–44). San Francisco: Jossey-Bass Publishers. Saperstein, A. (2006). Double-checking the race box: Examining inconsistency between survey measures of observed and self-reported race. Social Forces, 85, 57–74. Troyer, L., & Younts, C. W. (1997). Whose expectations matter? The relative power of first-and second-order expectations in determining social influence. The American Journal of Sociology, 103, 692–732. Troyer, L., Younts, C. W., & Kalkhoff, W. (2001). Clarifying the theory of second-order expectations: The correspondence between motives for interaction and actors’ orientation toward group interaction. Social Psychology Quarterly, 64, 128–145. Webster, M., & Driskell, J. E. (1978). Status generalization: A review and some new data. American Sociological Review, 43, 220–236.

COMPARISON PROCESSES IN SOCIAL EXCHANGE NETWORKS David R. Schaefer and Olga Kornienko ABSTRACT The comparison processes introduced by Thibaut and Kelly (1959) are fundamental to social exchange theories of power. However, research has focused almost exclusively on only one type of comparison – the comparison between alternative sources of valued rewards (CLalt) – which affects relationship commitment. Thibaut and Kelley also articulated a more general comparison level (CL) that determines relationship satisfaction. We propose that in exchange settings where relationships are not interdependent, the network structure can affect an actor’s CL, with subsequent effects on power use. Results of a laboratory experiment offer initial support for this hypothesis and call for greater research on comparison processes within exchange networks. Keywords: Social exchange; power; comparison level. Comparisons of one form or another are fundamental to theories of group processes and social psychology more generally. These take the form of social comparisons of one’s abilities with others’ (Festinger, 1954), comparisons between groups that shape social identity (Tajfel & Turner, 1979), and evaluations of rewards through referential structures (Berger, Advances in Group Processes, Volume 27, 185–204 Copyright r 2010 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0882-6145/doi:10.1108/S0882-6145 (2010)0000027010

185

186

DAVID R. SCHAEFER AND OLGA KORNIENKO

Zelditch, Anderson, & Cohen, 1972) or comparisons of one’s actual holding to a comparison holding (Jasso, 2006) that form the basis for justice evaluations. Comparison processes are no less important to the foundations of social exchange theory. In a rare point of convergence, power-dependence (P-D) and network exchange theory draw on Thibaut and Kelley (1959) to assert that the comparison between alternative exchange partners is a fundamental basis for power imbalances in exchange networks (Emerson, 1972; Lovaglia, Skvoretz, Willer, & Markovsky, 1995). Dozens of studies have demonstrated that actors with better alternatives for the resources they value have a power advantage over partners with poorer alternatives (see van Assen (2003) for a review). However, in Thibaut and Kelley’s original formulation, alternatives were only one component of the comparison process. Beyond the comparison between alternatives (CLalt), there is the internal comparison level (CL) that is used to compare rewards with one’s expectations for rewards. Although Thibaut and Kelley argued that both CL and CLalt are important to understanding behavior within relationships, social exchange theory has concentrated almost exclusively on the effects of CLalt (for an exception, see Molm, 1991). For instance, the research on power in negatively connected networks that has occupied social exchange investigations for the past 30 years is primarily concerned with the effects of alternatives. Yet, in many types of exchange structures actors do not have alternative exchange partners. In such settings, we propose that the CL maintains relevance for exchange outcomes such as power. Moreover, the operation of the CL allows exchange outcomes to affect one another across relationships that are otherwise independent. Thus, the structure of the network can systematically alter actors’ CLs, and hence their behavior during exchange, with consequences for the distribution of power in the network. This chapter considers how differences in network structure can affect exchange outcomes through comparison processes. In the following pages we begin by providing background on social exchange theory with an emphasis on how power develops in different types of exchange structures. We then review the concept of CL as articulated by Thibaut and Kelley (1959). An implication of the theory is that power imbalances can occur where no previously established structural basis for power exists. We develop a laboratory experiment that tests this assertion by manipulating participants’ CLs through network structure. Our analysis reveals the expected effects of network manipulations of CL on power use. We discuss implications of these findings for other types of exchange structures and outcomes.

Comparison Processes and Social Exchange

187

THEORETICAL BACKGROUND Social exchange theories have long been interested in how mutual dependence among actors can have both integrative and harmful effects on relationships (Blau, 1964; Homans, 1961). The outcome that has gained the most empirical attention has been power. Power is a relational outcome that can be differentiated into its potential, or structural power, and its occurrence during exchange, known as power use (Cook & Emerson, 1978). According to Emerson (1972), structural power is a function of actors’ dependence on one another for valued resources. One component of dependence is value – the more A values B’s resource, the more A depends on B (DAB), and the greater is B’s structural power advantage over A (PAB). When dependence is equal (DAB ¼ DBA) then structural power is balanced (PAB ¼ PBA). However, when an actor is more dependent on the relationship than the partner (i.e., DABWDBA), then that actor faces a structural power disadvantage (i.e., PABoPBA). Disadvantaged actors are more motivated to exchange and will sacrifice more to ensure its completion. Consequently, actors with a structural power advantage are able to use power over their disadvantaged partner during exchange and experience greater rewards at the partner’s expense. A defining point in the development of social exchange theory was Emerson’s (1972) proposition that exchange relationships are best understood not at the dyadic level, as had been the focus of prior theorizing, but by considering them within the context of the broader network of exchange relations. The nature of the interdependence between relations has implications for dependence within them. Thus, because dependence is the basis of power, networks have the capacity to affect power within a relationship. According to Emerson (1972), networks exist when relationships are interdependent, or, in his terms, connected to one another. The type of connection that exists between two relations is determined by the resources controlled by one’s partners. Resources can belong to the same or different domains. Two resources are in the same domain when the acquisition of one decreases the value of the other due to satiation (Emerson, 1972). Thus, resources in the same domain are substitutable for one another (Yamaguchi, 1996). When one’s partners control resources in the same domain, then the connection between the relations is negative. Exchange in one relationship decreases the frequency of magnitude of exchange in the other (Emerson, 1972). Other times, one’s partners may control resources in different domains, but given the nature of the resources, exchange in one relationship increases the frequency or magnitude of

188

DAVID R. SCHAEFER AND OLGA KORNIENKO

exchange in the other. In that case the connection between the two relations is positive (Emerson, 1972). Lastly, one’s partners may control resources in distinct domains, where the acquisition of one has no effect on the value of the other. If exchange in one relation is independent of exchange in another relation, the two relations are not connected. This is the case even though the same actor belongs to both. Connection types are important because they determine the nature of dependence, whether structural power imbalances exist and, if so, the mechanism responsible for power use. In negative connections, power use occurs through exclusion when one actor has better alternative sources for resources than the other (Markovsky, Willer, & Patton, 1988). Better alternatives reduce an actor’s dependence on the partner. For instance, if A can exchange with either B1 or B2, but not both, and the Bs do not have other partners, then B1 and B2 are completely dependent on A. A’s dependence on any one B is lower because A can obtain resources from either B1 or B2. To avoid being excluded from exchange, the Bs must try to outbid one another for exchange with A, which gives A a power advantage. Better alternatives can take the form of higher value alternatives (Molm, Peterson, & Takahashi, 2001; Bonacich & Friedkin, 1998) or more available alternatives (Cook, Emerson, Gillmore, & Yamagishi, 1983; Markovsky et al., 1988). By contrast, in positive connections, actors do not have alternatives. Rather, power imbalances occur because one actor values exchange more than the other. For example, consider the student (S) who must complete an internship with I to secure employment with company J. The contingency of exchanges creates a power imbalance through ordering in I–S, the first exchange that must be completed (Corra & Willer, 2002). Because the internship with I is a prerequisite for the job with J, the student is highly dependent on I. All else being equal, exchange in the I–S relationship is more valuable to S, which gives I a power advantage. S is more dependent and will make greater sacrifices to ensure that her internship responsibilities are completed, opening the door to the job she covets with J. Observe that positive connections lead to power imbalances even though the value of alternative partners is the same for advantaged and disadvantaged actors – in this case zero. Power is imbalanced because the value of completing the exchange is greater for S because it facilitates a second exchange. Emerson’s insight to the importance of networks for exchange outcomes sparked decades of research aimed at uncovering the basis of power in social exchange networks and refining methods for predicting its use. Such research has explored differences across network structures

Comparison Processes and Social Exchange

189

(see Van Assen, 2003), forms of exchange, such as negotiated versus reciprocal (Lawler, Thye, & Yoon, 2008; Molm, Collett, & Schaefer, 2007), and types of exchange resources (Schaefer, 2009). In addition, alternative conceptualizations of connection types have emerged (Patton & Willer, 1990). However, throughout this research, the focus has remained on interdependent, connected relations as the only context where structurally induced power can emerge. Social exchange theories agree that when relations are not connected, or the connection type is null, there is no basis for structural power imbalance (Emerson, 1972; Patton & Willer, 1990). The current research challenges this assumption by returning to the roots of social exchange theory and Thibaut and Kelley’s (1959) concept of CLs.

Overview of Comparison Levels Thibaut and Kelley (1959) introduced comparison levels to explain how individuals evaluate their relationships. Their dual conception distinguishes a general CL from the comparison level for alternative relationships (CLalt). The CL is defined as an individual’s expectation regarding the outcome the individual feels he or she deserves. The CL is a general standard that is applicable to multiple relationships. Individuals compare their experienced outcomes to their expected outcomes, embodied in the CL, to determine how attractive or satisfactory the relationship is (Thibaut & Kelley, 1959, p. 21). Outcomes that provide rewards exceeding one’s CL are satisfactory, while those providing lower rewards are unsatisfactory. Moreover, the same level of reward provides more satisfaction when CL is low versus high. For example, a $1,000 raise is more valuable to someone who is accustomed to a salary of $6,000 versus $60,000. Second is the CLalt, which refers to the most attractive alternative to a given relationship. For any relationship, CLalt is the best outcome that can be expected from exchanging in another relation, or not at all. In contrast to the CL, the CLalt is specific to a relationship – the alternatives to one relationship differ from the alternatives to other relationships. CLalt forms the basis of commitment to the relationship. Actors will remain in a relationship so long as the rewards the relationship provides exceed his or her CLalt. Should an alternative relation offer greater rewards, then there is the incentive to abandon the current relationship in favor of the higher value alternative. For any relationship, CL and CLalt operate jointly to determine the attractiveness and survival of relationships (Thibaut & Kelley, 1959).

190

DAVID R. SCHAEFER AND OLGA KORNIENKO

The outcome in a relationship must exceed one’s CLalt in order for the relationships to persist. So long as the outcome exceeds CLalt the relationship will persist, though the relative standing among outcomes, CL, and CLalt have different implications. When outcomes exceed one’s CL, actors will be generally satisfied; however, the level of satisfaction can vary depending on the relationship between CL and CLalt. When CL exceeds CLalt (OutcomeWCLWCLalt) then the actor is satisfied with the relationship and will remain committed to it. When CL falls below CLalt (outcomeWCLaltWCL), the relationship is attractive, but the actor is not particularly invested in it. Should outcomes fall below one’s CL (CLWoutcomeWCLalt), then the actor will be dissatisfied with the relationship, but remain in it due to lack of alternatives.

Comparison Levels in Prior Research The focus on negative connections in social exchange research has led to the primacy of CLalt in theories of power. The concept of CLalt has been a key component of the two most long-standing traditions within social exchange theory – P-D theory (Emerson, 1972) and network exchange theory, or NET (Willer, 1999) – both of which explicitly incorporate comparisons to help understand when structural power imbalances will arise in negative connections. Specifically, power use occurs when some actors have better alternatives than others. In such cases, one actor has a higher CLalt than the other. For instance, in the negatively connected B1-A-B2 network, A’s best alternative to exchange with each B is whatever the other B is offering. By contrast, the B’s only alternative to exchange with A is nonexchange, or zero reward. Hence, in each relationship, A has a higher CLalt than B, which gives A a power advantage. Whereas CLalt has been widely used to understand how structural power imbalances arise, the incorporation of CL has been limited. Part of this limitation stems from Emerson’s narrow interpretation of the inputs to CL. In their original formulation, Thibaut and Kelley noted several factors that could affect an actor’s CL. These include (1) exchange outcomes within the relationship being evaluated, (2) prior outcomes in other relationships, and (3) second-hand information about the outcomes experienced by others. Still, not all these inputs carry the same weight when comparing the outcome in a relationship to one’s CL. Thibaut and Kelley (1959) suggested that the relative inputs to CL from others’ experiences versus one’s own vary over time. Early on in a relationship, the CL is based on generalized

Comparison Processes and Social Exchange

191

experiences. However, as interpersonal experiences cumulate and the relationship develops, behavior in the specific relationship becomes most salient. Thus, over time the CL shifts from a global construct to a relationship-specific construct. As such, the CL within a relationship has both an exogenous and endogenous component. Emerson emphasized only the endogenous element of CLs by focusing on the outcomes of recent exchanges within the relationship that is being evaluated (Emerson, 1972; Cook & Emerson, 1978). This definition treats the CL as a relation-specific construct, thereby eliminating effects across relationships by definition. However, Thibaut and Kelley (1959) originally proposed the CL as a general construct that is applied to multiple settings. They drew upon research finding that satisfaction in prior relationships sets expectations for future relationships and that successful performance in one set of endeavors is transferred into high goals in future endeavors (pp. 96–97). Thus, the CL is informed by relations outside of the relationship being evaluated. Others have since followed Thibaut and Kelley’s original formulation of CL and emphasized the general nature of CLs by considering the inputs from multiple relationships. For example, sociologists have argued that comparison processes are likely to draw upon referents within individual’s social circles (Alwin, 1987) or local social networks (Gartrell, 1987). At an even further remove, economics research has found that market information can influence negotiation outcomes (Blount, Thomas-Hunt, & Neale, 1996; Fum & Del Missier, 2001). We retain Thibaut and Kelley’s original conceptualization of the CL as a general construct that is informed by multiple sources. Hence, when actors are embedded in a social network, each relationship has the potential to influence all others through the CL.

Structural Power through the Comparison Level According to Thibaut and Kelley, both CL and CLalt have the capacity to affect relationship outcomes. We propose that in the absence of alternatives, or where CLalt is equivalent for all actors (i.e., zero), the network structure can affect power use through an actor’s CL. This occurs because the CL affects the perceived value of exchange, which subsequently affects dependence and power. Prior research has already established that differences in the value of a relationship can affect power. This is how power operates when relationships are positively connected, where differences in value alone have the capacity to produce power imbalances.

192

DAVID R. SCHAEFER AND OLGA KORNIENKO

While positively connected relationships are interdependent and thus one could expect structural power to arise, we go one step further and propose that in the absence of relational interdependence the structure of the network can still affect the balance of power in relationships. That is, null connected relations can affect the balance of power in one another through the CL. Because value is at the root of dependence and power (Cook et al., 1983), variations in CL that affect value ultimately affect power. CL is thus related to structural power if the structure of relations in a network systematically affects actors’ CL and, in turn, exchange outcomes. Thus, we must demonstrate that the structure of null connections can affect the perceived value of exchange in a relation. To begin, an actor’s CL is informed by all relations, including the focal relationship and external relationships. Consequently, when the CL is employed to evaluate the focal relationship, the focal relationship’s perceived value will depend on the value of external relationships. If external relationships offer high value, thereby creating a high CL, then the perceived value of the focal relationship will be less than if external relationships and the CL were low value. Differences in the value of external relationships thus produce differences in the value of the focal relationship. Because value is a component of dependence, such differences have the capacity to affect power. Consider two scenarios. In the first, the Y-X-Z network is null connected, meaning Y and Z control resources in distinct domains and X can exchange with both partners. In the absence of alternatives or other structural sources of power, the X-Y and X-Z relations should be power balanced. If the relations carried a total value of 24, then each member could expect a reward of 12. In the second scenario, the relationships differ in value. X-Y is still worth 24, but X-Z is worth 6. In this situation, we expect that X’s CL will be lower than in the first scenario. The value of the additional relationship provides context to the focal relationship that can affect the CL and subsequent power use. The same level of reward offered by Y would look better in the second scenario where the CL is lower as a consequence of the lower value X-Z relation. Hence, X requires fewer rewards from Y in the second scenario to produce the same level of satisfaction from exchange with Y. Because X becomes satisfied earlier in the second scenario, X is expected to settle for lower rewards, resulting in more balanced power. If the CL operates as we propose, then it should allow the balance of power in a relationship to be affected by an actor’s additional relationship, even if that relationship is not ‘‘connected’’ to the first. .

193

Comparison Processes and Social Exchange

Empirical Test We test the effects of CL by focusing on a null connection that is situated within a larger structure that is positively connected. As we explain in the Discussion, this particular setting is where we most expect to find the effects of CL on power. Thus, this setting serves to test the plausibility of the hypothesized effects of CL. If successful, then additional research can refine the mechanisms behind CL and test for effects in a wider range of structural conditions. We consider three different positively connected networks in which exchange in the X-Y relation must precede exchange in the Y-Z relation (Fig. 1). Our interest is in how X’s CL differs across the network structures and implications for power in the X-Y relationship. The positive connection produces a power imbalance in the X-Y relationship. The contingency of exchanges facing Y creates a power imbalance through ordering in the first relation (X-Y) that advantages X over Y. Were the X-Y relation powerbalanced, each actor could expect to receive half the points (12). However, Y is willing to accept less than half because completing an exchange with X enables exchange with Z. On the basis of Corra and Willer’s (2002) resistance equations, X can expect to earn 14.3 of the 24 points (60%), with the remaining 9.7 going to Y. Within the standard positively connected structure shown in Fig. 1(a), X does not have other relations. Thus X’s CL develops based on the value of Y and ensuing high rewards obtained through exchange. Were X to have an additional relation, then the outcome of exchange in that relationship would

Y

Y 24

X

24

24

Z a. Standard

X

Y 24

24 b. CL-high

24

Z

X

24

6

Z

c. CL-low

Fig. 1. Network Structures with Different Comparison Levels for X. Notes: An arrow between relationships indicates a unilateral positive connection (exchange in the X-Y relationship must precede exchange in Y-Z). Relationships without arrows are null connected.

194

DAVID R. SCHAEFER AND OLGA KORNIENKO

also inform X’s CL. Thus, we consider adding a relation between X and Z that is null connected to the other relations (meaning X and Z can exchange regardless of whether exchange transpires in their respective relationships with Y). We suggest that the value of the X-Z relation will inform X’s CL. If the X-Z relation provides lower rewards than exchange with Y, then X’s CL will decrease and reduce X’s power advantage over Y. We consider two networks that include an X-Z relationship that is either high or low value (Figs. 1(b) and (c)). Adding the X-Z relationship gives X an additional basis for evaluating outcomes of exchange with Y. Fig. 1(b) presents a high CL network, in which the X-Z relation is worth 24 points. Because the value of X-Z is the same as X-Y, there should be no initial effect of its value on X’s CL. However, over time X will experience lower rewards from Z than from Y because only the latter exchange is power imbalanced. The lower rewards in the X-Z relationship are expected to reduce X’s CL (compared to X’s CL in Fig. 1(a)). As the CL drops to less than in the standard network (1a), X will require lower levels of reward from Y to experience the same level of satisfaction. Thus, in the CL-high network, the X-Y relationship will display weaker power use than in the standard network. The effects of CL are expected to be stronger in the Fig. 1(c) network, where the value of the X-Z relation is lower. The lower value of X-Z is immediately apparent to X, which should have an immediate effect on X’s CL. The outcomes from exchange, where X is expected to receive 3 of the 6 points, will further reduce X’s CL. As a result, a given level of reward from Y will be more satisfying to X in the CL-low network. In turn, X’s power use over Y should be even lower in the CL-low network than in the other two networks.

METHODOLOGY We investigated the effect of CLs on power with a laboratory experiment that examined the three networks shown in Fig. 1. We do not manipulate the CL directly; instead, we manipulate network structure and test for the theoretically derived effects of differences in network structure on exchange outcomes. Participants were 90 undergraduate students who participated in the experiment as a means to earn money. We randomly assigned students to a network and position (X, Y, or Z), where they remained for the duration of the experiment. All aspects of the experiment took place through computers using a program developed with z-Tree software

Comparison Processes and Social Exchange

195

(Fischbacher, 1999). On arrival, students read detailed instructions and completed several practice exchanges prior to beginning the experimental exchange phase. During the exchange phase, participants negotiated with one another over points that were converted into payments at the end of the experiment. The experiment contained 10 cases per condition, with conditions balanced by gender.

Network Structure In each network, the X-Y and Y-Z relations constituted a unilateral positive connection. Participants knew the letter corresponding to their partner(s) but not the overall structure of the network or the ordering of the relations. Only participants in position Y knew they were required to complete an exchange with X before being able to exchange with Z. Participants in position X were not aware that an X-Y agreement would enable an additional exchange for Y. Participants in position Z knew they had to wait for Y to complete an exchange in another relation before they could exchange with Y but did not know that Y’s other relation was with X.

Exchange Process The exchange phase contained 40 periods, each with up to 8 rounds of negotiation. Within relations, negotiation took the form of splitting a pool of 24 points (6 points in the X-Z relation of the CL-low network). Participants did not know they were dividing points between one another. Rather, they were told the maximum number of points they could request in the relation (24 or 6). At the beginning of each period, participants entered their request for points. The computer transformed requests into an offer to the partner (24 points minus the request). Participants reached an agreement if their requests were less than or equal to their partner’s offer (e.g., the sum of their requests was r24 points). Participants then received their share of the points. To avoid equity effects (Cook & Emerson, 1978), participants were not told how much their partner received. If an agreement was not made in a round, the next round would begin – participants learned the offers others made to them and could respond by accepting the offer, making a counteroffer, or repeating their previous request. This process repeated until exchange in all relations occurred or the eight rounds ended.

196

DAVID R. SCHAEFER AND OLGA KORNIENKO

In the X-Y and X-Z relations, participants began negotiating in round 1 and had up to eight rounds to reach an agreement. Negotiation in the Y-Z relation did not begin until X and Y reached an agreement. In the CL-high and CL-low conditions, the X-Z relation was null connected to the other relations. X could thus negotiate simultaneously with Y and Z, and Z could simultaneously negotiate with X and Y (given X and Y had previously exchanged).

Exchange Outcomes Our indicators of the effects of comparison processes are power use and the initial requests X made from Y. We measured power use in each relation as the total number of points a participant received in a relation relative to the total number of points exchanged in the relation. For example, X’s power relative to Y was computed as: PowerXY ¼

PointsXY PointsXY þ PointsYX

PointsXY is the total number of points X received from Y and PointsYX is Y’s points from X. Since X’s power over Y is directly related to Y’s power over X (PowerXY ¼ 1PowerYX) we only report one ratio for each relation. We captured the initial request that X made to Y during the first round of the first period. Initial requests provide a complementary indicator of CL that is unaffected by the history of negotiations with Y. X’s initial request from Y is solely the product of the network structure. Initial requests are measured in their original metric and range from 1 to 24. Lastly, for descriptive purposes, we calculated agreement frequency in each relationship as the proportion of the 40 periods in which participants reached an agreement.

RESULTS We begin by providing an overall picture of how exchange proceeded and power emerged in each relationship. We then evaluate our hypotheses regarding the effects of CL. Given the networks we studied were composed of positive and null connections, exchange was possible in every relation. The only provision was that the unilateral positive connection required Y to exchange with X before exchanging with Z. The agreement frequencies shown in Table 1 indicate that X reached agreements with Y and Z in

197

Comparison Processes and Social Exchange

Table 1.

Means and Standard Deviations of Exchange Outcomes. Network Structure Standard

CL-high

.925 (.085)

.978 (.025)

.953 (.085)

Y-Z

.597 (.252)

.733 (.224)

.613 (.273)

X-Z

—

.953 (.042)

.938 (.071)

.750 (.225)

.689 (.161)

.574 (.183)

Z’s power use over Y

.587 (.222)

.599 (.242)

.521 (.230)

X’s power use over Z

—

.437 (.244)

.507 (.164)

Agreement frequency X-Y

Power usea X’s power use over Y

Initial requests X’s request from Y

20.6 (2.84)

20.1 (2.92)

CL-low

18.5 (5.25)

po.05; po.01; po.001 (two-tailed tests). a

t-test comparisons to balanced power (0.5).

upwards of 90% of the periods, which was to be expected absent barriers to exchange. However, agreement rates in the Y-Z relation were substantially lower, ranging from 60% to 73% across conditions. We next examine how the introduction of the X-Z relation affected the outcome of X’s negotiations with Y. We hypothesized that the nullconnected X-Z relationship would provide a comparison that would decrease X’s requests from Y. Moreover, we expect the effects of X-Z on CLs to be proportional to the value of the X-Z relation. A high value partner should set a high CL and produce greater inequality than a low value partner. Thus, X’s power use over Y should decrease from the standard to the CL-high network and decrease further in the CL-low network. As shown in Table 1, X received 75% of the points in the standard network, falling to 69% in the CL-high network and 57% in the CL-low network. According to t tests, there is no difference in power use between

198

DAVID R. SCHAEFER AND OLGA KORNIENKO

the standard and CL-high networks. Thus, while the difference is in the expected direction, we do not find that the presence of the additional relation itself reduced X’s power use. Turning to the CL-low network, we find less power use than both the standard (t ¼ 1.92, df ¼ 18, one-tailed p ¼ .04) and CL-high networks (t ¼ 1.50, df ¼ 18, one-tailed p ¼ .08), though the latter difference is only marginally significant. When the X-Z relation was worth only 6 points, X’s power use over Y weakened relative to the other conditions to a level that did not significantly differ from balanced power. Thus, the combination of a second relation that was lower valued led to a decrease in X’s power use relative to the standard condition where X had no external relationship. Lastly, we test how the initial requests that X made from Y were affected by CLs. We do not expect initial requests to differ between the standard and CL-high networks because the values of the relations are identical. Instead, we expect the lower value of X-Z in the CL-low network to reduce X’s initial requests from Y compared to the standard and CL-high networks. Table 1 reports the initial requests from Y across networks. X made greater initial requests from Y in the standard and CL-high networks than in the CL-low networks. A comparison of initial requests reveals a marginally significant difference between the CL-low network and the other two networks (t ¼ 1.61, df ¼ 28, one-tailed p ¼ .06). These results follow the same pattern as for power use and indicate that the additional relation had an effect on the structurally imbalanced X-Y relation when its value was low. As exchange commenced, X immediately sought less from Y when the X-Y relation offered greater relative value.

DISCUSSION This study extends a long line of social exchange research that aims to understand how networks of relationships affect power. We depart from prior research in our consideration of structures in which relationships are not interdependent. We focus on null connections, which have been treated as the equivalent of disconnected dyads within social exchange theories. The conventional wisdom is that null connections contain no structural basis for power (Emerson, 1972; Patton & Willer, 1990). Yet, despite the lack of an instrumental association between null connected relations, they share a common actor. We argue that the presence of a common actor allows one relationship to affect the outcome of exchange in another.

