Interpersonal Networks in Organizations: Cognition, Personality, Dynamics, and Culture (Structural Analysis in the Social Sciences)

This page intentionally left blank

Interpersonal Networks inOrganizationsCognition, Personality, Dynamics,and Culture

This book brings a social network perspective to bear on topicsof leadership, decision making, turnover, organizational crises,organizational culture, and other major organizational behaviortopics. It offers a new direction for organizational behavior the-ory and research by drawing from social network ideas. Acrossdiverse research topics, the authors pursue an integrated focuson social ties both as they are represented in the cognitions ofindividuals and as they operate as constraints and opportunitiesin organizational settings. The authors bring their twenty years’worth of research experience together to provide a programmaticsocial network approach to understanding the internal function-ing of organizations. By focusing a distinctive research lens oninterpersonal networks, they attempt to discover the keys to thewhole realm of organizational behavior through the social net-work approach.

Martin Kilduff is the Kleberg/King Ranch Centennial Professorof Management at the University of Texas at Austin. He is alsoeditor of Academy of Management Review (2006–8) and coau-thor of Social Networks and Organizations (with Wenpin Tsai;2003). He has served on the faculties of Penn State and INSEAD,and he has been a visiting professor at Cambridge University,London Business School, Keele University, and Hong Kong Uni-versity of Science and Technology.

David Krackhardt is Professor of Organizations at the HeinzSchool of Public Policy and Management and at the TepperSchool of Business at Carnegie Mellon University. Prior appoint-ments include faculty positions at Cornell’s Graduate School ofManagement, the University of Chicago’s Graduate School ofBusiness, INSEAD (France), and the Harvard Business School(Marvin Bower Fellow).

Structural Analysis in the Social Sciences

Mark Granovetter, editor

The series Structural Analysis in the Social Sciences presents approachesthat explain social behavior and institutions by reference to relationsamong such concrete entities as persons and organizations. This con-trasts with at least four other popular strategies: (a) reductionist attemptsto explain by a focus on individuals alone; (b) explanations stressing thecausal primacy of such abstract concepts as ideas, values, mental har-monies, and cognitive maps (thus, “structuralism” on the Continentshould be distinguished from structural analysis in the present sense);(c) technological and material determination; and (d) explanations using“variables” as the main analytic concepts (as in the “structural equa-tion” models that dominated much of the sociology of the 1970s), wherestructure is that connecting variables rather than actual social entities.

The social network approach is an important example of the strategyof structural analysis; the series also draws on social science theory andresearch that is not framed explicitly in network terms but stresses theimportance of relations rather than the atomization of reduction orthe determination of ideas, technology, or material conditions. Thoughthe structural perspective has become extremely popular and influentialin all the social sciences, it does not have a coherent identity, and noseries yet pulls together such work under a single rubric. By bringingthe achievements of structurally oriented scholars to a wider public, theStructural Analysis series hopes to encourage the use of this very fruitfulapproach.

Recent Books in the Series

Philippe Bourgois, In Search of Respect: Selling Crack in El Barrio(Second Edition)

Nan Lin, Social Capital: A Theory of Social Structure and ActionRoberto Franzosi, From Words to NumbersSean O’Riain, The Politics of High-Tech GrowthJames Lincoln and Michael Gerlach, Japan’s Network EconomyPatrick Doreian, Vladimir Batagelj, and Anujka Ferligoj, Generalized

BlockmodelingEiko Ikegami, Bonds of Civility: Aesthetic Networks and Political Ori-

gins of Japanese CultureWouter de Nooy, Andrej Mrvar, and Vladimir Batagelj, Exploratory

Social Network Analysis with PajekPeter Carrington, John Scott, and Stanley Wasserman, Models and

Methods in Social Network AnalysisRobert C. Feenstra and Gary C. Hamilton, Emergent Economies, Diver-

gent PathsAri Adut, On Scandal: Moral Disturbances in Society, Politics, and Art

Interpersonal Networksin OrganizationsCognition, Personality, Dynamics,and Culture

MARTIN KILDUFFUniversity of Texas at Austin

DAVID KRACKHARDTCarnegie Mellon University

CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo

Cambridge University PressThe Edinburgh Building, Cambridge CB2 8RU, UK

First published in print format

ISBN-13 978-0-521-86660-6

ISBN-13 978-0-521-68558-0

ISBN-13 978-0-511-42908-8

© Martin Kilduff and David Krackhardt 2008

2008

Information on this title: www.cambridge.org/9780521866606

This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press.

Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Published in the United States of America by Cambridge University Press, New York

www.cambridge.org

paperback

eBook (EBL)

hardback

Contents

Acknowledgments page ix

1 Introduction 1

I. Perceiving Networks2 A Network Approach to Leadership 133 An Analysis of the Internal Market for Reputation

in Organizations 394 Systematic Biases in Network Perception 595 Effects of Network Accuracy on Individuals’

Perceived Power 84

II. The Psychology of Network Differences6 Social Structure and Decision Making in an MBA Cohort 1017 The Social Networks of Low and High Self-Monitors 1318 Centrality in the Emotion Helping Network: An

Interactionist Approach 157

III. Network Dynamics and Organizational Culture9 Network Perceptions and Turnover in Three Organizations 181

10 Organizational Crises 20811 The Control of Organizational Diversity 236

12 Future Directions 259

References 275

Index 305

vii

Acknowledgments

We drew upon a number of published articles in preparing this book. Weare happy to acknowledge the sources of these articles here. We thank ourcoauthors on these articles for their contributions and thank the journalsfor permission to reuse these materials. We also thank Ranjay Gulati,David A. Harrison, and Ajay Mehra for helpful comments during thepreparation of the book.

Chapter 2 draws from Balkundi, P., and Kilduff, M. 2005. The tiesthat lead: A social network approach to leadership. Leadership Quar-terly, 16: 941–61. Chapter 3 includes material from Kilduff, M., andKrackhardt, D. 1994. Bringing the individual back in: A structural anal-ysis of the internal market for reputation in organizations. Academy ofManagement Journal, 37: 87–108. Chapter 4 (and parts of Chapter 1)draws from Krackhardt, D., and Kilduff, M. 1999. Whether close or far:Social distance effects on perceived balance in friendship networks. Jour-nal of Personality and Social Psychology, 76: 770–82. © 1999 by theAmerican Psychological Association. Adapted with permission. Chapter5 contains material reprinted from Krackhardt, D., Assessing the politicallandscape: Structure, cognition, and power in organizations, Administra-tive Science Quarterly, 35 (2) by permission of Administrative ScienceQuarterly, © 1990 Cornell University. Chapter 6 draws from the follow-ing three articles: Mehra, A., Kilduff, M., and Brass, D. J. 1998. At themargins: A distinctiveness approach to the social identity and social net-works of under-represented groups. Academy of Management Journal,41: 441–52; Kilduff, M. 1990. The interpersonal structure of decision-making: A social comparison approach to organizational choice. Orga-nizational Behavior and Human Decision Processes, 47: 270–88; andKilduff, M. 1992. The friendship network as a decision-making resource:Dispositional moderators of social influences on organizational choice.Journal of Personality and Social Psychology, 62: 168–80. © 1992 by theAmerican Psychological Association, adapted with permission. Chapter 7

ix

x Acknowledgments

contains material reprinted from Mehra, A., Kilduff, M., and Brass,D. J., The social networks of high and low self-monitors: Implicationsfor workplace performance, Administrative Science Quarterly, 46 (1) bypermission of Administrative Science Quarterly © 2001 by Cornell Uni-versity. Chapter 8 draws from Toegel, G., Anand, N., and Kilduff, M.2007. Emotion helpers: The role of high positive affectivity and highself-monitoring managers. Personnel Psychology, 60: 337–65. Chapter 9draws upon the following two articles: Krackhardt, D., and Porter, L. T.1986. The snowball effect: Turnover embedded in communication net-works. Journal of Applied Psychology, 71: 1–6 © 1986 by the AmericanPsychological Association (adapted with permission); and Krackhardt,D., and Porter, L. T. 1985. When friends leave: A structural analysis ofthe relationship between turnover and stayers’ attitudes. AdministrativeScience Quarterly, 30 (2) © 1985, adapted and reprinted by permis-sion of Administrative Science Quarterly, Cornell University. Chapter 10includes material from Krackhardt, D., and Stern, R. 1988. Informalnetworks and organizational crises: An experimental simulation. SocialPsychology Quarterly, 51: 123–40. Chapter 11 draws from two sources:Krackhardt, D., and Kilduff, M. 1990. Friendship patterns and culture:The control of organizational diversity. American Anthropologist, 92:142–54; and Krackhardt, D., and Kilduff, M. 2002. Structure, cultureand Simmelian ties in entrepreneurial firms. Social Networks, 24: 279–90. Finally, Chapter 12 includes material adapted from the followingsources: Ibarra, H., Kilduff, M., and Tsai, W. 2005. Zooming in andout: Connecting individuals and collectivities at the frontiers of organiza-tional network research. Organization Science, 16 (4): 359–71. © 2005,the Institute for Operations Research and the Management Sciences, 7240Parkway Drive, Suite 310, Hanover, MD 21076, USA, reprinted by per-mission; and Kilduff, M., Tsai, W., and Hanke, R. 2006. A paradigm toofar? A dynamic stability reconsideration of the social network researchprogram. Academy of Management Review, 31: 1031–48.

1

Introduction

Human beings are social creatures who depend on links to others toaccomplish many of life’s tasks. The networks of relations within whicheach person is embedded include family, friends, and acquaintances. Theembeddedness of human activity in such networks is true not just forprimal activities such as child-rearing but also for economic activitiessuch as finding a job (Granovetter, 1974).

Indeed, business organizations themselves are held together not onlyby formal relations of authority but also by informal links that connectpeople across departmental and hierarchical boundaries. Starting with theHawthorne studies (Roethlisberger and Dickson, 1939), researchers haveinvestigated the importance of informal networks for job satisfaction(e.g., Roy, 1954), organizational conflict (e.g., Whyte, 1948), workeroutput (e.g., Jones, 1990), organizational power (e.g., Brass, 1984), andmany other aspects of social and organizational life (see Kilduff and Tsai,2003, for a review).

Only recently, however, has research attention focused on actors’ per-ceptions of the structure of relations in social settings and on how actors’individual differences may affect the network positions they occupy.These topics – actor perceptions and actor individual differences – pro-vide the inspiration for our book. Actors’ perceptions of social networkswithin which they are embedded affect the decisions they make (see thediscussion in Burt, 1982, chapter 5), and these perceptions are subjectto considerable bias (Krackhardt, 1987a). How other people perceivethe structure of relations surrounding the individual affects not onlythe individual’s power to act (Krackhardt, 1990) but also the individ-ual’s reputation (Kilduff and Krackhardt, 1994). Actor individual dif-ferences can range from the visible attributes of ethnicity and gender(shown to affect patterns of centrality and exclusion in social networks –e.g., Mehra, Kilduff, and Brass, 1998) to specific personality character-istics such as self-monitoring (Snyder, 1974) that may be particularly

1

2 Interpersonal Networks in Organizations

predictive of individuals’ network positions (Mehra, Kilduff, and Brass,2001).

This is a book about the cognitive and personality distinctiveness ofindividuals and the ways in which such distinctiveness affects relation-ships in organizations. The different chapters in the book unfold sto-ries about how people perceive themselves and others in networks offriendship and advice, the biases people exhibit in their mental repre-sentations of who is connected to whom, the rewards and penalties thatpeople experience as a result of such biases, and ways in which individ-ual differences affect network positions and outcomes. We draw fromcognitive network theory and an emerging personality approach to socialnetwork positions to examine (in organizational contexts) perceptionsof networks, the psychology of network differences, and the dynamics ofsocial network turnover, crisis, and culture. Unlike conventional networkstudies that tend to focus on interchangeable position holders, our focusthroughout is on individual human beings and their distinctive patternsof network thinking and interaction. Across diverse research coveringorganizational behavior topics, we pursue an integrated focus on socialties both as they are represented in the perceptions of individuals and asthey relate to individual differences.

Perceiving Networks

Cognitions concerning social networks are important to the extent thatpeople are uncertain concerning who is connected to whom. People maytry to reduce such uncertainty by paying particular attention to theconnections of those who are prominent. Savvy network entrepreneurscan take advantage of such uncertain knowledge to create social capitalthat may be merely fleeting but can, nonetheless, be valuable. The follow-ing story illustrates how a prominent banker used his visibility tobestow social capital that could be traded by his protege for financialcapital:

At the height of his wealth and success, the financier Baron deRothschild was petitioned for a loan by an acquaintance. Reput-edly, the great man replied, “I won’t give you a loan myself; butI will walk arm-in-arm with you across the floor of the StockExchange, and you soon shall have willing lenders to spare”(Cialdini, 1989: 45).

The baron in this story assumed that perceivers scan the social net-work connections of individuals for signals concerning difficult-to-discernunderlying quality – such signals including connections to prominent

Introduction 3

others. Research interest in such cognitive interpretations of networkconnections increased throughout the 1990s, concurrent with the cog-nitive turn in sociological approaches more generally (e.g., DiMaggio,1997; Schwarz, 1998). Research concerning interorganizational relation-ships has increasingly focused on how network links affect perceivedreputation and status (Zuckerman, 1999). Social networks are not justpipes through which resources flow; these networks are also potentiallydistorting prisms through which actors’ reputations can be discerned(cf. Podolny, 2001).

Interest in a cognitive approach to social networks developed ear-lier in organizational behavior approaches than in more macro-orientedapproaches. Pioneering work suggested that organizations and envi-ronments interacted as networked cognitions in the minds of partici-pants: “what ties an organization together is what ties thought together”(Bougon, Weick, and Binkhorst, 1977: 626). Social equals in organiza-tions tend to change their perceptions to establish consensus concerningenvironmental changes, whereas people connected to high-status individ-uals tend to be overly influenced by these high-status individuals’ percep-tions of environmental change (Sampson, 1968; Walker, 1985). A recentreview of the relationship between network connections and perceptionsof the environment suggested that “knowledge emergence, as opposed toknowledge transfer, may occur . . . between social equals from differentsocial circles, rather than between dyads divided by differences in mutualesteem and power” (Ibarra, Kilduff, and Tsai, 2005: 366).

Building on this legacy of work in organizational behavior, our cog-nitive emphasis in this book is predicated on the finding that differentindividuals looking at the same networks tend to see different sets ofconnections (cf. Krackhardt, 1987a). To the extent that each individualoccupies a specific position in a social network, the complexity of thenetwork is likely to be viewed differently by each individual (Kilduff,Tsai, and Hanke, 2006). Some of these idiosyncratic views are likely tobe more accurate in terms of mapping more closely on to a consensuallyvalidated representation of the network determined by the agreement ofmembers of each interacting dyad. Such accuracy with respect to theorganizational advice network can correlate with the power to influenceothers (Krackhardt, 1990).

The perception of social networks begins as soon as an individual entersa new organizational context. People are motivated to generate an over-all picture of a social group that they have joined, they seek to identifysubgroups that might complicate or facilitate their putative plans, andthey look for others to whom they can attach themselves (cf. von Hecker,1993). Seeing the new interacting group into which they have just steppedas a distinct social system (cf. Campbell, 1958), people bring with them


preconceptions from their interactions in previous social systems concern-ing the expectation that friendship overtures are likely to be reciprocatedand the “transitivity” expectation that people who share a mutual friendwill be friends themselves (Heider, 1958). These expectations bias thecognitive maps that people develop to represent social networks (Krack-hardt and Kilduff, 1999). But people are not slaves to default expectationsabout friendship reciprocity and transitivity – people can also learn fromvivid experiences (Ahn, Brewer, and Mooney, 1992) that social life can beriven with gaps where one might have expected ties (Janicik and Larrick,2005).

Each of us brings to our organizational sense making a different recipefor constructing representations of social networks. To the extent thatpeople are cognitive misers who try to economize on memory demands(Fiske and Taylor, 1991), their mental representations of social networksare likely to exhibit simplifications such as excess clustering of peopleinto densely connected groups and overattribution of popularity to peo-ple perceived to be central (Kilduff, Crossland, Tsai, and Krackhardt,forthcoming).

Individualizing Networks

In bringing the individual back into social network research, we empha-size in this book not only the importance of individual cognition but alsoraise questions concerning how different types of people establish differ-ent network positions and experience different network outcomes. Peoplediffer with respect to whether or not they occupy brokerage positions insocial networks (Burt, 1992) and outcomes from brokerage include bothbenefits such as higher job performance ratings (e.g., Mehra, Kilduff, andBrass, 2001) and potential costs such as reputation loss (e.g., Podolny andBaron, 1997). Despite this exploration in prior literature on the outcomesof brokerage, we still know relatively little about why some individualsrather than others are more central in social networks and occupy bro-kerage positions (Burt, 2005: 28).

We explore in this book the likelihood that the patterning of socialrelations in organizations – including the elevation of some individuals topositions of centrality and brokerage – derives from stable individual dif-ferences. Visible individual differences such as ethnicity and gender func-tion as bases for identification and network formation (Hughes, 1946).People tend to interact with similar others in organizations and this isparticularly true for relations, such as friendship, that are more expres-sive than instrumental (Blau, 1977). Together with exclusionary pressuresfrom the majority, this preference for similar, or “homophilous,” others

Introduction 5

may contribute to segregation within informal networks (Brass, 1985)and the marginalization of minority members.

But going beyond this emphasis on demographic differences, we alsoexplore in the book the likelihood that brokerage is related to self-monitoring personality orientation. Those high in self-monitoring resem-ble successful actors in their ability to play different roles for differ-ent audiences (Snyder, 1987). Self-monitoring, in comparison to otherpersonality variables, may be particularly relevant to the prediction ofbrokerage because of the theoretical (Day and Kilduff, 2003) and empir-ical (Flynn, Reagans, Amanatullah, and Ames, 2006) emphases on howpersonal identity affects the structuring of relationships. Other major per-sonality variables tend to suffer from limited predictive validity when itcomes to explaining why some individuals are more central than others(see, for example, the exemplary investigation of the network correlatesof the Big Five personality constructs by Klein, Lim, Saltz, and Mayer,2004). Because high self-monitors compared to low self-monitors tend toadapt their underlying personalities to allow themselves to become part ofdistinct social groups (Snyder and Gangestad, 1982), self-monitoring ori-entation is one key factor in understanding how individuals span acrosssocial divides in organizations.

Positioning the Book

Interest in social networks has increased rapidly over the past decade, butfew books focus specifically on interpersonal networks within organiza-tions, and none pursue the topics we cover here. There are some excellentrecent research monographs. The book by Noah Friedkin (1998) entitledA Structural Theory of Social Influence is unusual in bringing a socialpsychology approach to bear on questions of influence from a social net-work perspective. But there are few topics of overlap between that bookand our own. The recent book by Peter Monge and Noshir Contractor(2003) entitled Theories of Communication Networks takes a program-matic approach in synthesizing the authors’ collaborative research indeveloping a multitheoretical and multilevel model. Again, there are fewtopics of overlap here. We see both of these books as companions toour own rather than as rivals. One of us has coauthored a recent socialnetwork book that critiques and extends social network theory in generalin the context of offering a theoretical and methodological introduction(Kilduff and Tsai, 2003). There is one chapter in that book (pages 66–86) that urges researchers to pursue structural research from a cognitiveand individual difference perspective in pursuit of questions that haveoften been neglected. We see that chapter as whetting the appetite for


a more programmatic and comprehensive treatment of such topics asthey apply to organizational behavior. Finally, the recent edited book byCross, Parker, and Sasson (2003) focuses on networks in the knowledgeeconomy with a particular appeal to managerial rather than researchconcerns.

In studying interpersonal networks within organizations, we embracea realist philosophy of science in terms of a research focus on three levels:the actual, the perceived, and underlying structures. The actual networkof relationships in an organization can be perceived and experiencedby individuals in many different ways, and, thus, actual and perceivednetworks can be discrepant with each other for any specific individual.In terms of the tendency for perceived networks to adhere to structuralpatterns, we know that perceptions of social relations tend to be shaped bycognitive heuristics such as the balance schema that individuals employ tomake sense of complex realities (Krackhardt and Kilduff, 1999). Actualnetworks are also structured by underlying tendencies – for example,the tendency for people to cluster themselves together on the basis ofsimilarity on dimensions that are considered important (such as ethnicityand gender – cf. Mehra et al., 1998). Our research engages all three levelsof analysis, and investigates the discrepancies and tensions between theselevels.

We anticipate that the book will advance theory and research con-cerning organizational behavior and also push forward the social net-work research program itself. In tackling issues at the microbehaviorlevel within organizations, we bring a traditionally sociological approach(structural social network theory) to dwell on topics (such as organiza-tional turnover) typically studied from a more psychological approach.This book synthesizes research interests across the micro–macro divideto open new arenas for social network theory and methods. In bringinga distinctive research lens focused on interpersonal networks, we hope tounlock the whole realm of organizational behavior to the social networkapproach.

Overview of the Book

This book emphasizes the importance of interpersonal networks, particu-larly friendship networks, for understanding people’s behaviors in organi-zations. There are three major sections, following this introduction. In thefirst part – “Perceiving Networks” – we focus on how individuals perceivenetworks in organizations, and explore the consequences of such percep-tions. In the second part – “The Psychology of Network Differences” –we analyze how individuals differentially draw upon network resources,with particular attention on how network position and individual

Introduction 7

personality contribute to performance outcomes. In the third part – “Net-work Dynamics and Organizational Culture” – we study how individualsin organizations respond to network influences, looking at turnover, cri-sis, and culture. A common theme runs throughout the book: We arebringing the importance of individual cognition, personality, and actionback into a network research area that has tended to neglect if not com-pletely ignore the importance of the microfoundations of structural con-straint.

Chapter 2, “A Network Approach to Leadership,” is a key resourcefor the whole book in providing a focused review of our major themesand their relevance for leadership in organizations. In this chapter, wearticulate four interrelated principles that generate network theories andhypotheses and present a theoretical framework of leader effectivenessfrom the perspective of cognitive network theory. In Chapter 3, “AnAnalysis of the Internal Market for Reputation in Organizations,” weaddress whether perceptions of networks matter more than reality. Weaddress how network perceptions are aggregated to create “real” net-works, and how misperceptions of networks affect competitive outcomesin organizations, such as the reputations of individuals as good perform-ers. We look specifically at the question of whether, if you are perceivedby others in the organization to have a prominent friend, will this affectothers’ perceptions of your job performance. In Chapter 4, “SystematicBiases in Network Perception,” we continue our focus on the systematicbiasing of perceptions of organizational networks. We develop the themeof whether boundedly rational people in organizations tend to rely onheuristics to establish the friendship boundaries around themselves andothers. In Chapter 5, “Effects of Network Accuracy on Individuals’ Per-ceived Power,” we examine the consequences of accurate perceptions ofsocial networks in relation to individuals’ political power in organiza-tional settings.

The second part of the book, “The Psychology of Network Differ-ences,” focuses on the use of networks with respect to decision making,individual performance in organizations, and helping behaviors. In termsof bringing the individual back in, we examine the possibility that indi-viduals’ network positions are, to some extent, an expression of individ-ual personality. In Chapter 6, “Social Structures and Decision Makingin an MBA Cohort,” we address the issue of how people make deci-sions about complex issues, drawing upon network research that inves-tigated such decision making in an environment overflowing with rela-tively complete information. We trace how individuals group themselvesinto clusters on the basis of ethnicity and gender, examine the extent towhich the cohesion and structural equivalence perspectives predict theseindividuals’ decision making, and look at whether self-monitoring per-sonality orientation offers a basis for understanding why some people


relative to others tend to draw more heavily upon network resources inmaking complex decisions. We follow up this self-monitoring theme inChapter 7, “The Social Networks of Low and High Self-Monitors,”in our examination of whether low and high self-monitors build distinctlydifferent network structures and whether self-monitoring and networkposition combine to affect individual performance in organizations. Thischapter continues our attempt to open a productive new seam of struc-tural research that brings together psychological richness at the individ-ual level and sociological context at the network level. In Chapter 8,“Centrality in the Emotion Helping Network: An InteractionistApproach,” the last chapter in Part II, we again examine the twin effectsof network position and personality, this time with respect to centralityin the emotion helping network in organizations.

The third part of the book, “Network Dynamics and OrganizationalCulture,” takes a more dynamic perspective concerning how individualsin organizations are influenced in their behaviors and attitudes by thoseto whom they are connected either cognitively or actually. Chapter 9,“Network Perceptions and Turnover in Three Organizations,” investi-gates, from a social network perspective, the process and consequencesof people leaving organizations. If someone occupying a role similar tomy own leaves, how likely am I to also quit the organization? If I decide tostay despite the fact that a friend has left, what will be my attitude towardthe organization – more or less committed? Chapter 10, “OrganizationalCrises,” continues the theme of network influence between and withinorganizational units in focusing on how internal and external friendshipties affect organizations’ responses to crises. Chapter 11, “The Controlof Organizational Diversity,” advances a distinctive approach to orga-nizational culture as a cognitive system developed and supported withinlocal social networks. From this perspective, the organization resembles amagnetic field within which individual components attract and repel eachother, with friends establishing mutually reinforcing interpretive systems.Our emphasis is on the local construction of cultural meaning within anoverarching set of shared cultural understandings and the extent to whichindividuals’ cultural attitudes are controlled by their network ties.

Finally, Chapter 12, “Future Directions,” looks forward to furtherresearch in terms of new approaches and phenomena to be addressedwithin the evolving research program that we have articulated.

Motivation for Writing This Book

Because of the eclectic nature of the social network field, our researchhas appeared in leading journals in a variety of different areas including

Introduction 9

anthropology, psychology, sociology, and management. Indeed, we knowof no other research program that has encompassed such different audi-ences. We have not had a chance to bring our different contributionstogether to emphasize the programmatic nature of our research inter-ests. In this book, we bring our research themes under one overarchingumbrella so that the significance of the work can be appreciated as awhole rather than in the particular fragments that happen to show up ineach discipline’s journals.

Rather than just reprinting articles, however, we have integrated mate-rial from different sources, updated our arguments, reduced redundancy,and emphasized core themes throughout. A major motivation for us inwriting this book is the opportunity to comment on the different themesthat we have been working on together for twenty years. We have syn-thesized and edited so that the book adds value beyond what has alreadybeen published.

The book offers a theoretical and empirical alternative for organiza-tional behavior research that often gets lost in the intricacies of microlevelattitudes at the expense of perceived and actual social context. Social net-work research has often critiqued other approaches in the social sciences.But it is time we went beyond critique to offer our fellow researchersa clear alternative that addresses topics they hold dear. In this book,we provide a blueprint for how theoretically motivated research can beaccomplished on both traditional topics such as turnover and organi-zational culture as well as new topics such as the perception of socialrelations.

Target Readership

This book is targeted at the research community of scholars interestedin social network research. A primary audience consists of professors inschools of management, psychology departments, and sociology depart-ments who want an up-to-date, theory-driven treatment of networkresearch on organizational behavior topics. The book will also be ofinterest to doctoral students in the same areas. We are honored to havethis book included in the distinguished Structural Analysis in the SocialSciences series edited by Mark Granovetter.

In summary, the potential synergy between micro-organizationalbehavior research and social network approaches is huge. A focus onthe social networks – both cognitive and actual – of organizational mem-bers is likely to enhance our understanding of organizational behavior,given the importance of social structures of interaction to the under-standing of attitudes and behavior. The social network perspective has


traditionally avoided a focus on specific people, preferring to examinesystematic patterns of interaction. Our aim in this book is to bring theindividual back into the picture – to account for the cognitions and per-sonalities of individuals in connection with the structural patterns thatconstrain and enable.

I

Perceiving Networks

2

A Network Approach to Leadership

Good administrators sometimes fail to understand social structure andfail to anticipate its consequences for organizational survival. Thiscan leave organizations vulnerable to manipulation by skilled politicalentrepreneurs. In one example, the entire top management team of amanufacturing company learned from a network analysis that the bombthreats, shootings, and vandalism threatening the future of the companywere instigated by partisans of a lower-ranking manager, who had sys-tematically recruited family, friends, and neighbors into the companyover a thirty-year period. In a district desperate for jobs, these partisansfelt loyalty to the informal leader, who had provided them informationthat allowed them to be first in line for vacancies on Monday morning.The CEO, confronted with an analysis of the deep cleavages existing inthe social structure of the organization resulting from the informal pat-terns of recruiting over decades, had this to say about those who hadbeen hired: “. . . they just seemed like waves of turtles coming over thehill; hired as they made it to our door” (Burt, 1992: 1).

This story illustrates the gap at the heart of our understanding oforganizational behavior. It illustrates how important it is for managersand would-be leaders to accurately perceive the network relations thatconnect people, and to actively manage these network relations. This storyalso illustrates how informal leaders who may lack formal authority canemerge to frustrate organizational functioning through the manipulationof network structures and the exercise of social influence.

Our goal in this book is to investigate the implications of new directionsin network theory that emphasize networks as both cognitive structuresin the minds of organizational members and opportunity structures thatfacilitate and constrain action. In this chapter, we emphasize the impor-tance of individual cognition for understanding social networks. We dothis through an exploration of how the cognitions in the mind of the indi-vidual influence the network relationships negotiated by the individual

13

14 Perceiving Networks

and how this individual network contributes to leadership effectivenessboth directly and through informal networks. We understand “leader-ship” to be a general concept applicable at many different levels in theorganization, and to include both formally designated leaders as well asinformal leaders. We link together social cognitions and social structure toforge a distinctive network approach to leadership that builds upon, butextends, previous work in both the network and the leadership realms.

Organizational Network Research Core Ideas

The organizational network perspective is a broad-based research pro-gram that continually draws inspiration from a set of distinctive ideas toinvestigate new empirical phenomena. The “hard-core” ideas at the heartof network research define its special character and distinguish it fromrival research programs (cf. Lakatos, 1970). What are these ideas famil-iar to all organizational network researchers? At least four interrelatedprinciples generate network theories and hypotheses: the importance ofrelations between organizational actors, actors’ embeddedness in socialfields, the social utility of network connections, and the structural pat-terning of social life (Kilduff et al., 2006).

An emphasis on relations between actors is the most important distin-guishing feature of the network research program. As a recent historicaltreatment of social network research (Freeman, 2004: 16) pointed out,a core belief underlying modern social network analysis is the impor-tance of understanding the interactions between actors (rather than afocus exclusively on the attributes of actors). An early treatment of net-work research on organizations stated that “the social network approachviews organizations in society as a system of objects (e.g., people, groups,organizations) joined by a variety of relationships” (Tichy, Tushman,and Fombrum, 1979: 507), whereas the importance of understandingrelationships as constitutive of human nature was stated as follows in arecent book: “Human beings are by their very nature gregarious crea-tures, for whom relationships are defining elements of their identitiesand creativeness. The study of such relationships is therefore the studyof human nature itself” (Kilduff and Tsai, 2003: 131). Our networkapproach locates leadership in the relationships connecting individuals.

The second principle that gives organizational network research its dis-tinctiveness as a research program is the emphasis on embeddedness. Fororganizational network researchers, human behavior is seen as embed-ded in networks of interpersonal relationships (Granovetter, 1985; Uzzi,1996). People in organizations and as representatives of organizationstend to enter exchange relationships not with complete strangers but

A Network Approach to Leadership 15

with family, friends, or acquaintances. Embeddedness at the system levelcan refer to a preference for interacting with those within the communityrather than those outside the community. We emphasize that people’s per-ceptions of others as leaders are reflected through the sets of embeddedties within which people are located.

The third driving principle of social network research is the beliefthat network connections constitute social capital that provides value –including economic returns (Burt, 2000). As a previous review of networkresearch on leadership pointed out, “Social capital is at the heart of socialnetwork analysis” (Brass and Krackhardt, 1999: 180). Depending uponthe arrangement of social connections surrounding an actor, more or lessvalue can be extracted (Burt, 1992; Gnyawali and Madhavan, 2001). Atthe system level, a generalized civic spirit emerges from and contributesto the many interactions of trust and interdependence between individualactors within the system (Coleman, 1990; Portes, 2000). Leadership, fromthe network perspective we develop, involves building and using socialcapital.

The fourth leading idea distinctive to the social network research pro-gram – the emphasis on structural patterning – often leads social net-work research to be referred to as the “structural approach.” Networkresearchers look for the patterns of “connectivity and cleavage” in socialsystems (Wellman, 1988: 26). Not content with merely describing thesurface pattern of ties, researchers look for the underlying structural fac-tors through which actors generate and re-create network ties. At thelocal level surrounding a particular actor, the structure of ties can bedescribed, for example, as relatively closed (actors tend to be connectedto each other) or open (actors tend to be disconnected from each other)(Burt, 1992). At the system level, organizational networks can be assessedfor the degree of clustering they exhibit and the extent to which any twoactors can reach each other through a short number of network con-nections (e.g., Kogut and Walker, 2001). To understand who is a leaderfrom a network perspective is to investigate the social-structural positionsoccupied by particular individuals in the social system.

These four leading ideas – the importance of relationships, the prin-ciple of embeddedness, the social utility of network connections, andthe emphasis on structural patterning – provide the common culture fororganizational network research that allows the diversity of viewpointsfrom which fresh theoretical initiatives emerge (cf. Burns and Stalker,1961: 119). Network research is also characterized by vigorous develop-ment of methods and analytical programs to facilitate the examinationof phenomena highlighted by theory (see Wasserman and Faust, 1994,for a review of methods; and the UCINET suite of programs – Borgatti,Everett, and Freeman, 2002 – for statistical software).