Comparison Processes and Social Exchange

199

Our theory draws on Thibaut and Kelley’s (1959) concept of CL as a basis for evaluating exchange possibilities and outcomes. CLs affect the perceived value of exchange, which in turn affects dependence within the relation and, ultimately, power use. Comparison levels were central to the earliest theories of social exchange networks (Emerson, 1972), however only one dimension (CLalt) has played a prominent role. Effects of the more general CL have not been further developed since Thibaut and Kelley (1959). For actors entering a new relation, there is no history with the partner to guide their CL. Actors may know the range of rewards a relationship can provide, but they do not have the experience necessary to establish a CL that reflects realistically attainable rewards. Rather, in establishing a CL, actors incorporate information from multiple sources, which includes one’s past or concurrent relationships. We proposed that actors with multiple relationships would engage in comparison processes, with consequences for structural power. We tested this assertion using a laboratory experiment that manipulated the comparisons available to participants by altering network structure. We observed how the presence and value of an additional null connected relation (X-Z) affected power use within a structurally imbalanced relationship (X-Y). The smaller reward in the external relation reduced the advantaged actor’s expectations, which led to weaker power use. As the value of the additional relation decreased, the CL dropped and power use weakened further. Thus, we conclude that comparison processes operating across relations can affect power use. Given that our proposition contradicts current understandings of power and structure, we focused on a situation where the likelihood of observing the hypothesized effects was greatest. We posit that external relations can affect the CL of actors in any type of connection and position. However, subsequent effects on power use will only appear in certain circumstances depending on connection type, network position, and how CL changes. There are three aspects of the structure investigated that we propose make it the most amenable to observing CL-mediated effects of structure on power. First, we investigated how CL can diminish a power imbalance (rather than intensify power use). We chose a positive connection because the mechanism that produces the power advantage is value-based (e.g., Y values exchange more than X). Prior research has demonstrated that value-based mechanisms are secondary to alternatives (CLalt) in producing power imbalances (Corra & Willer, 2002). Because the CL operates through value we do not expect it to counteract power produced through differences in alternatives (e.g., in negative connections). With negative connections, advantaged actors necessarily have two (or more) partners, which already provide a reference

200

DAVID R. SCHAEFER AND OLGA KORNIENKO

point for establishing a CL in the form of CLalt. It is the advantaged actor’s ability to compare the rewards offered by alternative partners who are competing to avoid exclusion that produces imbalanced outcomes (Cook & Emerson, 1978). It is unlikely that providing the advantaged actor an additional, balanced basis of comparison would reduce power use because it would not eliminate the escalating offers of disadvantaged partners driven by exclusion. Likewise, adding a null connected relation to the disadvantaged actors would not eliminate their competition with one another for exchange in the negatively connected relations. Thus, we suggest that when CLalt is relevant, because a negative connection exists and some actors have better alternatives than others, it takes precedence over CL in driving exchange outcomes. Instead, we propose that any power mechanism that operates through value can be affected by CL. This includes power through ordering in positive connections, as we studied, as well as power in inclusively connected networks. Inclusion exists when an actor must exchange in N relations before experiencing any benefit, while the N partners have no such requirement, placing the actor at a power disadvantage (Patton & Willer, 1990; Szmatka & Willer, 1995). In a simple inclusively connected network, the N partners have only a relationship with the disadvantaged actor, not to one another or other actors. Thus, they occupy a similar, peripheral position as X in the standard unilateral positive connections studied here. It is likely that in an inclusively connected network, giving the advantaged actors a powerbalanced external relation that is null-connected to the imbalanced relation would mitigate power use through the same process observed here. Second, we expect that external relations that lower one’s CL will have a stronger effect than those that raise one’s CL. In the former case, a lower CL makes one more conciliatory or agreeable during exchange, which partners will readily accept. In the latter case, a higher CL makes one more demanding during exchange, which will meet with greater resistance from the partner. The acceptance of lower rewards during exchange should be more successful than attempts to take more. Third, we expect stronger effects of CL for the actor with the greater capacity adjust their rewards. We expect increases in CL to have a stronger effect on disadvantaged actors than on advantaged actors, who are closer to the maximum reward. Conversely, we expect decreases in CL to have a stronger effect on advantaged actors who have more room to make concessions than disadvantaged actors. Thus, overall we expect the strongest effects when external relations provide an advantaged actor with a low CL. The network we studied combined these three elements. In such a

Comparison Processes and Social Exchange

201

setting, the advantaged actor becomes more satisfied with lower rewards and more agreeable to the partner, resulting in decreased power use. We found that in this situation comparison processes can operate across relations and influence exchange behavior. Still, future research is needed to delineate the structural conditions in which comparison processes through CL alone can affect power use. The effects of comparison processes that we found offer insight to prior social exchange research. For instance, Lovaglia et al. (1995) found that the number of exchange partners (e.g., degree) can influence the terms of exchange an actor negotiates. They suggested that local information on the number of potential exchange partners affects one’s confidence that a successful exchange can be completed. Expectations for being included (or excluded) were thus theorized to lead to greater (or lower) rewards. Their experiment found that a measure of power based on degree performed better than competing measures. Subsequently, Lovaglia and Willer (1999) suggested that the observed effect was due to a nonlinear effect of exchange likelihood and referred to any effect of degree itself as bias. Nevertheless, we propose that such effects, operating through comparison processes, have real effects on structural power. Because comparison processes alter the perceived value of exchange, they affect actors’ dependence on one another, which serves as the basis for structural power (Emerson, 1972). An open question is when outcomes in external relations feed into a CL and affect power use in a relationship. Must there be some association between relationships or is it enough for two relationships to share a common actor? On one hand, there are several means through which two null connected relationships may be associated through the CL. There may be similarities between the partners in each relationship, such as status characteristics or roles, which lead an actor to equate the outcomes in each. CL effects may also be more likely when relationships share the same context, such as the workplace or school. The outcomes of exchanges within a common environment, whether rewarding or not, are more likely to have implications on other relationships within the same environment. It is also likely to matter whether the exchanges in the null connected relationships are temporally proximate. The CL is informed more by recent events than past events (Thibaut & Kelley, 1959); thus, CL effects are likely constrained to within a bounded timeframe. On the other hand, research in psychology suggests that a common actor may be enough for relationships to impact one another. For instance, when actors come to believe that their actions have no effect on the outcomes they experience then they cease acting to change their circumstances (Alloy, Abramson, Peterson, & Seligman, 1984). Such learned

202

DAVID R. SCHAEFER AND OLGA KORNIENKO

helplessness becomes a general state that reduces the motivation to act, even in circumstances that differ from the types of situations that produced the state (where one’s actions may have potency). Other research suggests that ‘‘hopefulness’’ may also be learned. Positive social experiences provide psychological empowerment that can carry over to other relationships (Zimmerman, 1990). Although learned helplessness is an extreme case of the type of process we describe, these findings are consistent with our proposition that otherwise independent relationships can affect one another if they share a common actor. It also remains to be seen how these findings extend more broadly to situations where a dyad is embedded in a broader web of relations such as found in the natural world. For instance, does upward or downward mobility bring concomitant changes in expectations that alter the balance of power in enduring relations? Actors occupying positions rich in social capital may become accustomed to experiencing high value in their exchanges, leading to a high CL. When such actors are advantaged in a positively connected relation they are likely to use the power their position provides. If placed in a disadvantaged position they are likely to attempt to use power that their position does not afford. In contrast, actors with poorer access to social capital may be less accustomed to high value exchanges and maintain a lower CL. If placed in an advantaged position, they will exert less power use, even though their position affords the opportunity. Implications of this research go beyond power use to include additional outcomes that are themselves conditioned on power use, such as emotions (Lawler, 2001), trust (Molm, Schaefer, & Collett, 2009), and cohesion (Schaefer, 2009). To the degree power use elicits negative emotions that threaten trust and cohesion, variation in network structure that allow comparison processes to mitigate power use have the potential to generate more positive and harmonious relational outcomes. Understanding the importance of these results for integrative relational outcomes is a valuable next step.

ACKNOWLEDGMENTS We thank Linda Molm for comments on a previous version of this chapter and Wanani Levatau and Allison Moore for their assistance with the research.

Comparison Processes and Social Exchange

203

REFERENCES Alloy, L. B., Abramson, L. Y., Peterson, C., & Seligman, M. E. P. (1984). Attributional style and the generality of learned helplessness. Journal of Personality and Social Psychology, 46, 681–687. Alwin, D. F. (1987). Distributive justice and satisfaction with material well-being. American Sociological Review, 52, 83–95. Berger, J., Zelditch, M., Jr., Anderson, B., & Cohen, B. P. (1972). Structural aspects of distributive justice: A status value formulation. In: J. Berger, M. Zelditch, Jr. & B. Anderson (Eds), Sociological theories in progress (2nd ed., pp. 119–146). Boston: Houghton Mifflin. Blau, P. (1964). Exchange and power in social life. New York: Wiley. Blount, S., Thomas-Hunt, M. C., & Neale, M. A. (1996). The price is right – or is it? A reference point model of two-party price negotiations. Organizational Behavior and Human Decision Processes, 68, 1–12. Bonacich, P., & Friedkin, N. (1998). Unequally valued exchange relations. Social Psychology Quarterly, 61, 160–171. Cook, K. S., & Emerson, R. M. (1978). Power, equity and commitment in exchange networks. American Sociological Review, 43, 721–739. Cook, K. S., Emerson, R. M., Gillmore, M. R., & Yamagishi, T. (1983). The distribution of power in exchange networks: Theory and experimental results. American Journal of Sociology, 89, 275–305. Corra, M., & Willer, D. (2002). The gatekeeper. Sociological Theory, 20, 180–207. Emerson, R. M. (1972). Exchange theory, part II: Exchange relations and networks. In: J. Berger, M. Zelditch, Jr. & B. Anderson (Eds), Sociological theories in progress (Vol. 2, pp. 58–87). Boston: Houghton-Mifflin. Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7, 117–140. Fischbacher, U. (1999). z-Tree-Zurich toolbox for readymade economic experiments-experimenter’s manual. Working Paper No. 21. Institute for Empirical Research in Economics, University of Zurich. Fum, D., & Del Missier, F. (2001). Modeling counter offer behavior in dyadic distributive negotiation. In: E. M. Altmann, A. Cleermans, C. D. Schunn & W. D. Gray (Eds), Proceedings of the fourth international conference on cognitive modeling (pp. 79–84). Mahwah, NJ: Lawrence Erlbaum Associates. Gartrell, C. D. (1987). Network approaches to social evaluation. Annual Review of Sociology, 13, 49–66. Homans, G. L. (1961). Social behavior: Its elementary forms. New York: Harcourt Brace and Jovanovich. Jasso, G. (2006). The theory of comparison processes. In: P. J. Burke (Ed.), Contemporary social psychological theories (pp. 165–193). Palo Alto, CA: Stanford University Press. Lawler, E. J. (2001). An affect theory of social exchange. American Journal of Sociology, 107, 321–352. Lawler, E. J., Thye, S. R., & Yoon, J. (2008). Social exchange and micro social order. American Sociological Review, 73, 519–542. Lovaglia, M., Skvoretz, J., Willer, D., & Markovsky, B. (1995). Negotiated exchanges in social exchange networks. Social Forces, 74, 123–155.

204

DAVID R. SCHAEFER AND OLGA KORNIENKO

Lovaglia, M., & Willer, D. (1999). l2, an alternative for predicting weak power. In: D. Willer (Ed.), Network exchange theory (pp. 185–192). Westport, CT: Praeger. Markovsky, B., Willer, D., & Patton, T. (1988). Power relations in exchange networks. American Sociological Review, 53, 220–236. Molm, L. D. (1991). Affect and social exchange: Satisfaction in power-dependence relations. American Sociological Review, 56, 475–496. Molm, L. D., Collett, J. L., & Schaefer, D. R. (2007). Building solidarity through generalized exchange: A theory of reciprocity. American Journal of Sociology, 113, 205–242. Molm, L. D., Peterson, G., & Takahashi, N. (2001). The value of exchange. Social Forces, 80, 159–185. Molm, L. D., Schaefer, D. R., & Collett, J. L. (2009). Fragile and resilient trust: Risk and uncertainty in negotiated and reciprocal exchange. Sociological Theory, 27, 1–32. Patton, T., & Willer, D. (1990). Connection and power in centralized exchange networks. Journal of Mathematical Sociology, 16, 31–49. Schaefer, D. R. (2009). Resource variation and the development of cohesion in exchange networks. American Sociological Review, 74, 551–572. Szmatka, J., & Willer, D. (1995). Exclusion, inclusion and compound connection in exchange networks. Social Psychology Quarterly, 58, 123–131. Tajfel, H., & Turner, J. (1979). An integrative theory of intergroup conflict. In: W. G. Austin & S. Worchel (Eds), The social psychology of intergroup relations (pp. 94–109). Monterey, CA: Brooks-Cole. Thibaut, J., & Kelley, H. H. (1959). The social psychology of groups. New York: Wiley. van Assen, M. A. L. M. (2003). Exchange networks: An analysis of all networks up to size 9. In: S. R. Thye & J. Skvoretz (Eds), Advances in group processes: Power and status (Vol. 20, pp. 67–103). Oxford, England: Elsevier. Willer, D. (1999). Network exchange theory. Westport, CT: Praeger. Yamaguchi, K. (1996). Power in networks of substitutable/complementary exchange relations: A rational-choice model and an analysis of power centralization. American Sociological Review, 61, 308–332. Zimmerman, M. A. (1990). Toward a theory of learned hopefulness: A structural model analysis of participation and empowerment. Journal of Research in Personality, 24, 71–86.

THE INCIDENCE OF STRONG POWER IN COMPLEX EXCHANGE NETWORKS Marcel A. L. M. van Assen ABSTRACT The present study increases our understanding of strong power in exchange networks by examining its incidence in complex networks for the first time and relating this incidence to characteristics of these networks. A theoretical analysis based on network exchange theory (e.g., Willer, 1999) suggests two network characteristics predicting strong power; actors with only one potential exchange partner, and the absence of triangles, that is, one’s potential exchange partners are not each other’s partners. Different large-scale structures such as trees, small worlds, buyer–seller, uniform, and scale-free networks are shown to differ in these two characteristics and are therefore predicted to differ with respect to the incidence of strong power. The theoretical results and those obtained by simulating networks up to size 144 show that the incidence of strong power mainly depends on the density of the network. For high density no strong power is observed in all but buyer–seller networks, whereas for low density strong power is frequent but dependent on the large-scale structure and the two aforementioned network characteristics.

Advances in Group Processes, Volume 27, 205–238 Copyright r 2010 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0882-6145/doi:10.1108/S0882-6145 (2010)0000027011

205

206

MARCEL A. L. M. VAN ASSEN

Keywords: Network exchange; strong power; complex networks; small worlds; scale-free networks.

INTRODUCTION Exchange became the object of research of economists in the late nineteenth century (Edgeworth, 1881) and was first investigated experimentally by economists in 1948 (Chamberlin, 1948). Anthropologists (e.g., Malinowski, 1922) and later also other social scientists such as Homans (1958), Thibaut and Kelley (1959), and Blau (1964) recognized that social interaction can be conceived as social exchange. Citing Blau (1964, p. 88), ‘‘neighbors exchange favors; children, toys; colleagues, assistance; acquaintances, courtesies; politicians, concessions’’. In many markets and social situations, actors cannot freely exchange with all other actors. Factors that might prevent two actors from exchanging with each other are geographical proximity, costs of linking, lack of information (i.e., not knowing of each other’s existence or that they can exchange), and not having complementary interests. The set of all exchange opportunities can be represented by an exchange network. The presence of a link between two actors signifies these actors can exchange and both can benefit from it, whereas the absence of a link between two actors signifies they cannot. A well-known example of an exchange network is a buyer– seller network. In a buyer–seller network, there are links between buyers and sellers, but not among buyers and not among sellers because they have no complementary interests. One of the most prominent questions in research on exchange networks is how an actor’s position in the network affects his exchange outcomes. This question has been addressed in many studies in sociology since the seminal paper of Cook and Emerson (1978) (see Molm, 1997, and Willer, 1999, for an overview), and more recently in economics as well (e.g., Kranton & Minehart, 2001; Corominas-Bosch, 2004). Already the first studies on exchange in sociology uncovered a phenomenon that is presently known as ‘‘strong power’’ (Stolte & Emerson, 1977; Cook & Emerson, 1978; Markovsky, Willer, & Patton, 1988; Willer, 1987). Strong power refers to exchange in which one actor obtains (almost) all surplus, whereas the other actor in the exchange obtains (almost) nothing. Strong power, although not called as such, is also well known in economics. For instance, in a monopoly the monopolist usually extracts most of the surplus.

The Incidence of Strong Power in Complex Exchange Networks

207

Strong power brings to light that an actor’s network position has an immense effect on the actor’s exchange outcome. Moreover, strong power sometimes brings about reactions of unfairness; why does (s)he obtain almost everything, and I get nothing? For these reasons understanding strong power and its causes should be a major endeavor in the social sciences. The present study attempts to increase our understanding of strong power by examining its incidence in complex exchange networks and relating this incidence to characteristics of these networks. Van Assen (2003) found in his review of research on exchange networks in sociology that of the 51 exchange networks considered until that time, 20–22 (about 40%), depending on the theory assumed, contain strong power. Although this rate suggests that strong power is abundant, Van de Rijt and Van Assen (2008, see their tables 2 and 4) show that in theory only about 7% of all 12,112 exchange networks up to size 8 contains strong power and that the percentage of networks containing strong power seems to decrease in the size of the network. The present study examines whether the incidence of strong power indeed decreases in network size of the network, and if so, why, depending on the structure of the network. The assumptions on exchange and the network made in the analyses reported in this chapter are commonly made in sociological (e.g., Van de Rijt & Van Assen, 2008, p. 259) and economic research (Vega-Redondo, 2007, pp. 28–29). First, it is assumed that the network structure is exogenous. Second, exchange is assumed to be bilateral. That is, one link in the network represents one mutually profitable exchange opportunity of the two actors connected by this link. Third, each actor can exchange only once, which is also known as the ‘‘one-exchange rule.’’ The implication of the third rule is that not all exchange opportunities can be employed if there are actors with more than one exchange opportunity. Note that this is the case in any connected network of more than two players. Fourth, all exchange opportunities are equally valuable. Fifth, all actors are ‘‘substitutable,’’ meaning that if actors switch positions in the same network, then the distribution of outcomes over positions is not affected. In this chapter, it is also assumed that the network is connected, that is, there is a path from each actor to any other player in the network. Consequently, if there are N actors there are at least N1 links in the network. An implicit assumption of the analyses is that the actors ‘‘know’’ whether they can be excluded from exchange or not, presuming rational forward looking actors or backward looking actors having experienced many rounds of exchange in the network. I use Network Exchange Theory (NET; Willer, 1999) to derive predictions on strong power in exchange networks. In the next section, ‘‘Strong power,’’

208

MARCEL A. L. M. VAN ASSEN

I define strong power using NET and I explain why I use NET to derive my predictions. This chapter focuses on complex exchange networks. But what are complex networks? One can attach several meanings to ‘‘complex networks’’ (see Vega-Redondo, 2007, pp. 19–23). At any rate, complexity goes together with a network of large size. At present exchange networks consisting of more than about 12 actors are not yet examined. Reasons for the lack of attention to larger exchange networks are that existing algorithms to generate predictions are not programmed to deal with large networks or it takes too much time to generate a prediction. Of all theories and algorithms published at present, X-net (Markovsky, 1995) is the least size limited; it can be applied to networks as large as 25 nodes (Willer & Emanuelson, 2008). In this chapter, I use a method recently developed by Willer, Van Assen, and Emanuelson (2010) to generate predictions on strong power in networks containing up to 144 actors. According to Dunbar (1993), this network size provides an approximation of the upper limit of the size of natural social groups. Human social groups had a group size of around 150 in prehistory, a number predicted on the basis of the size of the human neocortex and confirmed by evidence from the ethnological literature (Dunbar, 1993). Hence by examining networks up to size 144 most real-life social networks are included, at least with respect to size. It is important to emphasize that complex or large networks does not simply mean ‘‘more’’ on the ‘‘number-of-actors dimension.’’ Large networks introduce novel and important considerations in the analysis that are not present in networks of small size (Vega-Redondo, 2007, p. 22). One such consideration not present in small networks is structure in a specific sense. I examine the incidence of strong power in large networks of different structures that are known to occur in different large field applications. A default structure to compare to other structures is the uniform network. In the uniform network, all pairs of actors are equally likely to have an exchange opportunity. Five other well-known large-scale structures are investigated: small worlds, exponential networks, scale-free networks, trees, and buyer–seller networks. These large-scale structures and their relevance for the social sciences are explained in the section ‘‘Complex networks.’’ The contribution of this chapter to the social sciences in general and the field of exchange networks specifically is fourfold. First, it adds to the study of complex networks, which has become a booming field of research in the past decade (Vega-Redondo, 2007, p. 1), by providing an example of how complex networks can be incorporated in the study of social behavior. The present study shows how the incidence of strong power in exchange networks is

The Incidence of Strong Power in Complex Exchange Networks

209

affected by the network structure (uniform, small worlds, exponential networks, scale-free networks, trees, and buyer–seller). Second, the study presents a welcome addition to the literature on network exchange since it describes and applies a method developed by Willer, Van Assen, and Emanuelson (2010) that, as opposed to other methods, can be applied to large exchange networks. Third, since many interesting applications of exchange theory consist of more than a dozen actors, the method and results presented here are relevant to these applications. To mention a few applications, consider a social network of pupils in a class, networks of social relations between employees within a department or company, and buyer–seller networks. Finally, and most specifically, this chapter adds to our understanding of strong power in exchange networks and the factors affecting it. The chapter reveals whether the incidence of strong power decreases in network size, as suggested by the results of Van de Rijt and Van Assen (2008), and how the incidence depends on global and local structure of the network. I examine the proportion of networks containing strong power, and the number of high power positions in each of the large-scale structures, as a function of the size and average degree of the network, and two properties of the network that are shown to be relevant for strong power. An actor’s degree is his number of potential exchange partners. Average degree or density is then the average number of potential exchange partners actors have. The two properties relevant to strong power are the proportion of actors in the network with degree 1 and the proportion of actors with clustering 0. Clustering addresses whether one’s exchange partners are also each other’s exchange partners. If clustering is 1, all one’s exchange partners are also each other’s partners. If clustering equals 0, then one’s exchange partners cannot exchange with each other. Clustering is dissimilar in different large-scale structures of different size. For instance, if friendship networks across the world were uniform, the probability that two of my friends are each other’s friends is very small (low clustering), but in small worlds this probability is much higher (high clustering). Hence large-scale structure likely affects the incidence of strong power at least partly since these structures differ in characteristics associated with strong power. The largescale structures, their properties, and how they differ on these properties, are discussed in the section ‘‘Complex networks.’’ In the section ‘‘Effects of network characteristics on strong power,’’ the two properties ‘proportion of actors with degree 1’ and ‘proportion of actors with clustering 0’ are shown to be positively associated with the incidence of strong power and the number of high power positions. Whereas network size has only a negative effect on clustering and not on the proportion of actors

210

MARCEL A. L. M. VAN ASSEN

with degree 1, average degree of the network has a negative effect on both of these properties in uniform networks, small worlds, exponential and scalefree networks. Consequently, size is shown not to be a strong predictor of strong power in these large-scale structures, but average degree is; in the simulation hardly any strong power was detected in large-scale structures with an average degree of six or higher. Comparing the large-scale structures, less strong power was observed in small worlds than in exponential networks, a bit more in uniform networks, and most in scale-free networks, conditional on size and average degree. Buyer–seller networks are shown to be peculiar networks. Increasing average degree has a positive effect on strong power if the number of sellers is different from the number of buyers, but this effect is negative effect if these numbers are equal. Although most of these results are anticipated, the results were also illustrated by simulating networks of all large-scale structures of different size and average degree. The design of the simulation is described in ‘‘Design and Algorithm of Simulation Study’’ section, and the results of the simulation are described in ‘‘Results of the Simulation’’ section. The final section discusses the results and concludes the chapter.

STRONG POWER In the introduction, strong power is loosely described as referring to exchange in which one actor obtains almost all surplus, whereas the other actor in the exchange obtains almost nothing. Before defining strong power precisely, I show that theories of exchange in economics and sociology do not agree on which exchange networks contain actors receiving almost all surplus. I use the so-called 7-line in Fig. 1(b) to illustrate this. The 7-line is a tree and buyer–seller network, for instance, with S1, S2, S3 representing the sellers and the other actors B1–B4 the buyers. Two well-known models of exchange in networks in economics (Kranton & Minehart, 2001; Corominas-Bosch, 2004) assume that a maximum number of exchanges take place. Consequently, both theories assume that all three sellers in the 7-line exchange, that is, the exchanges are B1–S1, B4–S3, and either B2–S2 or B3–S2. Although the two theories assume different bargaining processes (an auction in Kranton and Minehart versus public offers in Corominas-Bosch), both predict each seller to get all surplus in his exchange. Sociologists Bienenstock and Bonacich (1992) applied the core concept from cooperative game theory to exchange in networks. The core also assumes a maximum number of exchanges taking place and

211

The Incidence of Strong Power in Complex Exchange Networks

A

S1

A

S2

S3

B A

B1

A

1a: Hourglass

B2

B3

B4

1b: 7-Line B1

S1

B2

S2

B3 1c: Fully connected buyer-seller network with two sellers and three buyers.

Fig. 1. Three Exchange Networks.

leads to the same prediction that the sellers in the 7-line get all surplus. However, the assumption that three exchanges take place can be violated in the 7-line. If buyers B2 and B3 exchange with sellers S1 and S3, respectively, seller S2 is excluded and therefore gets nothing. To prevent becoming excluded, seller S2 could offer part of the surplus to B2 or B3. In either way, S2 does not obtain all surplus suggesting that S2 in the 7-line might not be a high power actor. Indeed, Lucas, Younts, Lovaglia, and Markovsky (2001) found empirical evidence that exclusion of B2 in the 7-line decreased this actor’s profits in his exchanges. Some theories of network exchange take into account that seller S2 can be excluded and does not obtain all surplus. In economics, the model of Polanski (2007) assumes a bilateral bargaining process predicting that the sellers in the 7-line get three-fourths of the surplus. In sociology, Buskens and Van de Rijt’s (2008) ‘‘Sequential Power-Dependence Theory’’ and some of the 10 theories of network described by Willer and Emanuelson (2008) also predict S2 not to obtain all surplus. Two well-known examples are expected value theory and NET. Expected value theory (e.g., Friedkin, 1995) is known for not predicting strong power when other theories do (Van de Rijt & Van Assen, 2008). NET encompasses a set of models developed by Willer, Markovsky, Skvoretz, and others (for an overview see Willer, 1999; Emanuelson, 2005). I define strong power using NET, since NET is the only theory that is adapted to handle complex networks

212

MARCEL A. L. M. VAN ASSEN

(Willer, Van Assen, & Emanuelson, 2010). Willer and Emanuelson (2008) found that NET is the simplest and second most precise theory of all 10 theories examined by them. NET assumes maximal exchange patterns, meaning that after exchanging no connected dyads remain. Note that the assumptions of maximum number of exchanges and maximal exchange patterns are different. For instance, exchange pattern {(B2–S1), (B3–S3)} in the 7-line is a maximal exchange pattern, but the number of exchanges is not. There are three maximal exchange patterns in the 7-line. Actors that are excluded in at least one maximal exchange pattern cannot be high power actors in NET. Seller S2 is not included in the maximal exchange pattern {(B2–S1), (B3–S3)} and therefore cannot be a high power actor. Denote actors that are excluded in at least one maximal exchange pattern as E, and those that are always included as I. NET then defines a strong power (sub)network as a (part of the) network with the following three characteristics: (i) There are more E than I actors in the subnetwork. (ii) Each E actor is connected only to one or more I actors within the subnetwork. (iii) Each I actor is connected to at least two E actors in the subnetwork. The I actors may also be connected to other actors outside the strong power subnetwork. The I actors in the component are high-power positions and are assumed to get (almost all of) the surplus. Not all I actors are high power actors, as the 7-line example shows. B1, B2, S2, B3, B4 are E actors, whereas S1 and S3 are I actors. But S1 and S3 are no high-power actors, since the E actors B2 and B3 are connected to another E actor S2 [violating (ii)]. In the present study an algorithm is used to locate high-power positions. The algorithm is developed by Willer, Van Assen, and Emanuelson (2010). In the first step it is determined for each actor if he is an E or I actor. Next, step-by-step all relations are removed that cannot occur within strong power subnetworks. With their removal, only strong power subnetworks remain as connected positions. The following relations are removed subsequently: 1. I–E iff I–E–E These relations cannot be in a strong power subnetwork because of (ii). 2. I–I I–I relations are removed because they do not define strong power networks.

The Incidence of Strong Power in Complex Exchange Networks

213

3. E–E These relations violate (ii) as well. After applying (1) to (3) only E–I links remain. 4. I–E–I iff either I has degree 1; this is repeated until no such I–E–I remains in the network. According to (iii) each I should be connected to at least two E actors. After (4) no I actor with degree 1 remains. 5. Relations in subnetworks with an equal number of E and I actors. These relations are removed because they violate (i). If no relation is present after applying (1)–(5), then no strong power is present in the network. The only relations that remain are between highpower I actors, each connected to at least two E actors, where these I actors can be connected to others in the initial network but the E actors cannot. The number of high-power actors in the network is equal to the number of I actors that are still connected after removing the relations. Applying the algorithm to the examples of Fig. 1 results in an empty network in the case of the Hourglass, which contains only E actors, and the 7-line exchange network. The network in Fig. 1(c) is a strong power network; S1 and S2 are two I actors that are linked to more than two E actors that are not linked to other actors. For our analysis we classify an exchange network as a strong power network if it contains at least one strong power subnetwork. A strong power network has at least one high-power actor. For each network the number of high-power actors is counted in the analyses.

COMPLEX NETWORKS A detailed and mathematical treatment of the structures discussed in this section (except buyer–seller networks) can be found in the book Complex Social Networks by Vega-Redondo (2007). For each large-scale structure, I subsequently describe their origin, how it is generated in the present study, and its properties. While discussing their properties, I focus on two properties that are shown later on to be associated with strong power, the proportion of actors having degree 1 and the proportion of actors having clustering 0. I also address other properties such as connectedness and the diameter of the network. I start by defining each of the properties in the next subsection. Then I discuss each of the large-scale networks in separate subsections, and how the two properties are affected by average degree and

214

MARCEL A. L. M. VAN ASSEN

size of the network. This section is rather technical at some places. For those who want to skip technicalities, the properties of the large-scale structures are summarized at the end of the section, whereas the properties themselves are discussed in the subsequent subsection. Properties Average degree d is the number of links M divided by the number of actors N. An actor’s degree di is simply his number of links. Although I fix the networks’ degree, the large-scale structures differ in their distribution of di. One aspect of this distribution shown to be relevant for strong power is the proportion of actors that has degree 1, which I denote by pd 1 . For the Hourglass network in Fig. 1(a) the actor degrees are 4 for A and 2 for the B actors, hence pd 1 ¼ 0. For the 7-line in Fig. 1(b) the actor degrees are 2 for all actors except for B1 and B4 that have degree 1, hence pd 1 ¼ :29. For each actor i having at least two neighbors, clustering coefficient Ci is defined as the proportion of pairs of neighbors that are linked to each other. The clustering index C for the whole network is defined as the average of Ci across those actors for which Ci is defined. The large-scale structures differ a lot in the average and dispersion of their distribution of Ci. Relevant to strong power is the proportion of actors with Ci ¼ 0, which I denote by pc0 . Turning to the examples, in the Hourglass CA ¼ 1/3 because of all six possible pairs of A’s neighbors only two pairs are linked to each other, and CB ¼ 1. Hence C ¼ .87 and pc0 ¼ 0. In the 7-line C B1 and CB4 are not defined since they have no neighbors, whereas the other actors have clustering 0, yielding pc0 ¼ :71. The 7-line example makes clear that clustering is not defined for those actors with degree 1. Hence pd 1 þ pc0 ¼ 1 for the 7-line. All networks considered in this study are connected, meaning that there is a path from one actor to any other actor in the network. The diameter of a network is the maximum of the distances between pairs of actors. The distance between two actors equals the number of links on the shortest path between these two actors. The diameter of the Hourglass equals 3, and equals 6 of the 7-line. Uniform Networks By uniform networks I mean networks in which the probability of a link between any two actors is equal for all pairs of actors (see also Wasserman & Faust, 1994, Section 13.4). Two models generating networks satisfy this

The Incidence of Strong Power in Complex Exchange Networks

215

+ ´ nyi model. In this model property. First, there is the binomial or Erdos–Re every possible link occurs independently with fixed probability p. The second model assigns equal probability to all networks with exactly M ¼ N(N1)p links. The difference between the two models is that the first represents drawing with replacement and therefore does not fix the number of links in the network, whereas the second represents drawing without replacement thereby fixing the number of links in the network to M. Since I want to fix the number of links, I use the second model to generate uniform networks. The uniform network is generated in two steps. In the first step, a uniform network is generated with M links without actors with degree 0. It starts by randomly determining the links of the first actor with all other actors. If the first actor has no links, the generation starts once again, otherwise it continues by randomly determining the links of the second actor with the third to last actor. If the second actor has no links, the network is discarded and the process starts from scratch, otherwise it continues with the third actor, etc. In the second step, it is checked if the network is connected. If it is not, the network is discarded and the generation starts once again at step 1. Uniform networks are very likely connected if the average degree of the network exceeds ln(N) (Bolloba´s, 2001). The generation process described earlier requires a lot of time if the average degree d is considerably smaller than ln(N), because in that case only a small fraction of fraction of uniform networks is connected. Therefore, uniform networks of large size with low degree are not generated in this study. The diameter of the uniform network grows at rate ln(N)/ln(d) (Vega-Redondo, 2007, p. 39) for large size networks. Proportion pd 1 obviously decreases in d and can be shown to increase slightly in N to a value bounded above by de–d.1 C equals d/N and hence increases in d and decreases in N. Proportion pc0 increases sharply in N, conditional on d.2

Small Worlds Small world networks were first coined and examined in the seminal paper by Watts and Strogatz (1998). Whereas large and sparse uniform networks (i.e., networks with low average degree relative to size) are decentralized (no actors with high degree), have a small diameter, and have low clustering, large and sparse small worlds are also decentralized and have a small diameter, but have high clustering (Watts, 1999). Hence the distinctive feature compared to uniform networks is high clustering, or in other words,

216

MARCEL A. L. M. VAN ASSEN

a much higher likelihood that ‘‘my friends are friends’’ that seems to characterize many social networks. Examples of real social networks that have these characteristics are for instance collaboration of feature-film actors and file sharing communities (see Schnettler, 2009, for more examples and an overview of small-world research). Small worlds are generated starting from a ring of N actors in which each actor is connected to (1/2)d of his neighbors on one and (1/2)d on the other side. For actor i ¼ 1 to n the link that connects i to i þ 1 is rewired with probability m.3 To be exact, with probability m the link is replaced by a link from i to an actor to which i was not connected in the initial ring.4 This process is repeated for links between i and i þ j, for j ¼ 2 to (1/2)d. The complete process guarantees that each actor has diZ(1/2)d, but not that the network is connected. In the unlikely case that the network is not connected, the generation process is started once again. The characteristic properties of small worlds are not obtained for all values of m. If mNdW1 networks are generated with high clustering but with a small diameter as in uniform networks (Vega-Redondo, 2007, p. 64). For m ¼ 1, the small world is similar to a uniform network, with the restriction that each actor’s degree is at least (1/2)d. In the present study, I study small worlds with m ¼ .1 and m ¼ .5. The property mNd W1 is satisfied for all values of d and N examined in the simulations. For m ¼ .1 clustering is very high and the network still resembles a ring network, whereas for m ¼ .5 the small world resembles more the uniform network with much less clustering but that is still larger than that of a uniform network (Watts & Strogatz, 1998). With respect to the small world’s properties, all actors have degree d for m ¼ 0, whereas the degree distribution approaches the degree distribution of the uniform distribution for m ¼ 1 with the important difference that pd 1 ¼ 0 for dZ4, since diZ(1/2)d for all actors. For small values of i clustering coefficient C is close to its value in the initial ring, which equals (3/4)[(1/2)d1]/[(1/2)d(1/2)], and is independent of N (Watts & Strogatz, 1998). Note that C in the initial ring is larger than 1/2, which is considerably larger than the value of C ¼ d/N in large and sparse uniform networks. Proportion pc0 ¼ 0 for m ¼ 0 and becomes positive for large values of m but remains smaller than pc0 in uniform networks. Exponential and Scale-Free Networks Although small worlds can account for both high clustering and a small diameter of the network, it cannot account for the heterogeneity present in

The Incidence of Strong Power in Complex Exchange Networks

217

many real-life networks (Albert, Jeong, & Baraba´si, 1999). Actors in uniform networks and small worlds are similar with respect to their clustering and degree, whereas in real-life networks these properties can differ widely across actors.5 Examples of such real-life networks are for instance the world wide web, citation networks, and networks of sexual partners in humans. These networks are characterized by some highly connected actors (called hubs) and clustering that is considerably lower than in small worlds with low m but larger than in uniform networks. The exponential and scale-free networks are generated dynamically using a process developed by Baraba´si and Albert (1999). The process has two mechanisms, growth and linking, implemented in the following way in the present study. The process starts with d þ 1 fully connected actors with degree equal to (1/2)d. Then one actor is added (growth) that links to d actors (linking) already present in the network. The linking mechanism is different for exponential and scale-free networks. Linking is unbiased in exponential networks, meaning that each link to any actor already in the network is equally likely. Linking in scale-free networks is characterized by preferential attachment; the probability of a link to an actor is proportional to that actor’s degree.6 In the context of social networks preferential attachment reflects that, when a newcomer enters the community, he is more likely to become acquainted with more ‘‘visible’’ people. The process guarantees connectedness of the network. The diameter of a scale-free network is known to be ln(N)/ln[ln(N)] for large N, which is larger than the diameter of uniform networks. Exponential networks are characterized by a geometric degree distribution, whereas scale-free networks are characterized by a power-law degree distribution (Vega-Redondo, 2007, pp. 64–71). Both degree distributions have the same average degree d, but the preferential attachment mechanism produces a larger standard deviation of degree for scale-free networks than for exponential networks. The process guarantees that diZ(1/2)d, hence pd 1 ¼ 0 if dZ4. Average clustering is less than in small worlds with low m and more than in uniform networks. Clustering is larger in scale-free networks than in exponential networks, since the probability of having at least two links to highly connected actors that are linked to each other is higher in scale-free networks. For the same reason, pc0 is larger in exponential networks than in scale-free networks. The results of the simulations reported later show that pc0 of uniform networks is higher than of scale-free networks, but lower than of exponential networks for the degrees and network sizes studied. Intuitively, this result can be understood as follows. In uniform networks the dispersion of di and Ci are smaller; whoever an actor is linked to, the probability that

218

MARCEL A. L. M. VAN ASSEN

‘‘your friend’s friend is your friend as well’’ is constant. Now consider exponential networks. Actors that enter the exponential network late have low degree and are more likely connected to actors with low degree than in uniform networks, which explains the higher value of pc0 for exponential networks. Now consider scale-free networks. Because of preferential linking in scale-free networks, actors entering the network are more likely connected to those with high degree. The actors with high degree are more likely connected to each other, which explains why pc0 is lower in scale-free networks. Apparently, the effect of a higher probability to connect to those actors with a high degree is so strong in scale-free networks that pc0 is even smaller than that in uniform networks.

Buyer–Seller Networks Studies on exchange in economics typically distinguish between buyers and sellers; the buyers demand one indivisible unit of a good that is valuable to them, and sellers own one of such units (e.g., Kranton & Minehart, 2001; Corominas-Bosch, 2004). Buyer–seller networks are bipartite (and viceversa), which means that all actors can be divided into two disjoint sets (in this case buyers and sellers) such that links are only between actors from different sets. In this study buyer–seller networks are generated in the same way as uniform networks; links are independently drawn with a probability that is equal for all buyer–seller pairs. Proportion pd 1 depends not only on N and d but also on the relative number of buyers. Whereas pd 1 of buyer–seller networks with an equal number of buyers and sellers is similar in value to pd 1 in uniform networks, pd 1 increases in the relative number of buyers with pd 1 ¼ 1  ð1=NÞ in the extreme case with only one seller and N1 buyers. Bipartite networks have no cycles or loops of odd length. Clustering requires cycles of length 3, hence C ¼ 0. However, pc0 is less than 1, since Ci is not defined for actors with degree 1. Thus, pd 1 þ pc0 ¼ 1 in buyer–seller networks. Since pd 1 is higher in buyer–seller networks with less sellers than buyers, it follows that pc0 is higher in buyer–seller networks with an equal number of buyers and seller.

Tree Networks Tree networks are minimally connected networks with N1 links and average degree 22/N. Tree networks are bipartite and special cases of

The Incidence of Strong Power in Complex Exchange Networks

219

buyer–seller networks. In this study tree networks are generated using the algorithm for exponential networks with one linked pair in the initial step and linking one additional actor to one actor already in the current network in each consecutive step. The diameter of trees is generally considerably larger than that of the other network structures. Two actors have degree 1 when all actors are connected on a line, whereas pd 1 42=N for other trees. There is no clustering (C ¼ 0) and pd 1 þ pc0 ¼ 1. Characteristics of Large-Scale Structures: A Summary Table 1 summarizes the comparisons between large-scale structures on the network characteristics pd 1 , C, and pc0 , for dZ4. Trees are excluded, as well as networks with d ¼ 2 because all these networks are in fact trees with one additional link. Since small worlds are only defined for even degree, the most sparse networks that can be used to compare all networks have d ¼ 4. Proportion pd 1 is highest for buyer–seller networks with more buyers than sellers, about equal for uniform networks and buyer–seller networks with an equal number of buyers and sellers, and zero for all other structures. That is, only in buyer–seller and uniform networks there can be actors with at most one exchange partner. With the exception of small worlds with m ¼ .5, all structures can be ordered with respect to clustering C. Clustering is highest in small worlds with m ¼ .1, high in scale-free networks, lower in exponential networks, even lower in uniform networks, and absent in buyer–seller networks. If small worlds with m ¼ .5 resemble uniform networks more than rings, then its clustering is less than that of scale-free and exponential networks, otherwise its clustering is greater. The structures can also be ordered on pc0 . From high to low we have buyer–seller networks with an Table 1.

A Comparison of Large-Scale Structures with Respect to pd 1 , C, pc0 , for Average Degree More Than 2. Comparisons

pd 1

BSdWBSeBUWSW5 ¼ SW1 ¼ SF ¼ E ¼ 0

C

SW1WSW5WUWBSe ¼ BSd ¼ 0; SW1WSFWEWU

pc0

BSeWBSdWEWUWSW5WSW1; UWSFWSW1

Notes: The large-scale structures are uniform (U), random worlds with p ¼ .1 (SW1), random worlds with p ¼ .5 (SW2), exponential (E), scale-free (SF), buyer–sellers with an equal (BSe) and a different (BSd) number of buyers and sellers.

220

MARCEL A. L. M. VAN ASSEN

equal number of buyers and sellers, buyer–seller networks with more buyers than sellers, exponential networks, uniform networks, and small-world networks. pc0 of scale-free networks is in between that of uniform and smallworld networks with m ¼ .1, but can be lower or higher than that of small world networks with m ¼ .5; pc0 of small worlds with m ¼ .5 is larger than that of scale-free networks if they resemble uniform networks more than rings, otherwise it is less. The next section derives how differences in structures on these properties can affect the incidence of strong power networks.

EFFECTS OF NETWORK CHARACTERISTICS ON STRONG POWER NET’s definition of strong power and the algorithm to detect it in networks can be used to derive some effects of network characteristics on strong power in exchange networks. The first dependent variable in the analysis is the incidence of strong power, that is, the proportion of networks containing a strong power subnetwork. The second dependent variable is the number of high-power actors in a network. I show in the first subsection that two local or microlevel characteristics of the network at least partly determine whether an actor is a high-power actor, an E actor in a strong power subnetwork, or neither. These two local characteristics are actor’s degree and actor’s clustering. On the basis of these characteristics, I calculate the network statistics pd 1 and pc0 and show in the subsequent section that both are positively associated to strong power, conditional on N and d, in all large-scale structures except buyer–seller networks. The effect of macrolevel characteristics of the network C, N, and d on strong power are discussed in the section thereafter. The large-scale structures are compared to each other with respect to strong power using Table 1. In the penultimate subsection, I discuss the special cases of trees and buyer–seller networks. The final subsection concludes and poses some questions to be answered in the simulation. Actor Degree and pd 1 , Actor Clustering and pc0 Pivotal to strong power is the incidence of E actors that are only connected to I actors. An actor with degree 1 is an E actor that is connected to an I actor. If two actors with degree 1 are connected to the same I actor, then these three actors are part of a strong power network with I being a high

The Incidence of Strong Power in Complex Exchange Networks

221

power actor obtaining (almost) all surplus. The probability that two actors with degree 1 are connected to the same actor is increasing in pd 1 .7 Hence both a higher incidence of strong power and number of high power actors can be expected for higher values of pd 1 . Strong power can also exist in networks without actors with degree 1. Strong power (sub)networks can also arise by E actors that are not connected to each other, but collectively connected to a smaller group of I actors. For example, consider the strong power network in Fig. 1(c). This network can be conceived as a network with two sellers (S1 and S2) that are fully connected to three buyers (B1, B2, B3). Characteristic of this network is that there is no clustering; none of my friends’ friends is my friends. The higher the proportion pc0 , the higher the likelihood that at least two actors are collectively connected to a larger group of actors with Ci ¼ 0. The effect of pc0 on strong power is weak since Ci ¼ 0 is neither a necessary nor a sufficient condition for E actors to be collectively linked to a smaller group of I actors. The condition is not necessary, because if S1 and S2 are linked CiW0 for all actors in Fig. 1(c), but the network is still strong power. The condition is also not sufficient, since if one of the B actors is connected to a sixth actor that does not have high power, clustering of that B actor is still zero but the network is no longer strong power.

Effect of Macrolevel Characteristics C, N, d Clustering has a negative effect on strong power if it connects E actors. The more clustering, that is, the more 3-cycles, conditional on N and d, the more likely E actors become connected, which destroys strong power. The effect of N on the incidence of strong power, conditional on d, is positive. Whereas pd 1 is not much affected by N, the effect of N on C is negative and positive on pc0 . The probability of at least one strong power subnetwork increases in N because the values of C and pc0 are more favorable, and there are also more opportunities for these subnetworks since the number of possible combinations of actors increases in N. The effect of d is large and negative. Consider a certain network of size N, with certain values of pd 1 , pc0 , C. Adding one link to the network can only increase the values of these three network statistics and make conditions for strong power worse. If the original network contains strong power subnetworks, randomly adding one link reduces the number of high power positions with substantial probability.8

222

MARCEL A. L. M. VAN ASSEN

Strong Power in Large-Scale Structures The explanation of the effect of d clarifies that the effects of pd 1 , pc0 ,C on strong power are highly correlated because the three statistics are strongly affected by d. However, the large-scale structures differ systematically in their distribution of pd 1 , pc0 , C condition on d, hence a different incidence of strong power can be expected in the different large-scale structures. Table 1 compares the large-scale structures with respect to their distributions on pd 1 , pc0 , C for dZ4. Trees, networks with d ¼ 2, and buyer–seller networks are considered in the next subsection. Small worlds are characterized by pd 1 ¼ 0, low pc0 , and the highest C. Hence no or hardly any strong power can be expected in small worlds, but more so for small worlds with m ¼ .5. Strong power in uniform networks is more frequent than in small worlds because there pd 1 , pc0 , C are more favorable. Exponential and scale-free networks will also contain more strong power than small worlds with low m; pd 1 is the same in these networks, but pc0 and C are more favorable in exponential and scale-free networks. It is not clear how to order the incidence of strong power in uniform, exponential, and scale-free networks. Uniform networks have positive values of pd 1 , at least for d ¼ .4, but exponential networks have a higher value of pc0 . Since the effect of pd 1 on strong power is stronger than the effect of pc0 , a higher incidence of strong power can be expected in uniform networks than in exponential networks if d ¼ 4. Scale-free networks have lower pc0 and larger C than exponential networks, suggesting a lower incidence of strong power for scale-free networks. However, the exponential and scale-free networks not only differ in pc0 and C, but also in structure. The actors entering the scale-free network late are connecting with higher probability to the actors that entered very early (preferential linking). If d is small, it is likely that at least some of the early actors have high power and are connected to E actors that entered late. Smaller pc0 and higher C is not detrimental for strong power, since it is the I actors that are connected and not the E actors. This suggests that, even though pc0 and C seem to be less favorable, strong power can be more frequent in scale-free networks than in exponential networks or even uniform networks.

Strong Power in Trees, Networks with d ¼ 2, and Buyer–Seller Networks The effects of pd 1 , pc0 , C on strong power in trees and buyer–seller networks is different from those on the other large-scale structures. In these networks

The Incidence of Strong Power in Complex Exchange Networks

223

pd 1 þ pc0 ¼ 1 and C ¼ 0. Degree is minimal in trees and hence generally large values of pd 1 and high incidence of strong power are expected. The probability that the tree contains at least one strong power subnetwork is increasing in N, since the probability that at least two actors with degree 1 randomly connect to the same actor increases in N. I examine incidence of strong power and number of high power positions in trees as a function of N using simulation. The large-scale structures with d ¼ 2 are trees with one additional link. These trees with one additional link, however, are generated by different random processes. The small-world network will likely consist of lines and cycles and a small value of pd 1 . Exponential and scale-free networks will have higher values of pd 1 than uniform networks since they likely contain more hubs. As scale-free networks have larger hubs and larger values of pd 1 , the incidence of strong power is scale-freeWexponentialWuniformWsmall world with m ¼ .5Wsmall world with m ¼ .1. Trees were generated using the random process also used to generate exponential networks, hence the incidence of strong power in trees is comparable to that of exponential networks with d ¼ 2. Buyer–seller networks can be considered trees with links added to it such that only buyers and sellers are connected. The effect of degree on the incidence of strong power and the number of high power positions in the network depends on the relative number of buyers and sellers. I first consider minimally connected buyer–seller networks (trees). For convenience, assume that the proportion of buyers in the network is at least .5. The probability that the network contains a strong power subnetwork then increases in the proportion of buyers. The reason is that the higher the proportion of buyers, the higher the probability that at least two buyers with degree 1 are linked to the same seller. This probability increases for two reasons. First, the number of buyers with degree 1 increases. This number equals the difference in the number of sellers and buyers plus 1. Second, these buyers must be linked to a smaller number of sellers. It can be shown that, assuming NET,9 a buyer–seller tree with k sellers always contains a strong power subnetwork if it has at least 1.5k þ (1/2) buyers if k is odd, and 1.5k if k is even. 10 Hence in the limit, buyer–seller trees with a proportion of sellers equal to .4 or less always contain a strong power subnetwork. Consequently, all simulated buyer–seller networks with three times as many buyers than sellers contain a strong power subnetwork. To examine the degree of density on buyer–seller networks, I consider two cases; the case of an equal number of buyers and sellers and the case of more

224

MARCEL A. L. M. VAN ASSEN

buyers than sellers. In the first case, a network can contain a strong power subnetwork, but only if there are at least two of these subnetworks and if NZ6. The simplest example is the tree with one seller connected to three buyers of which one is connected to three sellers. If in the first case strong power subnetworks exist in a tree, then randomly adding links between buyers and sellers will likely connect E actors that belong to different strong power subnetworks. Hence increasing degree will decrease strong power in networks with an equal number of buyers and sellers, with no strong power in a fully connected buyer–seller network. In the second case, trees can also contain strong power subnetworks in which the buyer obtains all surplus. The simplest example is the tree with one seller connected to four buyers of which one is connected to three sellers. Randomly adding links to such trees has two effects. First, possibly existing strong power subnetworks with high power buyers will gradually disappear. Second, more sellers will become high-power actors, with all of them having strong power in a fully connected buyer–seller network. The intuition of the second result is that, whatever the order of exchanges between pairs of buyers and sellers, after each exchange a network remains in which the sellers cannot be excluded but the buyers can. To conclude, if the number of sellers and buyers is unequal, both the incidence of strong power and the number of high power positions increase in degree, as opposed to the effect of degree in the other large-scale structures.

Conclusions and Questions The results discussed in previous results show that degree and clustering are negatively associated to strong power in all large-scale structures except buyer–seller networks. However, there is more to strong power than simply degree and clustering. For instance, consider the network in Fig. 1(c) and connect B1 with B2. Then it is still a strong power network, although average degree is substantial (d ¼ 2.8), pd 1 ¼ 0, average clustering is very high (C ¼ .8), and pc0 ¼ 0. Hence we can expect to explain only part of the incidence of strong power with clustering and degree. To answer the question to what extent pd 1 , pc0 , C, N, d can explain the incidence of strong power, I simulate networks of the large-scale structures. I also use the results of the simulation to examine in which of the large-scale structures uniform, exponential, and scale-free the incidence of strong power is highest. Moreover, the simulations will reveal for each structure at what average degree strong power is a rarity.

The Incidence of Strong Power in Complex Exchange Networks

225

DESIGN AND ALGORITHM OF SIMULATION STUDY The dependent variables in the simulation are incidence of strong power networks or the proportion of networks containing a strong power subnetwork, and the frequency of high power positions. These variables are examined as a function of Nr50 for trees, by simulating 10,000 networks of each size. The independent variables manipulated in the simulation for the other largescale structures are network size and average degree. Network sizes examined are 12, 24, 36, 72, and 144, values of average degree are 2, 4, 6, 8, and 10, for the seven large-scale network structures; small worlds with m ¼ .1 and m ¼ .5, the two buyer–seller structures, and uniform, exponential, and scale-free networks. Networks were not generated for each of the 175 structure size  degree combinations. Some combinations were excluded since randomly generated networks were seldom connected, others were excluded because they were physically impossible or redundant.11 In total, the design contains 17517 ¼ 158 cells. For each of these cells, networks were randomly generated with the algorithms described in the ‘‘Complex Networks’’ section. One crucial procedure of the algorithm to locate the high-power positions in the network examines whether actors can be excluded or not and labels actors accordingly (E or I, respectively). I used the procedure also used by Willer, Van Assen, and Emanuelson (2010). For a given network, the following procedure was run 10,000 times to label the actors: (i) randomly choose an exchange relation to be carried out, (ii) delete all relations to these actors, (iii) if the network still contains relations, go back to step (i). The procedure finds a random maximal exchange pattern, that is, a sequence of exchanges that leaves no actors who can still exchange. If a position is included in all 10,000 randomly selected maximal exchange patterns, it is considered an I position, otherwise an E. The probability of making errors using this procedure is minimal. For instance, if the actual probability of inclusion of a position is .999, it will be falsely categorized as an I position with a very low probability equal to .000045. An additional justification for using this procedure instead of an exact procedure is that an actor who is excluded so infrequently will very likely not experience exclusion at all and will act and be treated as an actor that is always included.

RESULTS OF THE SIMULATION First I discuss the results on trees and networks with d ¼ 2 of different largescale structures. The second subsection discusses the results for the other

226

MARCEL A. L. M. VAN ASSEN

structures. In the third subsection, I ran ANOVAs and ANCOVAs to assess how much variance of strong power can be explained by pd 1 , C, pc0 , N, and d. Strong Power in Trees and Networks with d ¼ 2 The incidence of strong power as a function of N is in trees depicted in Fig. 2. The function seems erratic for small N. However, for NW7, 12 it actually shows that the incidence increases for both even and odd size but that the incidence of strong power starts out higher for trees of odd size. Recall that trees are buyer–seller networks and that strong power is more likely present in networks with more buyers than sellers, which is the case in networks with odd size. Among trees of odd size, at least 95% (99%) of trees of size 11 (23) contain strong power. For trees of even size, these percentages are obtained for

Fig. 2.

Incidence of Strong Power in Trees as a Function of Network Size.

The Incidence of Strong Power in Complex Exchange Networks

Fig. 3.

227

Average Number of High-Power Actors in Trees as a Function of Network Size.

size 18 and 24, respectively. Fig. 3 depicts the average number of high power actors in the network as a function of N. Except for No9 the relationship is linear; 99.2% of the variance of the average number of high power actors is explained by a linear function of N. The slope of this function is .177. Table 2 presents both the incidence of strong power and the number of high power actors for all other large-scale structures with d ¼ 2. Both variables show the same and expected pattern; the order in which strong power is present in the structures is buyer–seller with more buyersWscalefreeWexponentialWuniformWsmall world with m ¼ .5Wsmall world with m ¼ .1. The incidences show that, except for small worlds with low m, networks of size 24 or more with small degree very likely contain at least one strong power subnetwork. The results on the average number of high-power positions reveal that many of the actors in the exponential and scale-free networks are involved in strong power subnetworks; about 17% of the

228

MARCEL A. L. M. VAN ASSEN

Table 2. Incidence of Strong Power and the Average Number of High-Power Actors as a Function of N (in Columns), d (in Rows), and Large-Scale Structure (in Blocks). Buyer–Seller Networks with More Buyers Than Sellers 12

24

2

2.86 1.00

5.41 1.00

4

3.00 1.00

5.71 1.00

8.01 1.00

14.74 1.00

6

6.00 1.00

8.95 1.00

17.35 1.00

33.73 1.00

8

6.00 1.00

9.00 1.00

17.96 1.00

9.00 1.00

18.00 1.00

10

36

72

Buyer–Seller Networks with Equal Number of Buyers and Sellers 144

12

24

.96 .40

2.45 .82

.00 .00

.02 .02

.06 .05

.22 .17

.56 .36

.00 .00

.00 .00

.00 .00

.00 .00

.02 .02

35.69 1.00

.00 .00

.00 .00

.00 .00

.00 .00

35.98 1.00

.00 .00

.00 .00

.00 .00

.00 .00

Small worlds with m ¼ .1

36

72

144

Small worlds with m ¼ .5

2

.09 .05

.11 .06

.13 .08

.24 .13

.50 .26

1.10 .53

2.03 .78

2.97 .89

5.75 .99

11.68 1.00

4–10

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

Exponential

Scale-free

2

2.14 .85

4.21 .98

6.37 1.00

12.68 1.00

25.77 1.00

2.27 .98

4.24 1.00

6.10 1.00

11.71 1.00

22.92 1.00

4

.03 .01

.04 .01

.02 .01

.02 .01

.00 .00

.39 .14

1.18 .35

1.97 .50

4.37 .76

8.78 .95

6

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

.01 .00

.02 .01

.04 .01

.08 .01

8–10

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

Uniform 2

1.48 .69

2.88 .91

4

.00 .00

.05 .04

.11 .09

.27 .19

.61 .38

6

.00 .00

.00 .00

.00 .00

.00 .00

.01 .01

8–10

.00 .00

.00 .00

.00 .00

.00 .00

.00 .00

Notes: The first and second number in each cell represent the average number of high-power actors and incidence, respectively

The Incidence of Strong Power in Complex Exchange Networks

229

actors in exponential and scale-free networks has strong power, a percentage that hardly varies across network sizes.

Strong Power in Networks with dZ4 Table 2 shows the results on strong power for the large-scale structures with dZ4. The results on small worlds, buyer–seller, and uniform networks were as anticipated. No strong power was found in small networks, even at d ¼ 4. All buyer–seller networks with three times as many buyers than sellers contained a strong power network, and the number of sellers having strong power increased in d until (almost) all had it for d ¼ 10. Strong power was negatively associated to d in buyer–seller networks with an equal number of buyers and sellers; a very low incidence (.02) of strong power is observed in large networks (N ¼ 144) with d ¼ 6. Results for uniform networks were similar; whereas strong power is observed in a small proportion of uniform networks of large size with d ¼ .4, incidence is very low (.04) in large networks (N ¼ 144) with d ¼ 6. One open question is the incidence of strong power in exponential and scale-free compared to uniform networks. The incidence turned out to be lowest in exponential networks, for which hardly any power was observed for d ¼ 4. Incidence in scale-free networks was larger than that in uniform networks. The majority of scale-free networks of large size (NZ72) contained at least one strong power subnetwork, and there was even a low incidence (.01) of strong power subnetworks of large size at d ¼ 6.

Explaining Strong Power Before explaining variance of strong power, I examined the associations between its predictors, that is, I examined pd 1 , C, and pc0 as a function of N, d, and structure. Table 3 shows the average values of pd 1 , C, and pc0 as a function of N, d, and structure. With a few exceptions the values were as anticipated in Table 1. The exceptions involved the comparison of uniform networks and small worlds with m ¼ .5 with respect to C (not larger for small worlds of small size, i.e., Nr24) and pc0 (not larger for uniform networks or small size (Nr24) and d ¼ 4). Density is so high in these small networks that clustering in uniform networks is large and comparable to that in the small worlds. The fact that clustering in uniform networks was

230

MARCEL A. L. M. VAN ASSEN

Average Values of pd 1 , C, pc0 as a Function of N (in Columns), d (in Rows), and Large-Scale Structure (in Blocks).

Table 3.