The organizational network research program is progressive in thesense that new theory is constantly being developed from the metaphysicalcore of ideas that makes up the heart of the research program, highlight-ing new areas of application. It is the purpose of this chapter to highlightthe area of leadership from a network perspective. The four leading ideasthat comprise the intellectual source of theory development for organiza-tional network research are best understood as mutually reinforcing corebeliefs that, like the planks of a ship, keep the research program afloat –in terms of new theory development and exploration of new phenomena.At the level of network theory and research, all four ideas tend to be inex-tricably involved. We will invoke these ideas as appropriate throughoutthe chapter.

In contrast to network research, traditional leadership research hasfocused on human capital attributes of leaders and situational attributesof leadership contexts. Human capital attributes of leaders include traits(e.g., House, 1977; Kenny and Zaccaro, 1983) and behavioral styles (e.g.,Lewin, Lippitt, and White, 1939; Podsakoff, Todor, and Skov, 1982),whereas situational attributes of leadership contexts include task struc-ture (Fiedler, 1971), the availability of leadership substitutes (Kerr andJermier, 1978), the nature of the decision process (Vroom and Yetton,1973), and the quality of leader–member exchange (Dansereau, Graen,and Haga, 1975; Graen, Novak, and Sommerkamp, 1982). A social net-work perspective does not eclipse the valuable results of conventionalleadership research; rather, a network perspective can complement exist-ing work without repeating it. In particular, in this review we amplify thevoices that have called for a new understanding of leadership effective-ness to include leaders’ cognitions about networks and the actual structureof leaders’ ties (e.g., Hooijberg, Hunt, and Dodge, 1997; see also Bass,1990: 19).

As with all theoretical perspectives, the network approach has bound-ary conditions that limit its range of application. Social network processesare less likely to have the effects we discuss to the extent that organiza-tions are characterized by perfect competition between equally informedactors all of whom have the same opportunities (see the discussion inBurt, 1992). (Even under conditions of perfect information, however,some actors are likely to be more influenced by social networks thanothers – see Kilduff, 1992.) A further limiting condition is the extent ofwork interdependence: Under conditions of low interdependence betweenactors and little or no social interaction, network processes and theireffects will tend to be minimized.

In network terms, leadership embodies the four principles that we artic-ulated earlier. Leadership can be understood as social capital that collectsaround certain individuals – whether formally designated as leaders or


not – based on the acuity of their social perceptions and the structure oftheir social ties (cf. Pastor, Meindl, and Mayo, 2002). Patterns of informalleadership can complement or complicate the patterns of formal leader-ship in organizations. Individuals can invest in social relations withothers, can structure their social networks by adding and subtractingrelationships, and can reap rewards both in terms of their own personalperformance and organizational unit performance (Sparrowe, Liden,Wayne, and Kraimer, 2001). But embeddedness in social networksalways involves the paradox that social relations, particularly thoseoutside the immediate circle of the individual, may be difficult both toperceive accurately and to manage (cf. Uzzi, 1997). Thus, althoughthe social structure of the organization determines opportunities andconstraints for emergent leaders, the social structure is not within thecontrol of any particular individual.

Leadership and the Structure of Ties

We start our network approach to leadership theory with a discussion ofactor cognitions concerning networks, move out to the inner circle aroundthe actor, and then further zoom out to include progressively more of thesocial structure of the organization and the interorganizational realm.The theoretical framework is illustrated in Figure 2.1, and represents atentative model of leadership effectiveness from a network perspective.We provide an overview of the causal connections of the model beforezooming in to discuss in more detail the dynamics within each part of themodel.

As Figure 2.1 shows, the first step in the conceptual model indicates thatleaders’ cognitions about social networks affect the “ego networks” thatsurround each leader. Cognitive network theory (see Kilduff and Tsai,2003: 70–9, for a review) suggests that people in general shape theirimmediate social ties to others to be congruent with their schematic expec-tations concerning how relationships such as friendship and influenceshould be structured. The schematic expectations of leaders affect theirability to notice and change the structure of social ties (e.g., Janicik andLarrick, 2005). Thus, cognitions in the mind of the leader are the startingpoint for our theorizing concerning the formation of ties connecting theleader to others.

The network cognitions of leaders concerning such crucial organiza-tional phenomena as the flow of social capital within and across orga-nizational boundaries and the presence and meaning of social dividesare hypothesized to affect the extent to which leaders occupy strate-gically important positions in the organizational network. An accurate

Net

wor

k A

cuit

y

•Acc

urac

y

•Sch

emas

Ego

Net

wor

k

•Den

sity

•Ran

ge

•Coh

esio

n

Inte

rorg

aniz

atio

nal N

etw

ork

•Bou

ndar

y sp

anni

ng

•Alli

ance

s

Org

aniz

atio

nal N

etw

ork

•Cen

tral

ity

Lea

der

Eff

ecti

vene

ss

Org

aniz

atio

nal L

evel

•Sur

viva

l

•Gro

wth

•Inn

ovat

ion

----

----

----

----

----

----

----

----

----

--

Intr

a-or

gani

zatio

nal L

evel

•C

oalit

ion-

build

ing

•M

ento

ring

dis

trib

uted

le

ader

ship

•Bro

keri

ng

• •

Figu

re2.

1.T

heor

etic

alfr

amew

ork

linki

nga

lead

er’s

netw

ork

accu

racy

tole

ader

-rel

evan

tou

tcom

es.

18


perception of the informal influence network can itself be a base of powerin the organization (see Chapter 5) and can facilitate the leader’s ability toforge successful coalitions (Janicik and Larrick, 2005). We extend theseinsights to hypothesize that the acuity of leader cognitions will affect theextent to which a leader plays a strategically important role in the rele-vant interorganizational network. We know of no research bearing on thisthesis, although recent work concerning interorganizational relationshipsincreasingly concerns itself with hypothesized perceptual processes suchas organizational reputation and status (e.g., Podolny, 1998; Zuckerman,1999).

The extent to which a leader plays a role in these three actual net-works – the ego network, the organizational network, and the interorga-nizational network – is hypothesized to affect leader effectiveness. Thiscritical hypothesis derives from our basic understanding of how the fourguiding principles of the network approach extend leadership theory.Modern concepts of leadership identify the relational content of theinteraction between people as the key aspect involved in the structur-ing of situations and the altering of perceptions and expectations (e.g.,Bass, 1990: 19). Modern network theory suggests that individuals whoare central in the immediate networks around them and in the largernetworks that connect them to others throughout the organization andbeyond the organization are likely to acquire a particular type of expertpower: knowledge of and access to those few powerful others whosewords and deeds control resource flows and business opportunities (e.g.,Burt, 2005). Leaders may not be able to move into the center of everyimportant network, of course. Embeddedness in one social network maycome at the price of marginality in another network. There are trade-offsinvolved in building social capital, particularly when brokerage acrosssocial divides may engender distrust rather than gains.

One blow-by-blow account of an organizational power struggle con-trasted the networking strategies of two combatants for sole control ofthe CEO position they currently shared. Whereas co-CEO Louis Glucks-man was central within the Lehman Brothers organization as a wholeand occupied a particularly strategic position among the traders, his rivaland co-CEO Pete Petersen neglected internal networking in pursuit ofconnections with the leaders of other organizations (Auletta, 1986). Bothmen were effective leaders – Glucksman contributing to internal effective-ness and Petersen building and maintaining the external relationships thatbrought contracts to the partnership. But both had built quite differentsocial network bases of power.

The role of external affective ties with the representatives of otherorganizations in providing vital help to companies in financial troublehas been emphasized by research on the survival prospects of small firms


in the New York garment industry (Uzzi, 1996). More generally, theorganizational theory and strategy literatures have examined the extent towhich ties between organizations constitute a knowledge base importantfor outcomes such as firm growth (e.g., Powell, Koput, and Smith-Doerr,1996), new ties (e.g., Gulati and Gargiulo, 1999; Larson, 1992), andinnovation (Hargadon and Sutton, 1997). Thus, the extent to whichleaders are effective in terms of accessing important resources is likely todepend on the social-structural positions they occupy in the key networkswithin and between organizations.

What are the outcomes associated with leader effectiveness from asocial network perspective? We have thus far mentioned such aspectsof leader effectiveness as organizational growth, survival, and innova-tion. These are the responsibility of formal leaders and are outcomesat the organizational level of analysis. As Figure 2.1 summarizes, leadereffectiveness from the network perspective that we articulate would alsoinclude such components of internal organizational functioning as coali-tion building, mentoring, and brokering. These are intrinsically network-ing outcomes of both formal and informal leadership that can enhancecoordination across functions within the organization. We return to theseinternal measures of leader effectiveness later in the chapter.

The model outlined in Figure 2.1 necessarily simplifies the relation-ships between cognition, social networks, and leadership effectiveness.We neglect, for example, the ways in which occupancy of social-structuralpositions in networks affects individuals’ cognitions and expectationsabout networks (see Ibarra et al., 2005, for a review). The organizationand the environment within which it operates can be jointly considered aset of cyclical processes captured in networks of cognitions (cf. Bougonet al., 1977). We focus in this chapter on leadership, and therefore empha-size the proactive enactment of outcomes leading to leader effectiveness.

Network Cognition and Leadership

A key discovery of modern social network research is that cognitionsmatter (see Chapter 3), and thus we start the in-depth discussion ofthe theoretical framework with an emphasis on network cognition, atopic relatively neglected within conventional leadership research (butsee early leader-member exchange [LMX] work on whether peers withinunits accurately perceive the quality of dyadic leader–subordinate rela-tions – Graen and Cashman, 1975). Depending upon how the boundaryis drawn around a particular individual in an organization, that individ-ual may or may not appear to be influential in the eyes of others. Thatimplicit leadership theories may be triggered by the structural position


of certain individuals in the eyes of others is a possibility hinted at inrecent leadership theory (Lord and Emrich, 2001) but which has yet tobe systematically examined. From the perspective of perceivers located insmall groups, certain actors may appear influential, but perceivers survey-ing the larger context of the whole organization may dismiss these sameactors as relatively inconsequential (see the discussion in Brass, 1992).Conversely, people who seem relatively powerless within one local groupmay be revealed to have close connections with powerful others outsidethe group. Thus, we organize our discussion by progressively zooming outfrom individuals’ network cognitions to include expanding social circleswithin and beyond the organization.

From a network perspective that emphasizes the importance of rela-tionships, embeddedness, social capital, and social structure, the ability offormal or would-be informal leaders to implement any leadership strat-egy depends on the accurate perception of how these principles operatein the social context of the organization. To be an effective leader of asocial unit is to be aware of (a) the relations between actors in that unit,(b) the extent to which such relationships involve embedded ties includ-ing kinship and friendship, (c) the extent to which social entrepreneursare extracting value from their personal networks to facilitate or frus-trate organizational goals, and (d) the extent to which the social structureof the unit includes cleavages between different factions. The accurateperception of this complex social reality is fraught with difficulty, and,therefore, network cognition is an arena for innovative research.

If a leader wants to use social network ties to lead others, the leadermust be able to perceive the existence, nature, and structure of these ties –not just the ties surrounding the leader, but the ties connecting others inthe organization both near and far. Actors who are perceived to havepower in terms of the structure of their social ties to others may wieldinfluence even though they seldom or never exercise their potential power(Wrong, 1968; see the discussion in Brass, 1992: 299). To a considerableextent, organizations and environments exist as cognitions in the mindsof leaders and followers within organizations (Bougon et al., 1977; Kil-duff, 1990) and in the interorganizational arena of reputation and status(Podolny, 1998; Zuckerman, 1999).

Thus the question arises, how do people perceive network ties withinand between organizations? How does anyone tell whether, for example,two individuals are personal friends? Even a small organization of fiftypeople represents a considerable cognitive challenge in terms of tryingto perceive accurately the presence or absence of 2,450 friendship linksbetween all pairs of individuals, links that may well be relatively invisibleexcept to the individuals concerned. To create and manage the networksthat promote leadership effectiveness, it may be necessary to possess an


accurate representation of network links involving not just friendship andkinship, but also advice, communication, and other important networkties.

What happens when formal leaders pay no attention to the four prin-ciples we have enunciated as representing the network approach to lead-ership? Is there any penalty consequent upon leader ignorance of socialrelations inside organizations, leader blindness to the embeddedness ofworking relationships in extra-organizational arrangements such as kin-ship, leader neglect of the extent to which social entrepreneurs manip-ulate embeddedness for their own ends, and leader unconsciousness ofthe social cleavages within the organization? The answer, provided in thecase study alluded to in the opening paragraph of this chapter, is shockingin its illustration of diseased social capital. When the management fired,in a routine cost-cutting exercise, the informal leader to whom so manypeople were beholden not just for jobs but for the references necessary toactually get jobs inside the industrial plant, deep trouble ensued betweenemployees loyal to the informal leader and those helping the managementkeep the industrial plant solvent. Shootings, bomb threats, and leakingsof confidential management documents were the order of the day. Theformal leadership team had no comprehension of what was happening,not having noticed that the workforce included so many people withstrong social ties to a particular individual. (For the full case study, seeBurt and Ronchi, 1990.)

The CEO in this case was a good administrator and a skilled engineerwho failed to understand the necessity of keeping track of the social struc-ture of competition within and outside the organization. Social networksinterpenetrate the boundary between employees and nonemployees, andthe management of this boundary has important consequences for organi-zational functioning. Job applicants with social contacts (such as friends)inside the organization can exploit social capital advantages to extractcritical information at both the interview and job offer stages. Thesereferred individuals (compared to those who are not referred by currentorganizational members) tend to present more appropriate resumes andto apply when market conditions are more favorable (Fernandez andWeinberg, 1997). Referred individuals have a significantly greater like-lihood of being offered a job as a result of these advantages. Further,referrals (relative to nonreferrals) can use inside knowledge to boost theirstarting salaries in the negotiation process.

Thus, what might appear to a corporate leader as a systematic pro-cess of institutionalized racism involving higher starting salary increasesto ethnic majorities relative to ethnic minorities can be revealed throughsocial network analysis as a function of who has friends inside the organi-zation (Seidel, Polzer, and Stewart, 2000). The fairness of a hiring process


may be fundamentally compromised because it is invisibly embedded inkinship and friendship networks.

The perception of this otherwise invisible process of homophilous hir-ing is crucial to any effort by the leadership team to increase workforcediversity. The explicit management of external ties to recruit new mem-bers who are known to existing members of the organization can enhancethe organization’s economic returns (Fernandez, Castilla, and Moore,2000). If leaders comprehend the social network relationships not justamong organizational employees but also between employees and thoseoutside the organization, then leaders can build the social capital of theorganization by putting individuals’ personal social networks to work forthe organization’s benefit.

Typically, managers are busy people whose work is fragmented andinterrupted (Mintzberg, 1973). Much of our research in organizationtheory focuses on the formal arrangement of titles, offices, and reportingrelationships, whether with respect to the integration and differentiationof the organization (e.g., Lawrence and Lorsch, 1967), the inertia of theorganization (e.g., Hannan and Freeman, 1984), or the ceremonial facadecreated to be isomorphic with institutional demands (Meyer and Rowan,1977). Leadership research, to the extent that it has considered socialnetwork relations, has also focused overwhelmingly (from an LMX per-spective) on managers and the extent to which subordinates, for example,established networks that mirror those of their formally appointed man-agerial leaders (Sparrowe and Liden, 2005).

The cognitive revolution in leadership research has focused not onthe cognitions of leaders, but on leadership factors in the minds of fol-lowers (Eden and Leviatan, 1975; Lord and Emrich, 2001). There is anopportunity to extend both LMX research and cognitive approaches toleadership from the perspective of cognitive network theory (see Kilduffand Tsai, 2003: 70–9, for a review) with a focus on how leaders andfollowers comprehend (a) the structure of social relations (cf. Chapter 5),(b) the embeddedness of economic action in affect-laden networks (cf.Uzzi, 1996), and (c) the opportunities for social entrepreneurship acrossstructural divides (Burt, 2005). A greater understanding of how lead-ers and followers comprehend the social structure from which actionin organizations proceeds can enhance research on the management ofrelationships.

Accuracy

From a cognitive network theory perspective, leadership involves not justsocial intelligence (i.e., the accurate perception of social relationships inorganizations) but also the management of others’ perceptions. First, let


us consider accuracy. People perceive the same network differently, withsome individuals achieving a high degree of accurate perception, whereasother individuals lead their organizational lives in relative ignorance ofthe actual network of relationships within which work is accomplished(Chapter 3).

In general, perceptions of networks involving sentiment relations suchas friendship suffer from a series of predictable biases. People prefer tosee their own relationships as reciprocated – they prefer not to perceivetheir friendship overtures as unrequited. Similarly, people prefer to believethat their friendship circles are transitively complete – they like to believethat their own friends are friends with each other (Heider, 1958). Thiscognitive balance schema operates also as a default mechanism for fillingin the blanks concerning ties between relative strangers at the individual’sperceived organizational network’s periphery. In the absence of contraryinformation, people tend to assume that friendship ties of others arereciprocated, and that two friends of a distant stranger are themselvesfriends (Freeman, 1992; Chapter 4).

These cognitive distortions can affect leadership emergence. Peoplein organizations see themselves as more popular than they actually are(Krackhardt, 1987a), a tendency that can, perhaps, lead some individ-uals to neglect the vital process of maintaining their social capital (onthe assumption that they are already popular), whereas other individ-uals, through a self-fulfilling prophecy process, may transform the illu-sion of popularity into actual friendship links that initially did not exist.Assuming that others like them, some people may reciprocate nonex-istent liking and thereby create friends. Slight initial differences withrespect to how people perceive their connections to others can poten-tially lead to cumulative advantages through this self-fulfilling prophecyprocess.

Further, there may be a tendency to perceive popular actors as beingeven more popular than they really are (Kilduff et al., forthcoming).Human beings, in their perceptions of social networks, are “cognitivemisers” (Chapter 4) who may tend to simplify networks by perceivingthem as dominated by a few central actors even if the actual network hasno dominant cluster. A misattribution of popularity to a few actors canresult in these actors actually increasing their popularity. An emergingleader who is perceived to be popular may benefit from a bandwagoneffect: People may want to associate with someone perceived to be arising star. On the other hand, the perception that a social network isdominated by an elite group of leaders may discourage those who per-ceive themselves on the periphery from attempting to pursue leadershipoptions.


Schemas

New research (Kilduff et al., forthcoming) suggests that individuals maytend to perceive friendship networks in organizations as small worlds.Small world network structures are unusual in that they exhibit bothhigh local clustering and short average path lengths – two characteristicsthat are usually divergent (Watts and Strogatz, 1998). Clustering refersto the extent that actors are connected within local groups, whereas pathlengths refer to the number of network connections between one actorand another in the network. A small world network resembles the hub-and-spoke structure of the U.S. commercial air traffic system: local hubswith lots of connections and short average path lengths because journeysfrom one city to another are routed through the hubs. (Compare this withthe distinctly non-small world of the U.S. interstate highway system.)

The small world effect, investigated originally in the 1960s by Milgram(1967), has become a burgeoning area of organizational social networkresearch (e.g., Kogut and Walker, 2001; Uzzi and Spiro, 2005). As socialnetworks become larger and more global, the discovery that some of thelargest social networks such as the World Wide Web exhibit small worldproperties has excited considerable research interest (see Dorogovtsev andMendes, 2003, for a review). Leadership within extremely large networksis a neglected topic but one that seems tractable from a small worldperspective, given that small world networks are organized for efficientcommunication and coordination.

We focus here on the possibility that some individuals more than othersmisperceive the extent to which organizational networks resemble smallworlds (Kilduff et al., forthcoming). Such a bias has distinct implicationsfor leadership research. Simplifying perceptions to perceive a friendshipnetwork as a small world offers a considerable advantage to the aspir-ing informal leader in terms of reducing the cognitive load required tokeep track of so many different relationships. The rules for creating acognitive map of the friendship network are relatively simple from thisperspective: Put similar people (with similarity defined on some relevantdimension such as demography or interests) into clusters and connectthe clusters. Further research is needed to examine the extent to whichthe match between the “small worldedness” of the individual’s cognitivenetwork and the small worldedness of the actual network predicts leadereffectiveness.

Cognitive network schemas play a significant role in one importantaspect of leadership, namely coalition building (cf. Stevenson, Pearce, andPorter, 1985). Leaders are constantly involved in appointing people totask forces and committees. Ensuring that these teams consist of the right


balance of people can make the difference between gridlock and effectiveaction. In a pioneering set of studies, researchers found that individualswith experience of networks characterized by disconnections – struc-tural holes – were better at perceiving the potential to bridge structuralholes by identifying suitable collaborators – a key to successful coalition(Janicik and Larrick, 2005). By making sure that different constituenciesare represented at the top of the organization, the leader may facilitatethe engagement of widely different groups in the organizational mission.But in order to make these representative appointments, the leader mustfirst be able to accurately perceive any existing social system cleavages.

This recent research on the structural hole schema is interesting in sug-gesting that people are able to move beyond reliance on default modesof thinking (such as balance) when trying to make sense of the socialnetwork in organizations. People learn from experience to expect certainpatterns in the social world, and tend to see new situations in the lightof their anticipations. Thus, the leaders of an organization, familiar withthe patterns of activity taking place from day to day, may impose onthese patterns of interaction their own preconceptions of who shows upfor meetings. Leaders anticipate that regular attendees will show up andremember these people as having showed up even if they did not, whileforgetting that more peripheral members of the organization were actu-ally present on a specific occasion (cf. Freeman, Romney, and Freeman,1987). Further, people in general tend to perceive themselves to be morecentral in friendship networks in organizations than they actually are(Krackhardt, 1987a). Thus, network cognition can depart from actualpatterns of network activity, with consequences for the leader’s ability touncover political conflicts, spot communication problems between cul-turally divided groups, avoid reliance on problematic individuals for thetransmission of important resources, achieve strategic objectives throughthe appointment of key people to influential positions, and manage rela-tions within and across departments (Krackhardt and Hanson, 1993).

Leaders who perceive important social networks accurately in theirorganizations are likely themselves to be perceived as powerful (Chap-ter 5). This perceived power can itself represent an important supplementto formal authority. But for those who want to span across structuralholes and gain the reputed benefits of this activity, it may be cruciallyimportant to be perceived by others as not pursuing personal agendas(Fernandez and Gould, 1994). Social perceptions take place within rep-utational markets (Chapter 3) and, in the subtle battle to achieve promi-nence, individuals may strive to appear to others to be associated withleaders of high status. The perceived status of exchange partners can actlike a distorting prism to filter attributions concerning the focal individual(cf. Podolny, 2001).


Individuals move in and out of organizational contexts, and as they doso, their structural positions change. In one context, someone assumes aleadership position, but the same individual may be a follower in anothercontext in the same organization. Partly this is due to shifting percep-tions. Individuals self-perceive themselves as powerful in some contextsand as less powerful in other contexts, and their self-attributions may beconcordant with or discrepant with others’ attributions. Actors in orga-nizations may exert power without having to request compliance withtheir demands, simply on the basis of possibly false perceptions:

Just as players can successfully “bluff” in poker, employees canalso act as if they control scarce resources, as if they were poten-tially powerful. . . . Persons who are in a position to control infor-mation can withhold, disclose, and modify it in order to influenceothers’ attributions of power (Brass, 1992: 299).

Thus, the importance of perceptions of leadership emergence and indi-vidual influence may reside in the extent to which they are never tested. Inone recorded instance of a battle between dual CEOs for the exclusive con-trol of the Lehman Brothers investment banking house, Louis Glucksmanconvinced his rival Pete Petersen that Petersen had lost friendships withboard members, whereas Glucksman had retained their regard. But nei-ther rival checked to see whether their perceptions of their social relationswith the all-important board members were accurate (Auletta, 1986).

To summarize our general ideas concerning the importance of acu-ity in leaders’ perceptions of social networks, we indicate in Figure 2.1that accuracy is likely to improve the extent to which a leader occupiesa strategic position in three social network structures relevant to orga-nizational behavior: the ego network, comprising the individuals imme-diately connected to the leader; the complete organizational network,comprising not just direct connections but also the leaders’ indirect con-nections to everyone in the organization; and the interorganizational net-work of relationships important to the leader’s work outside the focalorganization.

In Figure 2.1, we also include the role of cognitive schemas in deter-mining the match between leaders’ perceptions of networks and actualnetworks. We need more research concerning the extent to which cog-nitive schemas help or hurt leaders develop accurate maps of the socialnetworks within which they operate. Whereas research on cognitive short-cuts implies that perceivers who rely on such shortcuts tend to make errors(Kahneman and Tversky, 1973), others see positive benefits deriving fromthe use of such schemas (e.g., Taylor and Brown, 1988), including greatersatisfaction in close relationships (Murray, Holmes, and Griffin, 1996;


see Kenny, Bond, Mohr, and Horn, 1996, for evidence concerning theeffects of relational schemas on accuracy).

We have spent considerable time on the social cognition of networksof relationships given the growing recognition within leadership researchof how leader cognitions affect leader behaviors with implications forboth leader effectiveness and organizational effectiveness (e.g., Hooijberget al., 1997). Leadership research has long recognized the importanceof implicit leadership schemas in the minds of followers (see Lord andEmrich, 2001, for review). Building on this emphasis on cognition andcognitive schemas, we seek to extend leadership research from a distinc-tively network emphasis on social relations, embeddedness, social capital,and social structure.

The Ego Network

Moving on from the network cognitions in the head of the individual,we now consider the social circle of relations actually surrounding theindividual. A strong argument could be made that it is this ego networkthat fundamentally affects all the other network relationships a leaderforms and influences – hence the centrality of the ego network in Figure2.1. It is this personal network that forms the basis of, for example, theinfluential structural hole perspective (Burt, 1992, 2005). A major taskof future research is to assess whether the structure of direct connec-tions leaders have with colleagues is as important as the structural holeapproach implies, or whether more indirect connections involving inter-mediaries can dampen or enhance leadership effectiveness, as implied inembeddedness research (Burt, 2007; Uzzi, 1996).

Density

A key theoretical concept concerning how direct connections within theego network relate to leadership is density, as indicated in Figure 2.1.Individuals whose social contacts are themselves connected to each otherhave dense social circles, whereas individuals whose social contacts havefew connections among themselves have sparse social circles (Wassermanand Faust, 1994). Members of a dense network tend to share similarattitudes and values toward the leader of the organization (Krackhardt,1999).

From a network perspective, whether the members of a dense net-work tend to enhance or neutralize the leader’s effectiveness is likely todepend upon whether the shared attitudes toward the leader are positiveor negative. A dense network of people favorably disposed toward the


leader represents a pool of social capital available to the leader. Mes-sages communicated to this group are likely to be favorably received andexpeditiously transmitted. A dense network of people negatively inclinedtoward the leader represents a potentially distorting prism, likely to takeany message or initiative from the leader and cast it in the most unflatter-ing light. More research is needed on the ways in which dense networksdistort or enhance leadership initiatives.

Range

Structural-hole theory (Burt, 1992), following on from the weak-tiehypothesis (Granovetter, 1973), suggests that individuals whose personalcontacts include a diverse range of disconnected others gain benefits.These benefits (including faster promotions – Burt, 1992) derive fromthe information and control possibilities of being the “third in the mid-dle” between other individuals who must pass resources and informa-tion through the focal individual. Thus, the focal individual has accessto diverse communications within his or her immediate contacts. If theindividual (conventionally referred to as “ego” in network research) isembedded in a clique, then the diversity of information and resourcesreaching ego from immediate contacts may be low. Further, the oppor-tunity for ego to play an informal leadership role, distributing ideas andother valued resources throughout the immediate social circle, vanishesif ego is simply one more person in a highly connected group.

As simple as the implied principle appears to be – connect oneself todiverse others who themselves are not connected to each other in order toenhance leadership potential in the informal network of relationships – itis much harder to realize than might at first be apparent. The principle ofembeddedness operates strongly in this context. Simply stated, individualsprefer to associate with homophilous others – those who are similar tothemselves (McPherson, Smith-Lovin, and Cook, 2001). This tendency islikely to be just as strong among putative leaders as it is among people ingeneral – even economic transactions at the firm level tend to be embeddedin kinship and friendship networks (Uzzi, 1996).

Homophilous networks represent information restriction (Popielarz,1999). Individuals embedded in such networks, established not just interms of kinship but also on the basis of proximity (Festinger, Schachter,and Back, 1950), ethnicity, or gender (Mehra et al., 1998), are likelyto experience strong cohesion (many ties among the similar others) butalso information restriction. Groups as powerful as the dominant coali-tion (Cyert and March, 1963), the top management team (Hambrick andMason, 1984), and the board of directors (Palmer, 1983) may exhibitin-group homogeneity under the pressures of ease of communication,


shared backgrounds, and demographic similarity (see the review in West-phal and Milton, 2000). Social capital advantages are likely to be signif-icantly diminished as leaders embed themselves in homogenous groups,leading to negative effects on market share and profits (Hambrick, Cho,and Chen, 1996). Business survival prospects tend to be better for thosebusinesses whose owners establish a large range of personal contacts withimportant representatives of the task environment relative to those own-ers who establish a smaller range of such contacts (Oh, Kilduff, and Brass,2005).

Cohesion

The cohesiveness of a dominant coalition may be sharply increased ifthe coalition perceives it is challenged by a set of actors (pursuing ahostile takeover, for example) or by negative outcomes of previous deci-sions (Kilduff, Angelmar, and Mehra, 2000). This increased homophily,while facilitating coordinated action by the top management team, mayadversely restrict decision-making options. The extent to which leadersturn to their personal contacts for advice following poor firm perfor-mance predicts subsequent tendencies to minimize changes in corporatestrategy (McDonald and Westphal, 2003).

There are strong pressures in organizations for people to agree withtheir personal friends concerning important values and ideas. For aninformal leader, embedded in a coalition of like-minded individuals, tochallenge the hegemony of the official culture is always possible. But itis much more difficult for an informal leader to resist the social pressurefrom within his or her social circle to agree with close friends concerninghow to interpret widely shared core values (Chapter 11).

It is interesting to note that, from a network perspective, the socialpressure on ego differs little irrespective of the size of the clique withinwhich ego is embedded, given that the clique contains people who all haveties to each other within the clique but no common ties to those outside.Whether ego is embedded in a three-person clique or a larger clique, egostill experiences group pressure to conform (Simmel, 1950). This pressurebecomes powerful as soon as a dyadic interaction (between two people)expands to include three people. To the extent that a leader belongs totwo or more of these cliques (of three or more people), the leader isvulnerable to cross-pressures from the different cliques to which he orshe belongs. Different cliques tend to reinforce different interpretationsof reality, and these discrepant interpretations may place the leader, wholinks the two different cliques together and who may play a brokerage rolebetween these different groups, in a complicated situation. Each clique


may present the leader with demands that, considered jointly, may bedifficult to meet.

One case study described how an informal leader who strongly favoredthe ongoing unionization drive in an entrepreneurial company foundhimself unable to use his influential position in his personal social circleto influence others. This individual was a member of eight different three-person friendship cliques and was thus “frozen by the set of constraintsimposed by the numerous cliques” (Krackhardt, 1999: 206). Three ofthis person’s cliques contained vociferous opponents of unionization. Sounpleasant was his position in his social circle that he resigned from thefirm ten days before the unionization vote was taken and rejoined thefirm two days after the vote had failed. This individual’s apparent powerin the social circle of personal friends was stultified by his embeddednessin cohesive, but mutually discrepant, cliques.

Informal Leadership Emergence

Within the social circle surrounding the formal leader, there are likely tobe some individuals who play informal leadership roles. These informalleaders tend to spring up in teams in which formally appointed leadersplay little or no role in the coordination of team activity (perhaps becausethe formal leaders are focused on activities external to the team). Thus,informal leadership is likely to be a feature of teams in which formal lead-ership is, relatively speaking, absent. One study of leaderless teams foundthat informal leaders disproportionately influenced team efficacy – theextent to which team members evaluated their abilities to perform spe-cific work-related tasks (Pescosolido, 2001). Such informal leaders alsoplay a role in regulating team members’ emotions (Pescosolido, 2002).Key process variables, such as team efficacy and team emotions, affectteam performance (Barsade, 2002; Gibson and Vermeulen, 2003).

Given the potential power of these informal leaders to manage thecognitions and emotions of group members, even in the absence of anyformal authority, formally appointed leaders’ relationships with theseinformal leaders become more important than perhaps approaches thathave focused on leader–member exchange relations have recognized. Wesuggest that within the leader’s in-group there are some ties that are morecrucial for leader effectiveness than others; and, outside the leader’s in-group, neglect of individuals with considerable social influence is likelyto imperil leader effectiveness.

To summarize this section is to recognize that structural hole theory(Burt, 1992) suggests that would-be leaders should structure their inter-personal networks to reach diverse constituencies, using relatively few ties


to expand the range of information and resources accessed. An effectivenetwork strategy, according to this interpretation of structural-hole the-ory, is likely to involve leaders building links to a variety of different con-stituencies and delegating to trusted “lieutenants” the task of managingrelationships with the other members of each constituency. Informationwould flow to leaders through the trusted lieutenants from all around theorganization. It is with each trusted lieutenant that the informal leaderdevelops and maintains a strong tie (as suggested in the dyadic approachto leadership – see Dansereau, 1995, for a review). It is this emphasis onextending the leader’s ties throughout the organization that we turn tonext.