Buyer–Seller Networks with More Buyers Than Sellers 12

24

2

.53 .00 .47

.54 .00 .46

4

.02 .00 .98

.12 .00 .88

.14 .00 .86

.16 .00 .84

6

.01 .00 .99

.03 .00 .97

.04 .00 .96

.05 .00 .95

8

.00 .00 1.0

.00 .00 1.0

.01 .00 .99

.00 .00 1.0

.00 .00 1.0

10

36

72

144

Buyer–Seller Networks with Equal Number of Buyers and Sellers 12

24

36

72

144

.33 .00 .67

.35 .00 .65

.01 .00 .99

.04 .00 .96

.05 .00 .95

.07 .00 .93

.07 .00 .93

.00 .00 1.0

.00 .00 1.0

.00 .00 1.0

.01 .00 .99

.01 .00 .99

.01 .00 .99

.00 .00 1.0

.00 .00 1.0

.00 .00 1.0

.00 .00 1.0

.00 .00 1.0

.00 .00 1.0

.00 .00 1.0

.00 .00 1.0

.00 .00 1.0

Small worlds with m ¼ .1

Small worlds with m ¼ .5

2

.09 .02 .89

.09 .01 .91

.09 .00 .91

.09 .00 .91

.09 .00 .91

.30 .06 .62

.30 .02 .67

.31 .01 .68

.30 .00 .69

.30 .00 .69

4

.00 .43 .02

.00 .39 .03

.00 .39 .04

.00 .38 .04

.00 .38 .05

.00 .32 .14

.00 .18 .35

.00 .14 .45

.00 .10 .56

.00 .08 .63

6

.00 .55 .00

.00 .48 .00

.00 .46 .00

.00 .45 .00

.00 .45 .00

.00 .52 .00

.00 .26 .03

.00 .19 .08

.00 .13 .18

.00 .10 .27

8

.00 .69 .00

.00 .53 .00

.00 .50 .00

.00 .49 .00

.00 .48 .00

.00 .72 .00

.00 .34 .00

.00 .24 .00

.00 .15 .03

.00 .12 .08

10

.00 .90 .00

.00 .56 .00

.00 .53 .00

.00 .51 .00

.00 .50 .00

.00 .91 .00

.00 .42 .00

.00 .29 .00

.00 .18 .00

.00 .13 .01

.66 .01 .30

.66 .00 .31

Exponential 2

.48 .17 .27

.49 .06 .38

.50 .03 .42

Scale-free .50 .01 .46

.50 .00 .48

.59 .20 .16

.64 .06 .23

.65 .03 .26

231

The Incidence of Strong Power in Complex Exchange Networks

Table 3.

(Continued )

Exponential 4

Scale-free

.00 .48 .20 .00 .67 .01

.00 .24 .43 .00 .35 .11

.00 .16 .53 .00 .24 .20

.00 .08 .69 .00 .12 .38

.00 .04 .80 .00 .06 .55

.00 .61 .11 .00 .73 .00

.00 .40 .30 .00 .49 .05

.00 .31 .40 .00 .37 .10

.00 .20 .56 .00 .20 .23

.00 .12 .68 .00 .15 .38

8

.00 .83 .00

.00 .46 .01

.00 .31 .05

.00 .16 .16

.00 .08 .32

.00 .85 .00

.00 .56 .00

.00 .44 .02

.00 .28 .07

.00 .17 .17

10

.00 .95 .00

.00 .55 .00

.00 .38 .01

.00 .20 .05

.00 .10 .16

.00 .95 .00

.00 .64 .00

.00 .49 .00

.00 .32 .02

.00 .20 .07

6

Uniform 2

.35 .07 .55

.36 .02 .61

4

.03 .36 .14

.06 .17 .35

.06 .11 .46

.07 .05 .63

.07 .03 .75

6

.00 .54 .01

.01 .26 .07

.01 .17 .15

.01 .08 .33

.01 .04 .53

8

.00 .73 .00

.00 .35 .01

.00 .23 .03

.00 .11 .11

.00 .06 .27

10

.00 .91 .00

.00 .43 .00

.00 .29 .00

.00 .14 .03

.00 .07 .11

Notes: The first, second, third number in each cell correspond to pd 1 , C, pc0 , respectively.

sometimes even higher than in small worlds is the result of the difference in the algorithms to generate the two structures. I ran three two-way ANOVAs with N (5 levels) and d (5 levels) as factors, and with pd 1 , C, and pc0 as dependent variables, for each large-scale structure separately. Hence in total 7 (structures)  3 (dependent variables) two-way ANOVAs were conducted. These analyses revealed that, as anticipated, pd 1 , C, and pc0 are very highly correlated with N and in particular d. The lowest explained variance (.73) was attained for pd 1 in the small world structure with m ¼ .1. The explained variances in the other 20

232

MARCEL A. L. M. VAN ASSEN

ANOVAs were at least .92. Two conclusions can be drawn from the results of these ANOVAs. First, conditional on the large-scale structure, network characteristics N and d determine to a very large extent the properties pd 1 , C, and pc0 that are shown to be relevant for strong power. Second, when predicting strong power with pd 1 , C, pc0 , N, and d there will be multicollinearity and each variable’s unique contributions to the explanation of strong power will be small. Because of high multicollinearity and low unique contributions, I report only the total variance explained of strong power in a two-way ANCOVA with factors N and d and covariates pd 1 , C, and pc0 . I ran one ANCOVA for each structure, for both dependent variables incidence and number of high power actors, resulting in 14 analyses. The results for both dependent variables were similar. I report the results on incidence between brackets directly after the results on the number of high power actors. Only 19.2% (21.6%) of the variance of strong power in the small world with m ¼ .1 was explained. These low percentages can be explained by the absence of strong power in this structure, that is, there was hardly any variance in strong power to be explained. Substantially more variance of strong power, 65.4% (63.8%), was explained for the uniform networks. A little less variance was explained for the buyer–seller networks with an equal number of buyers and sellers, 57.6% (55.7%). In small worlds with m ¼ .5, 85.1% (87.5%) was explained of the variance. Very high percentages of the variance were explained for scale-free, exponential, and buyer–seller networks with more buyers: 93.4% (83.6%), 95.8% (95.8%), 99.7%, 13 respectively. To conclude, the macrolevel (N, d ) and microlevel (pd 1 , C, pc0 ) network characteristics explain together a substantial to large part of the variance of the incidence of strong power and the number of high power actor positions.

DISCUSSION AND CONCLUSION Research on exchange in networks brings to light that an actor’s network position has an immense effect on the actor’s exchange outcome. If the network structure favors some actors that obtain (almost) all surplus in their exchanges with others, it is said to contain strong power. Research on network exchange, which only examined small networks, suggests that the incidence of strong power is low and decreases in network size. The present study attempted to increase our understanding of strong power by examining its incidence in complex exchange networks of large size and

The Incidence of Strong Power in Complex Exchange Networks

233

relating this incidence to characteristics of these networks. I assumed NET in this examination. The theoretical and simulation results reveal that the effect of size on the incidence of strong power is positive in trees and in sparse networks with average degree equal to 2, whereas the effect of size is absent or weakly positive for more densely networks, conditional on degree. The reason that the incidence of strong power seemed to decrease in size in previous research is that previous research did not control for degree; a lower incidence of strong power was observed for larger networks, since the proportion of networks with low degree decreases in network size. Two microlevel properties of the network, actor’s degree and actor’s clustering, are shown to determine, at least partly, whether an actor has high power. The proportion of actors in the network with degree 1, pd 1 , and the proportion of actors with no clustering, pc0 , are positively associated to strong power, conditional on N and d, in all large-scale structures except buyer–seller networks. Of these two properties pd 1 is most important, since clustering is neither a necessary nor a sufficient condition for strong power. Both N and d affect the two properties, but d has a much stronger effect. Increasing d negatively affects both properties, and hence destroys strong power in all large-scale structures except buyer–seller networks. Properties pd 1 and pc0 are also affected by the large-scale network structure; hence, the incidence of strong power also varies across large-scale structures. The large-scale structures investigated are found in large real-life social networks and include small worlds, buyer–seller, exponential, and scale-free networks. The simulation results on very sparse networks of these large-scale structures show that strong power can be found in most networks of size 24 or larger, except in small worlds that resemble ring networks. Strong power was observed much less frequently in networks with average degree of 6 or more, with the exception of buyer–seller networks in which strong power was ubiquitous. For each structure, except small worlds resembling rings, most variance of strong power could be explained by the network characteristics based on degree and clustering. One obvious question is to what extent the obtained results are robust to the theory with which the analyses were carried out and to the algorithms with which the simulations were carried out. Many theories of network exchange predict strong power, and all concern the probability to be included in an exchange and, consequently, whether an actor has degree 1, and if an actor together with other actors can collectively exchange only with a smaller group of other actors. Hence the two identified properties pd 1 and pc0 are relevant for strong power, independent of the theory considered.

234

MARCEL A. L. M. VAN ASSEN

Predictions on the incidence of strong power, however, do depend on the theory assumed. For instance, core theory (Bienenstock & Bonacich, 1992) and theories of network exchange in economics (Kranton & Minehart, 2001; Corominas-Bosch, 2004) assume that only maximal exchange patterns occur with a maximum number of exchanges. Assuming a maximum number of exchanges, more actors can have high power, which results in a higher incidence of strong power. Predictions for these theories can be generated for large networks using the algorithm described in ‘‘Effects of Network Characteristics on Strong Power’’ section by labeling actors E or I only using exchange patterns with a maximum number of exchanges. The results of the simulation are not robust to specifications of the algorithm used to generate the networks. I used common procedures to generate small worlds, and exponential and scale-free networks. These procedures specify that each actor in the network has at least (1/2)d links. Changing the linking mechanism will have a strong effect on the incidence of strong power in these networks, if the change increases the proportion of actors having only one link. For instance, if the algorithm for exponential and scale-free networks is changed such that some actors link with only actor, then more strong power will be observed in these networks. Hence the results of the simulation should be interpreted with caution; they mainly provide an illustration of the effect of degree and size on strong power in different structures. The main conclusion of the present study is robust to the theory assumed and to specifications of the algorithms; the incidence of strong power is high for small degree, but low for large degree in all large-scale structures except small worlds resembling rings (no strong power) and buyer–seller networks (much strong power). Hence the incidence of strong power in real-life social exchange networks mainly depends on the density of the network. For low density, the large-scale structure of the network matters. Density and structure of real-life social networks are determined by many factors, such as geographical proximity, not having complementary interests, costs of linking, and the process of linking. It was assumed in the analyses that the network structure is exogenous. However, in real life, actors can to a certain extent manipulate how many links they have and to whom they are linked. Obviously, actors will attempt not to be taken advantage of and avoid being an E actor in a strong power subnetwork. Van Assen and Van de Rijt (2007), assuming expected value theory (Friedkin, 1995), showed that adding (deleting) a link increases (decreases) the gains of actors involved if before addition (deletion) they together earned in total not more (less) than the surplus generated by the link.

The Incidence of Strong Power in Complex Exchange Networks

235

Dog˘an, Van Assen, Van de Rijt, and Buskens (2009) show that, in a network with myopic actors freely changing their links to maximize their payoffs, strong exchange networks are never stable, that is, E actors in strong power subnetworks change their position such that they get more of the surplus. Dog˘an et al. (2009), however, also showed that the average degree of stable exchange networks is small when there are small costs of linking. Since in reallife applications, actors cannot freely link with each other for aforementioned reasons, and some costs of linking are involved, it is not evident that all actors can ‘‘escape’’ being an E actor in a strong power subnetwork. The only way to examine the incidence of strong power in large-scale social applications is to investigate social networks in the field. For instance, what is the incidence of strong power in exchange networks of pupils in a class, and of exchange networks of employees within a department or company? The algorithm developed by Willer et al. (2010) makes the application of NET to these larger social networks possible and brings closer the much wanted step from the lab to the field. Examining these networks will provide the first test of both real-life social networks and large exchange networks.

NOTES 1. Applying the binomial approximation yields a probability pd 1 equal to (N1)p(1p)(N2). Substituting p ¼ d/(N1) yields pd 1 ¼ d½1  d=ðN  1ÞðN2Þ . As N-N this can be simplified to ded, which decreases in d. The approximation ded is better for low d and large N. 2. Let p(di ¼ k) be the probability that actor i has degree k. Then, using the binomial approximation, pc0 ¼ pðd i ¼ 2Þð1  pÞ2 þ pðd i ¼ 3Þð1  pÞ3 þ . . . . Given d, increasing N results in a decrease of p, since p ¼ d/(N1). Decreasing p leads to an increase in each of the terms of pc0 , and hence in an increase of pc0 . As N-N proportion pc0 increases to its upper bound 1ded, which equals the probability that an actor has more than one link. 3. If i ¼ N, then i þ 1 ¼ 1. 4. This rewiring is not carried out if the actor has N1 links. 5. More specifically, in uniform networks and small worlds, the distributions of di and Ci are identical for all actors and have low variance. The distributions of di and Ci are different across actors for exponential and scale-free networks, since they depend on the order of appearance of i in the network. P 6. More formally, the probability of a link with actor k equals d k = Ii¼1 d i , where I denotes the number of actors that are already present in the network. The linking mechanism is sequential, that is, after linking with actor k the probabilities are updated by removing dk from the denominator. 7. The probability of a strong power subnetwork increases quite rapidly in pd 1 . Let a network have k actors with degree 1 and assume that these actors are randomly

236

MARCEL A. L. M. VAN ASSEN

connected to the remaining N-k actors in the network. Then the probability of a strong power subnetwork with at least two actors with degree 1 connected to the same I actor equals 1  Pki¼2 ½ðN  k  i þ 1Þ=ðN  kÞ. For example, consider a network with N ¼ 100 and k ¼ 5. Then the probability of no strong power subnetwork equals the probability that none of these five actors is connected to the same actors. This probability equals (94/95)  (93/95)  (92/95)  (91/95) ¼ 90, hence the probability of such a strong power subnetwork equals .10. If k ¼ 10 the probability of a strong power subnetwork is already .40. Note, however, that the networks are not generated by randomly connecting actors with degree 1 to other actors, but using another chance mechanism. Hence the probabilities derived in this footnote should be considered as an approximation. 8. Assume that the network contains k, E actors in strong power subnetworks. Randomly adding one link connects two of these actors with probability (k/N)  (k1/N1), which increases their power and possibly destroys the strong power subnetwork in which they are contained. If one of these E actors is connected to an actor outside a strong power subnetwork, strong power can also be destroyed if that other actor is not a high power actor after establishing this connection. 9. Most theories of exchange in economics (e.g., Kranton & Minehart, 2001; Corominas-Bosch, 2004) and also Core theory (Bienenstock & Bonacich, 1992) assume that a maximum number of exchanges is carried out. Consequently, they predict strong power if the number of sellers is not equal to the number of buyers. 10. The proof proceeds as follows; k sellers are connected to buyers on a line. Assume that sellers 1, 2, etc., are on position 2, 4, etc., of the line, respectively. The longest line that is a strong power subnetwork is a 5-line. The 5-line remains a strong power network if the high power sellers are connected to other buyers. Hence, to prevent strong power, additional buyers should be connected to a line of any length such that no strong power 5-line is created. This can be achieved by connecting one buyer to seller 3, one buyer to seller 5, etc. Adding more buyers will create a strong power subnetwork. 11. A combination is redundant if the buyer–seller network is fully connected. 12. All trees of size 3 and 5 (the 5-Line, the ‘‘T,’’ and the network with one actor connected to the other four) are strong power. 13. No analysis could be run on the incidence of strong power in buyer–seller networks with more buyers, since there was no variance; all networks contained a strong power subnetwork.

REFERENCES Albert, R., Jeong, H., & Baraba´si, A. (1999). Internet: Diameter of the world-wide web. Nature, 401, 130. Baraba´si, A., & Albert, R. (1999). Emergence of scaling in random networks. Science, 286, 509–512. Bienenstock, E. J., & Bonacich, P. (1992). The core as a solution to exclusionary networks. Social Networks, 14, 231–244. Blau, P. M. (1964). Exchange and power in social life. New York: Wiley.

The Incidence of Strong Power in Complex Exchange Networks

237

Bolloba´s, B. (2001). Random graphs (2nd ed). Cambridge: Cambridge University Press. Buskens, V., & Van de Rijt, A. (2008). Sequential power-dependence theory. The Journal of Mathematical Sociology, 32, 110–128. Chamberlin, E. H. (1948). An experimental imperfect market. The Journal of Political Economy, 56, 95–108. Cook, K. S., & Emerson, R. M. (1978). Power, equity and commitment in exchange networks. American Sociological Review, 43, 721–739. Corominas-Bosch, M. (2004). Bargaining in a network of buyers and sellers. Journal of Economic Theory, 115, 35–77. Dog˘an, G., Van Assen, M. A. L. M., Van de Rijt, A., & Buskens, V. (2009). The stability of exchange networks. Social Networks, 31, 118–125. Dunbar, R. I. M. (1993). Coevolution of neocortex size, group size and language in humans. Behavioral and Brain Sciences, 16, 681–735. Edgeworth, F. Y. (1881). Mathematical physics. Fairfield: Kelley. Emanuelson, P. (2005). Improving the precision and parsimony of network exchange theory: A comparison of three network exchange models. Current Research in Social Psychology, 10, 149–165. Friedkin, N. (1995). The incidence of exchange networks. Social Psychology Quarterly, 58, 213–222. Homans, G. C. (1958). Social behavior as exchange. American Journal of Sociology, 62, 597–606. Kranton, R. E., & Minehart, D. F. (2001). A theory of buyer-seller networks. The American Economic Review, 91, 485–508. Lucas, J. W., Younts, C. W., Lovaglia, M. J., & Markovsky, B. (2001). Lines of power in exchange networks. Social Forces, 80, 185–214. Malinowski, B. (1922). Argonauts of the western Pacific. New York: E.P. Dutton. Markovsky, B. (1995). Developing an exchange network simulator. Sociological Perspectives, 38, 519–545. Markovsky, B., Willer, D., & Patton, T. (1988). Power relations in exchange networks. American Sociological Review, 53, 220–236. Molm, L. D. (1997). Coercive power in social exchange. Cambridge: Cambridge University Press. Polanski, A. (2007). Bilateral bargaining in networks. Journal of Economic Theory, 134, 557–565. Schnettler, S. (2009). A structured overview of 50 years of small-world research. Social Networks, 31, 165–178. Stolte, J. F., & Emerson, R. M. (1977). Structural inequality: Position and power in exchange structures. In: R. L. Hamblin & J. H. Kunkel (Eds), Behavioral theory in sociology (pp. 117–138). New Brunswick, NJ: Transaction Books. Thibaut, J., & Kelley, H. H. (1959). The social psychology of groups. New York: Wiley. Van Assen, M. A. L. M. (2003). Exchange networks: An analysis of all networks up to size 9. Advances in Group Processes, 20, 67–103. Van Assen, M. A. L. M., & Van de Rijt, A. (2007). Dynamic exchange networks. Social Networks, 29, 266–278. Van de Rijt, A., & Van Assen, M. A. L. M. (2008). Theories of network exchange: Anomalies, desirable properties, and critical networks. Social Networks, 30, 259–271. Vega-Redondo, F. (2007). Complex social networks. New York: Cambridge University Press.

238

MARCEL A. L. M. VAN ASSEN

Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. Cambridge: Cambridge University Press. Watts, D. J. (1999). Networks, dynamics, and the small-world phenomenon. The American Journal of Sociology, 105, 493–527. Watts, D. J., & Strogatz, S. H. (1998). Collective dynamics of ‘small-world’ networks. Nature, 393, 440–442. Willer, D. (1987). Theory and the empirical investigation of social structures. New York: Gordon and Breach Science Publishers. Willer, D. (Ed.) (1999). Network exchange theory. Westport, CT: Praeger Press. Willer, D., & Emanuelson, P. (2008). Testing ten theories. Journal of Mathematical Sociology, 32, 165–203. Willer, D., Van Assen, M. A. L. M., & Emanuelson, P. (2010). Analyzing large scale exchange networks. Submitted for publication.

MULTIPLEX EXCHANGE RELATIONS Ko Kuwabara, Jiao Luo and Oliver Sheldon ABSTRACT A multiplex relation occurs when actors share different roles, actions, or affiliations that overlap in a relationship, such as co-workers who are also friends outside of work. Although multiplex relations are as varied as they are pervasive and often problematic, we know surprisingly little about when, under what circumstances, and exactly how overlapping ties affect social relations. Do they strengthen or weaken relationships? When do relationships become multiplex? How do they affect networks at large? In this chapter, we review notable studies that exist on this topic and suggest key questions and issues for future research. Our goal in particular is to suggest how exchange theory could contribute to these efforts.

A friendship founded on business is a good deal better than a business founded on friendship. – John D. Rockefeller Don’t think of him as a Republican. Think of him as the man I love, and if that doesn’t work, think of him as the man who can crush you. – Maria Shriver Luke, I am your father. – Darth Vader Advances in Group Processes, Volume 27, 239–268 Copyright r 2010 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0882-6145/doi:10.1108/S0882-6145 (2010)0000027012

239

240

KO KUWABARA ET AL.

From siblings running a business together to business associates entertaining clients on the golf course, and spouses belonging to different political parties to fathers who are mortal enemies, multiplex relations are as varied as they are pervasive. Multiplex relations occur when individuals share ‘‘multiple bases for interaction’’ with each other (Verbrugge, 1979, p. 1287; Marsden & Campbell, 1984; Wasserman & Faust, 1994), such as when the actors in a relationship play different roles (Ibarra, 1995; Zelizer, 2005), maintain different affiliations (Wheeldon, 1969), or engage in different types of activities and exchanges (Kapferer, 1969). These overlapping ties create a ‘‘coexistence of different normative elements in a social relationship’’ (Gluckman, 1962; also Foa & Foa, 1980; Clark & Mills, 1979), which can solidify or strain the relationship or alter the nature of the relationship altogether.1 On the one hand, it is commonly assumed that people enjoy thicker, stronger, and more durable relations when they share multiple ties to one another (Marsden & Campbell, 1984; Coleman, 1988; Hardin, 2002). The world of business is particularly rife with stories of weekend golf and extravagant wining-and-dining to build client relations. On the other hand, however, we are all too familiar with cautionary tales about mixing private and professional affairs. The tension that arises from incompatible norms can strain relationships by causing role conflict, miscommunication, or misalignment of mutual interests and expectations (Ingram & Zou, 2008). The trials and tribulations of mixing business and pleasure, teammates and rivals, or friends and lovers notwithstanding, we know surprisingly little about when, under what circumstances, and exactly how overlapping ties affect social relations. To date, research on multiplex relations has been largely correlational or descriptive, with little attention to the causal processes underlying multiplex relations: How do overlapping ties affect social relations? When do they strengthen or weaken relationships? When do relationships become multiplex? Our goal in this chapter is to suggest key questions and an agenda for leveraging theoretical and methodological tools of microsociology and groups research to advance our understanding of multiplex relations. To this end, the remainder of the chapter is organized as follows. First, we discuss why more scholarly attention to multiplex relations is needed. Second, we discuss in greater detail what we still do not know about multiplex relations and what questions remain unanswered. Finally, we conclude by discussing how exchange theory can contribute to future research on multiplex relations. We draw our inspiration for this chapter from Ingram and Zou (2008), who call for a more focused research program on multiplex relations in

Multiplex Exchange Relations

241

organizational contexts. Perhaps more than other types of multiplex relations, they argue, business friendships present an uneasy tension between the rational pursuit of instrumental goals and sharing affective ties. Our agenda in this chapter differs in two ways. First, whereas Ingram and Zou (2008) propose a compelling agenda for studying multiplex relations in the context of business and organizational life, our goal is to make a case for abstracting away the context to unpack the causal mechanisms underlying the formation of multiplex relations and their consequences for social relations and networks at large. Second, we note that business friendship represents only one type of multiplex relations, one in which affective and instrumental ties intermingle. However, relations can be multiplex on the basis of two instrumental ties or two affective ties if they introduce different norms of interaction (e.g., a parent and child who simultaneously think of themselves as close friends). Thus, without discounting the case for business friendship, we argue in this chapter that a richer understanding of multiplex relations requires a deeper level of analysis to zero in the causal and dynamic processes underlying multiplexity and broadening the scope of analysis beyond business contexts.

WHAT ARE MULTIPLEX RELATIONS? Following Verbrugge (1979, p. 1287), we use the term multiplexity to denote overlapping roles, actions, and affiliations within a relationship. Roles, actions, and affiliations define, and are defined by, different normative contexts of interaction. This definition distinguishes multiplex relations from repeated interactions. Although multiplex relations necessarily involve repeated interactions, merely repeating interactions does not constitute multiplex relations if the actors interact on the basis of the same roles, actions, or affiliations in every instance or on the basis of different roles, actions, and affiliations that do not overlap in time and space. Becoming friends with a co-worker after finishing a project together without sharing professional ties again is not a multiplex relationship in the full sense of its definition if the friendship and the business ties do not overlap. In other words, adjusting to different norms of interaction as role relations change is not the same as managing multiple norms of interaction simultaneously. An example of the latter is co-workers who stop working together to pursue a romantic relationship. In this case, the couple no longer interacts with each other as lovers and co-workers at the same time.

242

KO KUWABARA ET AL.

It is also important to note that multiplex relations can exist between people as well as organizations. Examples of multiplex relations between firms include multiple, overlapping interpersonal ties across upper echelons of organizations (e.g., interlocking directorates; Haunschild & Beckman, 1998), multiple interorganizational ties in which firms compete or cooperate in various functional areas (Powell, White, Koput, & Owen-Smith, 2005; Lomi & Pattison, 2006; Westphal, Gulati, & Shortell, 1997), or combinations of both (e.g., Burt, 1980; Beckman & Haunschild, 2002; Gulati & Westphal, 1999). In this chapter, we focus on relations in which actual human decision makers are directly implicated, consistent with the key idea that multiplex relations invoke ‘‘different normative elements’’ (Gluckman, 1962). Thus, we include research on interfirm relations that are based on personal connections, such as board interlocks. We mention research on interfirm relations that do not implicate interpersonal relations very selectively, only when it explicitly discusses the concept of multiplexity. We acknowledge that how ties with different relational norms overlap in multiplex relations between firms as opposed to between human actors is a critical question for future research.

WHY SHOULD MULTIPLEX RELATIONS BE STUDIED? Social scientists in general and sociologists in particular have long shown great interest in various manifestations of multiplex relations, though typically in the service of other research purposes. For instance, within networks research, multiple ties and networks are often sampled to map the network more completely (Marin & Hampton, 2007). Johnson and Miller (1986) note in their study of Alaska fisherman networks that measuring any single network (i.e., kinship, economic exchange or co-residence) alone will not accurately capture the social structure of the community. Similarly, Gould (1991) notes that the social movement that led to the Paris Commune could not be characterized adequately without having information on multiple types of ties. Others have studied multiplex relations only insofar as they have measurable consequences for the focal individuals or organizations, such as their economic performance (Ingram & Roberts, 2000; Valley, Neale, & Mannix, 1995a; Rangan & Sengul, 2009), antisocial behavior (Brass, Butterfield, & Skaggs, 1998; Krohn, Massey, & Zielinski, 1988), or organizational stability (Stern, 1979; Human & Provan, 2000). Very few

Multiplex Exchange Relations

243

studies so far have focused explicitly on multiplex relations themselves or the dynamics of interactions between actors who share overlapping ties as dependent variables (e.g., Uzzi, 1997; Lazega & Pattison, 1999; Lomi & Pattison, 2006; Lee & Monge, 2009).

The Challenges of Multiplexity in Everyday Life We suggest that multiplex relations warrant greater attention in and of themselves for several reasons. First, aside from academic debates and as a matter of practical and personal concern, people care profoundly about multiplex relations, both about their own and others. Thus begins Zelizer (2005, p. 1) in The Purchase of Intimacy: All of us sometimes gobble up the details of a famous couple’s divorce settlement, worry about whether certain children are suffering from their parents’ profligate spending, become indignant when someone close to us fails to meet important economic obligations, or complain about proposals to cut funding for day-care centers.

From mundane moments of ordinary life to those that threaten our very sense of social identity or economic welfare, economic exchanges and intimate relationships frequently intertwine in ways that complicate our decisions and their consequences. Nowhere, however, are the challenges of multiplex relations more apparent than in business (Ingram & Zou, 2008). More than ever, ‘‘people try very hard to draw a firm line between the people for whom they have genuine affection and those with whom they’re involved professionally y People wonder, would they be going to the ball game with me if we weren’t doing business? Do they care about me or is it reflecting some economic interest?’’ (Ingram quoted in Columbia Ideas @ Work 2009; http://www4.gsb.columbia.edu/ ideasatwork/feature/723969/BusinessþFriendships). Implicit in such concerns is uncertainty and anxiety surrounding the true motives of those with whom we share multiplex relations. Entrepreneurs are particularly aware of both the necessity and the inevitability of crossing professional and personal boundaries to connect with people who can provide valuable resources and support and the complications such ties can create (Dubini & Aldrich, 1991). A regular feature in INC, a popular periodical for young entrepreneurs, is a monthly column entitled ‘‘Balancing Acts’’ covering topics and anecdotes on managing work–family boundaries. One apparent reason why we are concerned about managing overlapping ties is that multiplex relations are ubiquitous, forcing us to constantly

244

KO KUWABARA ET AL.

negotiate the challenges they present. Admittedly, there is growing evidence that we, at least in the United States, are becoming more isolated and fragmented as members of the society. According to Putnam (2000), Americans are increasingly less likely to be members of voluntary organizations and local clubs. And according to McPherson, Smith-Lovin, and Brashears (2006), the average number of confidents that Americans maintain for personal or intimate conversations has declined, from roughly 3 in 1985 to 2 today, and the modal number of confidants has decreased from 3 to 0. Indeed, Smith-Lovin (2007) reports that only 6 percent of the general population is connected to no-kin confidants through multiple ties, whereas 58 percent of spouses have no other overlapping ties with each other outside of their marital ties. These trends notwithstanding, multiplex relations remain pervasive outside of very close relationships. Ironically, the same forces of increasing specialization and diversification that isolate us from each other in today’s complex society can also create more interdependencies and intersections where disparate domains collide and converge to create overlapping roles and ties. The increasing significance of organizational life in the contemporary society in particular has done much to blur the boundaries between the public and private sphere (Kacperczyk, Sanchez-Burks, & Baker, 2009). At work, organizations are taking down bureaucratic walls and creating flatter, more decentralized structures to promote richer social interactions among employees. Longer work hours also mean that people are more likely to form social relations within the confines of their work (Verbrugge, 1979). Additionally, advances in communication technology have significantly reduced the cost of creating and maintaining relationships while opening up new frontiers of interaction on the Internet. Through cell phone, email, and social networking sites, it is increasingly easier to befriend coworkers, clients, and superiors outside of professional contexts. At home, work–family boundaries are disappearing at an accelerating pace with the promulgation of ubiquitous computing technology that allows work to follow us home and enables us to work from home. At the same time, greater opportunities to form multiplex relations pose considerable challenges. Should one actually befriend coworkers and clients, or do such relations compromise our professional goals? Should spouses start a business together, or are ensuing dynamics likely to put too great a strain on the relationship? Questions such as these point to another compelling reason why people are wary of multiplex relations: we know relatively little about how overlapping ties actually affect our relationships, and in particular, whether they might strengthen or weaken them.

Multiplex Exchange Relations

245

Multiplexity and Tie Strength It is often assumed, both in networks research and in anecdotes, that multiplex relations are strong relations (Kapferer, 1969; Brass et al., 1998; Marsden & Campbell, 1984). The intuition is simple: when actors share multiple bases of interaction, they are more likely to share information across domains, exchange resources more frequently, and withstand external shocks to the relationship (Coleman, 1988; Ibarra, 1993; Uzzi, 1996, 1997). Multiplex relations are assumed to be particularly strong if personal or affective ties overlap with professional ties, imbuing instrumental exchange with symbolic or expressive value above and beyond the tangible value of resources changing hands. The idea of ‘‘thick’’ relationships (Hardin, 2002) dovetails with the image of relationships bound by multiple threads of connections. Similar views have been expressed by scholars of ‘‘socioeconomics,’’ who have espoused the notion that multiplex relations might not be stronger necessarily, but are seamlessly integrated and unproblematic inasmuch as they are governed by the unitary logic of rational action (e.g., Becker, 1976). Beyond interpersonal relations, research on interfirm relations has demonstrated that board interlocks, alliances, CEO association memberships, and investment ties can reinforce each other. For example, Beckman and Haunschild (2002) find evidence that strategic alliances and CEO association memberships reinforce interlocking ties to enhance learning and transfer of complex acquisition knowledge. The more multiplex relations a firm possess with its network partners, the lower the price that the firm paid for its acquisitions. An important counterpoint to these views is the observation that multiplex relations are more likely than simplex relations to create conflicts of interest and expectations that weaken relationships. Managing overlapping ties often requires reconciling conflicting role expectations and demands that arise from ‘‘the co-occurrence of different normative elements’’ (Verbrugge, 1979) in overlapping ties. For instance, research by Clark (1984) demonstrates that affective and instrumental ties (what she terms communal versus exchange relations) are governed by different norms about explicit record keeping to keep track of contributions and efforts to joint tasks. Paul Joseph, an entrepreneur, describes similar sentiments in an interview about starting a business with a friend: The adage about doing business with your friends is something we laughed at when we first got started, but I think two years into running a very profitable business together it really took its toll y because I was dealing with issues of questioning what we had done right, what we did not do right, who was responsible for what success, who was

246

KO KUWABARA ET AL.

responsible for what failure, who was living up to their end of the bargain, etcetera. So we definitely had some friction. (http://www.e-clips.cornell.edu/themes.do?id ¼ 220&clipID ¼ 3709&tab ¼ TabClipPage)

When these normative demands are incompatible with each other, they can undermine the relationship by engendering perceived norm violations and, in some cases, feelings of betrayal (Morris & Moburg, 1993). In particular, where instrumental goals and affective sentiments intersect, ‘‘many people feel that [y] economic activity – especially the use of money – degrades intimate relationships, while interpersonal intimacy makes economic activity inefficient’’ (Zelizer, 2005, p. 1). Underlying these beliefs is the pervasive assumption that economic action and affective ties are fundamentally incompatible and mutually detrimental and that failing to separate or compartmentalize them carefully leads only to tension and inefficiency. Data from the General Social Survey show that people are hesitant to make large purchases with friends and acquaintances out of similar concerns (DiMaggio & Louch, 1998). These concerns are hardly without empirical or anecdotal evidence. Chan (2009) finds that Chinese life insurance sales agents initially sold their products to close friends and relatives through high-trust relations when the idea of life insurance was relatively novel in China. As the public became more conscious of the economic gains the insurance industry was reaping, however, agents were forced to turn to moderate and weak ties for sales. Baker and Nelson (2005) find that newly founded small businesses grow faster if they avoid trading with their relatives and friends. The reason is that dealing with relatives and friends constrains the entrepreneur’s ability to pursue margin or migrate from poorer to richer client bases. Growth potential is restricted by social entanglement associated with transacting with one’s social contacts. Finally, in social psychology, a stream of research on negotiations has focused on the effects of personal relationships on negotiation processes and outcomes (McGinn, 2006). The primary finding is that personal relations compromise negotiation because people avoid negotiating with close friends or intimate partners.

Conceptualizing Multiplexity These conflicting views and evidences point to a critical gap in our understanding of interpersonal relations. Thus far, multiplex relations have been under-theorized. For instance, while researchers have attributed the macro trends in multiplex relations in the United States to a number of

Multiplex Exchange Relations

247

plausible factors, such as demographic changes or national culture, their findings have been almost exclusively correlational. Furthermore, there is little research to adjudicate exactly when multiplex relations results in stronger or weaker relations and to specify precisely when and how different normative elements are incompatible. Although the general idea of overlapping ties has been invoked in many research programs and discourses since Adam Smith’s (1759) treatise on moral sentiments, Becker’s (1976) theory of social and economic action, and Granovetter’s (1973) thesis on the strength of ties, our understanding of multiplexity remains largely correlational or descriptive in empirics and typological in theory. Scholars have made numerous distinctions between different types of ties, such as social versus economic exchange (e.g., Blau, 1964) or affective versus instrumental ties (e.g., Zelizer, 2005), yet little work has directly examined how they interact with each other. For instance, even while recognizing the importance of different types of exchange, exchange theorists have focused almost exclusively on pure exchange relations – that is, only negotiated exchange or only reciprocal exchange. Researchers have studied concepts similar to multiplexity, including role conflict, repeated exchange, and governance. However, important conceptual distinctions remain between these concepts and multiplexity, and these lines of research cannot fully account for what is interesting and unique about multiplex relations. For instance, as Verbrugge (1979) notes, multiplexity of relations is a structural property of dyads, not individuals. This distinguishes multiplex relations from role conflict (e.g., Cheek & Briggs, 1982; Leary, Wheeler, & Jenkins, 1986) that occurs around individual actors. Understanding the consequences of assuming multiple roles has been a central topic in sociology. Thoits (1983, 1986) examined how maintaining multiple roles affects psychological well-being, demonstrating that multiple identities and roles, such as spouse and employee, provide various benefits, including resource aggregation, justification for failing to meet certain role expectations, and buffering against role failure, which can help mitigate psychological stress. Zuckerman (1999) extends the idea of multiple-identities to the domain of economic sociology, arguing that for category- or role-spanning actors, the difficulty of managing different roles might outweigh potential advantages. Overlapping roles – particularly roles that do not conform to categories familiar to the audience – can invite confusion and lead to devaluation of the target individual or entity (Phillips & Zuckerman, 2001; Zuckerman, Kim, Ukanwa, & von Rittmann, 2003). In these lines of work, however, the focus of analysis is the role conflict that emerges from individuals playing

248

KO KUWABARA ET AL.

incompatible roles across multiple domains. In comparison, multiplex relations concern role conflict that occurs with particular relations between people who share different role relations with each other. This has important implications for both methodological and theoretical reasons. Methodologically, it requires sampling on dyads rather than individuals. Theoretically, it shifts the focus of conceptualization from the individuals to the pattern of interactions between dyads. Second, multiplexity consists of multiple ties that overlap across time or contexts. As noted, this differentiates multiplex relations from those in which actors share multiple roles, activities, or affiliations over time, but one at a time (i.e., repeated exchange). Becoming friends with a co-worker after finishing a project together without sharing professional relations again is not a multiplex relationship in the full sense of its definition if the friendship and the business relationship do not overlap. In other words, adjusting to different norms of interaction as role relations change is not the same as managing multiple norms of interaction simultaneously. Finally, the idea of overlapping ties distinguishes multiplexity from parallel research on how formal governance, such as contracts or explicit incentives, affects informal social relations (Bohnet, Frey, & Huck, 2001; Fehr & Rockenbach, 2003; Heyman & Ariely, 2004; Malhotra & Murnighan, 2002; Simpson & Eriksson, 2009; Kuwabara, 2010a). Formal governance is designed to sustain trust and cooperation between people interacting and exchanging under no binding terms. An ironic consequence, however, is that formal governance can crowd out trust and cooperation by signaling distrust or invoking external enforcement to ensure cooperation (Puranam & Vanneste, 2009). Although these observations closely parallel our interest in how the context of interactions (i.e., formal governance) affects the relationship (i.e., cooperation), they must be qualified by the caveat that one’s motivation to cooperate is neither a role, activity, nor an affiliation. More to the point, the research on governance also fails to inform explicitly how overlapping contexts of exchange interact with each other to affect the relationship.

WHAT SHOULD BE STUDIED? Against this backdrop of theoretical and empirical limitations, our hope is to draw attention to the need for greater clarity about the logic underlying how overlapping ties interact with each other. In this section, we identify a

Multiplex Exchange Relations

249

number of open questions and issues that we believe are critical for better understanding how multiplexity shapes relational dynamics.

Dynamics and Consequences of Multiplexity in Dyadic Relations Some of the most intriguing questions about multiplex relations concern the causality of their association with the strength of ties. Since Granovetter (1973), much discourse in sociology has been dominated by the importance of tie strength. On the one hand, strong relations provide a basis for trust, common identity, and reciprocity that facilitate mutual cooperation. On the other hand, strong relations tend to be structurally short, limiting one’s access to social capital outside of proximate social circles. This fundamental trade-off has important consequences for job search (e.g., Granovetter, 1973), organizational performance (e.g., Burt, 1992; McPherson, Popielarz, & Drobnic, 1992), educational success (e.g., Morgan & So¨rensen, 1999), health and well-being (e.g., Christakis & Fowler, 2007), and numerous other outcomes for individuals, organizations, and communities. Studying how multiplex relations strengthen ties is therefore an important step toward advancing our understanding of how social networks affect individual behavior. We have noted that multiplex relations have often been presumed to be strong relations in past research. At the same time, there are good reasons to question this assumption. One possibility, for instance, is that the association between multiplexity and tie strength could be a result of reverse causation. In some cases, a strong tie provides the basis for expanding the scope of interactions to other domains, such as when two colleagues who recognize that they enjoy working together become friends outside of professional contexts. Meanwhile, kinship and other types of strong relationships are frequently used to organize economic production or exchange (e.g., Johnson & Miller, 1986). Research has documented that, when the legal system is insufficient to guide social conducts or protect transactions, social relationships are more likely to take a front seat in deciding who transacts with whom (Guthrie, 2001; Guseva & Rona-Tas, 2001). Strong relations allow greater access to in-group connections, ensure trust and cooperation through mutual monitoring, and promote greater relational durability to withstand psychic and normative strains from introducing additional demands and expectations. Perhaps more likely, however, the causal relationship between multiplex relations and tie strength is dynamic and reciprocal, creating a

250

KO KUWABARA ET AL.

self-reinforcing feedback loop in which the relationship becomes stronger and more multiplex. The question then becomes: how do overlapping ties strengthen relations? For instance, are their effects additive or multiplicative? Does playing golf with a colleague strengthen the relationship more than, ceteris paribus, joining yet another project team with him? Furthermore, how complementary or compatible do overlapping ties need to be in order to reinforce relationships? While complementary ties are, perhaps by definition, less likely to create strain in the relationship, what remains unstudied is whether they necessarily strengthen relationships the more they overlap. In fact, ties that are too complementary or similar might do little to expand the scope of the interactions and reinforce relationships. Rather, some strain and conflict may be necessary for relationship building. Joint efforts to resolve relational conflicts can build people’s sense of efficacy in their abilities to work together and build relationships (De Dreu & Vliert, 1997; also Lawler, 2001). We can also draw insights from research on small group dynamics, which has shown that a moderate degree of cognitive conflict can enhance group performance by forcing people to rethink problems from different perspectives and arrive at more creative or mutually beneficial solutions (Amason, 1996; De Dreu & Weingart, 2003). This effect is particularly strong when the task involved is nonroutinized (Jehn, 1995, 1997). Finally, it may sometimes be the case that different types of ties can substitute for each other, creating a negative feedback loop in which adding a new tie ‘‘crowds out’’ or weakens existing ties. If so, overlapping ties might have little or no overall effect on the strength of the relationship. This substitution effect might occur for a number of reasons. Again, consider two colleagues who are also casual friends. Once appointed to the same project team, they might gradually drop their friendship to avoid ‘‘mixing work and play’’ or simply because they spend more time together at work and desire each other’s company outside of work less. Similarly, business partners might refrain from seeking friendship outside of work if they feel that their professional tie is already strong enough and that weekend golf will solidify their relationship only slightly. In organizational theory, Haunschild and Beckman (1998) studied interlock ties and association membership ties of CEOs as alternative sources of information about corporate acquisitions. They reasoned that, because CEO membership association and board interlocks are similar in content, having both creates overlapping ties that are too redundant. Consequently, the study found that CEO membership association ‘‘crowds out’’ the effect of board interlocks in the likelihood of involvement in acquisitions. In an historical example, Padgett and Ansell

Multiplex Exchange Relations

251

(1993) contend that the Medici family rose to political power in Renaissance Florence by avoiding multiplex relations to other banking families and only establishing simplex relations of either banking or marital ties to maintain strategic control and flexibility over its political and economic action. Thus far, there is very little research to tell us a priori when multiplex relations have additive, subtractive, or multiplicative effects on the strength of relationships and other outcome variables. For example, while in some studies multiplex relations of strategic alliances and CEO association memberships are found to complement each other (Beckman & Haunschild, 2002), they also substitute each other in other settings (Haunschild & Beckman, 1998). Meanwhile, in Gould’s (1991) study of the Paris 1871 uprising, the effect of overlapping ties is multiplicative, not additive: while informal (neighborhood) ties alone had no effect on mobilization, formal organization ties (joint enlistment) interacted with neighborhood ties to enhance the solidarity of social movement participants. In light of the observations above, greater efforts are clearly needed, both in theory and empirics, to move research beyond reporting post hoc correlations to capture the processes underlying the development of relational bonds. Given the range of possible outcomes already shown by disparate bodies of research, we suggest that, instead of asking whether different ties that invoke different relational norms are necessarily beneficial or detrimental to the relationship, a more productive approach might be to consider how actors negotiate and resolve the tension that arises from incompatible norms (Ingram & Zou, 2008). In this view, overlapping ties might strengthen relationships only to the extent that actors can manage conflicts constructively and meaningfully. Understanding this process, in turn, requires investigating on what basis different ties are perceived as compatible, complementary, counter-normative, or merely different. Besides the theoretical task of specifying when and how multiplex relations take on different effects, the main empirical challenge is collecting data that allow controlling for endogeneity problems to parse out reverse causality and selection biases that are insidious in cross-sectional data. Panel data with information on different types of ties between the same dyads over time will be necessary to investigate these questions. It will also be useful to sample broader measures of relationship strength and relevant mediators. Contemporary exchange theorists have made concerted efforts to examine a wider range of variables beyond objective measures of exchange (e.g., exchange rate or volume) to examine more subjective dimensions of solidarity, which concern actors’ ability to engage in mutually beneficial exchange to produce collective goods (Willer, Borch, & Willer, 2002).

252

KO KUWABARA ET AL.

Feelings of trust, regard, and cohesion, for instance, are emergent properties of repeated exchange that reinforce solidarity (Blau, 1964). Although such variables are correlated, they are nevertheless distinct dimensions of solidarity that are differentially affected by different exchange conditions and mediated by different causal pathways, such as perception of conflict, fairness and justice, or satisfaction (Lawler, Thye, & Yoon, 2000; Molm, Takahashi, & Peterson, 2003; Kuwabara, 2010b). In addition to this repertoire of variables, we also note Ingram and Zou’s (2008) suggestion to include measures of psychic stress to tap more directly into the difficulty of attending to different normative elements in relations. And lastly, another potentially fruitful avenue is to examine actual communication patterns to see how actors coordinate the logic of exchange or convey empathy and identification (McGinn & Keros, 2002) to resolve conflict between overlapping norms and expectations in ‘‘real-time.’’

Antecedents of Multiplexity Another understudied yet critical domain for researchers seeking to better understand multiplex relations is the microdynamics of multiplex relations, specifically what processes – exogenous and endogenous – push relationships to become more multiplex or less multiplex? What motivates people to expand their basis of interaction from simplex to multiplex or to contract multiplex relations toward simplex? What kinds of people and in types of relationships do people prefer multiplex or simplex relations? At the macrolevel, relationships can change as a direct result of policy changes or socio-structural engineering. Modern organizations are constantly implementing structural changes designed to promote more informal contact among their employees within and across units. Building and facilitating teamwork has become one of the most widely adopted management strategies today’s organizations (Staw & Epstein, 2000). For instance, it is now common practice for organizations to sponsor social activities outside of work to promote team-building. Gordon’s (1992) study reports that 82 percent of the companies studied facilitate some form of team development activity, and 68 percent of the Fortune 500 firms are using self-organized teams. These management techniques are designed to increase the likelihood of employees interacting in more varied and purportedly productive ways with each other. Relationships are also shaped profoundly by diffuse forces of the institutional environment. Zelizer (2005) examines the impact of law as an

Multiplex Exchange Relations

253

important context in which intimate ties intermix and often collide with formal or professional ties. Across a vast range of cases, from divorce to prostitution and familial disputes to lawsuits between business partners, courts must recognize, and often dictate, how personal relations are to be separated from or reconciled with economic interests and monetary transactions. In contrast, as noted earlier, Smith-Lovin (2007) suggests that the high degrees of differentiation and specialization in modern societies is ‘‘thinning’’ our relationships toward unidimensional, simplex relations, at least in very close and personal circles. In an intriguing historical analysis, Silver (1990) argues that, before the emergence of the market during the Scottish Enlightenment, the Scots did not have a clear concept of ‘‘strangers’’ – people who are neither friends nor enemies, people who could provide instrumental needs without affective exchange. However, the emergence of the market economy gradually led people to pursue economic interests outside of their immediate, close-knit circles and expand their scope of interactions to include unknown others. In turn, the idea of friendship became divorced from instrumental concerns and increasingly laden with affective and symbolic value. Thus, while pursing economic interests through affective ties was for the most part morally unproblematic in premarket societies, it became increasingly at odds with both the ‘‘iron cage’’ of rationality that came to pervade modern societies and the very idea of friendship that came to symbolize affective ties. Sanchez-Burks (2002) offers a similar argument with respect to national culture, showing that people in the United States, for instance, pay less attention to relational and affective cues in the workplace compared to nonwork settings. Sanchez-Burks (2002) attributes this finding to Protestant relational ideology in the United States that emphasizes the importance of restricting interpersonal interactions in the workplace and separating the domains of work and leisure. Morris and colleagues provide further evidence for cultural differences in how people build and use networks (Chua, Morris, & Ingram, 2009; Morris, Podolny, & Ariel, 2000). Morris, Podolny, and Ariel (2000) report that multiplex relations at work are particularly low in the United States compared to China, Germany, or Spain, consistent with Kacperczyk, Sanchez-Burks, and Bake (2010) more recent and more extensive comparisons. Chua et al. (2009) find that Chinese managers are more likely than American counterparts to report higher levels of trust toward others when they share both affective and instrumental ties. Finally, Yamagishi and colleagues (Yamagishi & Yamagishi, 1994; Yamagishi, Cook, & Watabe, 1998; Kuwabara et al., 2008; also Buchan, Croson, & Dawes, 2002) find that Japanese people are more likely

254

KO KUWABARA ET AL.

than Americans to trust ingroup members (with whom they share existing ties or common identity) than outgroup members. At a more micro level, whether relationships become multiplex might also be a function of individual differences in networking motivation, personality, or gender identity. For instance, a stream of research on social networks concerns identifying brokers who bridge multiple social circles that are otherwise disconnected from each other (Burt, 1992). Brokers have been associated with better job performance (e.g., faster job search, higher compensation, and earlier promotion) by virtue of access through their expansive ties to a wider range of information sources as well as gatekeeping positions that confer power and control over resource flows in the network. A critical trade-off is that effective brokering requires adeptly managing divergent role expectations, collecting and disseminating fast information, relaying various resources, and maintaining fragile relationships across disparate circles, all of which can tax the broker’s interpersonal skills and drain his emotional reserve. Brokering without carefully managing relations can therefore backfire. An emerging body of evidence suggests that individuals with high self-monitoring abilities – that is, behavioral and cognitive skills for regulating one’s actions to fit the norms of different social contexts (Snyder, 1974) – are particularly adept at occupying and managing brokering positions in organizations, which in turn affords them greater power, status, and information advantages (Mehra, Kilduff, & Brass, 2001; Flynn, Reagans, Amanatullah, & Ames, 2006). These findings thus suggest that self-monitoring might be an important personality variable linked to the transformation of simplex relations into multiplex relations. A crucial question with respect to multiplexity in particular is whether brokers are also more likely or motivated to maintain multiplex relations (which we elaborate in the next section). Finally, other lines of research have found that network structures differ as a function of gender (e.g., Brass, 1985; Burt, 1992). Men typically maintain homophilous networks, whereas women establish ties with each other for socioemotional support and ties to men for instrumental resources (Ibarra, 1992, 1993). Moreover, men tend to have more multiplex relations at work than women, because men tend to seek friendship from the same men who can provide them access to organizational resources (Ibarra, 1992). A situational or structural explanation of the observed differences in networks is that women are marginalized from important organizational positions so that they have a narrower range of network choices (Ibarra, 1993). However, it could also be the case that women and minorities seek ties with powerful actors in organizations – often men – for strategic or

Multiplex Exchange Relations

255

instrumental reasons. While both exclusionary and inclusionary forces are at play in varying degrees in shaping network structures of men and women (Brass, 1985), a recent study by Stallings (2008) has begun to disentangle the two forces and finds support for greater effects of structural constraints. The study found that women lack multiplex relations with female colleagues largely because of the shortage of senior women who can provide instrumental resources along with socioemotional support. With possible and important implications for understanding differential opportunities for men and women in organizations, further research is needed to understand how different sexes actually experience multiplex relations. Is one sex more likely to benefit from multiplex relations than the other?

Multiplexity and Network Dynamics Beyond the dyad, how do multiplex relations affect or interact with larger network structures? An important measure of network structure is closure (Coleman, 1988). Closed networks are characterized by ties to alters who are connected to each other, creating densely connected cliques in which ‘‘a friend of a friend is also a friend.’’ Dense connections foster trust and cooperation by promoting reciprocity and common identity through mutual monitoring in shared relationships. Open connections, on the other hand, involve ties to alters who are disconnected from each other, creating networks that are rife with opportunities for brokering. Open connections have been associated with numerous instrumental benefits that accrue to those who occupy what Burt (1992) has termed ‘‘structural holes.’’ How do multiplex relations affect the formation of closed versus open networks? How does network structure affect the formation of multiplexity? On the one hand, overlapping ties might reduce the need to form closed triads if they create strong relations. Conversely, people might be more likely to form multiplex relations in the absence of closed triads that sustain mutual trust and cooperation. On the other hand, closed triads might facilitate the formation of multiplex relations to the extent that actors in closed triads interact more often and share broader bases of exchange with one another. Lee and Monge (2009) examined these questions empirically, finding that triadic closure does not enhance multiplexity of dyadic relations. However, this result might be attributable to her particular research setting, which was a community of international development organizations in which the formation of collaborative ties on development projects is less organic. An interesting question, therefore, is to look further in other contexts at how

256

KO KUWABARA ET AL.

the institutional environment together with the overall network structure impacts the propensity for multiplex relations to occur. Broadening the scope of analysis to network structures also implicates observers outside of the dyad. As Podolny (2001) colorfully noted, networks are ‘‘pipes’’ that provide conduits for information and resources as well as ‘‘prisms’’ that reflect and signal the status, legitimacy, and identity of actors through their affiliation patterns. An important consideration is in what ways multiplex relations are affected by how they are perceived by others rather than how ties actually overlap in relationships. For instance, do people avoid mixing economic and social exchange because they are perceived to be culturally illegitimate by others (Chan, 2009), or because actors fear confounding professional and personal interests and undermining their relationships?

Measuring Multiplexity Finally, efforts to address the preceding conceptual issues are likely to fall short in the absence of greater clarity surrounding core methodological issues, such as measurement. For instance, at what level of analysis should overlapping ties be measured? Almost every relationship of some duration is multiplex to some extent, depending on how granularly one measures the relations and how broadly one defines the context of interactions. Some people may view playing golf and playing racquetball as invoking different norms (and hence multiplex) even when involving the same partners. Others might approach business partners and friends with more or less the same set of expectations and standards of conduct. Thus, it may be both impossible and even counter-productive to define what the level of measurement should be across all cases. Nevertheless, we suggest that more concerted efforts are needed to specify at what level of analysis relations should be defined and which overlapping ties are relevant based on theory. To this end, we propose two general objectives for future research. First, research needs to move beyond simple dichotomies of relations to consider other combinations of relations. Implicitly and explicitly, much of existing research has focused on various distinctions between economic and social exchange, including instrumental versus affective ties (e.g., Granovetter, 1985; Ingram & Roberts, 2000; Zelizer, 2005), tangible versus intangible resources (e.g., Bienenstock & Bianchi, 2004; Heyman & Ariely, 2004), self-interest versus regard for others (e.g., Clark & Mills, 1979), specific versus diffuse obligations (Blau, 1964), and, quite literally, economic versus

Multiplex Exchange Relations

257

social spheres (e.g., Becker, 1976). While these distinctions are interesting, they draw overly stark contrasts that overlook broader possibilities for different types of social ties to combine and intermix in a relationship. Two people can share overlapping ties, both affective-social in content yet each defined by different norms of interaction, such as two Asian-American friends who share both common white and common Asian friends and thus interact in different cultural contexts with each other (Mok, Morris, Benet-Martinez, & Karakitapoglu-Aygun, 2007). Conversely, two people may have overlapping ties that are both relatively instrumental in content, such as two managers’ whose companies compete with each other for market share while simultaneously cooperating to restrict further competition from entering the market (Ingram & Roberts, 2000). Nor need overlapping ties be purely affective or instrumental. It is not uncommon to express affection and gratitude in the context of professional relations. In fact, Lawler and colleagues (Lawler, 1992; Lawler et al., 2000; Lawler & Yoon, 1996) have shown that exchange tasks with ostensibly purely instrumental goals can produce feelings of attachment between actors through repeated interactions. Second, future research also needs to move beyond typologies to focus more clearly on how relationships actually overlap and how they interact with each other. In particular, how can we measure different normative elements that exist in overlapping ties? Research on multiplex relations typically infers overlapping ties from survey responses or archival data indicating the existence of multiple types of ties between people or organizations. However, simply measuring whether ties overlap leaves open the question of whether and to what extent they are actually compatible or incompatible. For instance, it is unclear whether going to musical activities together is compatible with the norm of parent–child ties (Krohn et al., 1988), and whether repeated buyer–seller ties and board interlocks are operating under compatible logics (Gulati & Sytch, 2007). One potential way to alleviate this concern may be to use attitudinal surveys to directly measure perceived normative tension of various bases of ties, and see how the perception impacts interaction behaviors. Lastly, to address these issues of definition and measurement, researchers will need to move their focus beyond the structure or patterns of interactions to the content of relations (Emirbayer & Goodwin, 1994; Burt & Schøtt, 1985). Increasingly, sociologists are noting the need to bring ‘‘the content of social ties back in, particularly culture and meaning that define, realize, and perform the relational content of embeddedness’’ (Chan, 2009, p. 713). Research on multiplexity should join these efforts by dedicating greater attention to the normative contexts of interaction.

258

KO KUWABARA ET AL.