The Organizational Network

There are some caveats to the “divide and conquer” strategy advocatedfrom the influential structural-hole perspective (Burt, 1992, 2005). Fromthis perspective, would-be leaders are recommended to divide social net-works in organizations into non-overlapping groups and to harvest socialcapital benefits from brokering information and other resources betweenthese groups. However, as structural-hole theory recognizes, there aresome groups (such as boards of directors) whose importance may requirea much more intensive relational strategy. To the extent that all the mem-bers of a particular group have power over ego’s leadership effectiveness,it makes sense for ego to invest in a personal relationship with everymember of the group. Second, the effectiveness of informal leadership islikely to depend not just on direct links to others but also on the patternof links beyond the immediate ties. The important idea here, then, is thatthe structural position of ego in the social network affects the leadershippotential of the individual in the organization, and this principle extendsbeyond the immediate social circle of the individual.

From an embeddedness perspective (Uzzi, 1996, 1997), an effectiveleadership network is a multistep process, only one step of which isunder the control of ego. First, ego needs to build ties to individuals whorepresent access to and from key constituencies within and outside theorganization. But, second, ego needs to monitor whether representativesof these key constituencies themselves have access to networks. And third,ego must monitor the interrelationships between these representatives(cf. Sherony and Green, 2002; Sparrowe and Liden, 2005). Leadershipsuccess can crucially depend upon these secondary networks and theinterrelationships between people beyond the leader’s ego network.

At present, we know little about how a leader within an organizationfunctions in the context of the social networks of informal leaders who


may or may not be occupying positions of official authority. Informalleaders, typically of lower rank than the primary leaders (to whom theymay or may not report directly), wield considerable influence derivedfrom their positions in the social network (Mechanic, 1962). We canglean some insight into how a leader at one level can benefit or sufferfrom the activities of socially well-connected informal leaders by consid-ering the literature on substitutes for leadership. Leaders whose subor-dinates possess expert power, for example, may find themselves to berelatively redundant. Subordinate expertise can act as a substitute forleadership in some cases and in other cases subordinates, representing theleader, can deputize for the leader (Gronn, 1999; Kerr and Jermier, 1978;Podsakoff and MacKenzie, 1997). This form of distributed leadership(Mayo, Meindl, and Pastor, 2003) is still poorly understood.

Mentoring Distributed Leadership

From the network perspective articulated in this chapter, leader effec-tiveness involves building social capital that benefits individuals in theorganization and extending the social networks of subordinates to facil-itate career advancement. One measure of leader effectiveness, there-fore, is the success of the leader in promoting the social networks andleadership potential of subordinates. By systematically sponsoring sub-ordinates’ development of social capital through introductions to keypeople in the organization and the environment, leaders can enhance theoverall leadership potential in the organization and groom their subor-dinates for organizational success. Hence the emphasis on the mentor-ing of distributed leadership as an aspect of leader effectiveness in Fig-ure 2.1. The perceived influence of proteges in the organization is likelyto be related to the extent to which the proteges build links across demo-graphic boundaries. Thus, helping a man build links to the network ofwomen or a woman build links to the network of men within an organi-zation can enhance the protege’s leadership potential measured in termsof perceived power (Brass, 1985).

Such sponsorship is likely to be especially important in the case ofmembers of underrepresented groups whose own attempts at brokerageacross social divides may rebound to hurt rather than help their careers,according to research in one firm (Burt, 1992). Members of underrep-resented groups tend to form homophilous networks among themselvesand may also experience discrimination from majority group members(Mehra et al., 1998). The mentoring of underrepresented group subor-dinates involves facilitating the development of the subordinates’ ownnetworks that may expand in directions not covered by the leader’s ownconnections (cf. Higgins and Kram, 2001). Research suggests that such


mentoring relationships can be successful even when the sponsor andthe protege are from different ethnic groups (Thomas, 1993). Networkleadership, then, can be measured in terms of how much social capital itcreates for others, especially those members of underrepresented groupswhose social network ties may be restricted because of in-group pressurestoward homophily and out-group bias (Chapter 6).

A particularly important test of network leadership occurs in the caseof isolates. G. K. Chesterton wrote, “There are no words to express theabyss between isolation and having one ally.” Members of work teamswho consistently fail to communicate with their colleagues may representwasted resources in today’s coordinated organizations whether or notthey suffer the “abyss” of isolation. Research in three high-technologymilitary organizations showed that isolates, relative to “participants,”tended to rely more on written and telephone communication, to with-hold information, to express less commitment to the organization, toexperience lower satisfaction with both communication and with theirjobs, and to be rated as lower performers (Roberts and O’Reilly, 1979).Clearly, such isolated individuals represent a networking challenge. Theextent to which such isolates are part of work groups may predict theextent of leader effectiveness in such groups. A related issue concernsthe extent to which workgroups exhibit disconnects between subgroups.Although recent work suggests that too few or too many structural holesin a team may adversely affect communication (Oh, Chung, and Labianca,2004) and team effectiveness, the question of how such structural holesaffect team performance and functioning remains unanswered (Balkundi,Kilduff, Barsness, and Michael, 2007).

Positive Emotion

Isolates and structural holes in groups tend to signal the existence ofemotional distress. Research attention has started to focus on the roleof formal leaders in the emotion management network in organizations(Chapter 8). Vertical dyad linkage theory alerted researchers to the bene-fits – emotional and vocational – associated with membership in theleader’s in-group (see Dansereau, 1995, for a review). Building on thislegacy, the positive psychology movement suggests that leaders haveresponsibility for maintaining the emotional health of all employees(Frost, 2003) rather than just those with privileged access to the leader.Yet, some people in formal leadership roles fail to attend to the toxicemotions created in organizational contexts and thereby fail to performas effective leaders (Maitlis and Ozcelik, 2004). The question of the man-agement of affective bonds and emotional health has been neglected inthe leadership and in the network literatures and begs for more attention.


The Interorganizational Network: BoundarySpanning and Alliances

Leaders, both formal and informal, can potentially network within theirorganizational units and outside their units. As representatives of theirorganizational units, leaders forge interorganizational links that mayor may not lead to or coincide with formally contracted relationships.Beneath most formal alliance ties between organizations “lies a sea ofinformal ties” (Powell et al., 1996: 120). Interpersonal friendships andother strong links such as kinship between CEOs can lead to businessalliances, just as business alliances can lead to warmth and trust betweenrepresentatives of different organizations (Larson, 1992; Uzzi, 1997).

One dramatic case study, referred to earlier in this chapter, highlightedthe danger of two individuals dividing the networking task between theminto its internal and external components (Auletta, 1986). Lehman Broth-ers was a venerable Wall Street investment banking firm in which partnerLouis Glucksman operated as the inside networker, maintaining cohesionand rapport with the company’s traders, whereas partner Pete Petersenoperated as the outside networker, responsible for bringing in new busi-ness from the rich and famous. When both partners were appointed asjoint CEOs, the ensuing battle for supremacy led to a financial crisis and atakeover by American Express, bringing to an inglorious end one chapterin the saga of a proud and independent institution. In the furious battlefor control between the inside and outside networkers, Glucksman hadthe upper hand, having developed social capital within the organizationamong the partners who controlled the firm through their votes.

As this example illustrates, managing the boundary between inside andoutside networking is a crucial task for formal leaders. The formal leadercan be considered a boundary spanner who manages not only an internalconstituency within the organization but who also represents the organi-zation in the community of organizations. Network links between orga-nizations tend to build from within the existing network. Organizationalleaders create stable relationships with trusted partners, and, over time,these stable ties accumulate into a network that provides to membersof the network information about future alliance partners (Gulati andGargiulo, 1999). Organizational leaders, for example, tend to recom-mend to one trusted partner the formation of a business relationship withanother trusted partner, thus creating a three-member clique (Larson,1992; Uzzi, 1996). With knowledge increasingly emerging from the inter-stices between hierarchical boundaries (Powell et al., 1996), leaders whopursue policies of splendid isolation are likely to see their organizationssuffer “the liability of unconnectedness” (Baum and Oliver, 1992) infailing to capture intellectual developments as they arise and expand.


An innovative organization such as Digital Equipment Company, oncefamed for its fortress-like culture and its devotion to in-house techni-cal development (Kunda, 1993), is likely to fade away in a knowledgeeconomy in which innovations are increasingly the product of industrialclusters rather than individual companies (Saxenian, 1990). Given theinertia of organizations relative to the speed of change in many environ-ments (Hannan and Freeman, 1984), even large and apparently dominantorganizations in knowledge-intensive industries need to build connectionswith a range of other organizations to access developing technology.

However, leadership effectiveness in this knowledge economy maydepend not just on the direct network links to other organizations underthe leader’s control, but also on the links beyond the leader’s control. Aswe noted with respect to networking within the organization, it is oftenthe links beyond the immediate social circle of the leader that affect manydesired outcomes. Research suggests that the survival of the organizationitself may be affected by the secondary links to organizations beyond theleader’s immediate control.

For example, in the New York garment industry, CEOs who developedstrong personal relationships with the heads of “jobbing” firms (thatdistribute work orders) increased the survival chances of their firms ifthey were able to access through these strong connections networks ofbalanced relationships. It was not just the primary ties to the jobbingfirms that were important for the focal firms. Survival was enhanced forthe firms of those CEOs strongly connected through a primary tie to a setof secondary ties that include a balanced mix of arm’s-length and closeties with a jobbing firm (Uzzi, 1997). Although the CEO may have somecontrol over whether to develop close, personal ties or more market-basedexchanges with heads of jobbing firms, the CEO may not even be awareof the types of business relationships that jobbers have with other firms.Thus, leadership effectiveness (and the survival of the organization) maydepend on second-order network links beyond the control of the CEO.

What of the leader’s centrality in the community of organizationalleaders? Research shows that organizational leaders tend to interact witheach other across a range of social events, with representatives of eliteorganizations tending to form their own elite social circles (Galaskiewicz,1985; Kilduff and Tsai, 2003: 22). However, centrality in this commu-nity of leaders may distract leaders from the strategic management oftheir own organizations. One study of an ethnic community of Koreanexpatriate entrepreneurs showed that the extent to which organizationalowners were central (in terms of spanning across divided social groupswithin the community) correlated negatively with performance and pre-dicted organizational demise (Oh et al., 2005). Of compelling interest,however, is the extent to which the leader’s ties to organizational leaders


outside the immediate community affect the flow of important resourcesand, thereby, organizational survival.

It may be in the interorganizational arena that new network methodsfocused on social network dynamics emerge, given the strong interestin understanding the evolution of strategic alliances (e.g., Gulati andGargiulo, 1999). Conventional wisdom suggests that networks tend tobe relatively stable, but this apparent stability can mask many types ofchange that can be captured in network “movies” showing the dance ofinteractions over time (Moody, McFarland, and Bender-DeMoll, 2005).

Conclusion

Leadership requires the management of social relationships. Starting withthe cognitions in the mind of the leader concerning the patterns of rela-tionships in the ego network, the organizational network, and the interor-ganizational network, social ties are formed and maintained, initiativesare launched or avoided, and through these actions and interactions, thework of the leader is accomplished. Building on the idea that networks areboth cognitive structures in the minds of individuals and actual structuresof relationships that link individuals, this chapter views organizationalnetworks as constructed and maintained by boundedly rational actors,subject to biases in their perceptions. Leadership research from a net-work perspective has the opportunity to forge a new understanding ofthe interplay between the psychology of individuals and the complexityof the networks through which actors exchange information, affect, andother resources.

Leadership research also has the opportunity to renew our understand-ing of how patterns of informal leadership complement or detract fromthe work of formally appointed leaders. If leaders rely solely on theirformally assigned authority and bring into their leadership circles like-minded others, they may isolate themselves from new ideas (as repre-sented by, for example, the slow learners investigated by March, 1991).Further, the influence of visible leaders, both informal and formal, islikely to be affected by network ties that may not show up at all in theorganizational chart. The members of governing coalitions, for example,are likely to be tied to powerful individuals temporarily removed frompositions of authority and deal makers who operate quietly to influenceorganizational outcomes. Only recently has research attention focusedon these virtual actors whose “ghost” ties constrain network change andaction (see, for example, Moody et al., 2005).

The network approach articulated in this chapter emphasizes theextent to which individuals’ thoughts and actions are embedded in their


perceptions of networks, in the immediate ego networks that surroundthem, in the organizational networks within which their ego networks areembedded, and in the interorganizational networks that connect them toleaders of other organizations. Leaders, we have emphasized, generateand use social capital through the acuity with which they perceive socialstructures and the actions that they take to build connections with impor-tant constituencies within and across social divides.

To understand leadership effectiveness from a social network perspec-tive is to study the individual’s position in the larger networks withinwhich the individual is located. The network approach, therefore, allowsa more macro focus on the full repertoire of network relationships thanhas been the case in previous leadership research. The network approachalso incorporates actors within the network who may or may not be con-nected with the leader, but whose actions, in creating new ties, for exam-ple, can affect leader outcomes by changing the structures within whichthe leader operates. Clearly, the network perspective – in its emphasis onsocial relations, embeddedness, social capital, and social structure – bothincorporates strands emphasized within previous leadership research andpoints in new directions.

This chapter has emphasized the importance of individuals’ percep-tions of network relations, and this theme is continued in the followingthree chapters. In the next chapter, we investigate the extent to whichindividuals who are perceived to have prominent friends gain advantagesin terms of performance reputation in organizations.

3

An Analysis of the Internal Market forReputation in Organizations

The basic idea we investigate in this chapter is whether the relative impor-tance, reputation, and value of any particular individual in an organiza-tion (in the eyes of others) are affected by the company the individual isperceived to keep. The assessment of reputation, we suggest, is likely tobe enhanced if the individual is perceived to have a high-status friend.Becoming the friend of a high-status person is not easy, and those whoare fortunate to enjoy such access are likely to gain considerable socialcapital. High-status people are carefully scrutinized to see who their asso-ciates are. This is not a new insight. Shakespeare’s Falstaff, an intimateacquaintance of Crown Prince Harry in the Henry IV plays, is depicted asreveling in reflected glory. Certainly, the Baron de Rothschild (accordingto the anecdote in Chapter 1) was in no doubt concerning the value hisapparent friendship would confer in terms of tangible financial capitalbecoming available from those impressed with his public social endorse-ment of the person with whom he walked “arm-in-arm.”

How Perceptions Affect Reputation

The theoretical framework within which we investigated the determinantsof reputation in organizational labor markets was balance theory (Heider,1958). From this perspective, someone perceived to be the friend of apositively valued other is also likely to be perceived positively: There is astrain toward cognitive balance in the perceptions of observers. We arguethat the performance reputations of people with prominent friends willtend to benefit from the public perception that they are linked to thosefriends.

This basking-in-reflected-glory effect has been hypothesized to involvea deliberate strategy on the part of individuals to garner positive

39


evaluations from those who perceive their ties to prominent others: “Itis our contention that people make known their noninstrumental con-nections with positive sources because they understand that observersto these connections tend to evaluate connected objects similarly” (Cial-dini et al., 1976: 374). Previous researchers have not investigated thebasking-in-reflected-glory effect in a performance context, despite thephenonemon’s apparent relevance to such issues as performance rating.

Building from this psychological base in balance theory, we propose toextend the study of the basking phenomenon to all the actors in a socialsystem. The structural approach to social networks suggests looking atperformance reputations in terms of the structure of relations in an entireorganization. Structuralists are familiar with the use of the metaphor ofthe market to describe any competitive situation in which people jockeyfor valuable outcomes, such as a reputation as a good performer (cf.White, 1970). In looking at an organization as a market for reputation,one’s focus is implicitly on the process of exchange. The higher an indi-vidual’s reputation, the more valuable he or she becomes in the internallabor market. In looking for signals of quality (Spence, 1973), people ask:Does the person hold a high position in the organization? Is the person afriend of a prominent leader? In this cognitive assessment process, bothindividual attributes and social ties may contribute to the determinationof performance reputation.

To recapitulate, we are predicting that an observer’s perception of anindividual’s performance will be significantly influenced by the degree towhich the observer perceives that individual to have a prominent friend.The basis for this prediction at the psychological level is the strain towardcognitive balance in the mind of observers. Within the internal labor mar-ket of an organization, people are assumed to be jockeying to increasetheir reputations as high performers by publicizing links to prominentothers. This assumption is supported by research on basking in reflectedglory (see Cialdini [1989] for a review) showing that people actively seekto enhance their public images by proclaiming bonds to successful others:“It is as if strains toward cognitive balance are at some level of conscious-ness understood to exist by observers and action is taken to exploit theconsequences of the balance process” (Cialdini and Richardson, 1980:414).

We assume that each individual is especially active in drawing attentionto his or her most prominent friend because this friend offers the indi-vidual the most opportunity for basking in reflected glory. Organizationmembers may also be alerting others to such relevant factors as their posi-tion in the organizational hierarchy and their organizational accomplish-ments. Further, members of the internal labor market are assumed to besearching for signals of the underlying performance quality of colleagues

An Analysis of the Internal Market for Reputation 41

and competitors (cf. Spence, 1973) in an ongoing social comparison pro-cess (cf. Festinger, 1954).

Formally stated, our first hypothesis is:

Hypothesis 1: The prominence of an individual’s most prominent friendwill influence the individual’s performance reputation in an organization,controlling for the individual’s formal status and job performance.

However, this proposition can be further refined in accordance withbalance theory to distinguish between perceived and actual friendshiplinks.

From a balance theory perspective (Heider, 1958), it is an individual’sperception of social relations, rather than the actual structure of socialrelations, that affects individual attitudes. What matters are the friendsa person is perceived to have, not actual friendships. Balance theorysuggests, therefore, that mental representations of patterns of relationswill be more important determinants of attitudes than the actual patternsof relations within which individuals are located. From this perspective,perceptual measures of network relations should be more predictive ofattitudes than objective measures.

Social network links can be derived either from each observer’s cog-nitive map of how he or she perceives the connections between organi-zational actors or from an aggregate map built up from the agreementof each party to each link. We follow Weick and Bougon (1986: 105–6) in using the term “cognitive map” to refer to an individual’s mentalrepresentation of relations within a system of connections. An individ-ual’s cognitive map of a friendship network, for example, consists of theindividual’s picture of who is friends with whom in a particular social sys-tem. Individuals are assumed to use these maps to negotiate their journeysthrough their social worlds.

An alternative to deriving a separate set of network links from eachobserver’s cognitive map is to use a map of the actual network assem-bled in a conventional structural fashion from the agreement of eachparty to each link. The network map in this case is not idiosyncraticto any one individual. The aggregate network map represents realitybecause it is compiled from the observations of all relevant observersrather than from the observations of just one observer (Krackhardt,1987a).

Hypothesis 2: Measures of perceived network relations will lead to bet-ter predictions of performance reputation than will measures of actualnetwork relations.


Methods

Silicon Systems (a pseudonym), the organization selected as the researchsite, was a small entrepreneurial firm located on the west coast of theUnited States in an area known for its many small startup companies. Sil-icon Systems specialized in the sale, installation, and maintenance of state-of-the-art information systems for clients such as local banks, schools,manufacturing firms, and research and development (R&D) labs. Notlong before this research began, giant competitors, such as the Inter-national Business Machines Corporation (IBM) and the American Tele-phone and Telegraph Company (AT&T), had begun to focus attentionon Silicon Systems’ market because of its perceived growth potential.According to the top managers of Silicon Systems, the small firm’s com-petitive edge remained its ability to respond more efficiently than its giantcompetitors to idiosyncratic customer demands.

Silicon Systems was wholly owned by its three top managers, eachof whom owned an equal share. All employees worked in the company’ssingle-floored building. The employees saw each other regularly and werefamiliar with each others’ work practices. The firm had grown from threeto thirty-six people in fifteen years, with much of the growth occurringin the five years prior to this study. During those years, the firm had beengenerally profitable, and the owners anticipated no downward trend intheir business.

Of the thirty-six people in the company (twenty-eight men and eightwomen), thirty-three people, or 92 percent, accepted $3 each from us tocomplete a lengthy questionnaire. We described the research as a study ofthe effects of networks in organizations. All respondents were promisedand given individual reports showing their cognitive maps of the networksand how those perceived networks compared to the actual networks.

Measures

NetworksNetwork Indexes: Friendship and Advice NetworksTo capture respondents’ cognitive maps of the friendship and advicerelations in this organization, we asked each person about his or her per-ceptions concerning every other person’s network links. For friendship,each person responded to the following question about every other personin the organization: “Who would this person consider to be a personalfriend? Please place a check next to all the names of those people whothat person would consider to be a friend of theirs.” For advice relations,the corresponding question was: “Who would this person go to for helpor advice at work?” Thus, for the friendship network, John Meredith wasasked a series of thirty-six questions of the form: “Who would Jane Asch


Jim

Pat

Ev

Steve

Ivo

Chris

ZoeFred UptonIrv

Figure 3.1. An employee’s cognitive map of the friendship relations atSilicon Systems.

consider to be a personal friend?” “Who would Jerry Bonavue considerto be a personal friend?” Each question was followed by the list of thirty-five employees’ names. Meredith then checked the names that indicated,for example, his perceptions concerning who Asch considered to be herpersonal friends.

Each respondent, then, gave us a complete cognitive map of his or herperceptions concerning who was friends with whom in the organization.To measure perceived friendship links, we used the following procedure:A friendship tie as perceived by person k existed between person i andperson j only if k responded on the questionnaire that i considered j afriend.

To measure actual friendship links, we determined the locally aggre-gated structure, or LAS (Krackhardt, 1987a) as follows: A friendship tieexisted between persons i and j only if person i claimed person j as afriend and person j agreed that person i claimed person j as a friend.Thus, a friendship link from i to j was defined as existing when bothparties agreed that it existed.

Figure 3.1 presents an example of a respondent’s cognitive map of thefriendship network. The striking aspect of this particular map is the rela-tively low number of connections that this individual perceived. Compar-ing this individual’s cognitive map with the actual structure of friendship


relations that emerged from the reports of all respondents (see Figure 3.2)shows that perceptions concerning friendship links can be considerablydiscrepant from actuality (Krackhardt, 1990). In this case, the discrep-ancy is the result of the individual not perceiving many friendship linksthat actually existed. In other cases, the discrepancy occurs because anindividual perceives friendship relations where none exist.

To measure a perceived advice link, we followed the following proce-dure: A person was considered to go to another person for advice if therespondent’s cognitive map showed that person as going to the other foradvice. That is, for respondent k, a perceived advice link existed betweenpersons i and j only if respondent k perceived that person i went to personj for advice.

To measure an actual advice link, we did this: A person was consideredto go to another for advice only if that person claimed that he or she wentto the other for advice. That is, person i was considered to actually go toj for advice only if i’s cognitive map showed i going to j for advice.This definition of an actual advice link is known as the row-dominatedlocally aggregated structure (Krackhardt, 1987a) and follows the stan-dard procedure in network analysis in that it relies on the self-reportof the individual concerned. This measurement preserves the asymmetryinherent in the relation, an asymmetry that is critical to our prominenceargument, as discussed in the following subsection.

Independent Variable: Friend’s Prominence MatrixWe hypothesized that each person’s performance reputation would beinfluenced by the extent to which each person had a prominent friendin the organization. We chose to focus on each person’s most prominentfriend rather than on an average of all friends’ prominence ratings becauseof the theoretical basis of the research. An average measure would notcapture an individual’s ability to bask in reflected glory. With an averagemeasure, two individuals might have the same friends’ prominence scoreeven though one person had no prominent friends whereas the other hadboth highly prominent and highly obscure friends.

We measured each friend’s prominence in four different ways andobtained four separate matrixes. Table 3.1 summarizes the differencesamong these four measurements. To contrast the predictive validity ofthe perceived and externalized structures, we measured prominence usingboth perceived and actual network links. To check for common methodvariance, we measured prominence both from questionnaire responses,as indegree centrality in the advice network – that is, the total number ofothers who went to the friend for advice (Scott, 1991: 72) – and from theorganization chart, as formal status in the organizational hierarchy. Thefour measures of friend’s prominence were, therefore, (1) the indegree

Ale

Dal

Mel

Rob

Gar

Wal N

an

Vic

Hug

Ivo

Tom

Chr

Ken

Ric

Dan

Ev

Ste

Ear

Pat

Jim

Irv

Car Fr

eH

al

Abe

Len

Zoe

Upi

Ger

Ben

Bob

Jac

Ovi

Figu

re3.

2.T

heac

tual

stru

ctur

eof

frie

ndsh

ipre

lati

ons

atSi

licon

Syst

ems.

45


Table 3.1. Summary of Research Variables

Variable Definition of Each Cell in Matrix

DependentPerformance

reputation matrixRespondent i’s perception of the job performance of person j,

rated on a seven-point scale, for all js not supervised by iIndependent

Perceived friend’sindegree centralitymatrix

Advice indegree centrality of person j’s most central friend,based on friendship and advice networks perceived by i

Actual friend’sindegree centralitymatrix

Advice indegree centrality of person j’s most central friend,based on aggregate (LAS) friendship and advice networks

Perceived friend’sformal statusmatrix

Level in organizational hierarchy occupied by person j’shighest-level friend, based on friendship network perceivedby i

Actual friend’s formalstatus matrix

Level in organizational hierarchy occupied by j’s highest-levelfriend, based on aggregate (LAS) friendship network

ControlJob performance

matrixSupervisor’s evaluation of the job performance of person j on a

seven-point scaleFormal status matrix Level in the organizational hierarchy occupied by person j

centrality of the perceived friend, (2) the indegree centrality of the actualfriend, (3) the formal status of the perceived friend, and (4) the formalstatus of the actual friend.

The indegree centrality measure of prominence was derived from theadvice network of relations. In social network research, “Prominent lead-ers are the objects of extensive relations from followers, while the latterare the objects of few relations” (Knoke and Burt, 1983: 199). To capturethis kind of prominence, therefore, an asymmetric measure was needed,one that counted nonreciprocated ties. Further, our theoretical assump-tion was that individuals publicize the existence of friendship links toprominent others and that perceivers scan an organization for clues asto who the prominent actors are. We needed, then, a measure of visi-ble prominence, one that emphasized direct ties rather than indirect ties.We wanted to capture the kind of prominence represented by someonewhose desk is often surrounded by people seeking help and advice ratherthan the kind of prominence represented by someone with relatively invis-ible influence based on indirect links. Because of our concern with asym-metry and visibility, we chose to measure informal prominence in termsof indegree centrality in the advice network, which refers to the extent towhich others seek help or advice about work-related matters from a focalperson. Technically, indegree centrality can be defined as the number of


other people who go to an actor for advice (Freeman, 1979). Indegreecentrality has been widely used in organizational research when direct,asymmetric ties are being measured (e.g., Burkhardt and Brass, 1990),although measures that include indirect ties may be more appropriate inother situations (cf. Ibarra, 1992, 1993b).

In the current research, we measured indegree centrality for both theperceived and actual networks. For the perceived measure, we lookedat each respondent’s cognitive map of perceived relations. Within eachcognitive map, we identified, for each person, the friend with the high-est indegree centrality rating. This rating was then recorded as the firstmeasurement of the independent variable.

The second measurement of the independent variable – the actualfriend’s indegree centrality – was based on the actual friendship andadvice networks aggregated from the responses of all respondents. Asdescribed previously, the existence of a friendship link in the actual net-work meant that the two people involved both agreed that the friendshiplink existed. Similarly, the existence of an advice link from, for exam-ple, John to Bill meant that John had indicated on his questionnaire thathe went to Bill for advice. For each person, therefore, we identified theactual friend with the highest indegree centrality rating and recorded thatvalue.

The third and fourth measurements of the independent variable werebased on the friend’s formal status rather than on the friend’s indegreecentrality. Because of potential common method variance, it was nec-essary to demonstrate that the critical variables in the study were notcorrelated simply because they were derived from the same source. Theobvious remedy for common method variance is to use different sourcesfor the independent and dependent variables, if doing so is consistent withthe conceptual framework of a study (Sackett and Larson, 1990: 474).

Following this strategy, we derived prominence ratings from the orga-nizational chart recorded in company records. In many organizations,those higher up in the hierarchy are also more prominent because manyothers go to them for help and advice about work-related matters. Formalstatus has been shown to be predictive of organizational power (Brass andBurkhardt, 1993; Krackhardt, 1990) and to correlate highly with networkcentrality (Ibarra, 1992; Krackhardt, 1990). Formal status, then, providesan alternative to perceived measures of prominence in organizations.

We were therefore able to test our hypotheses with the independentvariable ratings derived from company records and our dependent vari-able ratings derived from questionnaire responses. In this way, we avoidedthe problem of common method variance. Also, we were able to assessthe convergent validity of the independent variable by seeing whether


different definitions of the same variable produced the same results(cf. Campbell and Fiske, 1959).

At Silicon Systems, there were three levels of formal authority. Thethree owner-managers occupied the top level. Even though each owner-manager had different responsibilities and titles, all three were equalpartners, and they made all major company decisions as a team. Thenext level consisted of five managers, each of whom had supervisoryresponsibility over certain operational features in the organization. Theremaining twenty-eight employees had no formal supervisory titles orauthority. Formal status, then, was rated as follows: We gave each of thethree owners a status rating of 3, each of the five managers a rating of 2,and each of the remaining twenty-eight employees a rating of 1.

We assigned formal status to both perceived friends and actual friends.For the perceived measure, we looked at each respondent’s cognitivemap of perceived relations. Within each cognitive map, we identified,for each person, the friend with the highest formal status. This statusrating was recorded as the third measurement of the independent variable.The measure of the actual friend’s formal status was based on the realfriendship network aggregated from the responses of all respondents.For each person, we identified the friend with the highest status ratingand recorded the rating as the fourth measurement of the independentvariable.

In summary, we measured each friend’s prominence in four ways, pit-ting perceived and actual network measures against each other and pittinga network measure of prominence against an organizational chart mea-sure of prominence. For each of the four measures, we created a 36-by-36matrix, with cell entries representing the prominence ratings of friends.For example, for the matrix of perceived friends’ indegree centrality rat-ings, a “9” in a cell formed by the intersection of row 12 and column25 meant that, among all the friendships that person 12 perceived person25 to be involved in, 9 was the highest indegree centrality rating that anyof person 25’s friends achieved.

Dependent Variable: Performance Reputation MatrixEach respondent provided his or her perception of the job performanceof every person in the organization by circling numbers on a seven-pointscale next to people’s names. We collected these performance reputationratings in a 36-by-36 matrix. Each row in the dependent variable matrixrepresented the impressions in the mind of one respondent concerning theperformance of those others not actually under the respondent’s supervi-sion. Similarly, each column in the matrix represented the impressions ofone individual held by all those respondents not actually supervising that


individual. The performance reputation matrix contained the actual rawratings that the respondents provided. Social network analysts typicallyretain raw ratings in matrix form rather than seeking to perform analy-ses on average scores (see Kilduff and Tsai, 2003, for an introduction tosocial network analysis). We elicited the raw ratings with the followinginstructions: “If you think the person is performing their job extremelywell, you might circle the ‘7’ next to their name. If you think the person isperforming their job reasonably well, you might circle the ‘4’ next to theirname. If you think they are not performing their job at all well, you mightcircle the ‘1’ next to their name.” Each cell in the performance reputationmatrix contained the rating provided by one respondent concerning oneother person. For example, if person 12 rated person 25 as performingextremely well on the job, then a “7” would appear in the cell formed bythe intersection of row 12 and column 25.

Our reliance on a one-item measure of performance increases the pos-sibility of random error and makes significant results harder to find. Ourtests, therefore, are likely to be conservative assessments of the hypothe-sized basking-in-reflected-glory effect.

First Control Variable: Job Performance MatrixSupervisors’ ratings of subordinates’ performance were excluded fromthe dependent variable matrix previously described because these super-visory ratings constituted the measurement of job performance. Super-visors therefore used the same seven-point scale that was used for theperformance reputation measure. The job performance of people in orga-nizations is typically difficult to ascertain, especially for work with manydifferent aspects. However, one conventional measure of job performancein many companies is the supervisory rating: “The vast majority of per-formance ratings come directly from the immediate manager” (Bretz,Milkovich, and Read, 1992: 331). Previous research has shown thatperformance ratings obtained for research purposes are more reliableand valid than those obtained for administrative purposes (Wherry andBartlett, 1982), perhaps because issues other than ratee performance biasofficial performance ratings (Longenecker, Sims, and Gioia, 1987; Tsuiand O’Reilly, 1989: 410).

The reporting relations between supervisors and subordinates wereobtained from company records. The three owners of the company inthe present research had nobody above them in the organizational chartto provide a supervisory rating. For each owner, therefore, we used themean rating given by the other two owners as the supervisory rating. Theowners’ ratings of each other did not differ in any case by more than twopoints on the seven-point scale.


The supervisory ratings were collected in a column vector thirty-sixcells long containing values from 1 through 7, indicating the job per-formance of each person in the company. Thus, each cell (i,j) in thismatrix contained j’s supervisor’s rating of j’s performance. The vectorwas repeated thirty-six times to create a matrix of the same size (36 by36) as the other matrixes in the analyses.