RESEARCH AGENDA FOR EXCHANGE THEORY The crux of our discussion thus far has been how actors manage multiple bases of interaction that create normative tension or conflicts of interest when ties overlap. Of particular concern is when and in what ways overlapping ties are normatively compatible or incompatible with each other. The science of social relations and networks has yet to come to any adequate terms with the complexity of this pervasive phenomenon. To advance our understanding, we need to broaden the scope of analysis to look beyond affective-instrumental dichotomies while also zeroing in more directly on the processes underlying the interactions of different normative contexts. Exchange theory is well-poised to contribute to these efforts. Its theoretical focus on different forms of exchange in durable relations with repeated interactions offers a particularly promising point of entry for thinking about the nature of multiplex relations in their varied forms. Its well-developed experimental paradigm offers a number of methodological advantages as well. First and foremost among these is its use of random assignment. As we have elaborated in this chapter, establishing causal effects of multiplex exchange relations is fiendishly difficult. Random assignment to different conditions of exchange or relations will help tease out the effects of endogeneity and self-selection that confound field data. Second, the experimental paradigm has been used over and over in a number of important studies and is methodologically reliable. Exchange theorists have advanced their research programs progressively over the past decades by directly extending studies by one another to isolate variables of theoretical interest and refine their measurements (e.g., Molm, 1995; Willer, 1999; Cook & Cooper, 2003). Third, the experimental paradigm in exchange theory offers ready operationalizations of different types of exchange, which can be combined easily and flexibly to create multiplex relations in laboratories. Exchange theorists identify four different forms of exchange in particular: reciprocal, negotiated, productive, and generalized (see Lawler, Thye, & Yoon, 2008, and Molm, 1995, for review). The first two forms of exchange are direct exchanges that occur between two or more people who exchange benefits with each other directly. Reciprocal exchange involves unilateral and voluntary acts of giving resources by one actor to another without knowing when, how, or whether the other will reciprocate. Negotiated exchange involves bilateral decisions and agreements to divide resources between two people, often under binding terms of exchange. In comparison, productive exchange and generalized exchange occur at the group level. In productive exchange, individual members decide whether to contribute to

Multiplex Exchange Relations

259

the collective good, which provides benefits to them based on the amount of contributions pooled. In generalized exchange, actors provide benefits to others unilaterally, much as in reciprocal exchange, except the recipients can return benefits to others rather than eventually repaying the giver. The task for future research is to understand how these exchange forms co-occur and overlap within relationships under various conditions. How do they affect the emergence of multiplex relations, and how do they affect relational solidarity in turn?2 Over the past decade, social exchange theorists have made concerted efforts to compare how different forms of exchange produce solidarity (Molm; Lawler). So far, their research has largely focused on pure exchange relations, comparing those based solely on one type of exchange to another (Blau, 1964; Ekeh, 1974; Molm, Collett, & Schaefer, 2007; Lawler et al., 2008). However, their experiments can be easily extended to construct multiplex relations by exposing participants to a mix of exchange forms across exchange rounds, and existing measures can be used to provide direct comparisons to past research. Initial efforts are underway. Kuwabara (2010b) compared trust relations with fixed roles to those with alternating roles. In the fixed-role condition, one person was assigned to the role of the trustor and the other to the trustee for the entire duration of the exchange tasks, while in the alternating-role condition, participants alternated between the roles after each instance of exchange. The latter condition can be viewed as a case of multiplexity based on overlapping roles, even though participants engaged in the same exchange form in every round. The results showed that role multiplexity produces stronger relational bonds than fixed roles. Cheshire, Gerbasi, and Cook (2010) break new ground by allowing participants to transition from one type of exchange to another, both endogenously and exogenously. They examine the effects of transitioning from one form of exchange to another on trust and cooperation in exchange relations. In their experiment, participants undertake a series of ‘‘reciprocal exchanges’’ with each other in dyads or small networks, followed by a series of ‘‘negotiated exchanges.’’ The transitions are exogenous, manipulated by the researchers. The researchers find that the transitions matter: transitioning to binding negotiation can increase or decrease perceptions of trust. When the level of cooperation in the first series of exchanges is high, moving to negotiated exchange decreases trust [consistent with prior research by Molm, Takahashi, and Peterson (2000) that negotiation is more likely to undermine solidarity relative to reciprocity]. However, when the level of cooperation is low, negotiating binding agreements seems to boost trust.

260

KO KUWABARA ET AL.

Although such findings are informative, there are reasons to argue that transitioning from one form of exchange to another is different from what is meant by multiplex relations, that is, the co-occurrence of different exchange forms and norms. It is one thing for friends to drop friendship when they become co-workers, but it is quite another for them to remain friends as they continue to work together professionally. This raises important questions for future research in social exchange theory: First, to what extent – and why – might engaging in different types of exchange simultaneously be different from engaging in different types of exchange sequentially, one at a time? Is one a special form of the other, or are they qualitatively different, and how? Transitioning from one form of exchange to another poses challenges of change, that is, adopting new norms and logics of exchange, sometimes by abandoning the older norms. In comparison, multiplex relations create the difficulties of juggling different norms and logics of exchange simultaneously. Moreover, although engaging in different types of exchanges sequentially may be perceived as overlapping activities when brief periods of time exist between transitions, norms of exchange often emerge more gradually. Thus, more direct comparisons between simultaneous and sequential forms of multiplexity are needed to extend Cheshire et al.’s (2010) results to multiplex relations proper. Another important question for future research in social exchange theory is when and how do simplex relationships become multiplex? Many dyadic relations develop first on the basis of a single form of exchange. Are such relations more likely to become multiplex when they start out as reciprocal exchange relations or negotiated exchange relations? And what are the structural and psychological barriers to relationships expanding from simplex to multiplex, or vice versa? As already mentioned, a number of studies have examined the existence of reciprocal exchange relations among business partners. A notable example is Ingram and Roberts’ (2000) work on the Sydney hotel industry. As this study notes, reciprocal or noncontractual exchange relations can emerge between business competitors as both influence tactics and sources of efficiency where formal contractual exchange may be cumbersome. To our knowledge, however, little if any research has examined the actual process and dynamics of negotiated exchange relations expanding to include reciprocal exchange. Negotiations researchers have found that friends are more likely to avoid negotiating than strangers, or reach suboptimal outcomes when they do negotiate, suggesting that people in relationships based on reciprocal exchange have difficulty incorporating negotiated exchange into their repertoire. However, Valley, Neale, and Mannix (1995b) note that this finding may be due to an experimental protocol in which participants are

Multiplex Exchange Relations

261

asked to come to the laboratory with a friend to negotiate an artificial scenario, thus comparing real friendship to hypothetical negotiation. The experimental paradigm in exchange theory offers an alternative approach. Although reciprocal exchange tasks in lab settings cannot be directly equated with friendship, they can help capture key aspects of the elemental exchange structures of social relations and how they condition negotiated exchange or the relationship itself in ways that allow more direct comparison between forms of exchange. It will also be important to look beyond the exchange relation itself to examine what surrounding conditions might lead reciprocal exchange relations to incorporate negotiated exchange. Following Puranam and Vanneste (2009), we suggest three conditions that might lead actors to adopt negotiated exchange: the norm of negotiation, expertise in negotiation, and uncertainty. That is, actors in reciprocal exchange relations may be likely to also engage in negotiation when it is normative to do so, when they have legal or technical expertise to conduct negotiations successfully, and when the exchange task poses high levels of risk or uncertainty about its consequences. Finally, a critical step toward understanding the endogenous processes that mediate the development of solidarity in multiplex exchange relations will be to better specify to how conceptualize and measure normative compatibility between overlapping ties as the basis for how actors perceive the exchange relation. What are the structural features of overlapping exchanges that produce normative tension and conflict? That is, when are two exchange forms or relations perceived to be at odds with each other? One promising idea is to draw on the exchange-theoretic concepts of competition-cooperation (Kelley & Thibaut, 1978), on the one hand, and power-dependence (Emerson, 1976; Cook & Emerson, 1978), on the other hand. The first dimension describes the degree to which the actors’ interests are mutually aligned or opposed. If one actor’s preferred action harms the other, they are in competition; if their preferred actions are the same, they are in cooperation. In between these polar cases are situations in which mixed-motives exist. The second dimension describes the degree to which an actor’s dependence on the other for resources creates power differential or inequality. Both of these dimensions have been well-theorized to offer a potentially useful framework for understanding two important ways in which relationships are compatible. We suggest that how closely overlapping ties line up on these dimensions at least partially reflects their normative compatibility. In particular, we suggest that relational norms are compatible across overlapping ties if (1) actors have similar power relationships with

262

KO KUWABARA ET AL.

each other in both exchange relations/forms and (2) their incentives are mutually aligned. For example, two people share incompatible relations if one has power over the other in one domain (e.g., work), but not the other domain (e.g., home). Similarly, overlapping ties are incompatible if two people are at once competitors and cooperators. Compatibility is different from Molm et al. (2007) measure of perceived conflict in exchange tasks. Whereas perceived conflict measures actors’ subjective feelings about the difficulty of reaching mutually beneficial and satisfactory agreements with each other, compatibility concerns the normative or structural complementarity of overlapping ties. The former describes the exchange task and partner, whereas the latter concerns how different norms of exchange interact in the relationship. Actors might experience high levels of conflict over tasks that are normatively similar or low levels of conflict over tasks that are quite at odds with each other. For instance, a married couple may be able to negotiate and resolve the division of financial responsibilities amicably, yet still feel uncomfortable about negotiating with each other. Thus, measuring conflict alone fails to capture this broader source of tension that actors might experience at the relational level.

CONCLUSION Although social exchange theory has made significant advances toward understanding exchange relationships based on one type of exchange, many real-life relationships are multiplex, consisting of overlapping roles, norms, and activities. As such, laboratory work alone is ultimately inadequate. A fuller understanding of multiplex relations will emerge only from more concerted efforts to triangulate on the causal processes underlying multiplexity using experiments, mathematical modeling, ethnography, field data, and various other methods. To this end, we hope that our chapter serves as a call to action, motivating future efforts by sociologists and groups researchers of all methodological bents to tackle not only the questions and issues we have raised but also the many others we have surely omitted.

NOTES 1. In this chapter, we refer to a ‘‘relationship’’ to denote the relational structure and ‘‘tie’’ to denote the content or context of interaction, that is, whether two people are connected versus how they are connected. For example, two individuals have a relationship when they share a connection and interact with each other, where

263

Multiplex Exchange Relations

the relationship might consist of professional, social, and/or familial ties. Hence, a social relationship is multiplex when two people share multiple, overlapping ties with each other. 2. Exchange forms can overlap across both dyadic and group-based exchanges. For instance, two individuals in a church involved in fund-raising (productive exchange) can become friends (social exchange) on their own terms. In generalized exchange, there is almost always some potential for two actors to engage in reciprocal exchange between themselves. In this chapter, we restrict our discussion to dyads (Verbrugge, 1979) and preclude group-based exchanges, which do not necessarily involve direct exchange relations that overlap between actors.

REFERENCES Amason, A. C. (1996). Distinguishing the effects of functional and dysfunctional conflict on strategic decision making: Resolving a paradox for top management teams. The Academy of Management Journal, 39(1), 123–148. Baker, T., & Nelson, R. E. (2005). Creating something from nothing: Resource construction through entrepreneurial bricolage. Administrative Science Quarterly, 50, 329–366. Becker, G. S. (1976). The economic approach to human behavior. Chicago: University of Chicago Press. Beckman, C. M., & Haunschild, P. R. (2002). Network learning: The effects of partners’ heterogeneity of experience on corporate acquisitions. Administrative Science Quarterly, 47(1), 92–124. Bienenstock, E. J., & Bianchi, A. J. (2004). Activating performance expectations and status differences through gift exchange: Experimental results. Social Psychology Quarterly, 67(3), 310–318. Blau, P. M. (1964). Exchange and power in social life. New Brunswick, NJ: Transaction. Bohnet, I., Frey, B. S., & Huck, S. (2001). More order with less law: On contract enforcement, trust, and crowding. American Political Science Review, 95(1), 131–144. Brass, D. J. (1985). Men’s and women’s networks: A study of interaction patterns and influence in an organization. Academy of Management Journal, 28(2), 327–343. Brass, D. J., Butterfield, K. D., & Skaggs, B. C. (1998). Relationships and unethical behavior: A social network perspective. Academy of Mangement Review, 23(1), 14–31. Buchan, N. R., Croson, R. T. A., & Dawes, R. M. (2002). Swift neighbors and persistent strangers: A cross-cultural investigation of trust and reciprocity in social exchange. American Journal of Sociology, 108(1), 168–206. Burt, R. S. (1980). Cooptive corporate actor networks: A reconsideration of interlocking directorates involving American manufacturing. Administrative Science Quarterly, 25(4), 557–582. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Burt, R. S., & Schøtt, T. (1985). Relation contents in multiple networks. Social Science Research, 14(4), 287–308. Chan, C. S.-C. (2009). Invigorating the content in social embeddedness: An ethnography of life insurance transactions in China. American Journal of Sociology, 115(3), 712–754.

264

KO KUWABARA ET AL.

Cheek, J. M., & Briggs, S. R. (1982). Self-consciousness and aspects of identity. Journal of Research in Personality, 16(4), 401–408. Cheshire, C., Gerbasi, A., & Cook, K. S. (2010). Trust and transitions in modes of exchange. Social Psychology Quarterley, 73(1), 1–20. Christakis, N. A., & Fowler, J. H. (2007). The spread of obesity in a large social network over 32 years. New England Journal of Medicine, 357(4), 370–379. Chua, R. Y. J., Morris, M. W., & Ingram, P. (2009). Guanxi vs networking: Distinctive configurations of affect-and cognition-based trust in the networks of Chinese vs American managers. Journal of International Business Studies, 40(3), 490–508. Clark, M. S. (1984). Record keeping in two types of relationships. Journal of Personality and Social Psychology, 47(3), 549–557. Clark, M. S., & Mills, J. (1979). Interpersonal attraction in exchange and communal relationships. Journal of Personality and Social Psychology, 37(1), 12–24. Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94(Suppl.), S95–S120. Cook, K. S., & Cooper, R. M. (2003). Experimental studies of cooperation, trust, and social exchange. In: E. Ostrom & J. Walker (Eds), Trust and reciprocity: Interdisciplinary lessons from experimental research (pp. 209–244). New York: Russell Sage. Cook, K. S., & Emerson, R. M. (1978). Power, equity and commitment in exchange networks. American Sociological Review, 43, 721–739. De Dreu, C. K. W., & Vliert, E. V. d. (1997). Using conflict in organizations. London: Sage. De Dreu, C. K. W., & Weingart, L. R. (2003). Task versus relationship conflict, team performance, and team member satisfaction: A meta-analysis. Journal of Applied Psychology, 88(4), 741–749. DiMaggio, P., & Louch, H. (1998). Socially embedded consumer transactions: For what kinds of purchases do people most often use networks? American Sociological Review, 63(5), 619–637. Dubini, P., & Aldrich, H. (1991). Personal and extended networks are central to the entrepreneurial process. Journal of Business Venturing, 6, 305–313. Ekeh, P. P. (1974). Social exchange theory: The two traditions. Cambridge, MA: Harvard University Press. Emirbayer, M., & Goodwin, J. (1994). Network analysis, culture, and the problem of agency. The American Journal of Sociology, 99(6), 1411–1454. Emerson, R. M. (1976). Social exchange theory. Annual Review of Sociology, 2, 335–362. Fehr, E., & Rockenbach, B. (2003). Detrimental effects of sanctions on human altruism. Nature, 422, 137–140. Flynn, F. J., Reagans, R. E., Amanatullah, E. T., & Ames, D. T. (2006). Helping one’s way to the top: Self-monigors achieve status by helping others and knowing who helps whom. Journal of Personality and Social Psychology, 91(6), 1123–1137. Foa, E. B., & Foa, U. G. (1980). Resource theory: Interpersonal behavior as exchange. In: K. Gergen, M. S. Greenberg & R. H. Wills (Eds), Social exchange: Advances in theory and research. New York: Plenum. Gluckman, M. (1962). Les Rites de Passage. In: M. Gluckman (Ed.), Essays on the ritual of social relations. Manchester: Manchester University Press. Gordon, J. (1992). Work teams. How far have they come? Training, 29(October), 59–65. Gould, R. V. (1991). Multiple networks and mobilization in the Paris Commune, 1871. American Sociological Review, 56(December), 716–729.

Multiplex Exchange Relations

265

Granovetter, M. (1973). The strength of weak ties. American Journal of Sociology, 78(6), 1360–1380. Granovetter, M. (1985). Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91(3), 481–510. Gulati, R., & Sytch, M. (2007). Dependence asymmetry and joint dependence in interorganizational relationships: Effects of embeddedness on a manufacturer’s performance in procurement relationships. Administrative Science Quarterly, 52(1), 32–69. Gulati, R., & Westphal, J. D. (1999). Cooperative or controlling? The effects of CEO-board relations and the content of interlocks on the formation of joint ventures. Administrative Science Quarterly, 44(3), 473–506. Guseva, A., & Rona-Tas, A. (2001). Uncertainty, risk, and trust: Russian and American credit card markets compared. American Sociological Review, 66, 623–646. Guthrie, D. (2001). Dragon in a three-piece suit: The emergence of capitalism in China. Princeton, NJ: Princeton University Press. Hardin, R. (2002). Trust and trustworthiness. New York: Russell Sage. Haunschild, P. R., & Beckman, C. M. (1998). When do interlocks matter? Alternate sources of information and interlock influence. Administrative Science Quarterly, 43, 815–844. Heyman, J., & Ariely, D. (2004). Effort for payment: A tale of two markets. Psychological Science, 15(11), 787–793. Human, S. E., & Provan, K. G. (2000). Legitimacy building in the evolution of small-firm multilateral networks: A comparative study of success and demise. Administrative Science Quarterly, 45(2), 327–365. Ibarra, H. (1992). Homophily and differential returns: Sex differences in network structure and access in an advertising firm. Administrative Science Quarterly, 37(3), 422–447. Ibarra, H. (1993). Personal networks of women and minorities in management: A conceptual framework. Academy of Management Review, 18(1), 56–87. Ibarra, H. (1995). Race, opportunity, and diversity of social circles in managerial networks. Academy of Management Journal, 38(3), 673–703. Ingram, P., & Roberts, P. W. (2000). Friendships among competitors in the Sydney hotel industry. American Journal of Sociology, 106(2), 387–423. Ingram, P., & Zou, X. (2008). Business friendships. Research in Organizational Behavior, 28, 167–184. Jehn, K. A. (1995). A multimethod examination of the benefits and detriments of intragroup conflict. Administrative Science Quarterly, 40(2), 256–282. Jehn, K. A. (1997). A quantitative analysis of conflict types and dimensions in organizational groups. Administrative Science Quarterly, 42(3), 530–557. Johnson, J. C., & Miller, M. L. (1986). Behavioral and cognitive data: A note on the multiplexity of network subgroups. Social Networks, 8(1), 65–77. Kacperczyk, A., Sanchez-Burks, J., & Baker, W. E. (2009). Multiplexity and emotional energy in cross cultural perspective. Working paper. MIT. Kacperczyk, A., Sanchez-Burks, J., & Baker, W. E. (2010). Social isolation in the workplace: A cross-national and longitudinal analysis. Working paper. MIT. Kapferer, B. (1969). Norms and the manipulation of relationships in a work context. In: J. C. Michell (Ed.), Social networks in urban situations. Manchester: Manchester University Press. Kelley, H. H., & Thibaut, J. W. (1978). Interpersonal relations: A theory of interdependence. New York: Wiley.

266

KO KUWABARA ET AL.

Krohn, M. D., Massey, J. L., & Zielinski, M. (1988). Role overlap, network multiplexity, and adolescent deviant behavior. Social Psychology Quarterly, 51(4), 346–356. Kuwabara, K. (2010a). Do reputation systems undermine cooperation? Working paper. Columbia University. Kuwabara, K. (2010b). Solidarity, reciprocity, and the value of ‘‘doing things together’’: How economic exchange creates relational bonds. Working paper. Columbia University. Kuwabara, K., Willer, R., Macy, M. W., Mashima, R., Terai, S., & Yamagishi, T. (2008). Culture, identity and structure in social exchange: A web-based trust experiment in the United States and Japan. Social Psychology Quarterly, 70(4), 461–470. Lawler, E. J. (1992). Affective attachments to nested groups: A choice-process theory. American Sociological Review, 57, 327–339. Lawler, E. J. (2001). An affect theory of social exchange. American Journal of Sociology, 107, 321–352. Lawler, E. J., Thye, S. R., & Yoon, J. (2000). Emotion and group cohesion in productive exchange. American Journal of Sociology, 106(3), 616–657. Lawler, E. J., Thye, S. R., & Yoon, J. (2008). Social exchange and micro social order. American Sociological Review, 73, 519–542. Lawler, E. J., & Yoon, J. (1996). Commitment in exchange relations: Test of a theory of relational cohesion. American Sociological Review, 61, 89–108. Lazega, E., & Pattison, P. E. (1999). Multiplexity, generalized exchange and cooperation in organizations: A case study. Social Networks, 21(1), 67–90. Leary, M. R., Wheeler, D. S., & Jenkins, T. B. (1986). Aspects of identity and behavioral preference: Studies of occupational and recreational choice. Social Psychology Quarterly, 49(1), 11–18. Lee, S., & Monge, P. (2009). The coevolution of multiplex networks in organizational communities. Working paper. University of Pennsylvania. Lomi, A., & Pattison, P. (2006). Manufacturing relations: An empirical study of the organization of production across multiple networks. Organization Science, 17(3), 313. Malhotra, D., & Murnighan, J. K. (2002). The effects of contracts on interpersonal trust. Administrative Science Quarterly, 47, 534–559. Marin, A., & Hampton, K. (2007). Simplifying the personal network name generator: An alternative to traditional multiple and single name generators. Field Methods, 19, 163–193. Marsden, P. V., & Campbell, K. E. (1984). Measuring tie strength. Social Forces, 63(2), 482–501. McGinn, K. L. (2006). Relationships and negotiations in context. In: L. Thompson (Ed.), Negotiation theory and research (pp. 129–144). New York: Psychology Press. McGinn, K. L., & Keros, A. T. (2002). Improvisation and the logic of exchange in socially embedded transactions. Administrative Science Quarterly, 47(3), 442–473. McPherson, J. M., Popielarz, P., & Drobnic, S. (1992). Social networks and organizational dynamics. American Sociological Review, 57, 153–170. McPherson, J. M., Smith-Lovin, L., & Brashears, M. E. (2006). Social isolation in America: Changes in core discussion networks over two decades. American Sociological Review, 71, 353–375. Mehra, A., Kilduff, M., & Brass, D. J. (2001). The social networks of high and low selfmonitors: Implications for workplace performance. Administrative Science Quarterly, 46, 121–146.

Multiplex Exchange Relations

267

Mok, A., Morris, M. W., Benet-Martinez, V., & Karakitapoglu-Aygun, Z. (2007). Embracing American culture: Structures of social identity and social networks among firstgeneration biculturals. Journal of Cross-Cultural Psychology, 38(5), 629–635. Molm, L. D. (1995). Social exchange and exchange networks. In: K. S. Cook, G. A. Fine, & J. S. House (Eds), Sociological perspectives on social psychology (pp. 209–235). Needham Heights, MA: Allyn and Bacon. Molm, L. D., Takahashi, N., & Peterson, G. (2000). Risk and trust in social exchange: An experimental test of a classical proposition. American Journal of Sociology, 105(5), 1296– 1427. Molm, L. D., Collett, J. L., & Schaefer, D. R. (2007). Building solidarity through generalized exchange: A theory of reciprocity. American Journal of Sociology, 113(1), 205–242. Molm, L. D., Takahashi, N., & Peterson, G. (2003). In the eyes of the beholder: Procedural justice in social exchange. American Sociological Review, 68, 128–152. Morgan, S. L., & So¨rensen, A. B. (1999). Parental networks, social closure, and mathematics learning: A test of Coleman’s social captial explanation of school effects. American Sociological Review, 64, 661–681. Morris, J. H., & Moburg, D. J. (1993). Organizations as contexts for trust and betrayal. In: T. Sarbin, R. C. Carney & C. Eoyang (Eds), Espionage: Studies in trust and betrayal (pp. 219–241). New York: Praeger. Morris, M. W., Podolny, J., & Ariel, S. (2000). Missing relations: Incorporating relational constructs into models of culture. In: P. C. Earley & H. Singh (Eds), Innovations in international and cross cultural management. Thousand Oaks, CA: Sage. Padgett, J. F., & Ansell, C. K. (1993). Robust action and the rise of the Medici, 1400–1434. The American Journal of Sociology, 98(6), 1259–1319. Phillips, D. J., & Zuckerman, E. W. (2001). Middle-status conformity: Theoretical restatement and empirical demonstration in two markets. American Journal of Sociology, 107(2), 379–429. Podolny, J. M. (2001). Networks as the pipes and prisms of the market. The American Journal of Sociology, 107(1), 33–60. Powell, W. W., White, D. R., Koput, K. W., & Owen-Smith, J. (2005). Network dynamics and field evolution: The growth of interorganizational collaboration in the life sciences. American Journal of Sociology, 110(4), 1132–1205. Puranam, P., & Vanneste, B. S. (2009). Trust and governance: Untangling a tangled web. Academy of Management Review, 34(1), 11–31. Putnam, R. D. (2000). Bowling alone: The collapse and revival of American community. New York: Simon & Schuster. Rangan, S., & Sengul, M. (2009). The influence of macro structure on the foreign market performance of transnational firms: The value of IGO connections, export dependence, and immigration links. Administrative Science Quarterly, 54(2), 229–267. Sanchez-Burks, J. (2002). Protestant relational ideology and (in) attention to relational cues in work settings. Journal of Personality and Social Psychology, 83(4), 919–929. Silver, A. (1990). Friendship in commercial society: Eighteenth-century social theory and modern sociology. The American Journal of Sociology, 95(6), 1474–1504. Simpson, B., & Eriksson, K. (2009). The dynamics of contracts and generalized trustworthiness. Rationality and Society, 21, 59–80. Smith, A. (1759). The theory of moral sentiments (1st ed.). London: Printed for A. Millar, and A. Kincaid and J. Bell, in Edinburgh.

268

KO KUWABARA ET AL.

Smith-Lovin, L. (2007). The strength of weak identities: Social structural sources of self, situation and emotional experience. Social Psychology Quarterly, 70(2), 106–124. Snyder, M. (1974). Self-monitoring of expressive behavior. Journal of Personality and Cosial Psychology, 30, 526–537. Stallings, M. M. (2008). Whom would you approach for advice? Gender differences in network preferences. Unpublished manuscript, University of Pennsylvania. Staw, B. M., & Epstein, L. D. (2000). What bandwagons bring: Effects of popular management techniques on corporate performance, reputation, and CEO pay. Administrative Science Quarterly, 45(3), 523–556. Stern, R. N. (1979). The development of an interorganizational control network: The case of intercollegiate athletics. Administrative Science Quarterly, 24(2), 242–266. Thoits, P. A. (1983). Multiple identities and psychological well-being: A reformulation and test of the social isolation hypothesis. American Sociological Review, 48(2), 174–187. Thoits, P. A. (1986). Social support as coping assistance. Journal of Consulting and Clinical Psychology, 54(4), 416–423. Uzzi, B. (1996). The sources and consequences of embeddedness for the economic performance of organizations: The network effect. American Sociological Review, 61, 674–698. Uzzi, B. (1997). Social Structure and competition in interfirm networks: The paradox of embeddedness. Administrative Science Quarterly, 42, 35–67. Valley, K. L., Neale, M. A., & Mannix, E. A. (1995a). Friends, lovers, colleagues, strangers: The effects of relationships on the process and outcome of dyadic negotiations. Research on Negotiation in Organizations, 5, 65–93. Valley, K. L., Neale, M. A., & Mannix, E. A. (1995b). Friends, lovers, colleagues, strangers: The effects of relationships on the process and outcome of dyadic negotiations. Research on Negotiation in Organizations, 5, 65–93. Verbrugge, L. M. (1979). Multiplexity in adult friendships. Social Forces, 57(4), 1286–1309. Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. Cambridge, NY: Cambridge University Press. Westphal, J. D., Gulati, R., & Shortell, S. M. (1997). Customization or conformity? An institutional and network perspective on the content and consequences of TQM adoption. Administrative Science Quarterly, 42(2), 366–394. Wheeldon, P. D. (1969). The operation of voluntary associations and personal networks in the political processes of an inter-ethnic community. Social Networks in Urban Situations, 128–180. Willer, D. (1999). Network exchange theory. Westport, CT: Praeger. Willer, D., Borch, C., & Willer, R. (2002). Building a model for solidarity and cohesion using three theories. In: S. R. Thye & E. J. Lawler (Eds), Advances in group processes (Vol. 19, pp. 67–107). New York: Jai Press. Yamagishi, T., Cook, K. S., & Watabe, M. (1998). Uncertainty, trust and commitment formation in the United States and Japan. American Journal of Sociology, 108, 165–194. Yamagishi, T., & Yamagishi, M. (1994). Trust and commitment in the United States and Japan. Motivation and Emotion, 18, 9–66. Zelizer, V. A. R. (2005). The purchase of intimacy. Princeton, NJ: Princeton University Press. Zuckerman, E. W. (1999). The categorical imperiative: Securities analysts and the illegitimacy discount. American Journal of Sociology, 104(5), 1398–1438. Zuckerman, E. W., Kim, T.-Y., Ukanwa, K., & von Rittmann, J. (2003). Robust indentities or nonentities? Typecasting in the feature-film labor market. American Journal of Scciology, 108(5), 1018–1074.

CORRUPTION AS SOCIAL EXCHANGE$ Edward J. Lawler and Lena Hipp ABSTRACT This chapter applies social exchange theory to corruption. If two parties exhibit corrupt behaviors, secrecy becomes a new joint good, making the two parties more dependent on each other (an increase in total power). Since no external enforcement mechanisms are available in illicit exchanges, the initial reciprocal exchange pattern shifts toward negotiated or productive forms of exchange. Such forms of exchange, however, tend to leave traces, either because the amount of traded resources increases or the contingencies between the behaviors become more visible to the outside. Using the larger network structure, in which corrupt exchanges are embedded, to deal with the problem of detection also is Janus-faced. Adding more ties to the exchange increases either the competition between several potential exchanges partners (exclusively connected network) or the risk of nonreciprocity and whistle blowing (positively connected network). By showing that illicit relations are inherently unstable, we specify some of the scope conditions of social exchange theory.

$

An earlier version was presented at the Group Processes Conference in New York, August 2007.

Advances in Group Processes, Volume 27, 269–296 Copyright r 2010 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 0882-6145/doi:10.1108/S0882-6145 (2010)0000027013

269

270

EDWARD J. LAWLER AND LENA HIPP

Corruption remains a core governance problem around the world. It undermines public trust in the democratic process and damages individual lives.

The annual report of Transparency International, the worldwide nongovernment organization intended to fight corruption, identifies four important facts about corruption: (1) Corruption is widespread; (2) corruption occurs in politics as well as in private business; (3) corruption threatens democratic processes; and (4) corruption causes great costs to consumers and citizens. Its immense negative consequences have made corruption a relevant field of study for economists and political scientists (Della Porta & Me´ny, 1997; Della Porta & Rose-Ackerman, 2002; Lambsdorff, Taube, & Schramm, 2005). Corruption, moreover, is a social phenomenon of generic interest to sociologists (e.g., Ashforth & Anand, 2003; Granovetter, 2007). Several questions can frame or provide the focus for an analysis of corruption. Why do people engage in corruption? How can corrupt relationships be stable given that there are no external enforcement mechanisms? And, given this stability, why is it that corrupt exchanges are ultimately revealed? A common and convincing answer to the first question is that the temptations to engage in corrupt behaviors are too great to be resisted, especially in the absence of effective control and monitoring (e.g., Klitgaard, 1988). The second and third question, however, have received less attention in the academic research on corruption (Lambsdorff, 2002a, 2002b; Lambsdorff, et al., 2005 are notable exceptions). To answer these questions, this chapter theorizes corruption as a special form of social exchange (Emerson, 1972a, 1972b; Lawler, 2006; Molm & Cook, 1995) and analyzes the internal dynamics of corrupt relationships. Recent social exchange theory and research has made significant strides in understanding how social exchanges generate social relations. Social structures generate internal, endogenous dynamics that create relational ties by (a) reducing uncertainty and risk (Kollock, 1994; Molm, 1994; Molm, Peterson, & Takahashi, 1999) and (b) promoting positive emotions and affective sentiments about the social relation (Lawler & Yoon, 1996, 1998). We show how these findings apply to corrupt exchanges but suggest that relational ties in the context of corruption have the opposite effect, that is, they destabilize relationships. This has implications for illicit exchanges more generally, whether they involve corruption or not. A corrupt relation is conceptualized as an ongoing relational tie involving two or more people – A and B – who exchange valued resources or goods that are not their own. Thereby at least one of them misuses ‘‘entrusted power for

Corruption as Social Exchange

271

private gain’’ (Transparency International, 2008). This definition implies that at least one of the parties has an official position or role with power legitimized in a larger normative framework; and that party misuses the power granted to generate private benefits. Three dimensions distinguish corrupt exchanges from other types of social exchanges: (a) The exchange of benefits is illicit or illegitimate, given prevailing norms. (b) Once it occurs, parties have a joint interest in secrecy. (c) Finally, it generates negative externalities for third parties (i.e., imposes costs on), such as shareholders, constituents, or publics. Corrupt exchanges are involved when government officials have preferentially awarded contracts and received gifts or kickbacks, when financial analysts have exchanged insider, proprietary information about stocks to each other, and when universities have steered students seeking loans to select companies and received a share of profits from those loans. No organization is exempt from the potential for corruption involving an illegitimate exchange of benefits, services, or privileges. It is important to stress that this chapter focuses on the relational effects of corrupt exchanges rather than the sources or causes of corruption. We examine corruption emerging out of ongoing relationships rather than from random encounters between strangers or acquaintances. What issues or problems does a corrupt exchange pose for actors? How do they address or solve these problems? How do their efforts to solve problems associated with a corrupt exchange bear on the stability or instability of their larger relational tie? By posing these questions, this chapter is both an application of social exchange theory to corruption as a particular form of social exchange and an effort to extend the theory to social exchanges that are illicit, create a demand for secrecy, and have negative externalities for third parties. Our overall message is that corrupt relations contain the seeds of their own destruction. Corruption destabilizes legitimized relationships insofar as such exchanges unleash internal problems that actors solve by making themselves even more vulnerable to each other and to outsiders. Corrupt exchanges make people more dependent on each other while simultaneously balancing or equalizing their power, which should promote greater stability of the relation (Lawler, Thye, & Yoon, 2008; Lawler & Yoon, 1996). However, the new joint good (in the form of secrecy) produced by corrupt behaviors threatens this stability because of problems of enforcement and detection. Changes in power help the parties to corruption to solve the enforcement problem, by making their mutual interests and shared task more salient, but in doing so they are prone to move toward forms of exchange that entail more risk of detection. These risks of detection vary

272

EDWARD J. LAWLER AND LENA HIPP

with network configurations, in particular with how the actors are connected to others in the network. The chapter is organized into four sections. The first section reviews existing approaches to corruption. The second section introduces the social exchange framework. It describes the social context of corrupt exchanges and their fundamental problems of enforcement and detection. The third section elaborates on social exchange theory in more detail and analyzes (a) how corrupt exchanges change the power dependence relationship of the actors, (b) why corrupt relationships shift to forms of exchange that are riskier and more likely to be detected, and (c) how different network configurations exacerbate and mitigate risks of detection.

EXISTING APPROACHES TO CORRUPTION The vast majority of research on corruption aims to identify the main causes of corruption and the conditions under which it is likely to occur. The larger purpose is to develop policies for reducing or preventing corruption (see Donatella Della Porta & Rose-Ackerman, 2002; Donatella Della Porta & Vannucci, 1999, 2005; Klitgaard, 1988; Rose-Ackerman, 1999). There is a strong rationale for this applied focus. Corruption is generally perceived as a key threat to effective government and economic development (e.g., Klitgaard, 1988; Transparency International, 2008). Corruption has damaging effects on the modernization of underdeveloped nations, the costs of public projects, and the moral or ethical fabric of communities in which corrupt practices become standard, institutional patterns (e.g., Azfar, Lee, & Swamy, 2001; Rose-Ackerman, 1999; Transparency International, 2008). Overall, there has been little effort to develop theories of corruption or, in light of its relational property, to interpret corruption in sociological terms. Granovetter’s (2007) recent work on corruption is a notable exception. There are two primary approaches in previous work on corruption: (1) Economic (rational choice) approaches ask why individuals choose to engage in corrupt practices and target the incentives for one party, for example, a government official or an organizational representative, to accept bribes and for another party, for example, a client, a customers, or an ally to offer them. Principal–agent theory (Klitgaard, 1988) emphasizes the tie between the principal, that is, the damaged third party, and the agent, that is, the party engaging in corrupt behaviors. The newer transaction costs approach to corruption, in contrast, emphasizes the tie between the parties engaging in corrupt behaviors, for example, the briber and the bribed

Corruption as Social Exchange

273

(Lambsdorff 2002a, 2002b; Lambsdorff et al., 2005). (2) Sociological (institutional) approaches ask how corrupt practices become standard, taken-for-granted ‘‘ways of doing business,’’ for example, what norms or normative conditions promote or fail to prevent corruption. The emphasis is on the normative-institutional conditions under which professional or job-related effectiveness or success may promote or even require corrupt practices, despite a larger normative/legal framework casting them as inappropriate and illegitimate (e.g., Ashforth & Anand, 2003; Granovetter, 2007; Heidenheimer & Johnston, 2001). Below we briefly elaborate these two approaches. Principal–agent theory from economics is the dominant theoretical approach in existing literature on corruption. Its emphasis is the incentives of individuals (agents or officials) to engage in corrupt behaviors and the social-economic conditions that underlie these incentives. The ‘‘agents’’ of principal–agent theory are responsible and accountable to principals, but principals have difficulty overseeing the behaviors of agents. They have latitude and freedom to ‘‘cut deals’’ with those they serve without principals being aware of those private deals. Klitgaard (1988) argues that publicsector corruption is a result of public agents having monopoly power with high levels of discretion and little oversight. Using transaction cost economics, Lambsdorff and colleagues (2002a, 2002b, 2005) added a relational component to the analysis of corrupt relationships.1 Engaging in corrupt behaviors is only ‘‘worthwhile’’ when transaction costs of the relationship, that is, the costs of ensuring the desired outcomes and secrecy are lower than the expected gains. From these points of view, prevention efforts should be directed at changing the cost/benefit calculus of individual agents. This typically involves an enhancement of oversight and accountability by principals and the encouragement of whistle blowing. Norms, values, and the prescribed scripts of ‘‘how things are done’’ are the cornerstone in sociological and cultural explanations of corruption (Ashforth & Anand, 2003; Granovetter, 2007; Husted, 1999). The idea is that definitions and perceptions of what is a corrupt versus a legitimate behavior are socially constructed and therefore vary across time and place. Some behaviors may be common, tolerated practices in one context but considered corrupt in another. Paying a fee to a public official for receiving preferential treatment or asking a friend to give a job off to a relative despite his or her lack of qualification and skills, for example, may be viewed a legitimate thing to do in one country or group – or even an act of appreciation and loyalty – while it would be socially unacceptable and legally prosecuted in another. Sociological and cultural approaches to

274

EDWARD J. LAWLER AND LENA HIPP

corruption come out of empirical findings rather than from a unified theoretical framework (Banfield, 1958; Cohen, Pant, & Sharp, 1996; Husted, 1999). From this approach, prevention efforts should be directed toward the top leadership levels of an organization or a political elite, such that changes in values, norms, and expectations trickle down the organizational hierarchy of a public and private sector organizations. Economic and sociological approaches have distinct but not necessarily contradictory ideas about corruption. Economic (rational choice) theories focus on the individual-level incentive structure shaping the expected utility that agents ascribe to corrupt versus noncorrupt behaviors, whereas sociological (institutional) theories tend to emphasize how and why corrupt behaviors become widespread, standard practices. Rational choice theories may be well-suited to explain how corrupt behaviors start, whereas institutional theories may be well-suited to understand the implications of repetitive patterns of corruption becoming routine and taken for granted. An important difference is that in the former, actors make a choice to be corrupt based on cost/benefit calculations, whereas in the latter actors essentially ‘‘lack choice’’ because of institutional conditions and informal cultural expectations. Social exchange theory can incorporate elements of both rational choice and institutional approaches to corruption in the sense that it is an incentive-oriented theory, explicitly pointing to relational and, by implication, institutional effects of repeated patterns of interaction or exchange (Lawler, Thye, & Yoon, 2009, chap. 10). Recent social exchange theorizing is a foundation for elaborating the relational dimensions of corruption and analyzing the sources of stability and instability in relations involving corrupt exchange.

THE SOCIAL CONTEXT OF CORRUPTION A key tenet of social exchange theory is that people form and maintain relations with persons who they are dependent on for valued rewards or outcomes, and moreover that repeated exchanges are integral to social compared to economic exchanges (Emerson, 1972a, 1972b). Social exchange theory is built on the assumption that people respond to or seek increases in their outcomes, rewards, or profits, but that this need does not entail profit maximization (Molm & Cook, 1995). People ostensibly maintain an existing relational tie as long as it remains rewarding or beneficial, even though there may be better exchange opportunities available outside of the existing tie. They may stay in a relationship because of intangible or intrinsic benefits,

Corruption as Social Exchange

275

because they face less uncertainty and risk by remaining, or because the relational tie itself has expressive value. Using social exchange theory, we construe corruption as a repeated pattern of socially embedded exchanges, that is, corruption occurs within an ongoing network of social ties between parties. Close professional relationships may spawn occasional corrupt exchanges that over time expand in scope or become more common. The implication is that whereas corruption often involves an economic exchange (e.g., bribes for contracts), it is social exchange if there is an ongoing relational tie. Economic approaches to corruption tend to miss or downplay this social-relational dimension of corruption. In many settings, corrupt practices appear to develop in ongoing, normatively legitimate relationships within which two or more parties provide each other valued outcomes over time. Corrupt exchanges occur and are sustained if they affirm an ongoing legitimate relation and that larger tie helps to secure or guarantee the secrecy of the corrupt exchange. Consider an example. In 1989, Daniel arap Moi, then President of the Republic of Kenya, received a series of ‘‘personal donations’’ in connection with the construction of duty-free complexes in the airports in Nairoibi and Mombassa from the ‘‘World Duty Free Company.’’ At the International Centre for Settlement of Investment Disputes in June 2000, Mr. Nasir Ibrahim Ali, the CEO of Word Duty Free Company testifies: ‘‘As a leading businessman in Dubai, I met in normal business circles, one Rashid Sajjad, a Kenyan, who would come to Dubai frequently in order to buy goods for export to Kenya for his company [y] From my discussions with Sajjad, I came to understand that he was politically and powerfully connected in the Kenya Government. Wishing to diversify my business from perfumes into the duty-free market, I raised my interest with Sajjad and asked his advice on arranging the necessary licenses and authorization for the establishment of duty free complexes in Nairobi and Mombassa airports. Sajjad informed me that he would arrange meetings for me with the relevant officials in Kenya. [y] Sajjad informed me that although my concept for establishing duty-free complexes at Nairobi and Mombassa airports to an international standard would require heavy investment, which I believed would be for the national benefit of Kenya, protocol in Kenya required that I should in addition make a ‘personal donation’. I was given to believe, that this was payment for doing business with the Government of Kenya. [y] I felt uncomfortable with the idea of handing over this ‘‘personal donation’’ which appeared to me to be a bribe. However, this was the President, and I was given to understand that it was lawful and that I didn’t have a choice if I wanted the investment contract’’ (TDM, 2006, p. 37).

276

EDWARD J. LAWLER AND LENA HIPP

This quotation illustrates several points. First, Mr. Ali essentially slid down a ‘‘slippery slope’’ in the context of an existing, ongoing relationship in which legitimate exchanges took place. Some approaches assume that corrupt exchanges stem from pure, unbridled opportunism, but this is not necessarily the case when there is an ongoing, legitimate relation or association (see Ashforth, Gioia, Robinson, & Trevino, 2008, for a related discussion). People can inadvertently cross a line and not recognize it until afterwards. Second, those in an ongoing relationship are dependent on each other for valued social or economic goods, and thus, they have an incentive to sustain if not nurture their relationship, especially when the ‘‘shadow of the future’’ is long or indefinite. This is why Mr. Ali responded to Mr. Saijad’s requests. Third, by engaging in a corrupt exchange, the relationship of the parties changes in a significant way. Over time, Mr. Sajjad became more explicit in his requests for bribes, as Mr. Ali further describes in his testimony. ‘‘During the negotiations of the investment agreement, I was frequently travelling backwards and forwards between Dubai and Kenya. During this time I received several requests for gifts to bring for officials in the Kenyan Government. I was not given any money in return for these items and it appeared to me that it was expected that I would bring what they requested. I received a request by fax from Sajjad on 2 May 1989 with a shopping list for watches [y]. On 3 May 1989 I received a telex from Sajjad requesting a gift of the latest model Polaroid camera’’ (TDM, 2006, p. 38). These requests for more and more gifts show how corrupt exchanges in the context of a legitimate relation can evolve in subtle ways. Keeping the exchange hidden from others to combat detection is a new dimension in the relationship that needs to be ensured partners. Secrecy emerges as a new joint good, exerting additional pressure on the ongoing relationship. Moreover, the example illustrates that those who engage in a corrupt exchange face two fundamental problems: how to enforce or guarantee that the exchange occurs as agreed to, and how to avoid detection by outsiders to the exchange relation. The enforcement problem of corrupt exchanges stems from the fact that when these exchanges occur, there are no standard external enforcement mechanisms (e.g., judicial institutions) as there are for legitimate exchanges (Lambsdorff 2002a, 2002b). Neither party has a legitimate avenue of recourse if the other fails to live up to the implicit or explicit bargain or understanding. As a result, corrupt parties tend to develop their own enforcement mechanisms outside of legitimate channels but internal to the relationship (e.g., development of a close, personal tie, threats of violence).

Corruption as Social Exchange

277

The detection problem complements the enforcement problem, because what is at issue is not just whether the exchange occurs as agreed to, but also whether each party will work to maintain the secrecy of their corrupt exchanges. In corrupt exchanges, each party has an incentive to collaborate on its continual reproduction and reasons to worry whether the other will do likewise. Given that the tie between two corrupt exchange partners is embedded in a larger network, information can flow to others or those others may notice patterns that raise suspicion. Thus, theoretically, the confluence of enforcement and detection problems makes corrupt exchanges difficult to sustain, unless they are nested in a larger, ongoing relationship that provide parties with the capacity to solve these problems internally and endogenously. A larger, ongoing relationship helps to guarantee and sustain corrupt practices to the degree that it fosters high interpersonal trust, valued legitimate rewards that can be withheld in the future, and/or credible threats of harm. Yet, internal solutions, even in the context of socially embedded ties, may be unstable and fragile over time. Ideas from social exchange theory suggest this. Social exchange theory and research specifies several internal mechanisms whereby people may develop the capacity to mutually influence and control each other: trust conceived as ‘‘encapsulated interest’’ (Cook, Hardin, & Levi, 2005); relational cohesion or the sense of a unifying relational tie (Lawler & Yoon, 1996); or positive affective sentiments toward each other or that relational tie (Lawler, 2001; Molm, Schaefer, & Collett, 2009). We will elaborate each of these mechanisms in the following sections. The question guiding our argument is: How well can such internal solutions to enforcement and detection problems hold in the case of corrupt exchanges that demand or require secrecy? We theorize that parties to a legitimate relation who engage in an illicit exchange face a dilemma they cannot solve. The dilemma is that solutions to the enforcement and detection problems tend to operate at cross-purposes. If in solving the enforcement problem, parties strengthen interpersonal trust by developing a closer social tie, this could raise suspicion and make their corrupt behaviors more vulnerable to detection by outsiders. If the detection problem is solved by collaborating more explicitly to maintain secrecy, the actors become even more dependent on each other’s trustworthiness and multiple layers of secrecy then become critical. In general, mitigating enforcement problems tend to enhance the detection problems and vice versa. This has implications for how and why internal, endogenous effects of corrupt exchanges create the ‘‘seeds of relational destruction.’’

278

EDWARD J. LAWLER AND LENA HIPP

THE SOCIAL EXCHANGE FRAMEWORK Social exchange theory interconnects social structures, transactions between people, and relational effects or consequences. Structures of dependence and interdependence are the basis for incentives that people have to interact and exchange with some, but not others in their network; transactions are the efforts of those who are interdependent on each other to exchange valued benefits or goods; and relations are the result of repetitive transactions (exchanges) among the same people over time (Emerson, 1981; Lawler & Thye, 2006). From the theory, repetitive transactions are likely to persist as long as social structures entail stable incentives for two actors to interact; yet, relations that form under such conditions also can produce changes in the structure. This is especially so, if the relations where the most frequent interactions occur are set off from other less frequently enacted ties in the network, generating ‘‘pockets of cohesion’’ in the network (see Lawler & Yoon, 1998). Such effects should be common, if not exaggerated, in the case of corrupt exchanges, because joint interests in secrecy emerge with corrupt behaviors and make the relational tie even more distinctive from other ties. The development of exchange relations – that is, repeated, regularized patterns of interaction – has been tied to three fundamental conditions: (1) the power and dependence relations between the actors, that is, whether they have equal or unequal power (Lawler & Yoon, 1996); (2) the interconnections of the actors’ behaviors in the transferring benefits to each other, that is, whether the exchange is negotiated, reciprocal, productive, or generalized (Emerson, 1981; Lawler, 2001; Molm, 1994); and (3) the shape of the larger network, that is, how the relational ties within it are interconnected (Emerson, 1972b; Molm & Cook, 1995; Willer, 1999). Applied to corrupt exchanges, each of these conditions poses a central question: Whether and how corrupt exchanges shift the power relation of the people involved? Whether and how over time corrupt exchanges lead people to coordinate and connect their individual actions in different ways? Whether and how corrupt exchanges affect the prospects that others in the network will become aware of focal actors’ corruption? The following sections address each of these questions.

Power and Dependence in Corrupt Exchanges The social exchange approach to power assumes that power is a structurally based potential that is relational in form and grounded in interdependencies.

Corruption as Social Exchange

279

An actor’s (A) power over another (B) is based on the dependence of B on A (Emerson, 1972b). The amount of A’s power is determined by (1) the value B places on the outcomes or rewards provided by A, and (2) the availability of alternative As from whom those outcomes or rewards can be acquired by B. Applied to a corruption context, corruption tends to occur when A has exclusive control over a resource valued by one or many Bs and can command a high price for illicit services – for example, a procurement officer choosing an outside contractor among many who could do the job (see Klitgaard, 1988). Corruption may also occur in a context where there are many As and Bs available, and pairs of actors form relational ties and engage in illicit exchanges through which the actors gain an advantage vis-a`vis others – for example, trading insider information among financial analysts. By analyzing the dependencies and interdependencies in corrupt relations, new light can be shed on the problems of enforcement and detection explained earlier. A central implication of power dependence theory is that power is a nonzero sum phenomenon (Bacharach & Lawler, 1981; Lawler, 1992; Molm, 1987). Each person’s power can increase or decrease simultaneously, because mutual dependencies can and do grow or decline in ongoing relationships. An increase in one party’s power does not by definition imply a decrease in another’s power. This has led to a distinction between the ‘‘total power’’ in a relationship, that is, the sum or average power across actors (Bacharach & Lawler, 1998; Molm, 1987), and the ‘‘relative power’’ in that relationship, that is, the difference in power among the actors (Bacharach & Lawler, 1981; Lawler & Yoon, 1996; Molm, 1987). Total power can vary with relative power constant, just as relative power can vary within a given level of total power. Corrupt exchanges can change the power relationship along either or both of these dimensions, for example, by increasing the mutual dependence (total power) of the actors or by making them more or less equal in power. Power dependence theory helps to explain why many developing countries face persistent patterns of corruption in their public agencies. Here, private actors (e.g., individuals or companies seeking contracts) are often highly dependent on poorly paid, low-status government agents for valued resources and outcomes, but these government officials have substantial control over the allocation and distribution of valued resources (Donatella Della Porta & Vannucci, 1999; Klitgaard, 1988; Rose-Ackerman, 1999). Even though the agent (A) may occupy a small niche in a large bureaucracy and have relatively little status or influence within the organization or society, he or she may have significant power vis-a`-vis contractors in bidding

280

EDWARD J. LAWLER AND LENA HIPP

for work. The agent (A) has many potential contractors (from whom he can potentially accept bribes or kickbacks); B has few public officials (perhaps only one) with access to the contracts desired. The agent’s exclusive access to valued government contracts, combined with high discretion in allotting them, gives A significant power over B because B is highly dependent on A. In exchange network terms, this is a ‘‘strong power branch’’ network (Willer, 1999), where a central actor has resources valued by many others who do not have resources of similar value to the central actor or to each other. Such a branch network captures the main structural condition underlying Klitgaard’s (1988, p. 75) argument that corruption of public agents is a function of monopoly power of a resource, along with wide discretion to use that resource and a lack of accountability. The principal–agent explanation is that together these three conditions maximize information asymmetries between the agent and principal; the principal–agent tie accounts for the corrupt agent–contractor tie. Social exchange theory shifts the interpretive focus to the agent–contractor relation and suggests how corrupt exchanges change that relation.2 Research on social exchange indicates that increases in total power (mutual dependence) or equal, compared to unequal, power generate more cohesion in and commitment to an exchange relation (Lawler & Yoon, 1993, 1996, 1998; Thye, Yoon, & Lawler, 2002). Mutual dependencies (total power) generate stronger incentives for collaboration, and equal power makes exchanges easier by avoiding equity and justice issues that tend to occur in relations of unequal power. These general effects of ‘‘total power’’ and ‘‘relative power’’ help to disentangle the stabilizing and de-stabilizing features of a corrupt exchange. If corruption increases mutual dependence (total power) in the relationship, it provides corrupt actors with additional incentives to collaborate in the production of new joint goods. If corruption has a balancing (equalizing) effect on the power relationship, it makes it possible for corrupt actors to interact as full partners. Both of these conditions should promote trust, defined as ‘‘encapsulated interests,’’ that is, the belief that each person’s future behavior will take the other’s interests into account (Cook et al., 2005). The increase in total power is manifest in the ‘‘fact’’ that once two people with an ongoing relational tie engage in a corrupt exchange, they each become more dependent, each on the other. This signals a qualitative change in the relationship due to secrecy, becoming a new joint good that is salient, highly valued, and necessary to avoid the costs of detection. Each has a strong interest in continuing to produce and re-produce this joint good, and each knows that the other knows they are highly dependent on each other in

Corruption as Social Exchange

281

this respect. Moreover, each understands that is it truly a joint good that can only be generated, collectively and collaboratively. However, given a lack of certainty, they are likely to watch for or create opportunities to affirm each other’s intent to continue to cooperate and maintain secrecy. This may be done by engaging in new varieties of corruption that presume and test the trust of each in the other and by developing a closer personal relationship. Parties to corruption are also likely to make inferences about each other’s competence and willingness to maintain secrecy. If B is more careless than A or has close ties to other Bs who may suspect their corruption, this is a threat to the ongoing relationship between A and B. The upshot is that total power strengthens the joint interests of the corrupt actors, presents them with a new joint task (maintaining secrecy), and motivates them to cooperate in new ways to demonstrate their reliability and commitment to each other. Such changes, however, foster new vulnerabilities for each. One of the important findings from research on commitment in exchange relations is that relational ties are more cohesive if the jointly produced benefits are associated with positive emotions (i.e., energy or excitement as well as pleasure or satisfaction) and more frequent interactions or exchange (i.e., success at generating benefits for each actor). Those relations with higher total power or mutual dependence should generate more frequent interaction and exchange, and this repetition fosters more relational cohesion. The broad implication is that, all other things equal, corrupt exchanges should produce closer ties in which the frequency and domains of interaction and exchange expand. A purely professional tie may develop a personal dimension through more regular contacts outside of the professional setting, for example, informal meetings after work, helping each other with personal issues, getting to know each other’s families, and so forth. The relational tie thereby becomes multidimensional and, potentially, more visible to or detectable by outsiders. The underlying basis for the upward shift in total power – the need or demand for secrecy – also should have a power-balancing effect (Emerson, 1972b). In his classic statement of social exchange theory, Emerson (1972b) argued that exchange relations tend toward balance. Specifically, if an actor (A) has more power than another actor (B) within a relationship, A will use that power to extract more benefits or resources from B, which ultimately will lead toward a more balanced relationship. The power-balancing effect can occur either because A, the party that uses power, becomes more dependent on B, that is, those they receive the increasing benefits from, or those from whom they extract resources (B) become less dependent on A. We hypothesize that in the case of corruption, the former is more likely than

282

EDWARD J. LAWLER AND LENA HIPP

the latter, that is, an agent (A) with more power becomes more dependent on the other (B) as a result of a corrupt exchange. The importance of powerbalancing effects may be contingent on whether and to what degree the initial power relations among corrupt actors are unequal. We conjecture that the total power effects of corrupt exchanges are most fundamental and pervasive; power balancing effects are important especially in highly unequal power relations. To summarize, we propose that the insights of power-dependence theory are fruitful to understand how corrupt exchanges are sustained and how they change over time. Our main argument is captured by the following propositions on the impact of corrupt exchanges on power dependence relations between two actors: 1. Corrupt exchanges in an ongoing relation increase the mutual dependence (total power) of parties to that relationship insofar as it arouses a shared interest in secrecy (a new joint good). 2. A shared interest in secrecy promotes more frequent interactions within which actors affirm their intent to maintain secrecy and demonstrate their capacity to do so (competence). 3. Given 1 and 2, corrupt exchanges increase the relational cohesion of a corrupt social tie by further setting it off from other relational ties (pocket of cohesion effect). 4. When the initial relationship among corrupt actors is unequal, the total power effects are enhanced by concomitant shifts in power toward greater equality (power balancing effects). Given the above, we assert that collaborative responses to shared interests in secrecy may increase the likelihood that corrupt exchanges are detected over time. Why does a closer tie, based on high mutual dependence, not have the reverse effect – namely, protect the corrupt actors from detection by others? The reason is that collaboration responses to the secrecy problem shift the underlying form of exchange. To understand this, we apply standard forms of social exchange to corruption in the next section (Emerson, 1972b, 1981; Lawler, 2001; Molm, 2003b).