Second Control Variable: Formal Status MatrixThis variable controlled for the effects of formal status on the performancereputation of each focal person. In this small, organic organization, therewas little apparent status differentiation based on educational differencesor functional specialization. From our visits to the company, we con-cluded that the major status difference was between those who ownedthe company and those who only worked for it. Therefore, we definedformal status as the level in the organizational hierarchy that each personoccupied (3 = owner-manager, 2 = manager, and 1 = nonmanager).

The formal status scores were collected in a column vector thirty-sixcells long containing the numbers 1 through 3. Thus, each cell (i,j) inthis matrix contained j’s formal status. The vector was repeated thirty-sixtimes to create a matrix of the same size (36 by 36) as the other matrixesin the analysis.

Analysis

Social network data are often not amenable to standard statistical testsbecause the observations cannot be assumed to be independent. For exam-ple, in the current research, the matrix of friend’s indegree centrality rat-ings includes thirty-six ratings from each person in the study. Each ofthe thirty-six ratings within a row of this matrix derives from the samesource – the cognitive map of the respondent – and therefore exhibitssystematic interdependence. Indeed, in some of the matrices, the cell val-ues are repeated across rows. Krackhardt (1988) showed that such rowor column interdependence can bias ordinary-least-squares (OLS) testsof significance. The size of this bias is substantial: Results based on sam-ples drawn from a population for which the null hypothesis is true (thatis, there is no relationship between the independent and dependent vari-ables) have a 70 percent chance of appearing significant under standardparametric methods.

To deal with this problem of bias, we used the Multiple RegressionQuadratic Assignment Procedure (MRQAP) suggested by Krackhardt(1993). The procedure builds on earlier bivariate work done by Hubertand others (Baker and Hubert, 1981; Hubert, 1987; Hubert and Schultz,1976) and extended to the multiple regression case by Krackhardt (1987b,


1988). (Note that the use of MRQAP in this book generally avoids theserious problems outlined by Dekker, Krackhardt, and Snijders, forth-coming, concerning the dependent variable permutations often used insocial network research.)

The method is straightforward. First, OLS estimates of regression coef-ficients are calculated in the usual manner. Then the rows and columnsof the dependent variable matrix are permuted to give a new, mixed upmatrix. The OLS regression calculation is then repeated with the newdependent variable. This new regression produces different beta coeffi-cients and overall R2 values that are stored away. Another permutationof the dependent variable is then drawn, another regression is performed,and these new values are also stored.

This permutation-regression process is repeated an arbitrarily largenumber of times (in our case, one thousand). The distribution of thestored betas and R2s for each of the independent variables under the setof permuted regressions becomes the reference distribution against whichthe observed original values are compared. If fewer than 5 percent of thebetas derived from the permuted regressions are larger than the observedbeta, the beta is considered significant at the .05 level (one-tailed test). Iffewer than 1 percent of the betas are larger than the observed beta, it isconsidered significant at the .01 level.

The advantage of this simple procedure is that it is robust against vary-ing and unknowable amounts of row and column autocorrelation in thedyadic data. That is, if a sample is drawn from an autocorrelated popu-lation in which the null hypothesis is true, the probability that the resultswill appear significant by this MRQAP test is .05 (where alpha equals.05). This remarkable feature of the MRQAP occurs because the test is aconditional nonparametric test. That is, each permutation of the depen-dent variable retains the structure of the original dyadic data and thereforepreserves all the autocorrelation in each permuted regression; the test isconditioned on the degree of autocorrelation that exists in the data.

The permutation version of MRQAP (Krackhardt, 1993) differs fromthe earlier analytic version (Krackhardt, 1988). The analytic solution tothe multiple regression problem was based on Mantel’s formula (Mantel,1967) for the first two moments of the distribution of all permutations.The current version has several demonstrated advantages. First, it per-mits an unbiased test of the overall R2. Second, it is relatively powerfulin the face of missing data. Finally, whereas the analytic test necessarilycontains the assumption that the reference distribution of betas basedon the permutations is normally distributed, the permutation-based sam-pling procedure used here does not have such a requirement. PermutationMRQAP is now available in user-friendly form in the UCINET social net-work analysis package (Borgatti et al., 2002).


Table 3.2. Means, Standard Deviations, and Correlationsa

Variables Means s.d. 1 2a 2b 2c 2d 3

1. Performance reputation 4.93 1.392. Friend’s prominence

a. Perceived friend’s indegree centrality 6.70 8.86 .23b. Actual friend’s indegree centrality 7.87 7.00 .26 .22c. Perceived friend’s status 1.46 0.73 .28 .68 .30d. Actual friend’s status 1.55 0.83 .28 .15∗ .83 .30

3. Job performance 4.91 1.15 .33 .14∗ .31∗ .28 .494. Formal status 1.31 0.62 .30 .17 .54 .38 .65 .47

a All correlations are significant at p < .01, except for those with an asterisk.∗ p < .05

One of the advantages of the permutation version of MRQAP is that itcan handle missing values with much more statistical efficiency than theprior version. In the current research, several of the variables had valuesmissing, either because there were three nonrespondents or because weassigned cells in a matrix missing value status when defining the vari-able. For example, in the case of the dependent variable, performancereputation was only considered in (i,j) pairs in which i was not j’s directsupervisor, and a missing value was inserted when i was the direct super-visor of j.

Results

The descriptive statistics shown in Table 3.2 indicate a reasonably highlevel of performance at Silicon Systems, with both performance reputationand actual performance averaging around 4.9 on a seven-point scale.

The zero-order correlations in Table 3.2 show that the two measuresof perceived friend’s prominence – perceived friend’s indegree centralityand perceived friend’s formal status – were highly correlated (r = .68,p < .01), as were the two measures of actual friend’s prominence (r =.83, p < .01). Further, these correlations suggest that the actual friendsof high-status individuals tended to also be of high status (r = .65, p <

.01) and indegree centrality (r = .54, p < .01).Table 3.2 also shows that the dependent variable, performance reputa-

tion, was significantly correlated (at p < .01) with all four measurementsof the independent variable (friend’s prominence), as well as with bothcontrol variables (job performance and formal status). To answer thequestion of whether these significant correlations would remain signifi-cant when other variables were controlled for, we conducted a multipleregression analysis.


Table 3.3. Results of Multiple Regression Analysisa

Status Models Centrality Models

Variables 1 2 3 4 5 6 7

Friend’s prominencePerceived friend’s

indegree centrality .028∗∗ .026∗∗Actual friend’s

indegree centrality .024 .018Perceived friend’s

status .319∗∗ .315∗∗Actual friend’s status .109 .090

Job performance .296∗∗ .277∗ .273∗ .258∗ .286∗ .284∗ .277∗Formal status .407 .326 .281 .217 .263 .350 .245Intercept 2.93∗∗ 2.96∗∗ 2.74∗∗ 2.76∗∗ 2.99∗∗ 2.88∗∗ 2.92∗∗R2 .136 .139 .161 .162 .146 .166 .171

a Beta coefficients are unstandardized. Their significance was determined by MRQAP (Krackhardt, 1993).∗ p < .05, one-tailed test.∗∗ p < .01, one-tailed test.

Results of the first model, shown in Table 3.3, suggest that high perfor-mance on the job in this organization helped people achieve reputationsas high performers (p < .01) but that formal status did not significantlyaffect performance reputations. The two control variables explained 14percent of the variance in performance reputation. The question of inter-est, then, is whether the measures of the independent variable signifi-cantly increased explained variance above that already explained by thecontrol variables. Did the existence of a friendship link to a prominentperson boost individuals’ performance reputations in this organization,as hypothesis 1 predicts?

Table 3.3 shows that friendship with prominent others did boost indi-viduals’ performance reputations, but this effect depended on how thefriendship links were assessed. Recall that hypothesis 2 predicts that per-ceived friendship links will lead to better predictions of performancereputation than actual links. The results shown in Table 3.3 support thisprediction.

Models 2, 3, and 4 in Table 3.3 employed two different definitions ofthe status of the highest-status friend to measure the independent vari-able. Model 2 shows that entering the status of the actual friend intothe regression equation together with the control variables resulted in nosignificant increase in the variance explained. Model 3 shows that theintroduction of the status of the perceived friend did increase explainedvariance significantly (p < .01), from 14 to 16 percent. Entering both mea-surements of friend’s status simultaneously (model 4) confirmed that only


the perceived measure had a significant effect on performance reputation(p < .01).

In support of hypothesis 2, then, these models show that, with indi-viduals’ job performance and organizational status controlled, only theperceived measure of friend’s status had an effect on individuals’ perfor-mance reputations. Being perceived to have a friend at a high level inthe organization helped boost an individual’s reputation as a high per-former, whereas actually having such a friend had no significant effect onperformance reputation.

Models 5, 6, and 7 in Table 3.3 repeat the analyses already performedin models 2, 3, and 4, with a measure of the friend’s indegree centralityin the informal advice network substituting for a measure of the friend’sformal status. The results for models 5, 6, and 7 repeat the pattern seen inmodels 2, 3, and 4, indicating that the results favoring perceived friend-ship over actual friendship were not artifacts of the way that prominencewas measured.

Model 5 shows that entering the indegree centrality of the actual friendin the regression equation did not significantly increase explained vari-ance. Model 6 indicates that the introduction of the indegree central-ity of the perceived friend did increase explained variance significantly(p < .01), from 14 to 17 percent. Finally, model 7 confirms that whenboth the actual and the perceived measures of friend’s indegree centralitywere entered together, only the perceived measure had a significant effecton performance reputation (p < .01).

Paralleling the results from the status models, the results from the cen-trality models show that, with individuals’ job performance and organi-zational status controlled, only the perceived measure of friend’s indegreecentrality had an effect on individuals’ performance reputations. In otherwords, being perceived to have a friend to whom many others go forhelp and advice helped boost an individual’s reputation as a high per-former, whereas actually having such a friend had no significant effect onperformance reputation.

In summary, Table 3.3 shows that the status models (2, 3, and 4) andthe centrality models (5, 6, and 7) are similar both in terms of the superior-ity of perceived measures of friend’s prominence over actual measures andin terms of the variance explained by each set of models. These consistentresults support the convergent validity of our measures of perceived andactual prominence. The results show that it doesn’t matter whether theprominence ratings derive from the friend’s position in the organizationalhierarchy or from questionnaire items concerning who goes to whom foradvice at work. The robustness of the results across measures derivedfrom two different sources supports the conclusion that the significantcorrelations are not artifacts of common method variance.


One other concern, however, is that the effect of actual prominencemight have been suppressed as a result of the correlational structure ofthe data. Table 3.2 shows that, relative to measures of perceived promi-nence, measures of actual prominence were more highly correlated withthe control variable (job performance) that contributed significantly toexplained variance in all the models of Table 3.3. To check whether thehigh correlations between measures of actual prominence and the controlvariable distorted the results, we conducted the analyses again withoutcontrolling for job performance and found the same pattern of results(albeit with less explained variance): Only the perceived measures offriend’s prominence significantly predicted performance reputation. Themeasures of actual friend’s prominence continued to be nonsignificant inall models.

Discussion

In support of the hypothesized basking-in-reflected-glory effect, theresults show that performance reputation is partly a function of an indi-vidual’s job performance and partly a function of the individual having aprominent friend. Perhaps the most intriguing aspect of the results is thefinding that the actual existence of friendship links, recognized by bothparties to the links, had no significant effect on performance reputation.Rather, it was the perceptions in the minds of organization membersthat mattered. To explain outcomes such as performance reputation inorganizations, it may be necessary to explore the perceived networks thatinfluence the attitudes of organization members. Structure, as it exists inthe minds of individuals, may be more predictive of important outcomesthan has been recognized. Bringing the individual back into structuralanalysis, therefore, may enhance rather than detract from the effective-ness of a structural approach.

The results, then, support the utility of combining variables derivedfrom individuals’ cognitive maps with more conventional structural vari-ables. The thesis that psychological and structural approaches repre-sent incommensurable paradigms militates against the kind of cross-levelapproaches that appear well adapted to the complex realities of organi-zations. We used MRQAP as one way to combine levels of analysis. Thisstudy demonstrates the use of the procedure as a flexible tool for com-bining multiple observations from each individual’s cognitive map withsingle measures on each individual within the same analysis.

The theoretical basis for the current research is balance theory (Heider,1958), which has a long history of use within social network analysis;Davis (1979) reviewed the relevant research. Much of this previous work


modeled relations in social rather than cognitive space, following theinfluential mathematical extension of Heider’s ideas from the conceptof cognitive balance to that of interpersonal balance (Cartwright andHarary, 1956). Social network analysts continue to develop sophisticatedmathematical approaches to social structure (e.g., Boyd, 1991), but Blau’swarning remains pertinent: “There is a danger that the refined methodsthat network analysis . . . has developed will lead to sterile descriptivestudies” (1982: 279). In examining Heider’s predictions concerning thestrain toward cognitive balance, we have sought to return social networkanalysis to a theory-driven mode rather than a purely method-drivenmode.

The research presented here is both an example of how structural meth-ods can incorporate individuals’ cognitive maps and a contribution to theliterature on performance reputation. We have interpreted the results assupporting the idea that observers’ perceptions of individuals’ friendshiplinks to prominent others positively influence the observers’ evaluationsof the individuals concerned. This interpretation is compatible with bal-ance theory in general and with research on the basking-in-reflected-gloryeffect in particular

However, the data are cross-sectional and can support other causalarguments. For example, it is possible that individuals perceived by theircolleagues to be high performers are assumed to have prominent friends.Without more detailed observations on the process by which perceptionsconcerning performance and friendship links are formed, the presentresults must remain suggestive rather than conclusive. Future researchcould investigate how reputations change over time in response to impres-sion management techniques (cf. Tsui and Barry, 1986), and possiblepersonality differences between individuals in their impression manage-ment strategies. For example, high self-monitors – individuals who arehighly sensitive to social cues – may actively gather and use informationconcerning who is friends with whom; whereas low self-monitors – thosewho rely on their own attitudes and feelings for guidance – may be averseto trying to influence perceptions of their social relations (cf. Kilduff,1992).

A second limitation of the current research concerns the small size of theorganization studied and the correspondingly high degree of interactionamong its employees. Silicon Systems may be untypical because all theemployees were at least weakly tied to each other, if Granovetter’s (1973)definition of interacting more than once a year is used. The question ofwhether the results generalize to large organizations will be difficult toanswer given the methodological limitations of social network research.Typically, social networkers attempt to include the complete network ofpeople in a social setting. For research concerning people’s cognitive maps


of entire networks of relations, data collection and analysis constraintsdictate an upper bound of about fifty people (Krackhardt, 1987a).

However, in large organizations, where people may not know eachother as well as did the people in our study and where, therefore, specificinformation about others may be scarce, performance reputations may beeven more reflective of perceptions and impressions. Research has shownthat when decision makers lack information about an employee, theyrely on prevailing cognitions, such as stereotypes (Drazin and Auster,1987), and that halo errors are more likely to occur when raters areevaluating people with whom they are unfamiliar (Kozlowski and Kirsch,1987). Thus, we would expect individuals’ perceptions to be even moreimportant in determining others’ reputations in large organizations thanthey were in this small organization.

We assumed throughout this research that individuals act strategicallyto emphasize friendship links to prominent others. This assumption iscompatible with the basking-in-reflected-glory effect and with evidenceof wide variation with respect to how accurately people perceive net-work relations (Krackhardt, 1990). The relative opaqueness of friend-ship relations may provide opportunities for the strategic management ofimpressions.

Research on impression management suggests that individuals per-ceived to be linked to prominent others may be credited with the abilityto form powerful coalitions and the ability to influence higher-status per-sons (Tedeschi and Melburg, 1984). In other words, individuals perceivedto have prominent friends may gain important advantages in the marketfor power and influence in an organization. Research on these phenom-ena in organizational settings is lacking, although anecdotes abound. Forexample, in the struggle for the control of the Lehman Brothers invest-ment banking house, Louis Glucksman gained a crucial advantage byconvincing his rival Pete Petersen that he, Petersen, had lost friendshipswith board members that Glucksman had retained. Neither Petersen norGlucksman ever checked with the board members to see whether thoseimpressions were accurate (Auletta, 1986). More empirical research onthe impression management of friendship ties in organizations would beuseful.

The declared aim of structural analysis has been to reveal the struc-tural form beneath the apparent content of social relations. Accordingto structuralists, the unit of analysis is “the social network, never theindividual” (Mayhew, 1980: 349). Structuralists have tended to “shunthe ‘person’ construct as polluting” (White, 1992: 3). In this chapter, wehave challenged the notion that structure can be understood apart fromthe cognitions of individuals. Our argument is compatible with the cri-tique of structuralist claims by poststructuralists (see Agger, 1991, for a


general introduction). In particular, by including each individual’s cogni-tive map in the analysis, we follow poststructuralist writers in rejectingthe privileged status of any one particular interpretation of structure. Wehave also challenged the claimed incommensurability of individualismand structuralism by pointing to the influence on structural analysis ofthe psychology it has purported to reject and by providing an explicitdemonstration of how a cognitive theory can guide the use of structuralmethods.

In the next chapter, we continue our investigation of balance theoryand the perception of network relations, looking systematically at howpeople distort perceptions close to and far from their own positions infriendship networks in organizations.

4

Systematic Biases in NetworkPerception

In Chapter 3, we predicted and found that the perceptions in people’sminds concerning whether a target individual was a friend of a prominentperson significantly affected the target individual’s reputation concerningwork performance in an organization (Kilduff and Krackhardt, 1994).The actual existence of friendship links, recognized by both parties ineach link, had no significant effect on other people’s perceptions of anindividual’s reputation as a high performer. This research showed thatpeople’s perceptions of relations helped to determine reputations, whereasthe actual structure of relations had no effect.

In this chapter, we focus again on perceptions of the friendship net-work, this time investigating how perceptions are shaped by preexistingexpectations. We chose the friendship network to study because this net-work affects important choices individuals make. We ask, under what cir-cumstances are individuals’ perceptions of the friendship network shapedby schemas concerning how people typically behave in the friendshiprole?

The role of friend is well understood in society, as indicated by the highlevel of agreement within societies concerning how friends should act inrelation to each other (Argyle and Henderson, 1985: 92). People haveaccess to a schema or strategy that specifies how individuals typicallyact in this role (see the discussions in DiMaggio, 1991; Swidler, 1986).Cognitive psychologists have described schemas as mental structures thatenable people to anticipate the general features of recurring situations(Neisser, 1976: 51–78). Schemas allow people to search for and recognizerelevant features of the person, situation, or process. The schema thathas been most actively researched in the literature on friendship is thebalance schema (for reviews, see Crockett, 1982; Markus and Zajonc,1985; Wasserman and Faust, 1994, chap. 6).

According to Heider’s explanation of the balance schema (1958: 205),perceivers tend to treat positive sentiment relations such as friendship as

59


if they were symmetric and transitive. Symmetry refers to the perceiver’sassumption that friendship relations will be reciprocated. Thus, if theperceiver sees that A chooses B as a friend, the perceiver will anticipatethat B will also choose A as a friend. Transitivity refers to the perceiver’sassumption that friendship relations will be complete. Thus, if the per-ceiver knows both that A is friends with B and that A is friends with C,the perceiver will anticipate that B and C will also be friends. The balanceschema, then, consists of a set of cognitive expectations concerning thelikely structure of the social world in terms of reciprocity and transitivity.

Balance theory literature, however, offers several explanations for whypeople tend to perceive friendship relationships as balanced. We explorethree perspectives: an emotional tension model, a cognitive miser model,and a composite model that combines the predictions of the other two.

The Emotional Tension Model

The first model emphasizes the emotional tension that results from dis-crepant cognitions, such as the perception of unbalanced friendship rela-tions. People are motivated to resolve such cognitive discrepancies eitherby altering cognitions or by taking action in the world. Thus, if Jack per-ceives that his friendship overtures to his colleague Randolph are unre-ciprocated, the discrepant cognitions (e.g., “I’m Randolph’s friend, butRandolph doesn’t like me”) will prompt either a change in cognition(“Perhaps I’m not really his friend”) or a change in behavior (“I need towork harder to make this friendship work”).

Individuals who perceive that their friendship relations are unbalancedmay react with strong emotions rather than with cool analytical reason-ing. The balance schema, from this perspective, functions as a deep-seatedgoal of human interaction (see the discussions in D’Andrade, 1992; Fiske,1992). People strive to see friendship relations as balanced because theperception of unbalance induces feelings of uncertainty, instability (Fes-tinger and Hutte, 1954), and nervousness (Sampson and Insko, 1964). AsHeider suggested, ego’s perception of an unstructured region in the envi-ronment functions as a barrier that “makes action and therefore controldifficult if not impossible” (1958: 71).

The region closest to ego includes ego’s own personal friendships. Egohas power to directly influence whether these friendships are balanced ornot. If, for example, Jane finds that her attempts at friendship with Rubyare unrequited, then Jane can sever the friendship link or try even harderto elicit tokens of friendship from Ruby. Ego has considerable potentialpower to balance friendship relations through direct action of this sort.Similarly, if Jane finds that her friendships with Alice and Shirley havenot led Alice and Shirley themselves to become friends, then Jane can

Systematic Biases in Network Perception 61

endeavor to bring her two friends together – over lunch in the cafeteria,for example. Within the region of the network where ego is connected tohis or her friends, then, ego is potentially able to balance relationshipsthrough direct action.

As ego surveys the friendship relations of his or her friends and offriends of friends, however, ego’s power to impose balance becomes con-siderably weakened. Further, if an alter (i.e., another individual) at somedistance from ego is perceived to be involved in friendship relations thatare unbalanced, this can disturb ego, because alter is still part of ego’ssocial world and as such is part of the mutually shared environmentin which ego is involved every day. Perturbations affecting alter affectego because ego encounters alter and alter’s friends in the daily round.The region occupied by an alter with unbalanced relations is likely to beperceived as one of uncertainty and tension.

For relations close to ego, therefore, motivation is strong to balancerelationships, and ego has the power to impose balance. As ego sur-veys relations at distances farther and farther from ego’s own position,however, the power to act is diminished, but the emotional uncertaintyinduced by perceived imbalance is also likely to be diminished. The unbal-anced friendship relations of friends of friends represent areas of uncer-tainty and tension in the social world as perceived by ego. However,the prospect of uncertainty and tension derived from unbalanced distantrelations is likely to be less troublesome than the immediacy of uncer-tainty and tension derived from unbalanced relations within ego’s ownfriendship circle.

How quickly are the tension and uncertainty that are induced by per-ceived unbalanced friendship relations reduced as ego scans the socialrelations of friends, friends of friends, and so on out to the periphery ofthe social world? There has been some discussion of this issue. Heider(1958: 71) mentioned the problematic nature of unstructured regions any-where in the mutually shared environment, whereas Insko (1981: 322–3)suggested that ego is likely to suffer little tension to the extent that egohas low involvement with individuals whose relations are unbalanced.Our working assumption is that the emotional pressure to perceive rela-tions as balanced sharply diminishes but does not disappear as ego looksbeyond his or her own friendship circle.

Previous research has shown that people will alter relations or cog-nitions to preserve balance in close relations (Kumbasar, Romney, andBatchelder, 1994; Newcomb, 1961), and that individuals tend to seekout information that reduces dissonance and to avoid information thatincreases it (Ehrlich, Guttman, Schonbach, and Mills, 1957). Consid-erable evidence also indicates that people prefer balanced relations ingeneral, even when they themselves are not directly connected to the


Distance

Pro

port

ion

Figure 4.1. Illustration of the emotional tension model’s prediction thatthe proportion of relations perceived by ego as balanced declines withsocial distance from ego.

individuals concerned (De Soto, 1960; Freeman, 1992). In the everydayworld of social relationships, individuals are frequently brought into con-tact with acquaintances whose friendship relations may be unbalanced.In other words, individuals may be required to negotiate social pathwaysthat are perceived to be unstructured and therefore problematic. Avoidingpeople with friendship problems may be either not possible or not com-patible with, for example, a productive career. Thus, if Alice sees Johnas someone whose attempts at friendship are unreciprocated, she maywant to avoid John; interacting with him may be bothersome becauseof the presumed tension he is under. However, Alice may have to workwith John to accomplish her own tasks. Thus, the perceived imbalance inJohn’s friendships can affect Alice even if John himself is not a personalfriend of Alice. We assume that the effect of alter’s unbalanced relationson ego diminishes sharply but does not disappear as ego considers altersfarther and farther away from ego.

To summarize, an emotional tension perspective on perceived balanceleads to the prediction that ego’s close relations will tend to be perceivedby ego as balanced, because ego has both the motivation and the powerto arrange for them to be balanced. The curve in Figure 4.1 illustrateshow ego’s perception of balance may be affected by social distance. Asego looks beyond the immediate circle of close friends, the emotional


pressure to perceive relations as balanced sharply diminishes. Thus, asego assesses the likelihood of balance among strangers on the remotehorizon of ego’s social world, the degree of perceived balance shouldapproach its minimum.

The Cognitive Miser Model

A contrasting perspective that represents a more recent paradigm in thefield of social cognition views the person as a cognitive miser who, undercircumstances of “unavailability or indeterminacy of important informa-tion” (Taylor, 1991: 195), relies on short cuts or heuristics to fill in theblanks in knowledge. The cognitive miser perspective emphasizes thateven when people believe they are using complete information to formimpressions or make decisions, they may be relying on only one or twosalient cues (Dawes, 1976; Taylor, 1981; Taylor and Fiske, 1978). Peopleare cognitive misers in the sense that they tend to avoid devoting the timeand effort required to locate and use all relevant information prior toforming an opinion or perception.

Applied to perceptions of friendship networks, this perspective sug-gests that people utilize schemas to help make sense of the mass of poten-tial relations they observe. People may avoid expending cognitive energykeeping track of the potential relations characteristic of social groups. Tothe extent that an individual uses a well-developed schema, many detailsof the social world may be filled in by the schema rather than derivedfrom actual perception (see the review by Mandler, 1979). The balanceschema, then, provides ego with a way to infer the existence of relationswhen information is incomplete (Freeman, 1992). In particular, as peo-ple consider the friendship relations of those increasingly distant fromthemselves, they will have less and less knowledge of possible unbalancedrelations (cf. McPherson, Popielarz, and Drobnic, 1992: 155). The far-ther away the relationship, the less information ego has regarding it andthe more likely, therefore, ego is to assume that relations are balanced(Kuethe, 1962).

There is a further reason why, from a cognitive miser perspective,people may rely increasingly on assumptions concerning balance as theyscan distant relations: As people scan relations at greater and greaterdistances from themselves, they incorporate more and more people intotheir social world, and the number of possible relationships they mustkeep track of increases disproportionately. The effect of increasing groupsize on the number of possible relations in a group was dubbed “the lawof family interaction” in an influential article by Bossard (1945: 292). Asthe group increases from four to eight members, for example, Bossardcalculated that the number of possible relations increased from six to


twenty-eight. Other researchers have offered refinements concerning howquickly the number of possible relations increases as each additionalmember is added to the group (e.g., Kephart, 1950), but all echo Bossard’sobservation that the larger the group becomes, the more disproportionatethe increase in possible personal relationships between members. From acognitive miser perspective, people are likely to use the balance schemato fill in the blanks in their social knowledge rather than try to keep trackof the large numbers of possible relations involving people at farther andfarther distances from their own friendship circles.

The law of family interaction suggests, therefore, that ego faces a daunt-ing cognitive task in trying to keep track of relations of people at fartherand farther distances. From this perspective, ego may tend to rely on thebalance schema as a useful heuristic for making sense of distant relation-ships. Instead of keeping track of which relationships are reciprocated ortransitive, ego may tend to assume that distant relationships are generallybalanced.

Within ego’s own social circle, however, unbalanced relations concern-ing ego’s own personal friends may be hard to ignore. Research suggeststhat people are more likely to notice evidence of unbalance than balance.For example, angry faces are found more efficiently in happy crowds thanare happy faces in angry crowds (Hansen and Hansen, 1988). Heider sug-gested that unbalanced relations about which we have personal knowl-edge “stimulate us to further thinking” and “have the character of inter-esting puzzles” (1958: 180). In their review of the literature on the recallof schema-consistent and schema-inconsistent information, Markus andZajonc (1985) suggested that schema-inconsistent information is likelyto be recalled if it competes with the information in the schema and if thecognitive task requires the participant to make use of it. People are likelyto remember the existence of unbalanced friendship relations in whichthey themselves are involved because this imbalance competes with thestructures suggested by the balance schema, and the cognitive task ofmaking sense of the immediate social world requires people to keep trackof unbalance (cf. Janicik and Larrick, 2005). This is true, for example, inwork organizations where individuals are likely to see their friends everyworkday and are therefore reminded daily of the absence of reciprocityand transitivity.

In summary, the cognitive miser perspective suggests that people arelikely to notice unbalanced relations among those close to themselves.However, as people scan distant relations, they are likely to rely on thebalance heuristic to fill in the blanks of the relations of these distant oth-ers. Figure 4.2 offers one possible representation of the cognitive misermodel, depicting an increasing probability of perceived balance for rela-tions farther and farther away from ego.


Distance

Pro

port

ion

Figure 4.2. Illustration of the cognitive miser model’s prediction thatthe proportion of relations perceived by ego as balanced increases withsocial distance from ego.

The Composite Model

Despite the apparent contradiction between the emotional tension andcognitive miser models, there is a way to combine these two views: Thebalance schema may be imposed on close relationships (to avoid emo-tional tension) and attributed to the friendship relations of distant others(to fill in the blanks in social knowledge). If one accepts the extensive evi-dence that people are likely to suffer discomfort when they perceive theirown friendship relations as imbalanced, then the major results of theemotional tension model are accommodated. According to this model,people are relatively unaffected by the perception of imbalance amongthose with whom they have no friendship ties. The motivation to changeperceptions in favor of balance, then, is likely to affect mainly the per-ceptions of ego’s own friendship relations. If ego is not directly involved,little discomfort results from perceived imbalance.

According to the cognitive miser model, however, while casting one’sgaze outward over the friendship relations of those with whom one has nodirect links, one is likely to have less and less knowledge concerning suchdetails as whether the relations are reciprocated or transitive. The lessinformation that is available, the more one relies on the balance schemato fill in the blanks in one’s knowledge.

The composite model, then, suggests that the hypothesized graphs inFigures 4.1 and 4.2 can be joined to display a curvilinear relationshipbetween social distance and the degree of balance perceived. In summary,


according to the composite model, individuals perceive both their imme-diate friendship circle and the periphery of their social networks as morebalanced than social worlds of intermediate distance.

Method

At each of the four sites described in this section, participants werepromised and given an overview of the findings. At all four sites, thesame questionnaire was used, as described in the “Perceived FriendshipNetwork” section. Nonrespondents were excluded from all analyses. Therelatively high response rates (which varied from 86 percent to 100 per-cent) reduced problems associated with nonresponse bias.

Site 1: High-Tech Managers (HT)The participants at this site consisted of all twenty-one managers of a WestCoast entrepreneurial firm of approximately 100 people employed in themanufacture of high-tech machinery. The managers were all men. DavidKrackhardt collected the data as part of an effort to explore the effects of aprevious organizational development intervention conducted by externalconsultants. All twenty-one participants completed the questionnaire, andno compensation was offered to the participants. (See Krackhardt, 1987a,for further details.)

Site 2: Government Office (Gov)This workgroup consisted of thirty-six professional staffers in the federalbureaucracy. Their job included advising the executive branch concerningcourses of action that would facilitate the current public policy agenda.Each person in this workgroup had an advanced degree at the master’slevel or higher. The group’s composition changed yearly as new stafferswere added from different departments and others rotated out. The lead-ership of the group, however, had been in place for years. Thirty-oneof the thirty-six people in this office completed a questionnaire, and nocompensation was offered to the participants.

Site 3: Silicon Systems (Sil)The participants included all thirty-six employees (twenty-eight men andeight women) of a small entrepreneurial firm located in the Bay Area ofCalifornia. The employees were mostly semiskilled workers who installedcomputers and trained their clients in their use. Thirty-three of the thirty-six employees accepted $3 each to complete the questionnaire. (SeeKilduff and Krackhardt, 1994, for more details.)

Site 4: Pacific Distributors (Pac)The participants included all thirty-three supervisory and managerial per-sonnel (fifteen men and eighteen women) located at the headquarters of


a small, rapidly growing regional distributor of electronic components.The company employed 162 people in its headquarters and four branchoffices. All thirty-three participants accepted $10 each to complete thequestionnaire. (See Krackhardt and Kilduff, 1990, for more details.)

Measures

Perceived Friendship NetworkTo capture participants’ perceptions of friendship relations, we used thesame questionnaire across all four sites. At each site, every respondentanswered the following question about every other person in the organi-zation: “Who would this person consider to be a personal friend? Pleaseplace a check next to the names of those people who that person wouldconsider to be a friend of theirs.” For example, John Meredith of the Silsample was asked a series of thirty-six questions concerning the friend-ships of his thirty-six coworkers. The questions were in this form: “Whowould Jane Asch consider to be a personal friend?” Each question wasfollowed by the list of thirty-five employees’ names. John Meredith thenchecked the names that indicated his perceptions of who Jane Asch con-sidered to be her personal friends. (For more details, see Kilduff andKrackhardt, 1994.)

Each respondent, then, gave us a complete cognitive map of his or herperceptions concerning who was friends with whom in the organization.To measure perceived friendship links, we used the following procedure:A friendship tie as perceived by person k existed between person i andperson j only if k responded on the questionnaire that i considered j afriend.

Perceived ReciprocityWe measured the extent to which each person in the network was per-ceived by every other person in the network to be involved in reciprocatedfriendships. We created a matrix of scores for each site, with each cellin the matrix indicating (according to person i’s perceptions) the propor-tion of person j’s dyadic relationships that were reciprocated. More for-mally, the perceived reciprocity matrix was defined as follows: Sij = NSij/(NSij + NNij), where NSij is the number of reciprocated dyadic relationsthat i perceives j to be involved in, and NNij is the number of unrecipro-cated dyadic relations that i perceives j to be involved in.

Perceived TransitivityWe measured the extent to which each person in the network was per-ceived by every other person in the network to be involved in transitivefriendship relations. We created a matrix of scores for each site, with eachij cell representing (according to i’s perceptions) the proportion of j’s tri-adic relationships that were transitive. Given that we separately analyzed


perceptions of reciprocity from perceptions of transitivity, we chose theconservative path of considering only reciprocated ties in the transitivityanalysis. Although there is a tradition within social network analysis ofconsidering transitivity among unreciprocated relations as balanced (e.g.,Holland and Leinhardt, 1977), it is clear from Heider’s (1958: 206–7) dis-cussion of transitivity that he considered positive relations to be balancedonly if the transitive relations were also reciprocated.

Therefore, following Heider, when we refer to friendship relations inour definition of what constitutes a transitive triad, we are referring toreciprocated friendship relations. To compute transitivity, we temporarilysymmetrized ego’s perceptions of friendship relations using the intersec-tion rule that for a friendship relation between i and j to exist in ego’sperceptions, ego must perceive both a friendship link from i to j and afriendship link from j to i. Formally speaking, then, for any triple of actorsi, j, and k, given that i and j are friends and j and k are friends, the tripleijk is transitive if and only if i and k are friends. Transitivity is violated(i.e., the triple is intransitive) if, given that i and j are friends and j andk are friends, i and k are not friends. Vacuously transitive triples, triplesthat do not meet the conditional requirement that i and j are friends andj and k are friends, are not considered in this analysis.

The formula for computing each cell in the matrix T of perceivedtransitivity scores was as follows: Tij = NTij/(NTij + NTij), where NTij

is the number of transitive triples that i perceives j to be involved in, andNIij is the number of intransitive triples that i perceives j to be involvedin. If NTij + NIij = 0 (i.e., if none of the triads perceived by i that include jmeet the preconditions of transitivity), then Tij was set equal to a missingvalue.

Actual Friendship NetworkTo measure actual friendship links (distinct from perceived friendshiplinks), we determined the locally aggregated structure (Krackhardt,1987a) as follows: A friendship tie existed between persons i and j onlyif person i claimed person j as a friend and person j agreed that person iclaimed person j as a friend. Thus, an actual friendship link from i to jwas defined as existing when both parties agreed that it existed.

Actual ReciprocityWe created a matrix for each site in which each column in the matrixindicated the proportion of person j’s dyadic relationships that were actu-ally reciprocated (as validated by both parties to the friendships). Thus,whereas person i might perceive that person j’s friendships were 40 per-cent reciprocated, if j’s friendships were in fact reported by j and j’s friend-ship partners to be 60 percent reciprocated, the column of scores for j in


the actual reciprocity matrix would consist of .6 repeated in each cell ofthe column.

Actual TransitivityWe created a matrix for each site in which each column in the matrix indi-cated the proportion of person j’s triadic relationships that were actuallytransitive. We first symmetrized the actual friendship matrix, using theintersection rule that for a reciprocated tie to exist between persons iand j, person i had to report a friendship tie from i to j and person jhad to report a friendship tie from j to i. Then, following the procedureoutlined in the computation for perceived transitivity, we calculated theactual proportion of triads involving person j that were transitive. Eachcolumn in the matrix indicated the extent to which person j was involvedin transitive triads.

Social DistanceWe measured the extent to which each person in the network perceivedhimself or herself to be distant from every other person. Thus, our measureof social distance was a perceptual measure of how close or far egoperceived alter to be from ego. In a graph, the path distance betweentwo points is the length of the shortest path (or geodesic) that connectsthem (Harary, 1969). As Feld and Grofman (1989) pointed out, whennetworks are represented as graphs, the path distance between any twopoints is a good proxy for social distance. We measured the shortest pathdistance between each pair of individuals i and j as perceived by i. Thus,if respondent Sam Berkowitz perceived that the shortest path connectinghim to Alan Hobbs consisted of four lines, the distance from Berkowitzto Hobbs was measured as 4.

We treated social distance within any individual’s cognitive map as asymmetric concept. That is, if ego perceived the distance from ego toalter as equal to some value x, then this implied that ego also perceivedthe distance from alter to ego as equal to x. Thus, we symmetrized ego’sperception of the friendship network before calculating social distance.For example, if ego perceived that ego was a friend of j and also perceivedthat j considered k a friend, then we deemed the social distance betweenego and k equal to 2. Conversely, if ego perceived that k considered j afriend and that j considered ego a friend, then this was also deemed to bea distance of 2 between ego and k.

In calculating the social distance measure used in predicting the degreeof reciprocity in ego’s perceptions, we symmetrized ego’s cognitive map ofthe network using the union rule: A friendship relationship existed if egoperceived either of the people to have a friendship tie to the other. In con-trast to the reciprocity analyses, the transitivity analyses were conducted


on matrices that had already been symmetrized using the intersectionrule: A friendship relation existed if ego perceived both of the people tohave a friendship tie to each other. To be consistent, we calculated thesocial distance measure used in predicting the degree of transitivity inego’s perceptions on the matrix symmetrized using the intersection rule.

For both the reciprocity and transitivity tests, we considered distancesof infinity (indicative of the absence of paths between the two nodes) tobe missing values. We also performed analyses with the infinite distancesrecoded as distance N (the number of nodes in the network), and althoughthe meta-analysis results were the same, the large distances tended to actas outliers, obscuring the true underlying relationships. In this chapter,we restricted our presentation of the data to those cases where distanceswere “real” (that is, the actors were mutually reachable) and not infinite.Distance squared was calculated in a straightforward manner but wasmean-centered (i.e., the mean was subtracted from all values) before theterm was squared. This reduced collinearity problems in the regressionbecause the correlation between a variable and its mean-centered squaredterm is 0, whereas the correlation between a variable and its (non-mean-centered) square can be high, resulting in unstable coefficient estimates.

DensityWe assessed the density of each respondent’s cognitive map as the numberof lines in the map divided by the maximum possible number of lines(Scott, 1991: 74). Some respondents perceived many friendship links,whereas other respondents perceived few links. We controlled for thisvariation in density across perceivers’ cognitive maps as follows: Foreach respondent, we calculated a number between 0 and 1 that indicatedthe proportion of all possible friendship links that were perceived to exist.

As with the distance measure, the density measure was based on thespecific matrix that predicted either transitivity or reciprocity. That is,densities in the models used to predict transitivity were based on thesymmetrized friendship networks from which the transitivities were cal-culated. Densities in the models used to predict reciprocity, on the otherhand, were based on the nonsymmetrized friendship networks from whichthe reciprocity proportions were calculated.

Data Analysis

Social network data are often not amenable to standard statistical tests,such as ordinary-least-squares analysis, because the observations cannotbe assumed to be independent. For example, in the current research, thetransitivity matrix for the Pac site includes thirty-three scores from eachperson in the sample. Each of the thirty-three scores within a row of this


matrix derives from the same source (the cognitive map of the respon-dent) and therefore exhibits systematic interdependence. To deal with thisproblem, we used the Multiple Regression Quadratic Assignment Proce-dure (MRQAP), which has been explained in detail in Chapter 3 and inprevious work (e.g., Kilduff and Krackhardt, 1994; Krackhardt, 1987b,1988).

To assess the extent to which the combined results of our analysesacross four sites offered support to the expected relationships betweenvariables, we performed a meta-analysis (Hedges and Olkin, 1985). Typ-ically, the null hypothesis for a meta-analysis is that all samples are drawnfrom populations in which there is no relationship between the variablesof interest (these variables being, in our case, distance, distance squared,and proportions of balance). If the meta-analysis shows that the nullhypothesis is rejected, the conclusion follows that for at least one of thesamples there is a significant relationship between the variables of interest.However, finding an effect for only one of the samples scarcely ranks as“persuasive evidence of the efficacy of a treatment” (Hedges and Olkin,1985: 45).

However, the studies from four sites being combined in our analysisuse exactly the same measurement instruments and replicate exactly thesame regression model. In such a case, we can test the likelihood that thedistance-squared coefficients that we are interested in are not significantlydifferent across the four samples, and whether we can therefore interpretthe combined p value to refer to a common population.

Hedges and Olkin (1985) described many situations for which sucha test can be performed using a Q statistic. However, the data that wecollected, with their autocorrelated structure requiring a nonparametricMRQAP analysis, falls outside the situations described. This is unfor-tunate because the problems of autocorrelated data are common in thestudy of social networks. The arguments of Hedges and Olkin can never-theless be extended to our case by using the information in the permutedvalues of the regression coefficients to replicate the weighted mean esti-mates of the population beta and the estimate of the standard error ofthe beta. Because this is the first time, to our knowledge, that such a Qtest for meta-analysis has been applied to social network models such asours, we describe here in detail the procedure that we used in calculatingand testing Q.

Notation is drawn from Hedges and Olkin (1985). Q is a test statisticthat compares the observed betas β i from each sample i with a weightedestimate of the population beta β+:

Q =k∑

i=1

[(βi − β+)2

σ2

(βi)

](1)


where k is the number of samples, and σ (βi)2 is the estimated variance of

each beta, which has been estimated by calculating the variance of thebetas generated across all 999 permuted values of the dependent variableunder the null hypothesis. The Q statistic is asymptotically distributed asa chi-square with (k − 1) degrees of freedom:

β+ =

k∑i=1

[βi

σ2(βi )

]k∑

i=1

[1

σ2(βi )

] .

The β+ parameter is estimated by calculating a weighted average acrossthe k sites, where each weight is inversely proportional to the variance ofβ i in each sample. For both the reciprocity and transitivity analyses, thefollowing computations had to be made for each site i: β i, σ (βi)

2, β i−β+,(β i−β+)2, (β i−β+)2/σ (βi)

2.To test for the possibility (suggested by a colleague) that in calculating

transitivity we might have created a positive correlation with distance(and distance squared), we randomly generated raw friendship data. Forone hundred samples, the regression coefficients for distance squaredin the prediction of reciprocity and transitivity were not significantlydifferent from 0, and thus the distance-squared analyses appear to beunbiased. A slight but significant negative bias was uncovered for thecoefficient for distance in the transitivity model. To ensure that the resultswere not affected by this bias, we recalculated the significance levels of thiscoefficient against a null hypothesis of the mean of the simulated samples,using the standard error generated by the simulations. The results ofthe meta-analysis become even stronger when we use this conservativetest. Thus, the results that we report in the following section cannot beattributed to an artificial bias introduced by the analytic procedure itself.

Results

Table 4.1 provides a summary of the variables used in this analysis. Theproportions of reciprocity and transitivity perceived by individuals infriendship networks differed slightly across the four sites, as shown inTable 4.2. The mean proportion of perceived reciprocity ranged from.39 to .49, whereas the mean proportion of perceived transitivity rangedfrom .20 to .30.

Tables 4.3 and 4.4 present information addressing the issue of whetherthe regression coefficients differed significantly across the four sites andwhether it is therefore acceptable to interpret meta-analysis results as



Definition of Each Cell inVariable Matrix

Dependent MatricesPerceived reciprocity Respondent i’s perception

of the proportion ofpairs including j thatwere reciprocated

Perceived transitivity Respondent i’s perceptionof the proportion oftriads including j thatwere transitive

Independent MatricesDistance Length of the shortest

path between i and j asperceived by i

Distance squared Mean-centered distancesquared

Control MatricesDensity Number of links in

respondent i’s cognitivemap divided by themaximum number oflinks possible

Actual reciprocity and transivity Using the rule that bothindividuals must agreethat each considers theother a friend before afriendship link isestablished, each cellcontains the actualproportion of pairs (ortrials) including j thatwere reciprocated (ortransitive)

referring to a common population. The Q statistics at the bottom ofTables 4.3 and 4.4 are nonsignificant, indicating that the regression coeffi-cients for distance and distance squared did not differ significantly acrossthe four sites in either the reciprocity or the transitivity analyses. Forexample, in Table 4.3, the overall β+ for distance squared was .016,which was calculated by summing the four entries for distance squaredin the third column of the table. This yielded a nonsignificant Q of 0.476(p = .924, df = 3). Thus we conclude that the four distance-squaredcoefficients in the reciprocity model were not significantly different fromone another. On the basis of nonsignificant Q statistics for distance and


Table 4.2. Proportions of PerceivedBalance in Friendship Networks

Site

HT Gov Sil PacVariable n = 21 n = 31 n = 33 n = 33

ReciprocityM .39 .44 .44 .49SD .35 .35 .36 .35TransitivityM .20 .30 .25 .24SD .26 .31 .27 .30

Notes: HT = high-tech managers; Gov = government office;Pac = Pacific Distributors; Sil = Silicon Systems.

Table 4.3. Summary of Q Analyses Determining Whether ReciprocityRegression Coefficients Differed across Four Samples

Individual Site Statistics

(βi − β+)2 All Four Sites Combined

Variable βi Variance of βi Variance of βi β+ Q p

Site 1Distance −0.0894209 0.0017081 0.3814442Distance squared 0.0111083 0.0003310 0.0790133




All Four Sites Combined (df = 3)Distance −0.0638959 3.127 .373Distance squared 0.0162220 0.476 .924

distance squared in Tables 4.3 and 4.4, we can assume that the results ofmeta-analyses on these coefficients refer to a common population.

Figure 4.3 presents the results of the reciprocity analyses for the com-bined data across all four sites and for each site individually. The overallgraph shows a distinct curvilinear shape, indicating a tendency for ego toperceive close and distant relations as more reciprocated than relations in

Table 4.4. Summary of Q Analyses Determining Whether TransitivityRegression Coefficients Differed across Four Samples

Individual Site Statistics

(βi − β+)2 All Four Sites Combined

Variable βi Variance of βi Variance of βi β+ Q p

Site 1Distance −0.l242139 0.0039149 0.8911572Distance squared 0.0640057 0.0016400 1.0929966



Site 4Distance −0.0293083 0.0012974 0.9900603Distance squared −0.0029419 0.0003810 1.5893159

All Four Sites Combined (df = 3)Distance −0.0651483 3.448 .328Distance squared 0.0216671 4.274 .233

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

01 2 3 4 5 6 7 8 9

Pac GovOverall

Sil

HT

Distance

Pro

port

ion

Figure 4.3. Proportions of perceived reciprocity of people’s relations atfour sites as a function of social distance from ego. HT = high-tech man-agers; Gov = government office; Pac = Pacific Distributors; Sil = SiliconSystems.


0.5

0.4

0.3

0.2

0.1

01 2 3 4 5 6 7

Pac

Gov

Overall

Sil

HT

Distance

Pro

port

ion

Figure 4.4. Proportions of perceived transitivity of people’s relations atfour sites as a function of social distance from ego. HT = high-tech man-agers; Gov = government office; Pac = Pacific Distributors; Sil = SiliconSystems.

the middle distance of about two to five links from ego. This overall U-shaped curve provides support for a composite model that includes boththe emotional tension model’s prediction (higher perceived reciprocityfor close relations) and the cognitive miser model’s prediction (higherperceived reciprocity for more distant relations).

As Figure 4.3 shows, the graphs reached their minima at values ofsocial distance ranging from 2.12 (at the Pac site) to 5.8 (at the HT site).These minima were within the range of the data that we collected: Themaximum values for social distance for each site were as follows: Pac = 5,Gov = 7, Sil = 9, and HT = 7.

Figure 4.4 presents the results of the transitivity analyses for the com-bined data across all four sites and for each site individually. Again, theoverall graph shows distinct U-shaped curvilinearity, supporting the pre-dictions of the composite model. The graph for the Pac site, however,differs from the graphs for the other sites, showing a rather linear down-ward slope. At this site, therefore, the proportion of relations perceived astransitive tended to decrease with increasing distance from ego. As Figure4.4 shows, for the three sites exhibiting positive curvilinear graphs, thegraphs reached their minima at values of social distance ranging from 2.8(at the HT site) to 3.5 (at the Sil site), all within range of the data that wecollected. The maximum values for social distance for each site were asfollows: HT = 5, Gov = 5, Sil = 7, and Pac = 7.


Table 4.5. Summary of Multiple RegressionsPredicting Proportions of Perceived Reciprocity inFriendship Networks

Variable B β p

Site 1: HT (n = 21)Distance −8.942 −26.659 .984Density −2.443 −0.370 .502Distance squared 1.111 8.600 .275Actual proportion 9.768 9.895 .170

Site 2: Gov (n = 31)Distance −7.775 −21.294 .990Density 46.564 12.767 .109Distance squared 2.359 15.605 .069Actual proportion −3.454 −2.514 .686

Site 3: Sil (n = 33)Distance −7.174 −25.013 .999Density −27.498 −3.883 .693Distance squared 1.457 16.381 .034Actual proportion 17.616 16.718 .033

Site 4: Pac (n = 33)Distance −0.580 −1.603 .542Density 21.742 3.736 .449Distance squared 2.493 9.808 .169Actual proportion 14.174 12.649 .067

Note: Coefficients were multiplied by 100 for ease of presentation. Infi-nite distances are deemed missing values. All tests are one-tailed; thus, allnegative coefficients have p values greater than .5 HT = high-tech man-agers; Gov = government office; Pac = Pacific Distributors; Sil = SiliconSystems.

Table 4.5 presents the results of regression analyses predicting pro-portions of perceived reciprocity at each of the four sites, whereas Ta-ble 4.6 presents the equivalent results for perceptions of transitivity. Theseresults provide the basis for the meta-analysis results summarized in Ta-ble 4.7, where data from all four sites are combined. Consistent with theinformation evident in Figures 4.3 and 4.4, the meta-analysis confirmsthat the overall data exhibited curvilinearity (as assessed by the posi-tive distance-squared term): The coefficients for overall distance squaredwere significant for both reciprocity (Z = 2.432, p = .008) and transitivity(Z = 2.301, p = .011).

We can conclude from the evidence in the graphs and in the statisticalanalyses that the data are probably best fit by a U-shaped curve. In otherwords, the evidence lends support to a composite model that combinesthe downward curving prediction of the emotional tension model (lessperceived reciprocity as ego looks beyond his or her immediate friendship


Table 4.6. Summary of Multiple RegressionsPredicting Proportions of Perceived Transitivity inFriendship Networks

Variable B β p

Site 1: HT (n = 21)Distance −12.421 −45.997 .973Density 281.260 36.541 .058Distance squared 6.401 39.184 .065Actual proportion 6.666 4.952 .357

Site 2: Gov (n = 31)Distance −11.346 −38.802 .994Density 203.022 47.681 .004Distance squared 5.114 31.453 .014Actual proportion 47.936 28.180 .012

Site 3: Sil (n = 33)Distance −4.821 −21.331 .929Density 152.721 17.578 .118Distance squared 1.643 17.709 .167Actual proportion 21.258 14.661 .114

Site 4: Pac (n = 33)Distance −2.931 −13.123 .768Density 262.219 35.824 .122Distance squared −0.294 −2.999 .530Actual proportion 12.161 7.965 .278

Note: Coefficients were multiplied by 100 for ease of presentation. Infinitedistances are deemed missing values. All tests are one-tailed; thus, all neg-ative coefficients have p values greater than .5. HT = high-tech managers;Gov = government office; Pac = Pacific Distributors; Sil = Silicon Systems.

circle) and the upward curving prediction of the cognitive miser model(more perceived reciprocity as ego’s gaze includes the friendship relationsbetween dyads relatively unfamiliar to ego).

In the statistical analyses reported in the tables, we controlled not justfor the linear effects of social distance and the density of ego’s networkbut also for the actual proportion of balance in specific friendship dyads.We were able to focus explicitly on the question of how social distanceaffected perceptions of balance in friendship relations while controllingfor the possibility that perceptions might, in fact, align with reality. Con-trolling for people’s tendency to perceive balance where the members ofthe friendship pair confirmed that it existed, we found evidence for acurvilinear effect of social distance on perceived balance. In other words,the analyses allow us to reject the possibility that curvilinearity in thedata derives not from perceptions but from the distribution of actuallyoccurring friendship reciprocity and transitivity.


Table 4.7. Summary of Meta-Analysis Results across FourSites Predicting Proportions of Perceived Balance in FriendshipNetworks

Reciprocity Transitivity

Variable p, one-tailed Z p, one-tailed Z

Site I: HT (n =21)Distance .984 −2.144 .973 −1.927Distance squared .275 0.598 .065 1.514

Site 2: Gov (n = 31)Distance .990 −2.326 .994 −2.512Distance squared .069 1.483 .014 2.197

Site 3: Sil (n = 33)Distance .999 −3.090 .929 −1.468Distance squared .034 1.825 .167 0.966

Site 4: Pac (n = 33)Distance .542 −0.105 .768 −0.732Distance squared .169 0.958 .530 −0.075

All Four Sites CombinedDistance .999 −3.833 .999 −3.320Distance squared .008 2.432 .011 2.301

Note: All tests are one-tailed; thus, all negative coefficients have p values greater than .5.HT = high-tech managers; Gov = government office; Pac = Pacific Distributors; Sil =Silicon Systems.

Discussion

The results of both the graphical and statistical analyses suggest thatindividuals tend to perceive both close and distant relations as balanced.We found support for a composite model that includes both an emo-tional tension effect (close relationships tend to be perceived as balanced)and a cognitive miser effect (distant relationships tend to be perceivedas balanced). The results of the meta-analyses were consistent for bothperceived reciprocity and perceived transitivity and suggest a unifyingperspective on how individuals cognitively structure their social worlds.

Previous research has shown that people tend to prefer balanced ratherthan imbalanced relations in both perceived networks (De Soto, 1960;Freeman, 1992) and behavioral networks (Davis, 1979). The currentresearch across work organizations suggests that this general preferencehas specific significant effects on ego’s perceptions of friendship rela-tions both close to and distant from ego. Close to ego, the motivationto perceive one’s own interpersonal world as balanced may be to avoidemotional upset. Far from ego, the motivation to perceive the relationsof relative strangers as balanced may be to fill in the blanks in social


structure. Thus, the explanation of balanced structures in social networkscan combine both hot (emotional) and cold (knowledge) approaches tosocial cognition.

One of the puzzles of the current research concerns the anomalousresult for perceptions of transitivity in site 4. The sample for this organi-zation differed from the samples from the other organizations in that wecollected data only from the managerial core at headquarters, not fromall the people at headquarters. The composite model that we present maywell apply only to bounded groups such as those in sites 1, 2, and 3.

Site 4 was also different in being a relatively large organization com-pared with the other three sites. In larger groups, the task of organizingthe relations among the large set of alters may be difficult. Extensiveprevious research (reviewed by Moreland and Levine, 1992) has shownthat most natural groups are quite small, averaging two to three mem-bers and rarely exceeding five or six members. People appear to havedifficulty coordinating social interactions that involve more than five per-sons (Despartes and Lemaine, 1986). The minimum for reciprocity variedmore than the minimum for transitivity across the different sites. Tran-sitivity, as an indicator of balance, may be difficult to assess beyond adistance of four because it involves organizing six directional ties ratherthan the two ties involved in reciprocated relations. The task of mappingtransitivity relations is relatively complex even in small organizationsonce ego’s gaze travels beyond his or her familiar acquaintances.

Our study differs from previous research in the way that we measuredsocial distance. Whereas in the present study we used a continuous mea-sure of distance, previous work (Kumbasar et al., 1994) dichotomizedalters into those who were at a distance of one from ego and those whowere at a distance greater than one from ego. The dichotomization ofdistance prevents discovery of a curvilinear relationship even if one existsin the data.

In the present research, we aimed to go beyond the laboratory to studythe effects of schema use in actual social settings, in keeping with callsfor more field-based studies of human cognition (e.g., Funder, 1987). Wetested our models in four quite different social arenas rather than restingcontent with the standard single setting common in social network stud-ies (e.g., Burkhardt and Brass, 1990; Kilduff, 1992; Krackhardt, 1990;Walker, 1985). A further strength of the current research is the inclu-sion in the statistical tests of variables derived from naturally formednetworks of friendships as well as perceptions of those networks. Thisinclusion allowed us to focus on how schemas shape perceptions whiletaking account of the possibility that reality shapes perceptions.

One of the strengths of the current research – data collected from actualsocial settings – is accompanied by a potential weakness. Because we


have not experimentally manipulated the causal factor (social distance)to which we attribute the findings, the possibility remains that our causallogic could be reversed. Thus, it is possible that people tend to eitherdraw close to or keep quite distant from those whose relationships theyperceive as balanced. In the current cross-sectional analyses, we are unableto track dynamic processes of this sort, and this suggests that future workcould explore the effects of social distance on perceived balance in morecontrolled settings.

The attempt to understand transitivity in triads has a long but ratherconfused history. One of the leading researchers in the field was led todeclare that “after a decade of matrix grinding, I have no more idea of whytriads are transitive than I did when I began” (Davis, 1979: 60). Whereasreciprocity is well established as a defining feature of human society(Gouldner, 1960) and is especially evident among adults in the world ofwork (Gouldner, 1973: 268), transitivity appears most prominently ingroups of junior high school students (Davis, 1979: 61). Transitivity, ofcourse, involves an ordering of relations among a three-person group,whereas reciprocity involves relations among only two people at a time.Ego has greater control over whether a friendship link from alter to egois reciprocated than whether two of ego’s friends decide to complete thethird link of a transitive triplet. In other words, it is easier for ego toimpose reciprocity relative to transitivity on perceptions of friendshiprelations (Doreian, Kapuscinski, Krackhardt, and Szczypula, 1996). Thecurrent research, in proposing that perceived transitivity is a function ofsocial distance, offers a parsimonious explanation for when transitivityin social organizations is likely to be found, an explanation that worksequally well for transitivity as for reciprocity.

The importance of reciprocity and transitivity as structural principlesof group organization has been widely recognized. Gouldner (1960) sug-gested that reciprocity functions to counter bureaucratic impersonalityand to maintain the division of labor in work organizations. He quotedSimmel to the effect that all contacts among people “rest on the schema ofgiving and returning the equivalence” (Gouldner, 1960: 162 162). Recentwork examining social relations across diverse studies has confirmed thepervasiveness of the reciprocity heuristic in perceptions of liking (Kennyet al., 1996; Kenny and DePaulo, 1993). The principle of transitivity hasbeen described as the “key structural concept in the analysis of socio-metric data” (Holland and Leinhardt, 1977: 49–50). The expectationthat friendships will be balanced may serve to stabilize but not rigidifyorganizational systems in which patterns of interaction are reproduceddaily.

The assessment of perceived friendship in our research is two-valued,consistent with Heider’s (1958) discussion: People are either friends or not


friends. Transitivity, therefore, can be understood as an instance of thelogical principle known as the multiplicative rule. This rule implies that,for example, the multiplication of two positives (i.e., friendship relationsbetween A and B and between A and C) results in a positive (i.e., afriendship relation between B and C). Because a two-valued approachto balance theory is perfectly compatible with traditional logic (as Insko,1999, and Runkel and Peizer, 1968, point out), a preference for perceivingtwo-valued relations as transitive can be understood as a preference forperceiving the world as a logical, ordered place. One potentially valuableextension of the current research would be to examine preferences forbalance using a many-valued rather than a two-valued approach (for twoexperimental investigations, see Tashakkori and Insko, 1979, 1981).

In focusing on perceived balance in work organizations, we are helpingto uncover the ways in which people structure the social worlds wherecareers are established and where much of the business of the modernworld is conducted. The current research suggests that if people in orga-nizations perceive unbalanced relations close to themselves, they will actto balance these relations either by changing relationships or by changingcognitions. Further, people tend to perceive friendship relations far fromego as balanced because of increased reliance on the balance schema toorganize perception. It is in the middle ground – the area around theminimum – that ego is likely to be troubled by persistent imbalance. Inthis middle area, ego has no power to act decisively to change relation-ships, and ego may know too much about the relations of these peopleon the margins of ego’s world to be able to organize their relations usingthe principles of balance. Future research, then, could focus on this areaaround the minimum as the site of ego’s perceived dissatisfactions andopportunities. Ego is likely to be unhappy at work to the extent that heor she perceives relations in the middle distance as unbalanced. However,unbalanced relationships represent structural holes to be bridged (see thediscussion in Burt, 1992). To the extent that ego perceives, for exam-ple, that groups of individuals who should be communicating with eachother are not doing so or are doing so ineffectively, ego may be able toseize the initiative to bridge the gap and bring the people together. AsWeick (1995) has emphasized, perceptions have a way of becoming real-ity. Thus, perceived gaps in communication and friendship patterns maylead to actual movements of people to bridge perceived gaps, irrespectiveof whether such gaps actually exist. In a striking example of the “grass isgreener” effect, people in organizations may perceive more opportunitiesfor entrepreneurial action just outside their own friendship circles.

The reproduction and transformation of structure in social systemsdepend in part on the systemwide effects of the positions (Brass, 1984)and roles (DiMaggio, 1991: 94) occupied by individuals. A sense-making


perspective emphasizing the perception of relations can help explain howindividuals structure the social worlds to which they belong. Individualperceptions of social structure are important because such perceptionsshape reputations across internal labor markets (Kilduff and Krackhardt,1994). The process we have described connects the intimate world offriends and acquaintances with the distant world of relative strangers. Theextent to which the perceiver finds that the periphery of the social worldresembles the proximate may enable individuals to anticipate familiarpatterns of interaction across social boundaries and structural holes. Theindividual, then, in extending a vision of a balanced world to the relationsbetween comparative strangers, may sustain a logic of confidence thatpromotes action across social divides.

In the next chapter, we tackle the question of accuracy: Given thebiases that we have documented, does it help the individual to perceiveaccurately the friendship and advice relations in an organization? Whatbenefits flow to those who see more clearly the connections among others?

5

Effects of Network Accuracy onIndividuals’ Perceived Power

In the previous two chapters, we showed that perceptions of social net-works matter and that such perceptions are systematically biased. Butsome people are more accurate than others in perceiving network pat-terns. If this is so, do these accurate people gain benefits in organizationalarenas of competition and power? This is the theme we investigate inthis chapter. We expand the discussion to include perceptions of bothfriendship and advice networks, and investigate whether an accurate per-ception of the political landscape – including who are the central players –predicts who has power in the organization.

How does one assess the political landscape in an organization? Oneway of addressing this question is to identify the key political actors inthe organization (Pfeffer, 1981). But simply identifying the most pow-erful actors may not give sufficient information to anticipate the overalldynamics of resistance and support for political acts. Additional ques-tions about these actors come to mind: Are these powerful actors orga-nized such that they tend to act in unison? Do they represent differentpolitical constituencies? Precisely whom does each have influence over?Beyond knowing who is powerful, it is useful to know how the powerfuland powerless are organized or structured (Bailey, 1969: 108).

One way to approach the answers to these deeper questions about thepolitical landscape is to study access to and the control of informationflow in the organization (Pettigrew, 1973). As far back as 1965, Hubbellderived both a measure of the power of individual actors and an identifi-cation of powerful coalitions, using the actors’ networks of ties. Laumannand Pappi (1976) documented how power accrued to those in central net-work positions in a community of elites. Brass (1984) discovered that cen-trality in work-related communication networks was a robust predictor ofpower in a printing company. As Pfeffer (1981: 130) stated: “Clearly, thepower that comes from information control . . . derives largely from one’sposition in both the formal and informal communication networks.”

84

Effects of Network Accuracy on Perceived Power 85

More to the point, the study reported in this chapter suggests thatpower accrues not only to those who occupy central network positionsin organizations but also to those who have an accurate perception ofthe network in which they are embedded. An individual who has anastute knowledge of where the network links are can have a substantialadvantage. First, this information provides a good assessment of who ispowerful in the organization, because the central actors in the networkcan be easily ascertained. Knowing who the central – and powerful –actors are in the organization is essential political knowledge. Second,this information can be used to identify where the coalitions are in anorganization. Knowing where the coalitions are, how large they are, andwhere their support comes from gives one an edge in anticipating resis-tance and in mobilizing support for action or change. Third, an accurateassessment of the network can also reveal the weaknesses in politicalgroups by exposing holes, gaps, and locations of lack of support for anyparticular coalition. Thus, understanding the network provides a sourceof power independent of centrality in the network.

The central point in this chapter is precisely that: Cognitive accuracyof the informal network is, in and of itself, a base of power. Both theconcepts of power and cognitive accuracy are further developed in thischapter. In addition, we will argue that these two concepts are embeddedin a structural context that must be taken into account in any empiricalexploration.

Power

There has been much disagreement as to the precise meaning of power.Some writers have referred to it as the ability to get things done despitethe will and resistance of others, the ability to “win” political fights,or a capacity to outmaneuver the opposition (Bierstadt, 1950; Emerson,1962). Others (e.g., Kanter, 1979; McClelland, 1975; Roberts, 1986)have stressed the positive sum nature of power, suggesting that it is theraw ability to mobilize resources to accomplish some end (without spe-cific reference to organized opposition). Still others refer to power as theability to control premises of actions, such that power becomes almostunobservable (Bachrach and Baratz, 1962; Lukes, 1974; Mizruchi, 1983).Salancik and Pfeffer (1977) preferred to ignore these distinctions, notingthat, while academics may quibble over the definition of power, thoseactually experiencing the effects of power in the real world seem to exhibita consensus as to who has it.