Forms of Corrupt Exchange Corrupt relations involve one of three forms of social exchange: reciprocal, negotiated, and productive.3 Reciprocal exchange involves acts of giving by each actor, separated in time and undertaken without explicit expectations

Corruption as Social Exchange

283

of when and even whether the favors will be reciprocated (Emerson, 1972b; Lawler, 2001; Molm, 1994, 2003a, 2003b). Watering plants for neighbors when they are away or reciprocating dinner invitations are examples of everyday forms of reciprocal exchange, as is the case of a politician who does a favor to a major campaign contributor. Negotiated exchanges, in contrast, involve the explicit, agreed-upon barter of valued benefits. The prototype is two actors with diverging interests, such as a management and a union representative, who exchange offers and counteroffers to reach mutually acceptable agreements or accords. Here the connection between the giving behaviors of each is known and understood to them, that is, the terms of the exchange are explicit. Whether bribes involve negotiated or reciprocal exchange depends on how close and explicit is the connection between giving behaviors. Productive exchange entails a joint or collective product that actors have an incentive to produce and reproduce. Actors essentially work on a common project that neither can do alone and they succeed only if each does their part, as it is the case for scholars jointly writing an article, for example. Racketeering forms of corruption tend to have elements of productive exchange, along with reciprocal and negotiated forms. Reciprocal exchange is the most basic form of corruption because it is the easiest and most straightforward way for parties to create a disconnect between their giving behaviors. The disconnection of A’s and B’s corrupt behaviors is maximized if the giving behaviors are separated in time, involve an implicit or informal understanding, and do not necessarily fall within the same reward or value domain.4 Through reciprocal exchange, actors are most able to conceal and blur the contingencies between their corrupt behaviors. However, while reciprocal exchange is optimal for parties to corruption, it is unstable because of a significant trust problem; actors cannot be sure that the other will reciprocate benefits or do so in a timely way. In Molm’s (2003a) terms this is the ‘‘risk of non-reciprocity,’’ reflecting the enforcement problem endemic to such types of exchange, legitimate or corrupt. Corrupt reciprocal exchanges accentuate the risk of nonreciprocity and therefore the enforcement problem, in part because the necessity of secrecy adds another layer of exposure to risk. Thus, corrupt exchanges foster a multilayered trust problem: Each looks for assurance that the other will reciprocate, but also that the other can and will maintain the secrecy of their transaction. The above-described example of the bribes paid by World Duty Free Company Limited to the former Kenyan president and government officials illustrates this problem. The link between the duty free complexes in the Kenyan airports and the payment of ‘‘personal donations’’

284

EDWARD J. LAWLER AND LENA HIPP

needed to be concealed from the outside, while it was not yet certain that the actual construction, maintenance, and operation of the duty free shops would occur. Based on recent social exchange theorizing, there are several ways that actors can address the enforcement problems internally, that is, within the confines of their relationship. We emphasize two that entail a shift in the form of the exchange, that is, the connections between their giving behaviors: (1) Develop explicit, mutually acknowledged terms of exchange, which involves a shift from reciprocal to negotiated exchange; (2) collaborate in new ways that take advantage of their shared interest in secrecy and affirm or demonstrate their commitment to their relationship, which involves a shift from reciprocal to productive exchange. These movements away from a purely reciprocal form of exchange to negotiated or productive exchange can occur in any ongoing legitimate relationship, but, importantly, corrupt exchanges make such shifts likely as mechanisms for dealing internally with the enforcement problems associated with reciprocal forms of illicit exchange. Each shift in exchange form may help to assure actors that the exchange will occur as expected and each can rely on the other to maintain secrecy. By shifting from reciprocal to negotiated forms of exchange, actors develop a better understanding of what they will do for each other and they can more effectively coordinate the value and frequency of benefits provided. The actors identify the terms of the exchange, make clear each actor’s expectations of the other, and reduce miscommunication. This is a highly effective way to deal with problems posed by the risks of nonreciprocity inherent in reciprocal exchange. A private contractor is not likely to provide a significant gift (bribe) to a public official unless the contractor believes this will result in a contract; nor is the corrupt public official likely to favor a particular contractor unless the official is relatively sure that a bribe or kickback will be forthcoming. Illegitimate or illicit exchanges increase the risks of nonreciprocity while giving actors few options for handling these risks. A larger, ongoing, legitimate relationship may require less explicit understandings about what each does for the other, but purely reciprocal corrupt exchanges are still likely to be unstable over time without more explicitness about who does what and when. The main point is that an endogenous problem (enforcement or the risk of nonreciprocity) encourages a shift toward negotiated exchange; yet making the terms more explicit enhances and complicates the problem of secrecy. Negotiated forms of corrupt exchanges are likely to involve revealing behavioral indicators of the contingencies between what each does for the other, and as a result,

Corruption as Social Exchange

285

the actors become more vulnerable to detection by outsiders and also to defection by each other.5 In productive exchange, the interests between parties are aligned, meaning that the joint benefits from ‘‘teamwork’’ are greatest, as long as each continues to invest the necessary time, effort, or resources. Coordinating behaviors to maintain secrecy is likely to involve productive exchange. In fact, the secrecy problem sets the stage for an expansion of the corrupt practices or exchanges because it makes actors more aware of their capacity to work together. With productive exchanges, there are single joint products and the individual contributions to them are not always explicit or clear, as in negotiated exchange. In the productive form, exchanges are essentially person-to-group and group-to-person (Lawler, 2001). Racketeering, a particular form of corruption, tends to have elements of a productive exchange. Only through tight collaboration fixing prices, paying bribes, and sometimes even using violence, for example, have garbage cartels in New York and New Jersey ensured gigantic gains for trash collection. An important feature of productive exchange is that it generates relational or group ties and, thereby, more stability, cohesion, or solidarity (Lawler, 2001; Lawler et al., 2008, 2009). In the case of corruption, there are two noteworthy reasons for this. (1) The incentive structure provides for and makes apparent actors’ ‘‘encapsulated interests’’ (Hardin, 2001), because a valued good can only be produced collectively, and the risk of mutual deceit and nonadherence are minimal. Productive exchange is an internal solution to the enforcement problem because it creates a mutualassurance game where actors’ main challenge is to coordinate their behaviors to produce a collective good from which all receive their highest payoffs. Coordination is the major problem, not free riding. (2) Successful productive exchanges tend to foster a stronger sense of shared responsibility than other forms of social exchange (for supporting evidence, see Lawler, et al., 2008), and this strengthens the felt tie to the social unit (a relation or group). The person-to-unit tie enhances the degree that the relational tie among corrupt actors is set off from others in a larger network. This unique, shared relational tie addresses the enforcement problem internally, by affirming a group identity and fostering a collective orientation to the secrecy problem. If corrupt behaviors take on the form of a productive exchange, however, they are more likely to be detected, especially in the long term. Productive exchanges mitigate the secrecy problem but leave behavioral traces of the collaboration that is involved in maintaining that secrecy. This is one reason

286

EDWARD J. LAWLER AND LENA HIPP

that corrupt actors are often prosecuted for crimes ancillary to their core corrupt activities, for example, failure to pay taxes or violations of environmental regulation. Moreover, by demonstrating that the actors can work together, productive exchange encourages them to find new areas of corruption, through which they can jointly produce private gain, but it leads them to take risks that they otherwise would not take. All things being equal, broader and more expansive areas of corruption and pro-active efforts to maintain secrecy should foster even stronger person-to-group ties and enhance mutual dependencies, but at the expense of expanding the range of activities subject to detection. To summarize, the following propositions extrapolate implications of how the forms of corrupt exchanges bear on the enforcement and detection problems facing the actors involved: 1. For corrupt exchanges, the likelihood of detection is lower in reciprocal forms of exchange than in negotiated or productive forms of exchange. 2. Reciprocal forms of corrupt exchanges evolve toward either (a) negotiated or (b) productive forms of exchange as actors reduce the risks of nonreciprocity and attempt to deal with the problem of secrecy. 3. Negotiated forms of corrupt exchange reduce the risk of nonreciprocity without eliminating it, but increase the secrecy problem (risk of detection). 4. Productive forms of corrupt exchange both reduce the risk of nonreciprocity and mitigate the secrecy problem (the risk of detection) in the short term, without eliminating these problems in the long term. 5. The degree of trust (as ‘‘encapsulated interests’’) produced by repeated corrupt exchange is ordered as follows: reciprocityonegotiatedoproductive. 6. The severity of the secrecy problem (risk of detection) for corrupt relations is ordered as follows: reciprocityoproductiveonegotiated. Thus, the trust problem in reciprocal exchange leads to forms of corrupt exchange that may solve the ‘‘risk of non-reciprocity’’ but fail to satisfactorily solve the secrecy problem. Negotiated exchange enhances the detection problem, whereas productive exchange reduces it in the short, but not in the long, term. Because corrupt exchanges are embedded in a larger network, it is plausible that actors’ ties to others in that network have a bearing on the capacity of A and B to successfully deal with the secrecy problem. The next section takes up this issue.

Corruption as Social Exchange

287

Corrupt Exchange Networks and Detection A core principle of contemporary social exchange theory is that dyadic exchanges should be viewed in the context of the larger network within which they are embedded (see Emerson, 1972b). The underlying theoretical claim is that interactions in a dyad depend not just on the tie of the people in that dyad but also on the ties they have to others. The underlying empirical claim is that ‘‘isolated dyads’’ are rare and, if and when they develop, even these emanate from a larger set of network ties. A network is defined as an opportunity structure, that is, a set of possible exchange partners available to one or both actors in a focal dyad. Patterns of exchange (who interacts with whom) generate a realized network within this set of possible ties (the opportunity structure). Minimally, an exchange network requires three actors, at least two of whom have potential ties. Since different network structures pose distinct secrecy challenges for actors engaged in corrupt exchanges, basic ideas about networks of exchange help us to further examine the problems of secrecy in corrupt exchanges. Dyadic ties in exchange networks can be connected in different ways. The social exchange literature identifies three types of network connections (Emerson, 1972b; Willer, 1999): (1) exclusively connected exchanges, (2) inclusively connected exchanges, and (3) null-connected exchanges. In a ‘‘exclusively connected’’ network, an exchange in one dyad excludes an exchange in another, so only a subset of the possible exchanges in a network actually occur. In an ‘‘inclusive’’ network, the benefits from a given tie requires all the others to be consummated as well; thus, a more dense set of network ties form and removal of any one (e.g., breaking secrecy) makes all ties vulnerable. In a ‘‘null’’ network, there is no direct connection between ties, so few or many ties may occur depending on the actors. These network connections reflect different incentive structures for actors pursuing exchange with others. Each type of network connection captures a different context of corruption and raises somewhat distinct secrecy or detection problems. Consider again the example of a government agent, who decides which contractor wins a bid to undertake a major construction project for the state. This describes a branch network that is exclusively or negatively connected. The agent (A) is the central actor with ties to many prospective contractors and only one of them can be chosen, and thus others have to be excluded. The contractors are in competition. They have an incentive to build a strong tie to or bribe the agent but those who lose the bid have an incentive to ‘‘rat out’’ those who win with a bribe.

288

EDWARD J. LAWLER AND LENA HIPP

A fully connected, equal power network reveals similar dynamics. Here each actor can exchange with any of the other actors, but none are assured of an exchange. Consider a network of financial analysts who are willing to trade proprietary information to achieve market advantages. Each has an incentive to find a partner for corrupt exchanges of proprietary information, but those that do not find such partners have an incentive to report those that appear to. The competitive forces in a exclusively connected network make secrecy highly problematic for corrupt exchanges. A plausible solution to these detection problems is to spread the contracts or information around over time such that many or all contractors receive contracts or all the financial analysts in a network receive valued private information. However, such arrangements are complicated and unstable. They involve explicit understandings among the entire set of actors in the network, which may solve the enforcement problem but will accentuate the secrecy problem. A single dissenting voice can unravel such an informal arrangement; moreover, it is difficult to maintain a parity or equality of benefits across individual actors, and those who are disadvantaged have an incentive to divulge the corruption to authorities. In some specific contexts, actors in the network develop norms and institutions that define corrupt practices (bribes, kickbacks, insider trading) as a standard, necessary business practice, something that is most likely in the context of a weak overarching legal framework as sometimes found at an international level or within developing countries (see Klitgaard, 1988). Our main point is that exclusively connected networks make it difficult for two parties engaged corruption to conceal it from peers. A network is inclusive if all exchanges have to occur for any party to benefit. For example, in an inclusive network the central actor (C) receives benefits only if he or she interacts and exchanges with each of the other actors in a network. Whereas this central actor’s payoffs are contingent on an exchange with only one actor in a exclusively connected network, C’s payoffs are contingent on exchange with more than one actor in an inclusively connected network. The use of a middleman (C) for corrupt exchanges is an example.6 The tie between A and C and C and B enable A and B to maintain a separation between their corrupt giving behaviors, although the A-C and C-B ties have an explicit, negotiated form. This is exemplified by a lobbyist (C) who acts as an intermediary between a politician (A) and interest group (B) willing to purchase influence by contributing to A’s campaign. The actors, A and B, can arrange an exchange of favors for contributions through the lobbyist (C). This ostensibly conceals the direct tie between A’s and B’s exchange behavior.

Corruption as Social Exchange

289

Group-based corruption (e.g., organized or ‘‘enterprise’’ crime) is another example of an inclusive network. Here the benefits to each individual participant are contingent on their contributions, as well as those of others, to the collective enterprise. Inclusive networks are a structural foundation for productive forms of exchange. Shared interests in secrecy contribute to preserving it, but more actors are involved and, as noted earlier, people tend to undertake more risky varieties of corrupt behaviors under such conditions and develop more involved and detectable internal enforcement mechanisms to avert defection. With a middleman (C) both A and B are dependent on C and due to the problem of secrecy, C can play one side off against the other as the teritius gaudens. Thus, promised benefits, such as information, bribes, and gifts may not be assured. In a null network there are neither positive nor negative connections between the exchange ties. Dyadic ties are connected through central or common nodes, positions, or actors in the network. Thus, an agent may benefit from exchanges with several others, but any given exchange or tie has no bearing on the agent’s ties to others in the network. The actors in the network can treat their ties to others as distinct and independent. Consider a building or restaurant inspector who regularly inspects a group of buildings or restaurants and therefore has an ongoing tie with the managers of each of those units. The inspector can overlook violations, work to resolve them informally, or issue formal citations for violations, thereby imposing reputational and monetary costs on the building or restaurant. Bribes in exchange for ‘‘going easy’’ on a building or restaurant are difficult to detect as long as the inspector and manager keep the arrangement private. Thus, the secrecy problem is primarily internal to parties involved in the corrupt tie; the network establishes the basis for this. In null networks, the internal locus or source of the secrecy problem gives actors an incentive to make expectations for each other more explicit, but it also encourages them to develop a friendlier, more personal relationship that expands beyond the professional tie. The multidimensionality of the tie, however, may increase the likelihood of detection, in part because the inspector gains a ‘‘reputation’’ over time since building or restaurant managers are likely to have some ties with their counterparts. The comparison of exclusive, inclusive, and null types of network configurations distinguishes different structural contexts for corruption and points to the structural sources of the secrecy problem. In a exclusively connected network, secrecy is problematic because of the competitive environment and incentives of those disadvantaged to reveal the corruption of others. In inclusive-networks, secrecy is problematic because more

290

EDWARD J. LAWLER AND LENA HIPP

network ties are involved and thus there is greater potential for a weak link. Collaboration across many actors is more difficult to conceal from external sources. In null or unconnected networks, the secrecy problem is internal to a dyadic corrupt tie and a matter of whether the actors can trust each other to maintain secrecy, and this leads them to develop a larger social tie securing the corrupt exchange. In exclusive and inclusive networks, the threats to secrecy are external, whereas in null networks the threat is internal. The solutions also differ across networks. There is no viable solution to the secrecy problem in a exclusively connected network, due to the competition for advantage. Reciprocal exchanges are unlikely to evolve toward negotiated exchanges in response to the enforcement problem because maintaining sufficient distance between A’s and B’s corrupt behavior is difficult. Corruption in competitive networks, where some have to be excluded, is fragile unless it is institutionalized. Inclusive networks introduce a group-basis for corruption by creating shared interests that can only be achieved in collaboration with others in the network. Middleman structures are inclusive, as are most instances of enterprise (organized) crime. Unlike the competitive incentives of exclusive networks, inclusive ones entail cooperative incentives and promote productive forms of corrupt exchange. Unconnected network ties (null) lead to separate dyadic ties, connected only indirectly through a common actor. Reciprocal corrupt behaviors in this context are likely to evolve toward more explicit, negotiated understandings of who does what and when, with informal agreements to maintain secrecy. To summarize, several propositions are implied by the above discussion: 1. In exclusively connected exchange networks, corrupt exchanges remain reciprocal in form with little evolution toward either negotiated or productive forms, because the competitive context enhances the threats to secrecy. 2. In inclusively connected exchange networks, corrupt exchanges become productive in form, because actors’ mutual dependence and shared interests in collaboration mitigate threats to secrecy. However, they increase the enforcement problem due to the prospects of malfeasance by the central actor in the network or ‘‘whistle blowing’’ by one of the peripheral actors. 3. In unconnected (null) exchange networks, corrupt exchanges become negotiated in form, because their mutual dependence enhances trust and they face only internal sources of secrecy.

Corruption as Social Exchange

291

Overall, social exchange theory has important implications for understanding how corrupt exchanges change ongoing relational ties. First of all, a corrupt exchange in a legitimate relation increases mutual dependence by creating a new joint good (secrecy) that actors have an incentive to produce and that also sets their relationship apart from many others they are involved in. Second, concomitant with this secrecy problem is an enforcement problem, that is, how trust can be assured in the absence of readily available third parties that are typically present for legitimate, normatively sanctioned relations. Third, in acting to solve these dual problems – detection and enforcement – actors tend to expand or change the domains of their interaction or exchange and thereby create new areas where detection or enforcement are problematic; that is, they make themselves more vulnerable. In this sense, social exchange theory suggests how corrupt relations evolve over time and implies that they incorporate the seeds of their own destruction.

CONCLUSIONS This chapter conceives of corruption as social exchange, involving two or more actors. The opportunity for corrupt exchange stems from an alreadyexisting social tie. This could be a professional tie such as that between a politician and contributor, a building inspector and building manager, or a government procurement officer and vendor. This social-grounding or embeddedness of corrupt exchanges enables actors to secure their corrupt exchange, that is, come to trust each other to provide the benefits expected and maintain the necessary secrecy. There are two fundamental problems of corrupt exchanges that make such relationships unstable. First, there are no legitimate ‘‘third party enforcement’’ mechanisms that can be accessed if a party does not fulfill her or his side of the bargain. The parties themselves have to secure the exchange. Second, given the corrupt exchange is illegitimate or illicit, there is pressure on parties to carefully maintain the secrecy of their corrupt exchanges. Again, this is a task they have to resolve themselves without outside assistance. The need for secrecy not only makes it important for them to understand each other’s commitment to secrecy but also each other’s willingness and competence to accomplish it repeatedly over time. How can corrupt relationships build stability despite no external enforcement mechanisms? The answer offered in this chapter involves two main points. (1) Corrupt exchanges increase each actor’s dependence on the

292

EDWARD J. LAWLER AND LENA HIPP

other, making them more mutually dependent, but also more vulnerable to each other. In legitimate exchange relations, increases in mutual dependence strengthen relational ties by ‘‘encapsulating’’ actors’ interests, promoting positive affective ties, and the like. In illegitimate or corrupt relations, we argue that these effects are countervailed by problems of detection and enforcement. The key reason is that a new joint good emerges – the need for secrecy – when people engage in a corrupt exchange. They have and know they have a shared interest in collaborating to generate this good; yet, the more they collaborate to produce it the more vulnerable they become to each other and to detection from the outside. The mutual dependence of corrupt actors increases their incentives to deal with the secrecy problem, but it remains a pervasive, unresolvable source of uncertainty about each other. (2) People can reduce uncertainty about each other by demonstrating or affirming their commitment to their exchange and their relationship. Reciprocal exchanges that maintain a sharp separation between giving behaviors may evolve toward more explicit, negotiated forms of corrupt exchange where understandings of who does what are clearer, or toward collaborative efforts that are organized around the secrecy problem and generate productive forms of corrupt exchange. Given these stabilizing factors, why are corrupt exchanges likely to be eventually revealed? Under normal exchange conditions, adding a new type of exchange to an existing relation strengthens the relationship and makes the larger tie more cohesive, but in the case of a corrupt exchange, we argue that such effects are likely to be short term. The reason is that a pattern of corrupt exchanges unleashes endogenous processes that are a threat to the larger relationship. In the case of corrupt relations, negotiated and productive exchanges entail greater detection problems. By seeking to hide their connections, parties to corruption integrate further ties into the corrupt exchange. This way, however, the problem of secrecy and detection, is only solved for the short but not the long run. The overall implication is that corrupt exchanges face a dilemma that actors cannot solve. All the ways through which parties seek to deal with the problems of enforcement and detection inherently constitute the ‘‘seeds’’ of relational destruction. To conclude, corruption can be construed as a special class of social exchange. It has three distinct properties: (a) it is normatively illegitimate in the context; (b) it generates negative externalities for a third party (e.g., the community or public); and (c) secrecy is a joint good that is necessary for parties to receive the benefits from corrupt exchange. The problems that we have identified with corrupt exchanges apply to other types of illicit

293

Corruption as Social Exchange

relationships, such as extra-marital affairs and secret organizations; such relations entail enforcement and detection problems that are difficult for actors to resolve without creating new layers or problems of secrecy. Any solution is temporary or tenuous; any solution also requires clearer demonstrations of reliability and commitment by each party, resulting in new points of uncertainty for each.

NOTES 1. Transaction-cost theory points to the costs of corrupt actors finding reliable (and corrupt) exchange partners, with whom they can be confident of the secrecy of the transaction. The main idea is that a unique feature of corrupt exchanges is that they cannot rely on any external enforcement mechanism. This is generally consistent with implications of a social exchange approach. 2. Principal–agent theory addresses the incentives that initially lead to corrupt exchanges and how patterns of corruption develop given those incentives. Social exchange theory can address the initial incentives and development of repetitive patterns as well, but we suggest that its major contribution is to offer a unique way of understanding what happens to an ongoing exchange relation when a corrupt exchange occurs. 3. Generalized exchange is a fourth form of exchange. It is indirect, that is, there is no direct connection between a given pair of actors (A and B). Although collective or enterprise corruption (e.g., organized crime syndicates) can involve elements of generalized exchange, this form is not included in our analysis. To deal with the problems of detection and enforcement, strong social institutions or a common group affiliation between givers and receivers would be necessary in a corrupt generalized exchange context. 4. Resources or behavioral capabilities, that actors within an exchange relationship value can either fall in the same or different exchange domains (Emerson, 1972a). The contractor, for example, values exclusive information about a bid, whereas the government official is interested in new job opportunities or financial rewards (different exchange domains). Two financial analysts, in contrast, may both value exclusive information (same exchange domain). 5. Negotiated and reciprocal exchanges involve a mixed motive setting in which actors have incentives to cooperate and defect (give and not give). In their purest forms (see Molm, 2003a; Molm, Schaefer, & Collett, 2007), negotiated exchange makes the diverging interests more salient to actors than reciprocal exchange, and this makes it easier for people in reciprocal exchange to develop cohesive relations. We assert that the secrecy problem of corrupt exchanges actually reverse these patterns – specifically, more explicit exchanges are a stronger foundation for cohesive relations than less explicit, reciprocal forms. 6. What we consider a middleman in the context of a corrupt exchange is essentially what Corra and Willer (2002) define as a ‘‘gatekeeper’’. A gatekeeper controls the access to resources valued by others but does not own them. In the

294

EDWARD J. LAWLER AND LENA HIPP

above example, the lobbyist controls the access to a politician but cannot ensure changes to legislation.

ACKNOWLEDGMENTS The authors thank Shane Thye and Jeongkoo Yoon for helpful comments.

REFERENCES Ashforth, B. E., & Anand, V. (Eds). (2003). The normalization of corruption in organizations (Vol. 25). New York: Elsevier. Ashforth, B. E., Gioia, D. A., Robinson, S. L., & Trevino, L. K. (2008). Re-viewing organizational corruption. Academy of Management Review, 33(3), 670–684. Azfar, O., Lee, Y., & Swamy, A. (2001). The causes and consequences of corruption. ANNALS of the American Academy of Political and Social Science, 573(1), 42–56. Bacharach, S. B., & Lawler, E. J. (1981). Bargaining: Power, tactics and outcomes. San Francisco, CA: Jossey-Bass. Bacharach, S. B., & Lawler, E. J. (1998). Political alignments in organizations: Contextualization, mobilization, and coordination. In: R. M. Kramer & M. A. Neale (Eds), Power and influence in organizations (pp. 67–88). Thousand Oaks, CA: Sage. Banfield, E. C. (1958). The moral basis of a backward society. Glencoe, IL: Chicago Free Press; Research Center in Economic Development and Cultural Change, University of Chicago. Cohen, J. R., Pant, L. W., & Sharp, D. J. (1996). A methodological note on cross-cultural accounting ethics research. The International Journal of Accounting, 31(1), 55–66. Cook, K. S., Hardin, R., & Levi, M. (2005). Cooperation without trust? New York: Russell Sage Foundation. Corra, M., & Willer, D. (2002). The gatekeeper. Sociological Theory, 20(2), 180–207. Della Porta, D., & Me´ny, Y. (1997). Democracy and corruption in Europe. London: Washington. Della Porta, D., & Rose-Ackerman, S. (2002). Corrupt exchanges: Empirical themes in the politics and political economy of corruption. Baden-Baden: Nomos. Della Porta, D., & Vannucci, A. (1999). Corrupt exchanges: Actors, resources, and mechanisms of political corruption. New York: Aldine de Gruyter. Della Porta, D., & Vannucci, A. (2005). The governance mechanisms of corrupt transactions. In: J. G. Lambsdorff, M. Taube & M. Schramm (Eds), The new institutional economics of corruption (pp. 152–177). New York: Routledge. Emerson, R. M. (1972a). Exchange theory, part I: A psychological basis for social exchange: Exchange rules and networks. In: J. Berger, M. J. Zelditch & B. Anderson (Eds), Sociological theories in progress (pp. 38–57). Boston: Houghton-Mifflin. Emerson, R. M. (1972b). Exchange theory, part II: Exchange rules and networks. In: J. Berger, M. J. Zelditch & B. Anderson (Eds), Sociological theories in progress (pp. 58–87). Boston: Houghton-Mifflin. Emerson, R. M. (1981). Social exchange theory. In: M. Rosenberg & R. H. Turner (Eds), Social psychology: Sociological perspectives (pp. 30–65). New York: Basic Books, Inc.

Corruption as Social Exchange

295

Granovetter, M. (2007). The social construction of corruption. In: V. Nee & R. Swedberg (Eds), On capitalism (pp. 152–172). Palo Alto, CA: Stanford University Press. Hardin, R. (2001). Conceptions and explanations of trust. In: K. S. Cook (Ed.), Trust in society (Vol. 2, pp. 3–39). New York: Russell Sage Foundation. Heidenheimer, A. J., & Johnston, M. (2001). Political corruption: Concepts and context. New Brunswick, NJ: Transaction Publishers. Husted, B. W. (1999). Wealth, culture, and corruption. Journal of International Business Studies, 30(2), 339–359. Klitgaard, R. E. (1988). Controlling corruption. Berkeley, CA: University of California Press. Kollock, P. (1994). The emergence of exchange structures – An experimental-study of uncertainty, commitment, and trust. American Journal of Sociology, 100(2), 313–345. Lambsdorff, J. G. (2002a). How confidence facilitates illegal transactions: An empirical approach. The American Journal of Economics and Sociology, 61(4), 829. Lambsdorff, J. G. (2002b). What nurtures corrupt deals? On the role of confidence and transaction costs. In: D. Della Porta & S. Rose-Ackerman (Eds), Corrupt exchanges: Empirical themes in the politics and political economy of corruption (pp. 20–35). BadenBaden: Nomos. Lambsdorff, J. G., Taube, M., & Schramm, M. (Eds). (2005). The new institutional economics of corruption (Vol. 64). London: Routledge. Lawler, E. J. (1992). Power processes in bargaining. The Sociological Quarterly, 33(1), 17–34. Lawler, E. J. (2001). An affect theory of social exchange. American Journal of Sociology, 107(2), 321–352. Lawler, E. J. (2006). Exchange, affect, and group relations. In: A. J. Trevino (Ed.), George C. Homans: History, Theory, and Method (pp. 177–202). Boulder, CO: Paradigm Publishers. Lawler, E. J., & Thye, S. R. (2006). Exchange theory of emotion. In: J. Stets & J. Turner (Eds), Handbook of the sociology of emotions. New York: Springerforthcoming. Lawler, E. J., Thye, S., & Yoon, J. (2009). Social commitments in a depersonalized world. New York: Russell Sage. Lawler, E. J., Thye, S. R., & Yoon, J. (2008). Social exchange and micro social order. American Sociological Review, 73(4), 519–542. Lawler, E. J., & Yoon, J. (1993). Power and the emergence of commitment behavior in negotiated exchange. American Sociological Review, 58(4), 465–481. Lawler, E. J., & Yoon, J. (1996). Commitment in exchange relations: Test of a theory of relational cohesion. American Sociological Review, 61(1), 89–108. Lawler, E. J., & Yoon, J. (1998). Network Structure and emotion in exchange relations. American Sociological Review, 63(6), 871–894. Molm, L. D. (1987). Extending power dependency theory: Power processes and negative outcomes. In: E. J. Lawler & B. Markovsky (Eds), Advances in group processes (Vol. 4, pp. 178–198). Greenwich, CT: JAI Press. Molm, L. D. (1994). Dependence and risk: Transforming the structure of social exchange. Social Psychology Quarterly, 57(3), 163–176. Molm, L. D. (2003a). Power, trust, and fairness: Comparisons of negotiated and reciprocal exchange. In: S. R. Thye & J. Skvoretz (Eds), Power and status (Vol. 20, pp. 31–66). Oxford: Elsevier. Molm, L. D. (2003b). Theoretical comparisons of forms of exchange. Sociological Theory, 21(1), 1–17.

296

EDWARD J. LAWLER AND LENA HIPP

Molm, L. D., & Cook, K. S. (1995). Social exchange and exchange networks. In: K. S. Cook, G. A. Fine & J. S. House (Eds), Sociological perspectives on social psychology (pp. 209–235). Boston: Allyn and Bacon. Molm, L. D., Peterson, G., & Takahashi, N. (1999). Power in negotiated and reciprocal exchange. American Sociological Review, 64(6), 876–890. Molm, L. D., Schaefer, D. R., & Collett, J. L. (2007). The value of reciprocity. Social Psychology Quarterly, 70(2), 199–218. Molm, L. D., Schaefer, D. R., & Collett, J. L. (2009). Fragile and resilient trust: Risk and uncertainty in negotiated and reciprocal exchange. Sociological Theory, 27(1), 1–32. Rose-Ackerman, S. (1999). Corruptions and government-causes, consequences, and reform. New York: Cambridge University Press. TDM/Transnational Dispute Management. (2006). World Duty Free Company Limited (Claimant) v. the Republic of Kenya (Respondent) (ICSIC Case No. Arb/00.07). Transnational Dispute Management. Thye, S. R., Yoon, J., & Lawler, E. J. (2002). The theory of relational cohesion: Review of a research program. In: S. R. Thye & E. J. Lawler (Eds), Advances in group process (Vol. 19, pp. 139–166). Oxford: Elsevier. Transparency International. (2008). Global corruption report. Cambridge, UK: Cambridge University Press. Willer, D. E. (1999). Network exchange theory. Westport, CT: Praeger Publisher.