Without fully resolving this debate, it is reasonable to assume thatthe answer to the question of who has power depends in part on an


answer to the question, Power to do what? If the influence being soughtis within the routine operation of the organization, then people who are“experts,” people in “authority,” and, generally, people who know howthings work around the organization are likely to be seen as powerful. If,however, the influence entails a radical departure from prior operations,then the uncertainty that emerges is likely to arouse emotional responsesto influence attempts. Affect-laden issues such as trust, respect, or likingmay become important in evaluating who has the ability to mobilizesupport for the radical change (Chapter 10; Krackhardt and Stern, 1988).In such cases, the powerful person may be someone who has referentpower (French and Raven, 1959) or charisma (Bradley, 1987; Fiedlerand House, 1988; House, 1977) in the organization rather than someonewho simply has authority or expertise.

We included multiple kinds of power in this study (as recommended byPfeffer, 1981). The assumption is that some actors are powerful becausethey are acknowledged as adept at getting things done in the organiza-tion, despite some resistance (e.g., Brass, 1984) and that some actors areinfluential because of an ascribed individual trait that reflects intangiblequalities of trust and personal charm. These two different assessments ofpower are offered as ones that actors will readily recognize as influencebases in organizations: the ability to get things done in spite of resis-tance and the ability to influence people through personal appeal andmagnetism (which is termed charisma).

Cognitive Social Structure and Accuracy

The current study was motivated by the question, How closely does eachperson’s perception of the network approximate the “actual” networkand how does this relate to power? To address this question, two types ofaggregations were employed: The set of N individual perceived maps ofthe whole network, called “slices,” of Ri,j,k; and the “actual” network, asdefined by the two people actually involved in the relationship, referredto as the locally aggregated structure, or LAS (Krackhardt, 1987a).

Just as power itself is a multidimensional concept, network relation-ships may be assessed on several dimensions. But the specific questionis, What network relations are critical for the assessment of power? Forexample, a network composed of incidental communication links, suchas perfunctory “hello’s,” may not be as rich in power information as anetwork composed of critical advice relationships. The study reportedhere was based on the cognitive social structures for two different typesof networks that have been shown to be useful in understanding the


dynamics of informal organizations (e.g., Brass, 1984; Burt, 1982: 25;Krackhardt and Porter, 1985, 1986; Lincoln and Miller, 1979). First, theadvice network (who goes to whom for work-related advice) representsthe instrumental, workflow-based network in the organization. The sec-ond network assessed was the friendship network, or what Lincoln andMiller (1979: 186) called the “primary network,” which we have focusedon in the previous three chapters. The friendship network captures impor-tant affective and social bonds that can affect trust, especially in times ofchange (Chapter 10; Krackhardt and Stern, 1988).

Structural Influences

In pursuing issues of power, one cannot ignore critical contextual andstructural factors that also operate to give certain actors privilege andpower in an organization. Brass (1984) found that centrality in the infor-mal network itself predicts power. But centrality also has important the-oretical links to cognition (see Krackhardt, 1987a, for a comparison ofdifferent types of centrality). A series of studies has found that centralinvolvement in a social system increases one’s ability to “see” the socialsystem accurately (Freeman and Romney, 1987; Freeman et al., 1987).Freeman, Freeman, and Michaelson (1988) noted that “social intelli-gence,” the ability to discern social groups and boundaries, evolves overtime as participants gain experience in the social group. Freeman andRomney (1987) demonstrated that people’s ability to recall social struc-ture accurately was a function of whether they were members of thecore group or were peripheral, transitory members. These results, com-bined with Brass’s (1984) findings, suggest that centrality in the informalstructure can lead to both cognitive accuracy and power.

Another structural power base that cannot be ignored is the formalposition that a person holds in the organization. Clearly, those withmore authority will have more power, on the average, than those withless authority. In addition, those higher in the organizational chart areresponsible for a larger part of the organization. A first-line supervisoris responsible for the activities of his or her immediate subordinates. Amanager of several supervisors is responsible for these supervisors andultimately for the activities of all their subordinates. People’s positionsrequire them to pay attention to the way in which those under them worktogether and relate to each other. Thus, those higher up in the organiza-tion will have, by virtue of their position, a better opportunity to observeand take note of a larger part of the informal network. Consequently,they are likely to have a more accurate picture of the informal network.


STRUCTURE COGNITION POWER

Formal Power

CognitiveAccuracy

Reputational Power

Informal NetworkCentrality

Figure 5.1. Model relating structure, cognition. and power.

This should be particularly true in a small, entrepreneurial firm, wherethe owners-managers are known to be heavily involved in the details andday-to-day workings of the entire organization.

Those higher in the formal organization are forced to relate to a widerbase of people. A first-line supervisor must coordinate the activities ofa limited number of people, all of whom are likely to interact infor-mally with each other and be doing similar work. A top-level managermust coordinate the activities of supervisors and managers from differ-ent functions and sectors of the organization. This responsibility giveshigher-level managers more central positions in the formal organization,in that they will find themselves dealing with more issues that surfacebetween departments or groups. This formal role is likely, in turn, to leadto opportunities to be in the middle of the informal network, acting as abridge between groups of employees. Therefore, it is expected that formalhierarchical level will also contribute to network centrality.

There are thus both structural and cognitive power bases in an orga-nization. Although it is proposed here that cognitive accuracy is a powerbase in and of itself, one must take into account the fact that this cogni-tive power base is influenced by formal and informal structural factors.Because these structural factors are sources of power in their own right,these sources are explicitly included as part of the cognitive model ofpower presented here.

Figure 5.1 displays this model, which relates structure, cognition, andpower. Formal structure is shown as an exogenous variable leadingdirectly to informal structure, cognitive accuracy, and power. Informalstructure, in turn, contributes to cognitive accuracy and power. Finally,in accordance with the central theme of this chapter, cognitive accuracy


is predicted to contribute to power over and above the power alreadyexplained by the structural factors. This last link represents the mainproposition of this chapter:

Proposition: Controlling for formal and informal bases of power, cogni-tive accuracy of the informal network will be correlated with individualpower in the organization.

To test the model in Figure 5.1 and the proposition posed above, a net-work study was conducted of a small high-tech firm. Questionnaire andinterview data were collected from which the cognitive social structuresand “actual” networks were determined, and each employee providedassessments of how powerful and charismatic every employee was in theorganization. From these data, the central proposition and model weretested directly.

Method

The company studied and the sample were identical to that studied inChapter 3: thirty-six members of the high-tech company Silicon Systems,of whom thirty-three filled out our questionnaire.

Reputational Power

Previous work (Brass, 1984) established internal consistency and predic-tive validity for a reputational measure derived from ratings of supervisorsand peers. Building on this previous work, we asked each employee torate all the employees (including himself or herself) on the two dimen-sions of power: the ability to get things done despite resistance and theability to influence through personal magnetism (charisma). This proce-dure avoided the problem of availability bias, incomparable sources, anddichotomization. Moreover, with this multiple-source method, the inter-nal reliability of the two power scores can be estimated. Each person ratedeach other person on a seven-point Likert scale on both charisma and theability to get things done (potency). Two anchors were provided for eachscale: “Not at all charismatic” to “Highly charismatic” for charisma, and“Not at all powerful” to “Highly powerful” for potency.

To assess the reliability of the two measures, Cronbach’s alpha wascalculated for charisma and potency (for the formula, see Carmines andZeller, 1979: 44). Both charisma and potency had high reliability coef-ficients (Cronbach’s alpha = .96 and .99, respectively), demonstratingthat there was very high consensus in the organization on who was


influential on each of these dimensions. The correlation between the twopower indicators was .63, indicating considerable overlap between thetwo measures. For this reason, the two measures were combined into asingle dependent variable, overall power, using the factor scores from thefirst component of a principal components analysis of the two variables.

Formal Position

Although power derived from formal position may be ambiguous in somelarger organizations, this organizational base of power was quite clear inSilicon Systems. There were three distinct levels of formal authority. Atthe top level were the three owner-managers. Even though they took ondifferent responsibilities and had different titles, they were equal partnersand made all major company decisions jointly. The next level consistedof five managers, each of whom had supervisory responsibility over cer-tain operational features in the organization. The remaining twenty-eightemployees had no formal supervisory title or authority. Formal position,then, was scored as follows: Each of the three owners was given a for-mal position score of 3; the five managers were given a formal positionscore of 2; and the remaining twenty-eight employees were given a formalposition score of 1.

Cognitive Social Structure

The cognitive social structure is a three-dimensional array of linkages,Ri,j,k, among a set of N actors, where i is the sender of the relation,j is the receiver of the relation, and k is the perceiver of the relation.Using Krackhardt’s (1987a) methodology, a questionnaire was designedto assess the cognitive social structure of two relations in the organization:friendship and advice (see Chapter 3 for more details).

Actual Network

Although work in the area of recall of network relations has cast doubt onan informant’s ability to relate accurately to whom they actually talk onany given day (see Bernard, Killworth, Kronenfeld, and Sailer, 1984, fora review), Freeman et al. (1987) have shown that people are remarkablygood at recounting enduring patterns of relations that they have withothers. Thus, although people may not remember whom they talked totoday or this week, they can accurately tell you whom they are in thehabit of relating to over an extended period of time. Consistent withthese results, Brass (1984) found that the workflow network in his studyclosely corresponded to the network reported by respondents. Because it


is these enduring relational patterns that are of interest – as evidenced bythe wording in the questions – the locally aggregated structure, or LAS(Krackhardt, 1987a), was used as a proxy for the “actual” network. TheLAS is an aggregation defined by the local participants in the network. Itmimics the typical form in which network data are collected. (See Chap-ter 3 for technical details.)

Both i and j must agree that i goes to j for help and advice beforethe i → j link is recorded as existing in the “actual” advice network.Similarly, both i and j must agree that i considers j a friend before thei → j link is recorded as existing in the “actual” friendship network.Because the relationship is defined as existing when both parties agreethat it exists, this measure of the “actual” network is direct and hasobvious face validity. In the case of Silicon Systems, data were missing forthree of the thirty-six employees, so we adopted the following procedureto deal with these cases. If information concerning a link between twopeople was missing from one but not both parties, then the existence ornonexistence of a link is determined by the information provided by thereporting party. If information was missing from both parties, then a linkwas deemed to exist if five or more people in the network reported thatit existed.

Cognitive Accuracy

Each participant’s cognitive map of the network (representing each par-ticipant’s “perceived network”) was taken from the set of responses thathe or she selected on the network questionnaire. We then correlated cog-nitive maps of the network with the actual network to derive a measureof accuracy between perceived and actual networks for each participant.The measure that we used was the point correlation coefficient and isequal to the value obtained by computing a Pearson correlation coeffi-cient between the elements of the matrix representing each participant’scognition of the network and the elements of the matrix representingthe “actual” network. (See Gower and Legendre, 1986, for a review ofcorrespondence measures.)

Centrality

Of the many different ways to measure of centrality, betweenness is theone most closely aligned with the idea of power (see Freeman, 1979,for the formula). The individual who is in between other actors hasmore control over information flow from one sector of the network toanother. That person becomes a gatekeeper of information flow. More-over, betweenness is an indication of the nonredundancy of the source of


Table 5.1. Means, Standard Deviations, and Correlations amongVariables (N = 33)

Variable Mean Standard Deviation

Power 0 1.00Advice accuracy .406 .0624Friend accuracy .326 .0704Advice centrality 14.443 27.3039Friend centrality 17.014 25.1941Formal position 1.333 .6454

Correlation Structure

Variable Power 1 2 3 4

1. Advice accuracy .340∗2. Friend accuracy .146 .282∗3. Advice centrality .453∗∗ .210 .0314. Friend centrality .506∗∗ .172 .236 .2225. Formal position .656∗∗∗ .240 .041 .566∗∗∗ .161

∗ p < .05.∗∗ p < .01.∗∗∗ p < .001.

information. To the extent that a person is connected to different parts ofthe network and therefore has access to different, nonredundant sourcesof information, that person will have a wider variety of information at hisor her disposal. The higher the betweenness score of an actor, the greaterthe extent to which the actor serves as a conduit connecting others inthe network. Formally, betweenness centrality measures the frequencywith which an actor falls between other pairs of actors on the shortestor geodesic paths connecting them (Freeman, 1979: 221). Measures ofbetweenness centrality are difficult to interpret for nonsymmetric data.In preparing the matrices prior to the calculation of betweenness central-ity, the networks were symmetrized according to the rule that if eithermember of a pair nominated the other, the pair was considered to havea tie – reflecting the assumption that the presence of even an asymmetricrelationship represented an opportunity for exchange of information inboth directions.

Results

The means, standard deviations, and intercorrelations among all the vari-ables used in this study are presented in Table 5.1.

To test more completely the model in Figure 5.1, a set of hierarchicalregressions was performed on the dependent variable, overall power. The


Table 5.2. Hierarchical Regression Analysis of Reduced-Form Equationswith Reputational Power as Dependent Variablea

Equation

Independent Variables (1) (2) (3)

Formal position 1.107∗∗∗ (.210) .879∗∗∗ (.222) .782∗∗∗ (.210)Advice centrality .0015 (.005) −.0004 (.005)Friend centrality .161∗∗∗ (.0048) .0195∗∗∗ (.0049)Advice accuracy 5.091∗∗ (2.02)Friend accuracy −.559 (1.74)R2 .431 .597 .678

Hierarchical Test of Model�R2 .166 .082F 5.966 3.425df 2,29 2,27p .007 .047

a Standard errors are in parentheses.∗ p < .05∗∗ p < .01∗∗∗ p < .001.

results are presented in Table 5.2 as reduced-form equations (Cohenand Cohen, 1983: 361–6). Formal position explains 43 percent of thevariance in overall power. The two informal structure sources of power,centrality in the advice and friendship networks, add another 17 percentof explained variance (significant at the .007 level). Note, however, thatadvice centrality is not significant in the equation; only centrality in thefriendship network is significantly related to power when controlling forformal position (rho < .01). It appears, then, that any advantage that aperson has by being central in the routine advice network is attributableto his or her formal position of power in the organization.

In equation 3 of Table 5.2, cognitive accuracy in the advice and friend-ship networks explains an additional 8.2 percent variance (p <.047).Again, however, only one of the two added variables is significant: accu-racy on the advice network. Understanding the friendship network is notsignificantly related to one’s power reputation over and above being inthe center of the networks and having a position of formal authority.However, an understanding of the advice network is significantly relatedto one’s power reputation.

These results reveal an interesting juxtaposition of effects. Clearly,formal authority is correlated with reputed power, as expected, but thetwo networks relate in different ways to one’s power base. Centralityin the friendship network – not the advice network – is a key factor inreputed power; but it is cognitive accuracy of the advice network – and


not the friendship network – that adds a significant amount of explainedvariance to one’s power reputation.

A closer inspection of the simple relationships among the variablesin Table 5.2 provides a partial explanation for these findings. Advicecentrality and friendship centrality are both strong simple predictors ofpower (.45 and .51, respectively). But, although advice centrality is corre-lated with formal authority (.57), friendship centrality is not significantlyrelated (.16). Most of the variance in power explained by advice centralityis already explained by formal authority: Those central in the advice net-work are also those with higher authority. Because friendship centrality isnot related to formal authority, however, it provides a unique contribu-tion to power in the second step of the hierarchical regression. Knowledgeof the advice network does not significantly covary with formal authorityand therefore also provides a unique contribution to power in the thirdstep of the regression.

Discussion

The network analysis conducted on Silicon Systems confirms the majorproposition of this study: that an accurate picture of the informal networksignificantly correlates with power. But, the overall model presented inFigure 5.1, relating structural factors to cognition and power, receivedonly qualified support (see Table 5.1). As predicted, formal position is sig-nificantly related to power and advice centrality. Contrary to the model’spredictions, formal position does not significantly correlate with cogni-tive accuracy. Also, contrary to prior research (Freeman and Romney,1987), centrality was not directly related to cognitive accuracy. Becausethese simple correlations were not confirmed in Table 5.2, more elaboratetests of the path coefficients for Figure 5.1 were not necessary, beyondthe overall tests provided for by the hierarchical regressions reported inTable 5.2.

The question remains why the other relationships in Figure 5.1, whichform the theoretical building blocks for the basic proposition of this chap-ter, were not confirmed. One possible explanation for the lack of supportfor parts of the causal model may rest in the size of the firm. Because thefirm is small, people all know each other and are relatively better informedon each other’s relationships than they might be in a large organization.Thus, being in the center of the network or at the top of the formalhierarchy does not provide as strong an informational edge over others’vantage points. Perhaps in a larger organization, where many people arenot even aware of each other’s existence, these structural advantages may


prove more predictive of cognitive accuracy. Future research on this topiccould shed light on whether these results are generalizable to – or perhapseven enhanced in – larger firms.

Because the structural links to cognitive accuracy were not confirmed,the test of the major proposition of this study could have been reduced toa simple correlation between cognitive accuracy and reputational powerwithout controlling for formal or informal positional power. It was nec-essary, nonetheless, to present the analysis in full, controlling for thesetheoretically important sources of power. The amount of variance inpower explained by cognitive accuracy in the advice network (by itself inthe absence of other variables) was 11.6 percent (= .342; see the correla-tion in Table 5.1); the direct contribution of the two accuracy indicatorsin column 3 (Table 5.2) in the hierarchical analysis was 8.2 percent.This difference indicates that, as predicted by the theory, there is some,albeit small, spurious correlation due to structural effects. In the conser-vative approach taken here, the hierarchical test of the main propositionremoves this spuriousness.

The study showed that reputational power of the members of the firmwas significantly related to cognitive accuracy of the advice network, notthe friendship network. Perhaps this is an indication of the extent to whichpower surrounded those who were capable of handling relatively routineoperational problems. In answering questions about influence and power,employees were responding according to their experiences in their day-to-day lives in the organization. As mentioned previously, those peoplecentral in the advice network, the experts, are likely to derive power fromsuch routine situations. Had the organization faced a nonroutine situationsuch as a crisis, however, it is possible that an understanding of thefriendship network could have been more predictive of power in dealingwith the crisis. Dealing with crises does not require routine informationbut, rather, it requires trust (Chapter 10; Krackhardt and Stern, 1988).It is reasonable to speculate that understanding the friendship network,which better represents the trust relations in the organization, could provemore critical than understanding the advice network in such a nonroutinesituation. Of course, this is only speculation, because the data reportedhere do not involve anything but routine operations.

Caveats and Limitations

The theory presented at the beginning of the chapter argued that knowl-edge of the network is in its own right a base of power above and beyondthe power accrued through other formal and informal bases. This causalclaim leads directly to the prediction of association. However, as is always


the case in field studies such as this, one cannot infer the causal link fromthe data. There are three possible reasons for an association between twovariables, A and B: A leads to B, B leads to A, or there is a third variable(or set of variables) that leads to both A and B, in which case we say thatthe relationship is spurious. It is worth speculating about each of thesepossible reasons for the underlying association.

In the current study, the theoretical claim underlying the observed asso-ciation is that network knowledge leads to reputational power. But is itpossible that one’s reputation as a powerful person leads to a better under-standing of the social network? Perhaps, for instance, as one becomesreputed to have more power, one is fed differentially more social infor-mation. However, if this were the case, then it is likely that this differentialfocus of information would in turn lead to the actor becoming centralin the network, and partialling out network centrality would remove theassociation between reputational power and network knowledge.

A more serious concern is whether the observed relationship is spu-rious. Despite some statistical attempts to control for clear sources ofspuriousness, there are potentially an infinite number of variables thatare unaccounted for. For example, suppose that power reputation is anattribution based on the fact that certain people are closer to the action inthe organization. Suppose that being closer to the action also gives peoplecertain advantages in knowing the social network. Then one could arguethat the observed relationship between reputational power and networkknowledge is spurious. In part, one could also argue that being “closerto the action” is already controlled for by controlling for centrality inthe network; but then again, it may not control for all of it. We havecontrolled for the most obvious sources of spuriousness. But, clearly, onecannot conclude that all sources of spuriousness have been eliminated.However, the raw data from the analyses are available (Krackhardt, 1990)and we invite scholars to explore alternative models that might explainthe reported relationships.

Another important issue surrounds the use of the term “power” in thisstudy. The literature on power in organizations is extensive, and on atheoretical level, this study sheds light on only a small part of that liter-ature. The emphasis here is on the power induced through informationflows in the informal network. Dependencies and power in an organi-zation can emanate from many sources, not simply how information ispassed from one person to the next. For example, we have no infor-mation concerning workflow interdependencies (cf. Brass, 1984). Nordo we know who has the critical resources or who has control over them(cf. Pfeffer and Salancik, 1978). Nor do we have information on whois performing the important functions in the organization and the


exclusivity with which they perform them (see Dubin, 1957: 62, for thefirst explicit treatment of this two-component definition of “power”). Theuse of the term “power” in this study is relatively specialized and may notgeneralize to other conceptualizations or other contexts. Whether under-standing the social network has any bearing on whether an individual hascontrol over critical resources is an interesting question that must be leftto future research.

Second, there is a methods limitation, one discussed in full by Pfeffer(1981: 54–7). Using reputation to measure the relative power of an indi-vidual has potential biases. The measure assumes that the raters knowwho is powerful and that they are willing to tell the researcher hon-estly what they know. Despite these possible problems, Pfeffer (1981: 57)noted that when raters seem to agree on their power attributions, “thisconsensus and consistency in power ratings provides some evidence forat least a shared social definition of the distribution of power.” Given thehigh reliability scores of the components of the reputational measure inthis study, we share Pfeffer’s conclusion that that there is consensus inpower attributions.

Conclusion

This chapter demonstrates that knowledge of the relevant network isitself associated with reputational power, independent of other structuralbases of power. In particular, further work exploring the importance ofthe structure of different kinds of relations in organizations may provefruitful in understanding the dynamics of organizational behavior.

As Mintzberg (1983b: 1) put it, “Power is a major factor, one thatcannot be ignored by anyone interested in understanding how organi-zations work and end up doing what they do.” We have focused on aneglected aspect of organizational power, namely the power that derivesfrom accurate knowledge of the informal network of advice relations.Given all the attention paid to structural and resource bases of power, itis surprising that so few have investigated the power drawn from suchpolitical knowledge.

In the four chapters that comprise this part of the book, we havefocused on “bringing the individual back in” through systematic investi-gation of how perceptions of networks affect such outcomes as leadershipeffectiveness, individuals’ reputation, and individuals’ power. In the nextthree chapters that comprise the second part of the book, we extendthe current chapter’s introduction of an individual difference variable(charisma) in the context of a network study by looking systematically at


how a personality variable particularly relevant to the understanding ofnetwork structuring – self-monitoring – combines with network positionto help explain the differential uses of social networks. How do low andhigh self-monitors differ with respect to the social networks they createand are shaped by? This is one of the main questions we systematicallyinvestigate in the following chapters.

II

The Psychology of NetworkDifferences

6

Social Structure and Decision Makingin an MBA Cohort

The sample for the three related studies covered in this chapter consists ofindividuals enrolled in an elite master of business administration (MBA)program that functioned as one of the portals to management in corpo-rate America (Kilduff and Day, 1994). These managers-in-training madenetwork and social identity choices in a campus setting that imposed rel-atively few of the hierarchical constraints on interaction characteristic offormal organizations.

We examine how individuals’ networks are shaped by ethnicity andgender identifications and how individuals differentially respond to net-work influences in making complex decisions. In the first section of thechapter, we ask whether within-race and within-gender preferences canexplain patterns of network marginality for members of underrepresentedgroups or whether such marginality results from exclusionary pressuresfrom the majority. In the second perspective on this same MBA cohort,we examine the structure of social influence. Individuals, faced with theorganizational choice decision, tend to be influenced by others, accordingto theory. Are these others people perceived to be especially similar? Orare these others friends, or perhaps social rivals (i.e., occupants of thesame social position)? Following this examination of the social compari-son other, we provide a further analysis of this MBA cohort in examiningwhether some people more than others are likely to be influenced by net-work contacts in decision making. We push social network research inthe direction of incorporating personality, specifically the self-monitoringpersonality construct.

The student’s second year in a top MBA program is dominated by onequestion: Which organization should I join? The organizational choicedecision is the culmination of two years of social and academic training.These students are in transition between their previous careers as engi-neers, waitpersons, students, and so on, and their new careers as exec-utives. They spend two years constructing new identities for themselves

101

102 The Psychology of Network Differences

through continuous socialization by peers drawing on the culture of thebusiness school (cf. Van Maanen, 1983), preparing for the fateful choiceof an organization for which to work. During these two years, many stu-dents are relatively isolated from their families and previous social con-tacts. Further, the ambiguity of the organizational choice decision itselfin the absence of any clearcut scale on which to compare organizations,together with the importance of the decision in terms of future careersuccess, makes organizational choice an arena in which social compar-isons and pressures can be expected to operate. Indeed, social comparisonprocesses, concerning both academic and social prowess, are intense forMBAs at prestigious schools of business.

Study 1: Distinctiveness and Social Identity

What determines individuals’ identifications with others? Distinctivenesstheory (McGuire, 1984) suggests a parsimonious answer: People in asocial context tend to identify with others with whom they share charac-teristics that are relatively rare in that context. Thus, two African Ameri-cans in a crowd of whites will tend to notice and identify with each otherbecause of their common race; however, when in a group of other AfricanAmericans, the same two people are unlikely to notice or identify witheach other on the basis of race.

Drawing on distinctiveness theory, we predicted that, within the MBAprogram, members of numerically underrepresented groups, relative tothose in the majority, would exhibit a stronger tendency to identify withinthe group. But this prediction still left unanswered the question of whichof several possible underrepresented groups any particular individual willtend to identify with most strongly. For example, when is an AfricanAmerican woman more likely to feel strongly African American, andwhen is she more likely to feel strongly female? Distinctiveness theorysuggests that the relative rarity of a social category in a particular socialsetting will promote members’ use of that social category as a basis forsocial identification. In our MBA sample, members of racial minoritieswere numerically rarer than women. For racial minorities, we predictedthat race would be a stronger category for social identification than sex.However, for whites, the same reasoning suggested that sex, not race,would be a stronger category for social identification.

Similarly, we predicted that the salience of race relative to sex wouldhelp determine whether people more often chose same-sex or same-racefriends. Formally stated, the relative rarity of a social category in a par-ticular social setting will tend to promote members’ use of that socialcategory as a basis for friendship formation.

Social Structure and Decision Making in an MBA Cohort 103

Marginality

Members of underrepresented groups are likely to be less central in friend-ship networks than members of well-represented groups because of theformer’s tendency to select friends from the distinctive groups to whichthey belong rather than from the social network as a whole (see the discus-sion in Ibarra, 1993a). Previous theorizing has emphasized exclusionarypressures that tend to relegate underrepresented group members to themargins of social networks (Kanter, 1977). Thus, the structural marginal-ity of members of underrepresented groups may well be overdetermined;we predict that is due both to the friendship choices of underrepresentedgroup members and to exclusionary pressures and biases that focus onvisible demographic characteristics such as race and sex.

Method

SampleThe sample for this study consisted of a class of second-year MBA can-didates enrolled in a nationally ranked MBA program. Nonresidents ofthe United States were excluded from the sample (and from all question-naires) because the research design focused on the job choice process andincluded only those eligible to work in the United States. The average ageof the respondents was twenty-seven years. Of the 209 students sampled,181 (87 percent) completed mailed copies of the sociometric question-naire. Nonrespondents did not differ significantly from respondents withrespect to race or sex.

MeasuresFriendship and IdentityFriendship was measured by asking subjects to look carefully down alist of second-year MBAs and place checks next to the names of peo-ple they considered to be personal friends. To measure identity, subjectswere asked to look carefully down a list of second-year MBAs and placechecks next to the names of people they considered to be especially simi-lar to themselves. A pair of individuals was considered to be a friendshippair (or an identity pair) if at least one member of the pair nominatedthe other member as a friend (or as especially similar). Compared withthe various measures of the structural equivalence concept (e.g., Breiger,Boorman, and Arabie, 1975; Burt, 1976; White and Reitz, 1983) thatattempt to estimate who is structurally identical or equivalent to whom,the direct measure of identification with others is closer to the sub-jective perceptions emphasized by social comparison theory (Festinger,1954).


HomophilyIn measuring race and sex homophily in the identity and friendship net-works, we controlled for the relative availability of different groups (cf.Ibarra, 1992) because what may appear as a tendency on the part of,for example, women to form friendships with men may be attributableto the proportionally higher number of men in a group. The adjustedhomophily index, known as the point correlation coefficient (see Gowerand Legendre, 1986, for a review and Krackhardt, 1990, for the formula),ranged from −1 (indicative of extreme “heterophily”) to + 1 (indicativeof extreme homophily).

SexThis was coded as 1 for men and 0 for women.

RaceUsing photographs from the school directory and information from pub-licly available student resumes detailing membership in societies such asthe Black Students Association, two people independently coded respon-dents as either white, African American, Asian American, or Hispanic(these were standard categories used by the administration at this school).Agreement between the two coders was high (98 percent interrater agree-ment). Disputed cases were resolved through discussion and a search forfurther information in the resume book published by the school. For thehomophily and regression analyses, we dichotomized race as 0 for whitesand 1 for all others.

As a check on how reliably the coding reproduced individuals’ self-coding of race, our coding was compared with the official school recordson 113 individuals who had voluntarily reported their race. Only oneperson had been misclassified (our classification was white, but the self-classification was Hispanic). There was complete agreement between thetwo codings for all of those we had classified as minority group membersand for whom self-report records existed (seventeen people). We con-cluded that our coding of race reproduced self-ratings at an acceptabledegree of accuracy. The absence of questionnaire items concerning race orsex ensured that the questionnaire itself did not trigger salient categoriesfor reporting social identity or friendship.

Structural MarginalityThose on the margins have difficulty accessing the center of a networkeither through their own friends (direct ties) or through friends of friends(indirect ties). To capture both direct and indirect friendship ties, we usedan eigenvector measure (Borgatti et al., 2002) that computed centralityas the summed connections to others weighted by the centrality of thoseothers (see Bonacich, 1972, for the formula). Marginality was defined as


the converse of centrality: Those scoring low on centrality scored high onmarginality.

Because the eigenvector analysis program handled only symmetric data,for this analysis we symmetrized the friendship matrix, using the rulethat if either member of a pair nominated the other, then the pair wasa friendship pair. This operational definition preserved information onweak ties (cf. Granovetter, 1973) and produced the most robust indicatorof centrality as measured by the ratio of the largest eigenvalue to the nexthighest eigenvalue. To check whether the results were affected by thisdefinition of the friendship measure, we also symmetrized the matrixusing two alternate rules: (1) replace Xij and Xji by the minimum of (Xijor Xji) and (2) replace Xij and Xji by the average of (Xij or Xji). Thepattern of results remained the same.

MajorMost of the students in the sample had chosen one of two majors: finance(56 percent of the sample) or marketing (26 percent), with the remainingstudents (18 percent) choosing a number of other possible concentrations.Because we were interested in the core/periphery structure of the socialworld of the MBA students, we dichotomized choice of major to differ-entiate those students choosing popular majors (finance or marketing,coded as 1) from those choosing unpopular majors (coded as 0). Thisdichotomization resulted in a significant effect for the control variable inour analyses, whereas a coding representing all possible majors had nosignificant effects.

Analysis and Results

Missing data reduced the sample to 159 people, 76 percent of the originalpopulation. The final sample included ninety-five white men, forty-fourwhite women, ten racial minority men, and ten racial minority women.

The mean homophily values given in Table 6.1 show that, with avail-ability controlled for, individuals tended to identify with and form friend-ships with others of the same race. Similarly, individuals tended to identifywith and form friendships with others of the same sex. Individuals tendedto establish smaller identity networks than friendship networks, althoughthe two networks were significantly correlated (r = .28, p < .001).

Recall that we predicted that people would tend to identify with thosewith whom they shared a demographic characteristic that was relativelyrare. The mean homophily values in Table 6.2 provide support for thisidea. The results for the identity network presented in the top half ofTable 6.2 show that the tendency for minorities to identify within-groupwas significantly stronger (t = −2.03, df = 19.5, p < .05) than that of


Table 6.1. Descriptive Statistics and Correlationsa

Variable Mean S.D. 1 2 3 4 5 6 7

1. Majorb

2. Sexc .023. Raced −.094. Centrality 9.09 6.53 .21∗∗ .13 −.22∗∗∗

Race Homophily5. Friendship network 0.04 0.12 .15 −.00 .25∗∗∗ .16∗∗6. Identity network 0.04 0.11 .07 −.08 .32∗∗∗ .25∗∗ .47∗∗∗

Sex Homophily7. Friendship network 0.04 0.13 .12 −.13 −.19∗ .15 .07 .18∗8. Identity network 0.05 0.08 .03 −.27∗∗∗ −.06 .07 .05 .17∗ .34∗∗∗

Notes:a N = 159.b Finance and marketing = 1, other majors = 0.c Men = 1, women = 0.d Minorities = 1, whites = 0.∗ p < .05.∗∗ p < .01.∗∗∗ p < .001.

Table 6.2. Mean Homophily Values Showing Tendency to Choose PartnersSimilar to Selfa

Type of Homophily

Group n Sexb Racec t df

Identity NetworkWhites 139 0.06 (0.72) 0.02 (0.93) −3.99∗∗∗ 137Minorities 20 0.04 (0.68) 0.13 (0.36) 1.72∗ 18Men 105 0.04 (0.77) 0.03 (0.85) −0.96 103Women 54 0.09 (0.60) 0.06 (0.85) −1.45 52

Friendship NetworkWhites 139 0.05 (0.65) 0.02 (0.90) −2.62∗∗ 137Minorities 20 −0.06 (0.40) 0.16 (0.27) 4.63∗∗∗ 18Men 105 0.02 (0.69) 0.03 (0.86) 0.46 103Women 54 0.06 (0.41) 0.05 (0.82) −0.68 52

Notes:a Unadjusted homophily values are in parentheses.b Men = 1, women = 0.c Minorities = 1, whites = 0.∗ p < .05.∗∗ p < .01.∗∗∗ p < .001.

whites. Similarly, the tendency for women to identify within-group wassignificantly stronger (t = 3.82, df = 157, p < .001) than that of men.Further, the paired comparison t-tests in the first two rows of Table 6.2show that, as predicted, whites were significantly more likely to identify


Table 6.3. Summary of Regression Analyses PredictingHomophilya

Type of Homophily

Variable Sexb Racec

Identity NetworkMajor 0.01 (0.02) −0.01 (0.02)Sex −0.05∗∗∗ (0.01) −0.02 (0.02)Race −0.02 (0.02) 0.10∗∗∗ (0.02)

Model F 5.56∗∗∗ 6.80∗∗∗R2 0.10 0.11Friendship Network

Major 0.03 (0.03) 0.04 (0.02)Sex −0.05∗∗∗ (0.02) −0.01 (0.01)Race −0.11∗∗∗ (0.03) 0.14∗∗∗ (0.02)

Model F 6.34∗∗∗ 10.98∗∗∗R2 0.11 0.18

Notes:a N = 159. Values represent unstandardized coefficients; standard errors are in parentheses.b Men = 1, women = 0.c Minorities = 1, whites = 0.∗ p < .05.∗∗ p < .01.∗∗∗ p < .001.

with others on the basis of sex rather than race (t = −3.99, df = 137, p <

.001), whereas minorities were significantly more likely to identify withothers on the basis of race rather than sex (t = 1.72, df = 18, p < .05).

The patterns of friendship choices paralleled these results. Looking atthe bottom half of Table 6.2, the tendency for members of minority groupsto make friends within-group was significantly stronger (t = −3.28,df = 20.3, p < .01) than that of whites. Similarly, the tendency for womento make friends within-group was significantly stronger (t = 2.11, df =137, p < .05) than that of men. The paired comparison t-tests in rows5 and 6 in Table 6.2 show that, as predicted, whites were significantlymore likely to make friends with others on the basis of sex rather than race(t = −2.62, df = 137, p < .01), whereas minorities were significantly morelikely to make friends with others on the basis of race rather than sex(t = 4.63, df = 18, p < .001).

The results presented in Table 6.3 confirm that these univariate effectsof race and sex on the tendency to make in-group network choicesremained significant when control variables were introduced into theanalyses. The regression analysis results presented in the first column(labeled “Sex”) show that the tendency to identify with and make friendswith members of one’s own sex was stronger for women than for men,


with an individual’s race and choice of major controlled for. The regres-sion results under the second column (labeled “Race”) show that minori-ties were more likely than whites to identify and make friends within-group, with sex and choice of major controlled for.

We further suggested that members of underrepresented groups werelikely to be structurally marginal in the friendship network. This hypoth-esis was supported. Men were more central than women, and this dif-ference was significant (t = −2.02, df = 157, p < .05). Similarly, whiteswere more central than minorities, and this difference was also significant(t = 2.58, df = 157, p < .01).

The first regression model in Table 6.4 confirms that, with majorand sex controlled for, members of racial minorities tended to be moremarginal than whites (p < .05). This same model shows that womenwere only marginally less central than men (p < .10), with major andrace controlled for.

Going further, we suggested that the tendency to make in-group (i.e.,homophilous) friendship ties would be negatively related to the centralityof underrepresented group members and positively related to the central-ity of majority group members. The results of subsample analyses offeredsupport for this hypothesis. The subsample results presented in column 4of Table 6.4 show that sex homophily (the tendency to choose friends ofthe same sex) was positively associated with centrality for men. But theseparate analysis for women presented in column 5 showed no significanteffect for sex homophily. In an analysis not reported in the table, thepositive correlation between sex homophily and centrality for the malesubsample (r = .26, p < .01) was significantly higher (Z = 2.57, p < .05)than the negative correlation for the female subsample (r = −.17, ns).

Similarly, the subsample regression results shown in the last twocolumns of Table 6.4 show that race homophily (the tendency to choosesame-race friends) was positively associated with centrality for whites,but marginally negatively associated with centrality for minorities. Thecorrelation between race homophily and centrality for the white subsam-ple (r = .46, p < .05) was significantly higher (Z = 3.92, p < .001) thanthe same correlation for minorities (r = −.46, p < .05).

Finally, we suggested that visible demographic characteristics, such assex and race, would be negatively related to centrality for underrepre-sented group members and positively related to centrality for majoritygroup members. Model 2 in Table 6.4 shows that, with a marginallysignificant (p < .10) effect of sex homophily controlled for, sex had asignificant effect (p < .05) on centrality. This pattern of results suggeststhat women were less central in the friendship network not so muchbecause of their tendency to prefer woman friends, but more as a resultof their exclusion on the basis of gender. Model 3 in Table 6.4 shows

Tab

le6.

4.Su

mm

ary

ofR

egre

ssio

nA

naly

ses

Pre

dict

ing

Cen

tral

ity

inth

eFr

iend

ship

Net

wor

ka

Full

Sam

ple

Subs

ampl

es

Inde

pend

ent

Var

iabl

eM

odel

1M

odel

2M

odel

3M

enW

omen

Whi

tes

Min

orit

ies

Maj

orb

3.03

∗(1

.39)

3.07

∗(1

.40)

2.49

†(1

.37)

1.24

(1.8

4)5.

39∗∗

(1.9

7)1.

11(1

.47)

1.45

(2.6

8)Se

xho

mop

hily

7.13

†(3

.89)

12.2

4∗∗

(4.6

8)−7

.44

(6.9

7)Se

xc1.

86†

(1.0

7)2.

46∗

(1.0

8)R

ace

hom

ophi

ly15

.03∗

∗(4

.71)

32.7

7∗∗∗

(5.7

9)−1

2.72

†(6

.39)

Rac

ed−3

.33∗

(1.5

4)−5

.83∗

∗(1

.64)

Mod

elF

4.91

∗∗4.

44∗∗

7.48

∗∗4.

07∗

4.61

∗∗18

.50∗

∗2.

43R

20.

090.

080.

130.

070.

150.

210.

22

Not

es:

aV

alue

sre

pres

ent

unst

anda

rdiz

edco

effic

ient

s;st

anda

rder

rors

are

inpa

rent

hese

s.Fo

rth

efu

llsa

mpl

e,N

=15

9.Fo

rth

esu

bsam

ples

,n’

sar

eas

follo

ws:

men

,10

5;w

omen

,54;

whi

tes,

139;

min

orit

ies,

20.

bFi

nanc

ean

dm

arke

ting

=1,

othe

rm

ajor

s=

0.c

Men

=1,

wom

en=

0.d

Min

orit

ies

=1,

whi

tes

=0.

†p

<.1

0.∗

p<

.05.

∗∗p

<.0

1.∗∗

∗p

<.0

01.

109


that, with a significant effect (p < .01) of race homophily controlled for,race had a significant effect (p < .01) on centrality. These results suggestthat the marginality of members of racial minorities was due both to racehomophily and to exclusion on the basis of race. One caveat is in order:The significance of the race variable in model 3 (and similarly, of the sexvariable in model 2) indicates only that an individual’s race (or sex) tendsto contribute to the individual’s centrality. These results do not allow usto say that race (or sex) was used as a basis for friendship exclusion bymajority group members more than it was by underrepresented groupmembers.

To examine the structural network positions of whites and minoritiesin greater detail, we used multidimensional scaling (MDS) (Krackhardt,Blythe, and McGrath, 1994) on the unsymmetrized 159-by-159 friend-ship matrix. Figure 6.1 shows that the center of the network was occu-pied exclusively by whites, with a cluster of African Americans located inthe upper right of the graph and other racial minority members locatedaround the periphery. The MDS analysis depicted in Figure 6.2 showsjust the friendship patterns among racial minorities. African Americans(represented by ovals surrounding “Bill”) formed a relatively tight friend-ship group, with many links between members. However, the membersof other racial groups depended less on cohesive links among themselvesthan on the network-spanning activities of particular individuals. Forexample, the African American “Fay” represented a link to the Hispaniccommunity, and the Hispanic “Jen” linked the African Americans, theHispanics, and the Asian Americans.

Study 1 Discussion

The results show consistent support for a distinctiveness approach tothe patterning of social networks. The lower the relative proportion ofgroup members in a social context, the higher the likelihood of within-group identification and friendship. Previous homophily research (e.g.,Tuma and Hallinan, 1979) has shown that people tend to interact withsimilar others. Our research refined this general proposition by suggestingthat perceptions of similarity are based on distinctiveness within specificcontexts. Further, in extending distinctiveness theory from the realm ofidentity relations to the realm of friendship relations, we have shown howthis approach can help explain patterns of structural marginalization inorganizations.

The marginalization of racial minority members in the friendship net-work appeared to result both from exclusionary pressures and from thepreferences of the minorities for same-race friends. The marginalization

Figu

re6.

1.Fr

iend

ship

rela

tion

sam

ong

indi

vidu

als.

a,b

aL

ette

rsin

dica

tera

ceof

indi

vidu

al:W

=w

hite

,A=

Asi

anA

mer

ican

,B=

Afr

ican

Am

eric

an,a

ndH

=H

ispa

nic.

bT

opr

eser

vevi

sual

clar

ity,

som

ew

hite

sne

arth

ece

nter

ofth

eso

ciog

ram

are

not

show

nhe

re.

111

Figu

re6.

2.Fr

iend

ship

rela

tion

sam

ong

min

orit

ies.

a

aA

fric

anA

mer

ican

s’na

mes

are

encl

osed

inov

als,

Asi

anA

mer

ican

s’in

rect

angl

es,

and

His

pani

cs’

indi

amon

ds.N

ames

are

sex-

spec

ific.

112


of women in the friendship network appeared to result more from exclu-sionary pressures than from women’s preferences for woman friends.

Previous research has shown that the lower hierarchical rank of womenand minorities in many organizations exacerbates the difficulties theyencounter in integrating themselves into informal networks of influen-tial others (e.g., Ibarra, 1992). The results that we report from a samplelacking a formal hierarchy suggest that segregation in informal networksmay persist even in “delayered” organizational forms. Our results provideinsights into the social networks likely to emerge in educational organi-zations – specifically, competitive MBA programs in which a relativelylarge cohort of would-be executives are socialized together. Similar pat-terns may emerge in training programs for cohorts of new recruits in workorganizations. In addition, to the extent that people depend on friend-ships formed in MBA programs for job referrals and support throughouttheir careers, patterns established in these programs may have long-lastingeffects. Our research raises practical questions concerning whether net-work patterns formed in graduate or company training programs havelasting effects on interaction patterns in work settings.

To the extent that people belong to multiple groups, they have multi-ple bases of similarity on which to build bridges of social identificationand friendship. Simmel (1955: 125–95) discussed this issue. Our studydemonstrates that the relative rarity of a group in a social context is likelyto promote members’ use of that group as a basis for shared identityand social interaction. All people, at some point in their organizationalcareers, are likely to be members of underrepresented groups, whetherthis involves race, gender, working in a foreign country as an expatriate,or simply joining a cross-functional team composed mainly of those withdifferent expertise. From this perspective, organizations offer rich envi-ronments for identity development based on the shared characteristicsindividuals can discover. The discovery and promotion of shared bases ofidentification may be one of the most challenging tasks of management.

Study 2: Friends’, Rivals’, and Similars’ Influenceson Decision Making

We have shown that friendship patterns and identity patterns are formedaround ethnicity and gender, and that these patterns are particularlycohesive for members of underrepresented groups. But do these patternsof friendship and identity affect outcomes, such as decision making? Thisis the question we explore in study 2, using the same cohort of MBAstudents.


Decision-making research has been generally silent concerning socialinfluences on choices. Both the normative models, such as expectedutility theory (e.g., Becker, 1976), and the descriptive models, such asprospect theory (Kahneman and Tversky, 1979), consider individual deci-sion makers in splendid isolation from the force field of influences thatsurround them. Although studies of social influences on organizationalchoice are rare (but see Higgins, 2001), we do know that people gen-erally acquire information about job vacancies through their informalnetworks of friends, family, and acquaintances rather than through offi-cial sources such as advertisements or employment offices (Granovetter,1974; Reynolds, 1951; Schwab, 1982; Schwab, Reines, and Aldag, 1987:135–8). It would seem likely, therefore, that people rely on these samenetworks for help in evaluating potential employers. The present researchuses social comparison theory as a framework to study the effects of socialnetworks on the organizational choice process. According to Festinger’s(1954) formulation of social comparison theory, (1) human beings learnabout themselves by comparing themselves to others, (2) people choosesimilar others with whom to compare, and (3) social comparisons willhave strong effects when no objective nonsocial basis of comparison isavailable and when the opinion is very important to the individual (seeGoethals and Darley, 1987, for a review of social comparison research).

Sources of Social Information

Friends. Much research has focused on social influence processes betweenfriends and acquaintances (e.g., Coleman, Katz, and Menzel, 1966; Fes-tinger et al., 1950; Krackhardt and Porter, 1985; Newcomb, Koenig,Flacks, and Warwick, 1967). From a social comparison perspective,friends are readily available as comparison others. People are hypoth-esized to shape their opinions and decisions through direct discussionwith these important members of their social circle.

Structurally equivalent others. A perspective that disputes the relevanceof friendship to social comparison has focused on comparisons amongpeople who occupy similar positions in the social network (e.g., Lorrainand White, 1971). These individuals are competing with each other tomaintain and enhance their social positions. They are, therefore, keento adopt attitudes and behaviors that they see their rivals using success-fully. Those who are structurally equivalent in the network are hypothe-sized to “put themselves in one another’s roles as they form an opinion”(Burt, 1983: 272). Influence proceeds through symbolic communicationbetween these equivalent actors whether or not they interact with eachother (Burt, 1987; Knoke and Kuklinski, 1982: 60).


Similar others. Some structural equivalence research explicitly assumesthat individuals identified by the researcher as equivalent perceive eachother as such. As Burt has asserted: “[An actor’s] evaluation is affected byother actors to the extent that he perceives them to be socially similar tohimself” (1982: 178). This assumption has proved necessary in order toexplain how structurally equivalent individuals in a social system influ-ence each other. Interpersonal influence is hard to explain if individualsinteract neither personally nor cognitively.

To transform objective stimuli into subjective perceptions, Burt usesStevens’ (1957, 1962) law of psychophysics. Perceived similarity is cal-culated as a power function of objective similarity. But as Krackhardt(1987a: 112) has pointed out, the use of a simple translation formulato generate subjective social perceptions from so-called objective stimuliis questionable. The implication is that “those studying cognitive net-works should . . . measure perceptions of networks directly” (Krackhardt,1987a: 113).

Thus, there are three main questions that this research tries to answer.First, did pairs of friends tend to bid for interviews with the same orga-nizations? Second, did pairs who perceived each other as similar tend tobid for interviews with the same organizations? Finally, did those pairswith similar patterns of friendships tend to bid for interviews with thesame organizations?

Method

The sample consisted of the same class of 209 second-year MBA studentsas in the previous study. Of the 181 people who completed questionnaires,11 people either did not participate in the bidding or were excludedbecause they were foreign exchange students. Both questionnaire andbehavioral data were available for a total of 170 people (81 percent ofthe original sample).

The MBA Bidding ProcessOrganizational choice in the present study was measured in terms of thoseorganizations students tried to interview with over the five-month recruit-ing period. The business school used a computerized bidding system underwhich each student could spend a total of 1,300 points bidding for inter-views with the 119 organizations that recruited at the school. In general,those students who made the highest bids for particular interview slotswere automatically selected. The bidding data were sensitive to studentpreferences over a five-month period, and the collection of the data wasunobtrusive. Thus it was possible to monitor the behavioral preferences


of subjects and to compare, for example, the degree of bidding overlapbetween pairs of friends compared to pairs of non-friends.

MeasuresIndependsent VariablesFriendship and perceived similarity. As in the previous study, a pair ofindividuals was considered to be a friendship pair (or a similar pair) if atleast one member of the pair nominated the other member as a friend (oras a similar individual).

Structural equivalence. Structural equivalence was measured as the sim-ilarity in patterns of relations with other individuals in the friendship net-work. Thus, a pair of individuals who had exactly the same ties to otherindividuals (even though they had no ties to each other) would have ascore of zero, indicating no difference in their structural positions. A pairof individuals who had very different ties to other actors in the systemwould have a large difference score. The difference scores were calculatedas continuous measures in a Euclidean social space, where the distancebetween any two actors equals the square root of the sum of squared dif-ferences across all third actors (for the formula, see Knoke and Kuklinski,1982: 61).

Control VariablesFor each individual, it was possible to construct a vector of job preferencescomposed of zeros and ones, indicating, for each of sixteen job categories,whether or not the student had shown a preference for that type of job.The preference data were collected by the Career Services Center prior tothe recruiting season. A job preference correlation matrix was created bycalculating the Pearson correlation coefficients between the vectors of allpairs of individuals. This matrix contained information on how similarpairs of individuals were with respect to their job preferences.

To create a matrix that would show which pairs of individuals had thesame majors, we derived a list of seven categories of MBA majors fromthe academic concentrations claimed on student resumes. As reportedin study 1, MBA students overwhelmingly chose two majors: finance(chosen by 56 percent) and marketing (chosen by 26 percent). In the fewcases where it was impossible to identify an academic concentration fromthe available evidence, the students were categorized as miscellaneous.

For each student, then, it was possible to allocate a number from 1to 7 indicating the focus of his or her studies. We then created a majorssimilarity matrix; the 170–by-170 matrix consists of cell entries of 1indicating that two individuals had the same major and cell entries of0 indicating that the two had different majors. We used this matrix tocontrol for similarity in academic concentration in the regression analyses.



Variables Source of Data Operationalization

DependentOverlapping choices of

organizationsComputerized bidding

systemFor each pair, the correlation between

bidding patterns across 119organizations

IndependentFriendship Friendship questionnaire For each pair, whether either person

claimed the other as a friendStructural equivalence Friendship questionnaire The Euclidean distance between each pairPerceived similarity Perceived similarity

questionnaireFor each pair, whether either person

claimed the other as similarControl

Overlapping jobpreferences

MBA resume book For each pair, the correlation acrosschoices of 16 jobs

Overlapping majors MBA resume book For each pair, whether the individualschose the same of seven majors

Dependent VariableThe dependent variable was similarity in bidding behavior. Each indi-vidual could bid for interviews with 119 organizations. Thus, for eachindividual, it was possible to construct a bidding vector, 119 cells long,that showed for each organization whether or not a bid had been made.A bidding correlation matrix was constructed by correlating these bid-ding vectors for all pairs of individuals. The bidding correlation matrix,like the sociometric choice matrices, was 170 by 170 and consisted ofPearson correlation coefficients. These coefficients indicated how similarin their bidding behavior each pair of individuals had been. For example,a coefficient of .45 in cell (123, 81) indicated that the bids of persons 123and 81 were correlated at the .45 level. Table 6.5 provides an overviewof the variables and their measures.

Analyses

The basic analysis can be expressed as a multiple regression equationwith five regressors and one dependent variable, where Y is the biddingcorrelation matrix, Pref the preference correlation matrix, Maj the majorssimilarity matrix, Euc the Euclidean distance matrix, Sim the perceivedsimilarity matrix, Fr the friendship matrix, and ε the matrix of errorterms:

Y = β0 + β1(Pref) + β2(Maj) + β3(Euc) + β4(Sim) + β5(Fr) + ε.

Ordinary-least-squares (OLS) analysis is not appropriate for these databecause the error terms are autocorrelated within rows and columns. The


Table 6.6. Summary Statistics for Number of Friends, Numberof Similar Others, and Number of Bids

Maxa Median Mean Range SD

Similars 169 3 4.46 40 5.21Friends 169 13 17.49 70 14.50Bids 119 19 19.66 54 9.72

Note:a Maximum number of friends, similars, or bids that could be chosen.

OLS procedure requires that the observations be independent, whereasobservations concerning all possible pairs in a social network exhibitsystematic dependence. A solution to the problem of how to test thesignificance of the βs in a multiple regression equation when the data arestructurally autocorrelated has been demonstrated by Krackhardt (1988)in terms of the Multiple Regression Quadratic Assignment Procedure(MRQAP) and is followed here. (See also Baker and Hubert, 1981; Hubertand Golledge, 1981; Hubert and Schultz, 1976; Krackhardt, 1987b.)

Results

The descriptive statistics in Table 6.6 show that the social networksamong the MBAs were quite sparse. The median number of friends cho-sen was thirteen, compared to a median of three chosen as especiallysimilar, with each student choosing from 169 possible names. Table 6.6also shows that the mean number of organizations bid for was 19.66(out of 119 available). The mean number of successful bids (those thatresulted in interviews) was 16. The 84 percent success rate indicates thatthe bidding system was not characterized by cutthroat competition. Therewas little apparent incentive, in fact, for friends to collude in spreadingtheir bids among different organizations.

Table 6.7 shows that, comparing mean correlations, people’s job pref-erences were over twice as similar (r = .181) as their bidding behavior(r = .076). The base rate for bidding similarity was, then, quite low.In fact, the median bidding correlation between pairs of individuals wasonly .042.

The bivariate correlations in Table 6.8 indicate preliminary support forthe prediction that friends (relative to non-friends) and similars (relativeto non-similars) would tend to bid for the same interviews. Contraryto the expectations, however, those individuals who were structurallyequivalent – that is, friends with the same other people – were no morealike in their patterns of bids than those individuals who were friendswith different people (Z = −0.08, ns). Table 6.8 also reveals that the


Table 6.7. Means and Standard Deviations of Variables

Variable Min Max Mean S.D.

Biddinga −0.259 0.789 0.076 0.166Equivalenceb 2.449 13.491 8.730 1.668Friendshipc 0 1 0.146 0.353Similarityc 0 1 0.047 0.211Majorsc 0 1 0.385 0.487Preferencesa −0.540 1 0.181 0.310

Notes:Results are based on 28,730 dyadic observations.a Pearson correlations.b Euclidean distance scores.c Each observation was either 0 or 1.

Table 6.8. Significance of Zero-Order Correlations amongthe Variables

Variable 1 2 3 4 5 6

1. Biddingr – −.01 .11 .10 .35 .40Z – −.08 11.59∗∗ 13.05∗∗ 23.43∗∗ 32.08∗∗

2. Equivalencer – .07 .00 .01 .00Z – 2.90∗ −0.18 0.34 0.31

3. Friendshipr – .33 .08 .11Z – 33.24∗∗ 4.67∗∗ 7.40∗∗

4. Similarityr – .04 .07Z – 2.93∗ 6.91∗∗

5. Majorsr – .48Z – 22.05∗∗

6. Preferencesr –Z –

Notes:The Z scores were calculated by means of the Quadratic Assignment Procedure (QAP)and are based on 28,730 dyadic observations.∗ p < .005 (two-tailed).∗∗ p < .0001 (two-tailed).

control variables – similarity of job preferences and overlapping majors –were the most powerful predictors of bidding similarity.

The first model in the Table 6.9 summary of regression results confirmsthe importance of the control variables – those who had either similar jobpreferences or similar majors tended to bid for the same organizations.


Table 6.9. Five Multiple Regression Models Predicting Bidding Similarity

Model

Independent Variable 1 2 3 4 5

Equivalenceβ −0.002 −0.002Z −0.856 −1.026

Friendshipβ 0.030 0.021Z 5.787∗ 4.407∗

Similarityβ 0.059 0.047Z 8.503∗ 6.675∗

Majorsβ 0.072 0.072 0.071 0.072 0.072Z 11.371∗ 11.420∗ 11.286∗ 11.348∗ 11.360∗

Preferencesβ 0.160 0.157 0.157 0.160 0.156Z 19.119∗ 18.895∗ 18.888∗ 19.094∗ 18.722∗

Notes:The Z scores were calculated by means of the Quadratic Assignment Procedure (QAP).∗ p < .0001 (two-tailed).

The Z scores indicate significant relationships between bidding similarityand both similarity of preferences and similarity of majors (p < .000l).

The second model in Table 6.9 shows that, relative to non-similars,those who perceived each other as similar were significantly more likelyto bid for interviews with the same organizations, even controlling forsimilarities in preferences and majors (Z = 8.503, p < .000l). Friends, too,had similar patterns of bids, relative to non-friends, even when the controlvariables were included in the analysis (Z = 5.787, p < .000l), as model 3shows. Model 4, however, indicates that structural equivalence failed topredict bidding similarity when the control variables were included in theregression (Z = −0.856, ns), replicating the finding from the bivariateanalysis. Apparently, pairs who had similar patterns of friendship tieswere not significantly more similar in their choices of organizations thanpairs who had different patterns of friendship ties.

Model 5 shows that friendship and perceived similarity had indepen-dent main effects on bidding similarity. This was true despite the fact thatfriendship choices and choices of similar others were significantly inter-correlated. Model 5 in Table 6.9 also shows that structural equivalenceremained insignificant in the full model (Z = −1.026, ns).

The analyses indicated, as expected, that people with similar job prefer-ences and similar academic concentrations tended to choose to interviewwith the same organizations. More interesting were the results showing


that individuals who perceived each other as similar members of the MBAcohort, or who perceived each other as friends, tried to interview withthe same organizations. These results remained significant even control-ling for the powerful effects of overlapping job preferences and academicconcentrations. Finally, structural equivalence, which is often used asa measure of perceived similarity, failed to predict bidding similarity,although the effect was in the expected direction.

Study 2 Discussion

One of the surprises of this research was the failure of structural equiva-lence to predict patterns of homogeneity in bidding behavior. Early claimsconcerning the superiority of structural equivalence over explanationsbased on friendship and acquaintanceship (Burt, 1987) have been heavilydiscounted given reanalyses of the same data (Kilduff and Oh, 2006). Thegeneral argument made in favor of structural equivalence has been thatthe perception of similarity is crucial to the diffusion of influence, andthat perceived similarity may have little to do with direct social interac-tion. This general argument is certainly supported by the present research,but the most effective measure of perceived similarity was based on thedirect perceptions of the individual rather than on structural equivalentestimates of relative positions in the friendship network.

The relative superiority of perceived similarity and friendship overstructural equivalence must be kept in perspective. The best predictors ofbidding homogeneity were not the sociometric variables, but job prefer-ence similarity and similarity of majors. For the first time in the empiricalliterature on organizational choice (see Schwab et al., 1987, for a review),it was possible to focus explicitly on choices of organizations, controllingfor the confounding effects of job preferences and academic specializa-tion. The social influence independent variables operated on the marginsof choices, perhaps determining which organizations would be selectedby students for bids from the set of those offering the particular jobs thatinterested the students.

This rather peripheral role of social influences should not, however,lead us to dismiss them as unimportant. This research has examined socialinfluences only at the very end of a long decision process. The same kind ofanalysis could be applied to the actual choices of job preferences and MBAmajors. These previous choices restrict the extent to which later choicescan be influenced by social or other forces. But these previous choicesmay themselves have been influenced by friends, rivals, and family.

The present research differs from previous work (e.g., Granovetter,1974) that has found strong effects of social networks on the transmis-sion of job vacancy information in imperfect labor markets. The MBAs


in the present research made choices in a relatively perfect market char-acterized by full advance information concerning vacancies. Informationconcerning the characteristics of recruiting organizations was also widelydisseminated by means of special company presentations on campus andthrough the Career Services Center. The market appeared to work quiteeffectively, with about 70 percent of the students actually obtaining jobsthrough the school-organized system, receiving an average of three joboffers each.

In a relatively perfect market, therefore, with an abundance of infor-mation, social networks may help to create and validate choice criteria.Previous research has shown that friends do indeed mutually influenceevaluative criteria (Duck, 1973) and the way these criteria are used inorganizations (Krackhardt and Kilduff, 1990). The opinions of strangers,however, concerning trivial choices are unlikely to influence behavior(Kilduff and Regan, 1988). As social comparison theory would predict,only important and ambiguous decisions, such as organizational choice,motivate individuals to seek comparative information from peers.

Organizational choice was selected by Soelberg (1967) as an exampleof the kind of nonroutine decision making that is so little understoodand yet which “forms the basis for allocating billions of dollars worthof resources in our economy every year” (Soelberg, 1967: 20). Study 2suggests that in order to understand how individuals make such complexdecisions, it is essential to study their interactions in the social systems towhich they belong. What study 2 leaves unanswered, however, is whethersome people, more than others, are susceptible to social network influenceor whether network influence is such a strong force that it overwhelmsindividual personality differences.

Study 3: Does Self-Monitoring Moderate NetworkInfluences on Decision Making?

Missing from social network studies has been any discussion of dispo-sitional differences. In moving from the analysis of attributes towardrelation-centered analysis, network researchers appear to have lost sightof individual variability and its potential effects on the strength of rela-tions. Social comparison research in general has long been criticized forits neglect of individual differences (e.g., Radloff, 1968: 945). At its mostradical, the network view has suggested that the study of individuals is“a dead end” (Mayhew, 1980: 335) that leads to platitudes rather thanscientific explanations of human behavior.

Both the network approach and the dispositional approach can be usedto understand decision making. The network perspective emphasizes that


systems of relations influence decisions through such processes as cohe-sion among friends, social comparison among those who regard eachother as similar, or rivalry among occupants of equivalent social posi-tions (see study 2 for a discussion). The personality perspective empha-sizes that relatively stable underlying dispositions influence decisions bypredisposing individuals to prefer some outcomes rather than others. Forexample, the choice of a career may be partly determined by the degree ofcompatibility between the individual’s personality and the requirementsof the vocation (Holland, 1985).

Because of these different emphases, the two approaches are character-ized by different research strategies. The network approach tends to takesnapshots of decisions across the network at one moment in time (e.g.,Walker, 1985), whereas the dispositional approach is generating moreand more longitudinal research tracking individuals over the life course(e.g., Burns and Seligman, 1989; Gerhart, 1987; Staw, Bell, and Clausen,1986; Staw and Ross, 1985).

In summary, the two approaches offer opposing perspectives both interms of variables (relations in networks versus individual dispositions)and research design (snapshot across the network versus history acrosstime). In bringing the two approaches together, we may find it possible topose new questions that have not previously been considered. For exam-ple, are there systematic differences between individuals in the degree towhich they rely on friendship networks when making important deci-sions? To answer this question requires an examination of the effects offriendship networks on decision making for different personality types.

Self-Monitoring

The self-monitoring construct (Snyder, 1974, 1979; Snyder and Ganges-tad, 1986) distinguishes between those who are especially attuned to therole expectations of other people (high self-monitors) and those who insiston being themselves despite social expectations (low self-monitors). Thebasic idea is that compared with high self-monitors, “low self-monitorsrely less on social cues to direct behavior and more on introspection”(Caldwell and O’Reilly, 1982b: 125). High self-monitoring individuals“are more likely than low self-monitoring individuals to seek out rele-vant social-comparison information” (Snyder and Cantor, 1980: 223).

Low self-monitors (identified by their low scores on the Self-MonitoringScale) “are controlled from within by their affective states and attitudes”(Snyder, 1979: 89). High self-monitors (identified by their high scores onthe Self-Monitoring Scale) use cues from others as guidelines for monitor-ing (that is, regulating and controlling) their verbal and nonverbal self-presentation. Whereas high self-monitors are “highly responsive to social


and interpersonal cues of situationally appropriate performances,” thebehavior of low self-monitors reflects “their own enduring and momen-tary inner states” (Snyder and Gangestad, 1986: 125). High self-monitorsin a social situation ask the following: “Who does this situation want meto be and how can I be that person?”. By contrast, low self-monitors askthis: “Who am I and how can I be me in this situation?” (Snyder, 1979).

In summary, high self-monitors, relative to lows, are more likely toshape their behavior in accordance with cues supplied by the social circlesto which they belong. For the present research, which focuses on thechoices of MBAs at a prestigious school of management located in arelatively isolated rural area, the relevant social circle was taken to be thenetwork of personal friendships within the MBA cohort itself. Thus, wepredict that the organizational choices of high self-monitors will be morehighly correlated with the organizational choices of their friends.

Research has also suggested that high and low self-monitors differ in theevaluative criteria they bring to the choice process. Particularly relevantto the present research is the evidence that high self-monitors choose onthe basis of socially defined realities, whereas low self-monitors chooseon the basis of intrinsic quality. For example, whereas high self-monitorschoose products on the basis of the image they project, low self-monitorschoose on the basis of the products’ quality (Snyder and DeBono, 1985).

Adapting this research to the organizational choice process, one mightsuspect that compared with low self-monitors, high self-monitors wouldbe more interested in the reputation, public image, and prestige of orga-nizations. The image of the organization, like the image projected aboutconsumption goods in advertising, should have particular salience forhigh self-monitors. The organization that one joins becomes an integralpart of one’s self-image, and high self-monitors are very concerned withthe “images of self that they project in social situations” (Snyder andDeBono, 1985: 588). Furthermore, high self-monitors prefer job situa-tions that offer clearly defined roles (Snyder and Gangestad, 1982). It isas if high self-monitors wish to place themselves in situations that havestrong social norms.

By contrast, low self-monitors prefer situations that allow them thefreedom to be themselves (Snyder and Gangestad, 1982). Relative tohigh self-monitors, they will perhaps evaluate organizations on howmuch autonomy in work procedures is encouraged. Furthermore, onecan expect that low self-monitors will choose work compatible with thevalues and beliefs that are central to their self-identity.

On the basis of this discussion, we predict that high and low self-monitors will place different values on certain factors relevant to orga-nizational choice. The six factors listed in Table 6.10 were selected todiscriminate between the value systems of high and low self-monitors.


Table 6.10. Factors Important to Organizational Choice

Factor Source

Individual Freedom1. Freedom from pressures to conform

both on and off the job.Vroom, 1966.

3. The opportunity to determine my ownwork methods and procedures.

Lawler, Kuleck, Rhode, and Sorensen, 1975.

5. Work that is compatible with mypersonal values and beliefs.

Snyder and Gangestad, 1982.

Social Conformity2. Work that is of high status and prestige. Vroom, 1966; Lawler et al., 1975.4. A clear idea of exactly what my role in

the organization will be.Snyder and Gangestad, 1982.

6. The organization’s reputation andpublic image.

Pieters, Hundert, and Beer, 1968.

Factors 1, 3, and 5 can be categorized as individual freedom factors,likely to appeal to low self-monitors. These factors consist of freedomfrom pressures to conform both on and off the job, the opportunity todetermine one’s own work methods and procedures, and work that iscompatible with one’s personal values and beliefs. Factors 2, 4, and 6in Table 6.10 can be categorized as social conformity factors, likely toappeal to high self-monitors. These factors consist of work that is ofhigh status and prestige, a clear idea of exactly what one’s role in theorganization will be, and the organization’s reputation and public image.

Method

SampleThe sample consisted of the same 170 MBA students surveyed in study2.

Independent VariableFriendship, as measured in study 2, was the independent variable.

Moderating Variable: Self-MonitoringThis was measured on the questionnaire with the revised eighteen-itemtrue-false version of the Self-Monitoring Scale (Snyder and Gangestad,1986). In the present research, the scale’s reliability as measured by Cron-bach’s (1951) alpha was .75.

Dependent VariablesBidding SimilarityThis is the pairwise similarity in bidding behavior (as in study 2).


Factors Important to Organizational ChoiceOn the questionnaire, students were asked to rank the six factors inTable 6.10 in order of importance for their choices of organizations.

Analyses

Analyses were made using Quadratic Assignment Procedure (QAP) andRandom Assignment Procedure (RAP). We also employed nonparametrictests of ranked factors.

Quadratic Assignment ProcedureAs with study 2, we used QAP to provide a Z score of the significance ofthe relation between two matrices. To measure the strength of the cor-relations between the matrices, Goodman and Kruskal’s (1963) gammawas calculated, a nonparametric correlation coefficient for skewed andbinary data such as is contained in the friendship matrix (Hubert andSchultz, 1976).

Random Assignment ProcedureOur prediction concerning the moderating effects of self-monitoring onfriends’ influence was not amenable to analysis by the QAP procedurebecause the separate matrices for those high and low on the personalitymeasures were not square. For example, the friendship matrix for highself-monitors was rectangular because the subsample of highs had chosenfriends from the full sample of both lows and highs. That is, 70 highself-monitors selected friends from 170 high and low self-monitors.

To determine whether the observed difference in correlations betweenlow and high self-monitors could have resulted from chance alone, a dis-tribution of ten thousand possible correlation differences was generatedby randomly allocating people to the low and high self-monitoring cate-gories. In the randomly reordered matrices, the first one hundred peoplewere treated as if they were low self-monitors, whereas the last sev-enty people were treated as high self-monitors. The correlations betweenfriendship patterns and bidding patterns were calculated separately forthe lows and the highs, and the difference between the correlations wasstored in a file.

The random assignment of people to low and high categories wasrepeated ten thousand times to create a distribution of possible differencesin correlations. This distribution was then examined to see how manytimes the observed difference in correlations had been generated by chancealone. For example, if the observed difference in correlations was equaledor exceeded in 250 of 10,000 trials, then the observed difference wouldbe significant at the .025 level (one-tailed test).


Nonparametric Tests of Ranked FactorsTo test whether there were significant differences in the way high andlow self-monitors ranked the factors in Table 6.10, the ranks attributedto each set of factors by each individual were summed. The first set offactors (1, 3, and 5) were predicted to be preferred by low self-monitors.Each individual’s rank scores for factors 1, 3, and 5 were summed, givinga total that could range from 6 (i.e., 3 + 2 + 1) to 15 (i.e., 4 + 5 + 6).Similarly, each individual’s rank scores for factors 2, 4, and 6 (predictedto be preferred by high self-monitors) were summed. The mean rankscore for each set of factors was calculated for low and for high self-monitors. To test whether these means were significantly different fromeach other, a nonparametric analysis featuring the Kruskal-Wallis (1952)test (chi-square approximation) was performed.

Results

We know from the previous study, that friends, relative to non-friends,tended in general to bid for the same interviews. But was there a person-ality effect? We predicted that self-monitoring would moderate the signif-icant correlation between friendship ties and bidding similarity, and thisprediction was supported. The correlation between friendship and bid-ding similarity was higher for high (γ = .19) than for low self-monitors(γ = .13). To determine whether this difference was significant, RAP wasused to output a distribution of ten thousand possible differences betweenthe two groups. The observed difference was equaled or exceeded in 404of 10,000 trials. The result, then, was significant at the .04 level (one-tailed test).

Table 6.11 provides averaged information on a range of indicators,including the number of job interviews obtained, the number of bidsmade, and the number of friends. There were no significant differencesbetween low and high self-monitors on any of these indicators. Further-more, individuals’ self-monitoring scores were not significantly correlatedwith the number of friends (r = .11, p = .l7, ns).

High and low self-monitors may therefore adopt different interviewstrategies, but do they also tend to evaluate organizations differently? Wepredicted significant differences between low and high self-monitors withregard to the criteria by which they chose organizations. And, indeed,compared with low self-monitors, high self-monitors ranked more highlyall three factors in Table 6.10 concerned with social conformity. Con-versely, low self-monitors, compared with highs, ranked more highly allthree factors concerned with freedom from social pressures.

The difference between low and high self-monitors on the overall rank-ing of the social conformity and individual freedom factors was significant


Table 6.11. Summary Statistics for Number of Interviews, Numberof Bids, and Number of Friends by Self-Monitoring Category

No. ofInterviews No. of Bids No. of Friends

Median No.Group M SD M SD of Friends M SD

Low 15.77 7.22 19.39 9.12 12 16.20 12.59self-monitor

High 16.19 7.71 20.04 10.58 13 19.33 16.79self-monitor

Note: For the low self-monitor group, n = 100; for the high self-monitor group, n = 70.

(p < .02). The summed ranked scores for the three individual freedomfactors differed by approximately one scale point (8.95 for the lows,9.94 for the highs), as did the summed ranked scores for the three socialconformity factors (12.05 for the lows, 11.06 for the highs). It is, how-ever, interesting to note that the individual freedom factors were moreappealing than the social conformity factors to the sample in general.That is, although relative to the low self-monitors the high self-monitorsranked the individual freedom factors as less important, both highs andlows agreed that individual freedom was a more important criterion thansocial conformity in choosing an organization for which to work.

The results of the tests on self-monitoring can be summarized as fol-lows: Compared with those who tended to rely on their own counsel (lowself-monitors), those who were more sensitive to social information (highself-monitors) were more like their friends in their choices of employmentinterviews with organizations. Looking now at attitudes toward organi-zations, the results confirmed that low and high self-monitors differedwith regard to their rankings of criteria by which potential employersmight be judged.

Study 3 Discussion

In partitioning the friendship network by self-monitoring, study 3 posedthe following question: Do some people, more than others, tend to relyon their friends when making complex decisions? The results confirmedthat personality types hypothesized to differ in their preferences for socialcomparison information did differ significantly, both with respect to howmuch their decision patterns resembled those of their friends and withrespect to the criteria they used in the decision-making process.

Previous research suggested that people rely on social information inmaking ambiguous decisions (Pfeffer, Salancik, and Leblebici, 1976) andwhen information is scarce (Granovetter, 1974; Kunreuther, 1978). The


possibility that social influences on decision outcomes may depend notonly on situational factors but also on dispositional factors has beenneglected, despite a long history of research suggesting that individualsdiffer with regard to their susceptibility to social influence (McGuire,1968).

The reported differences between personality types in the degree towhich people’s choices overlapped with those of their friends weresmall but significant. In this relatively perfect labor market, differencesamong people in the degree to which they relied on social networks foractual information concerning job opportunities were minimized. Futureresearch could determine whether the relative differences between per-sonality types reported in this chapter would be accentuated in an envi-ronment in which information was scarce rather than abundant.

One of the goals of future research is to answer this question:Does reliance on the social network improve decision making? Previousresearch has suggested that people tend to rely on the advice of similarothers rather than on the best information available (Brock, 1965) andthat the benefits of using social networks are matched by the costs (Rook,1984). Individuals appear to differ in their ability to derive support fromsocial ties (Riley and Eckenrode, 1986). For the individuals studied in thischapter, we know that high self-monitors, relative to low self-monitors,tended to gain faster promotions in managerial careers (Kilduff and Day,1994). But we don’t know whether the faster promotions were due toself-monitoring personality characteristics directly or whether the effectof self-monitoring on outcomes such as early promotions was mediated bysocial network position. The competing and possibly combinative effectsof self-monitoring and social network position are examined in the nexttwo chapters of this book.

The research reported in this study supports the conclusion that evenin conditions approaching those of perfect information and equal oppor-tunity, individuals differ systematically in the extent to which they relyon the social network in making decisions. Self-monitoring was a signif-icant moderator of social influence even in an environment overflowingwith relevant facts. In a world in which such an excess of informationis increasingly becoming a burden to be borne, the social network, as adecision-making resource, may be as much an expression of personalityas it is a constraint on choice.

General Discussion

In relying on unobtrusive measures of social influences (cf. Pfeffer et al.,1976) rather than on systematic observation of influence processes, the


present research raises the question of whether the overlapping patternsof behavior actually resulted from social influences. Surely, students hadan interest in keeping quiet about their interview choices to reduce thelikelihood that their friends would compete for scarce interview slots.

In fact, the interview market was not characterized by cutthroat com-petition: Eighty-four percent of bids resulted in interviews. Students aver-aged sixteen interviews and three job offers each. There was little appar-ent incentive, then, for friends to hide their preferences from each other.Indeed, the job preferences of all students were published by the CareerServices Center before the start of the recruiting season. People startedoff, then, with a basic knowledge of who was likely to be interested inthe various opportunities arising.

Participant observations throughout the five-month interview periodconfirmed that students constantly exchanged information concerningtheir interview preferences. In the corridors, where upcoming recruitmentschedules were posted, students discussed company pros and cons. Thesediscussions continued in the student lounge, in the computer lab, and evenin the bathrooms. To check our interpretation of the present findings,we conducted interviews with several members of the sample after thedata collection. The consistent theme revealed by these interviews wasof a highly interactive MBA cohort, the members of which studied andsocialized together almost exclusively. Within this social world, friendsnot only discussed their interview preferences, but also, on occasion,jointly targeted selected companies. For example, one group of studentsnot only decided to focus on specific investment banking firms but alsodecided to evaluate these firms on a consensually agree-upon criterion:the number of hours per week employees actually had to show their facein the office.

As social comparison theory would predict, the evidence points to anintense and continuing exchange of opinions between friends concerningan important set of decisions. What was not being exchanged was anyinformation concerning which companies were visiting or which kindsof vacancies were available. It was possible, then, to conduct a relativelypure test of differential social influence in a context in which everyonehad complete information concerning vacancies.

In the following chapter, we follow up the intriguing possibility thatlow and high self-monitors differ with respect to how they arrange andbenefit from social networks.

7

The Social Networks of Low and HighSelf-Monitors

One of the enduring questions we face as human beings concerns whysome people outcompete others in the race for life’s prizes. In work organi-zations, for example, why are some people better performers than others?One answer to this question is provided by research on the importanceof structural position. Within each specific work context, some individ-uals occupy more advantageous positions in social networks than otherindividuals. These positions allow access to people who are otherwisedisconnected from each other. The individuals who act as go-betweens,bridging the “structural holes” between disconnected others, facilitateresource flows and knowledge sharing across the organization. Their con-tributions to organizational functioning may lead to enhanced rewards,including faster promotions (Burt, 1992) and higher performance ratings.

Research on structural position has emphasized the importance of beingin the right place (Brass, 1984) but has neglected both the possibility thatthe network positions occupied by individuals might be influenced bytheir psychology and the possibility that personality and social networkposition might combine to influence important outcomes such as workperformance. The structural approach to organizational dynamics tendsto emphasize the structure of positions in social space (Blau, 1993; Pfef-fer, 1991) and avoids dependence on difficult-to-measure psychologicalproperties of actors (e.g., McPherson et al., 1992). Recent calls for moreinsight into the origins of network positions and the importance of indi-vidual characteristics (e.g., Emirbayer and Goodwin, 1994) prompt usto investigate why some individuals occupy structurally advantageouspositions and how individual differences in psychology and structuralposition combine to determine performance in organizational contexts.

The structuralist approach is not alone in disregarding the possibleeffects of individual characteristics on social structures. Despite a longhistory of psychological research suggesting that individuals differ withrespect to social influence (e.g., McGuire, 1968; Riley and Eckenrode,

131


1986), there has been relatively little work in psychology on how individ-ual differences affect the structures of the social worlds in which peoplelive and work. Rather than neglecting either the structure of the socialworld or the psychology of the individual, we investigate how individualsstrive within social structures that both enable and constrain action. Wefollow in the tradition of those who recognize the importance of under-standing the microfoundations of structural patterns (e.g., Granovetter,1973; Ibarra, 1993a; Uzzi, 1996).

Earlier work by social network pioneers included personality measures(e.g., Newcomb, 1961; Sampson, 1968) and interpersonal orientations(e.g., Breiger and Ennis, 1979; see also recent work by Janicik and Lar-rick, 2005). In bringing the individual back into social network analysis,we build on this previous work. Rather than treat individual attributesand social attributes as separate realms of inquiry, we seek to under-stand how the social networks that significantly affect the performanceof organizational participants are shaped by the attributes of interactingindividuals.

Theory

The Structural Advantage

Individuals may outperform their peers because of differences in the net-works to which they belong. Links to friends and work partners can pro-vide the assistance and social support necessary for high performance, butnot all network configurations are likely to be equally helpful. Forminga large network, for example, may be less important than acquiring astructurally advantageous position within a network (Burt, 1992).

Social actors who connect disconnected others tend to gain both infor-mation and control benefits. Information concerning projects, crises,resources, and other contingencies flow from a diversity of social actors tothe central actor whose ties link disconnected others. Actors whose socialties are limited to one clique are less likely to receive diverse informationthan are actors whose ties span cliques because information that circulateswithin a clique of highly connected workers is likely to be redundant. Evi-dence for the benefits of structural holes comes from both small-group andorganizational research (see the review in Burt, Jannotta, and Mahoney,1998). Small-group experiments showed that people with exclusive rela-tions to otherwise disconnected contacts tended to gain greater resources(Cook and Emerson, 1978; Cook, Emerson, Gillmore, and Yamagishi,1983). One organizational study examined the importance for nonsuper-visory personnel of occupying high-betweenness centrality positions –that is, positions that enable occupants to act as potential go-betweensfor those not connected with each other. Results showed that the higher

The Social Networks of Low and High Self-Monitors 133

the betweenness centrality in the informal communication network, thegreater the social influence and the higher the likelihood of promotion tosupervisor within the following three-year period (Brass, 1984).

Occupying a position between disconnected others is important notonly for nonsupervisory personnel but also for those in the managerialranks. A study of the individual networks and achievements of seniormanagers in a high-tech firm showed that nonredundant contacts todiverse clusters of others were related to early promotions (Burt, 1992).Similar findings emerged in another study of mobility among employeesof a high-technology firm: People with sparse social networks that tiedthem to unconnected others tended to have high mobility (Podolny andBaron, 1997).

The accumulating evidence suggests that individuals with ties acrosssocial divides gain nonredundant information concerning opportunitiesand resources. The ability to obtain resources such as information isdirectly related to individual and group performance (O’Reilly, 1977;O’Reilly and Roberts, 1977a, 1977b). Further, actors who connect dis-connected others can facilitate the flow of information across the wholesystem of coordinated activity that makes up the organization, therebycontributing to the accomplishment of organizationwide goals. Giventhis, when we discuss individual performance in this chapter, we referto the extent to which individuals contribute to organizational purposes,building on the work in organization theory that emphasizes that jobperformance consists of individuals contributing to the tasks specificto the organization (Burns and Stalker, 1994: 97). Previous work hasfocused on the effects of structural position on outcome variables such aspower and promotions but has offered little conclusive evidence concern-ing performance in organizations. One of the few studies that did examinework performance found that employees occupying central positions inthe workflow network were no more likely to be high performers thanemployees occupying less central positions (Brass, 1981). In contrast,research on officers and enlisted men in three high-technology militaryorganizations showed that people with two or more network contactsperformed better than people with one or no network contacts (Robertsand O’Reilly, 1979). This research did not examine the importance ofnetwork centrality or ties that link disconnected others. Given these sug-gestive but inconclusive findings, it is useful to examine directly whetherstructural position predicts workplace performance.

Self-Monitoring

Individuals in organizations may outperform their peers not only becauseof differences in the networks to which they belong but also because ofindividual differences in personality. Of the many personality variables


that could potentially affect performance, self-monitoring, a variable cen-trally concerned with individuals’ “active construction of public selvesto achieve social ends” (Gangestad and Snyder, 2000: 546), stands outfor three reasons. First, self-monitoring theory provides compelling argu-ments linking individual differences in self-monitoring with a range of joboutcomes, such as performance in the workplace, leadership emergence inworkgroups, conflict management, information management, impressionmanagement, and boundary spanning (Kilduff and Day, 1994; Snyder,1987: 88–90). Second, self-monitoring theory makes clear predictionsconcerning the effects of self-monitoring orientation on how individualsshape social worlds (Snyder, 1987: 59–84). And, third, as one leadingstructuralist has noted, the cutting edge of personality research of inter-est to social networkers may lie in approaches that recognize individualdifferences in predictable patterns of variability across situations, as self-monitoring does (White, 1992: 206).

According to self-monitoring theory, individuals differ in the extentto which they are willing and able to monitor and control their self-expressions in social situations. Some people resemble successful actorsor politicians in their ability to find the appropriate words and behaviorsfor a range of quite different social situations. With chameleonlike ease,they present the right image for the right audience. Other people, bycontrast, appear to take to heart the advice Polonius gave to Laertes inShakespeare’s Hamlet, “To thine own self be true”: They insist on beingthemselves, no matter how incongruent their self-expression may be withthe requirements of the social situation. Research on self-monitoring hasprovided important insights into individual differences in how individualspresent themselves in social contexts (see Gangestad and Snyder, 2000,for a review).

In a social situation, high self-monitors ask, “Who does this situationwant me to be and how can I be that person?” (Snyder, 1979). By contrast,low self-monitors ask, “Who am I and how can I be me in this situation?”(Kilduff and Day, 1994; Snyder, 1979). Self-monitoring theory, therefore,provides new insight into the age-old question of whether behavior is afunction of consistent dispositions or strong situational pressures. Froma self-monitoring perspective, some individuals (the low self-monitors)are consistent in demonstrating behavior that expresses inner feelings,attitudes, and beliefs. Other individuals (the high self-monitors) are con-sistent in adjusting behavior to the demands of different situations.

Because high self-monitors rely on social cues from others to guide theirbehaviors rather than on their own inner attitudes and emotions, highself-monitors are more likely than low self-monitors to resolve conflictsthrough collaboration and compromise (Baron, 1989). Further, high self-monitors tend to emerge as group leaders (Zaccaro, Foti, and Kenny,


1991), particularly in situations calling for high levels of verbal interaction(Garland and Beard, 1979) and in normative climates that support theemergence of leadership (Whitmore and Klimoski, 1984).

High self-monitors tend to emerge as leaders perhaps in part becausethey are more skilled at social interactions (Furnham and Capon, 1983).One study found that low self-monitors attended more to internal cuesto produce effective work, whereas high self-monitors attended to situ-ational cues, including the leadership behavior of supervisors (Andersonand Tolson, 1989). High self-monitors are more active in conversations(Ickes and Barnes, 1977) and tend to talk about the other person (andother people) instead of talking about themselves (Ickes, Reidhead, andPatterson, 1986). High self-monitors are better than low self-monitors atpacing conversations (Dabbs, Evans, Hooper, and Purvis, 1980), usinghumor (Turner, 1980), and reciprocating self-disclosures during acquain-tance processes (Shaffer, Smith, and Tomarelli, 1982). In a review of stud-ies of interpersonal strategies used by high and low self-monitors, Snyderwrote that the “lubricating” techniques employed by high self-monitors“would have warmed the heart of Dale Carnegie” (1987: 42).

The social skills and leadership abilities of high self-monitors, therefore,may enable them to perform significantly better than low self-monitors inthe modern workplace, where cooperation with others to achieve organi-zational purposes is the norm and where leadership emergence is encour-aged (see the review by Baron and Markman, 2000). Although there is noreason to suppose that self-monitoring orientation affects the proficiencywith which individuals perform technical duties, contextual activities,such as cooperating with others and following procedures even whenthey are personally inconvenient, are also a major part of workplaceperformance (Borman and Motowidlo, 1993). Much managerial workinvolves communicating with others (Gronn, 1983), performing a varietyof different roles (Mintzberg, 1973), and relating to the needs of a largenumber of diverse people (Kotter, 1982). The social skills and leadershipabilities characteristic of high self-monitors may enable them to performbetter than low self-monitors in such contexts.

Previous research has shown that individual differences in how peo-ple approach social situations affect individual attainment in managerialcareers. Self-monitoring effects have been demonstrated on managerialpromotions over a five-year period: High self-monitors are more likely tobe promoted in managerial careers than low self-monitors (Kilduff andDay, 1994). Much of the pioneering work concerning the effects of self-monitoring on performance-related variables has consisted of laboratorystudies on students (e.g., Caldwell and O’Reilly, 1982a). The occasionalfield study has tended to focus either on the eventual outcomes of per-formance differences, such as early promotions (e.g., Kilduff and Day,


1994), or has focused on specific types of workers, such as boundaryspanners (e.g., Caldwell and O’Reilly, 1982b). It is important, therefore,to test whether self-monitoring predicts workplace performance acrossthe full range of organizational positions in an organization.

Three Models

Given the separate and unrelated literatures on social networks and per-sonality, the question is how structural position and self-monitoring com-bine to affect individual performance in organizations. We explore threeperspectives: a mediation model, an interaction model, and an additivemodel.

Mediation Model

Performance differences among individuals in organizations may be dueto the tendency of a particular personality type (the high self-monitor)to occupy structurally central positions that link otherwise disconnectedpeople and provide differential resources. Research across a range ofsocial relationships shows that high and low self-monitors tend to inhabitdifferent social worlds (Snyder, Gangestad, and Simpson, 1983; Snyderand Simpson, 1984; Snyder, Simpson, and Gangestad, 1986). Able totailor behavior to a range of different social situations, the high self-monitor tends to belong to a number of distinct social groups. The lowself-monitor, by contrast, prefers to belong to a clique within whichthe individual can express a characteristic disposition (Snyder, 1987:68–9).

The high self-monitor likes to have one friend for tennis, anotherfriend for basketball, and yet another friend for chess. High self-monitorsmaintain flexibility and make little emotional investment in relation-ships. Friends are chosen based on how closely their skills match activitydomains. As one high self-monitoring tennis player observed, “When Iwant to play tennis, I select a partner who can challenge me” (quoted inSnyder, 1987: 65). Low self-monitors, by contrast, tend to choose friendson the basis of liking, irrespective of whether the friends are proficient intennis, basketball, or chess. They like to be with the same friends acrossactivity domains (Snyder et al., 1983). As one low self-monitor com-mented about her choice of an activity partner, “Jan’s my best friend.Besides, she’s the most fun to be around, whatever the activity” (quotedin Snyder, 1987: 65).

Self-monitoring theory predicts, therefore, that high self-monitors, rela-tive to low self-monitors, will tend to develop friendship relations at work


with distinctly different people. Whereas low self-monitors will tend tooccupy relatively homogenous social worlds, high self-monitors will tendto develop relationships across groups, using their flexible identities toplay different roles in different groups. In a workplace, high self-monitorsare therefore likely to bridge social worlds, acting as conduits throughwhich otherwise unconnected people exchange information.

According to the mediation perspective, high self-monitors will occupycentral positions in social networks in organizations and reap the benefitsof access to diverse resource flows and information detailed by structuralsociologists (e.g., Burt, 1992). Because they tend to serve as go-betweensbetween disconnected others, high self-monitors will enhance their valueto the organization by facilitating resource flows and knowledge sharingacross the organization and thereby achieve superior performance. Thus,high self-monitors will tend to perform better than low self-monitors as adirect result of their differential success in occupying structurally advan-tageous positions in social networks. Complete mediation would suggestthat any effect of self-monitoring on work performance is due to theindividual’s structural position in social networks. Complete mediation,therefore, offers some support for the structuralist view (e.g., Burt et al.,1998) that individual dispositions can serve as proxies for the networkpositions that individuals are likely to occupy.

Interaction Model

The different, but not incompatible, interaction perspective suggests thatdifferent personality types may differentially take advantage of structuralpositions. High self-monitors may be more able and motivated than lowself-monitors to seek out and use the resources available from the differ-ent social groups accessible from bridging positions in social networks.The success of high self-monitors in organizations may occur not becausethe high self-monitors tend to occupy structurally advantageous positionsin social networks (the mediation argument) but because, irrespective ofwho happens to occupy the bridging positions in social networks, only thehigh self-monitors are willing and able to take advantage of the opportu-nities represented by such positions. The interaction model suggests thatboth a high self-monitoring disposition and a structurally advantageousposition in the social network are necessary for the individual to achievehigh work performance.

Numerous studies have confirmed that high self-monitors, comparedwith low self-monitors, tend to be more responsive to the specific char-acteristics of situations (see the review in Snyder, 1987: 33–46). Forexample, in one study, high self-monitors showed themselves acutely sen-sitive to the differing contexts in which social interaction took place.


The high self-monitors were conformist in social situations in which con-formity was the most appropriate interpersonal orientation and werenonconformist when reference group norms favored autonomy. By con-trast, low self-monitoring group members were virtually unaffected bytheir social settings (Snyder and Monson, 1975). This differential respon-siveness is likely to affect work performance. In a field study of peoplewhose jobs required that they interact with groups whose norms differedfrom one another, high self-monitors outperformed low self-monitors(Caldwell and O’Reilly, 1982b). This study, which focused on work-ers’ links outside the organization, provides support for the interactionmodel. Extending this research to the current study of workers withinthe organization, we might expect to find that only high self-monitors areable to take advantage of structurally advantageous network positions toenhance performance.

A further reason to expect performance differences for high and lowself-monitors occupying bridging positions relates to the detection ofuseful social information. High self-monitors are better at scanning thesocial world for information about people and their intentions. High self-monitors are more likely than low self-monitors to notice and remem-ber information concerning others (Berscheid, Graziano, Monson, andDermer, 1976), to be more successful at detecting people’s intentions(Jones and Baumeister, 1976), and to be more accurate at eyewitness iden-tification (e.g., Hosch, Leippe, Marchioni, and Cooper, 1984). If valuableinformation is available to those occupying bridging positions in socialnetworks, then it is more likely to be detected by high self-monitors thanby low self-monitors.

Additive Model

We have argued that high and low self-monitors may differentiallysucceed in organizations because they differentially occupy structurallyadvantageous positions in social networks (the mediation perspective)or because high self-monitors may be differentially able to capitalize onstructurally advantageous positions (the interaction perspective). A thirdpossibility is that structural position and self-monitoring may have rela-tively independent, additive effects on performance in organizations. Theadditive model involves twin predictions concerning work performance.The structural position prediction is that the greater the extent to whichindividuals act as potential go-betweens for those not connected to eachother, the higher the work performance. The self-monitoring predictionis that the higher the individual’s self-monitoring score, the higher theperformance. Support for the additive model would suggest two indepen-dent but not mutually exclusive ways for individuals to gain advantages


MEDIATION MODEL

INTERACTION MODEL

ADDITIVE MODEL

Self-Monitoring

Self-Monitoring

Self-Monitoring

Structural Position Performance

Structural Position

Structural Position

Performance

Performance

Figure 7.1. Three models of how self-monitoring and structural positionaffect individual performance in organizations.

in work performance: (1) occupying a structurally advantageous networkposition, and (2) possessing a high self-monitoring orientation.

Figure 7.1 summarizes the three models of the possible effects of struc-tural position and self-monitoring on performance that we tested in ourstudy.

Method

Bayou Corporation (a pseudonym) was a small high-technology com-pany involved in the chemical analysis of complex compounds. Employ-ees researched, produced, and marketed high-precision chromatographicequipment for laboratories and other clients that analyzed the chemicalcomposition of foods, fragrances, petrochemicals, pharmaceuticals, andother products. Bayou was founded in 1985 by an entrepreneur wholeft his job at a medium-sized chemical company to take advantage ofa business-incubator program at a major university. By 1998, Bayou


Corporation had grown to 116 employees, all located in one state-of-the-art facility. The company had won numerous awards for the qualityof its products and its environmentally conscious business practices. Theself-styled “head-coach” and founder of the organization had created anentrepreneurial culture that emphasized informality rather than bureau-cracy.

Bayou competed in fast-moving markets against much larger compa-nies such as Hewlett-Packard. The company founder emphasized theimportance of innovation and creativity as the keys to survival in thiscompetitive marketplace. Organizational structure was kept deliberatelyflat, with only three levels of hierarchy. Instead of establishing depart-ments, Bayou organized its employees into fluid workgroups that rangedin size from two to sixteen people. The company prided itself on being inthe forefront of equal opportunity employment and had won awards forits success in recruiting and promoting women.

We collected network and personality data by means of a question-naire sent to all 116 employees (68 men and 48 women). We collectedperformance-rating data by means of a separate questionnaire sent to all22 supervisors (17 men and 5 women). Data about reporting relation-ships, demography, and tenure came from company records.

The response rate was 88 percent for the questionnaire sent to allemployees and 100 percent for the questionnaire sent only to supervisors.Nonrespondents did not differ significantly from respondents with regardto sex, tenure, or performance. Missing data on self-monitoring reducedthe usable sample from 102 to 93 individuals for analyses involving thisvariable. Because there were no performance measures for the head ofthe company, analyses concerning both performance and self-monitoringused a sample of 92.

Measures

Social NetworksWe collected data on friendship relations and workflow relations usingthe roster method. For each network, we asked respondents to look downan alphabetical list of employees and place checks next to the names ofpeople they considered friends or work partners. Data for each relationwere arranged in 102-by-102 binary matrices. In each matrix, cell Xij cor-responded to i’s relation to j as reported by i. For example, if i reportedj as a friend, then cell Xij in the friendship matrix was coded as 1; other-wise, Xij was coded as 0. Each matrix contained 10,302 observations onall possible pairs of people.

For each network question, respondents were free to nominate as manynetwork contacts as they deemed appropriate. This format is preferable to


a fixed-choice design in which respondents are asked, for example, “Listyour four best friends,” because it is unlikely that all people have exactlyfour best friends. Limiting respondents to a fixed number of choicestends to introduce measurement error into network data (Holland andLeinhardt, 1973).

We depart from recent research on structural holes in ego networks(e.g., Burt, 1997) by including in our sampling all the actors in the orga-nization rather than just the actors mentioned by the focal individual. Inego-network research, the individual (or “ego”) is the source of infor-mation concerning whether ego’s contacts are themselves connected ordisconnected. Research has shown that individuals are reliable sourcesof information concerning the membership of stable networks to whichthey themselves belong (Freeman et al., 1987), but ego’s responses con-cerning possible interconnections between people to whom ego is tiedare subject to systematic bias (Krackhardt and Kilduff, 1999; Kumbasaret al., 1994). Thus, ego-network data used to assess structural holes arepotentially distorted by perceptual biases.

Comparing Workflow and Friendship NetworksAs research on social networks has pointed out (e.g., Roethlisberger andDickson, 1939: 493–510), in considering the importance of network posi-tion in an organization, researchers must consider two types of networks:the workflow and informal networks. The workflow network is the for-mally prescribed set of interdependencies between employees establishedby the division of labor in the organization. Work flows through the orga-nization as workers exchange inputs and outputs. A successful interactionin the workflow network enables the flow of work from one person toanother (Brass and Burkhardt, 1992: 197).

By contrast, informal social networks, such as the friendship network,derive from mutual liking, similarity of attitudes, or personal choice.Compared to the workflow network, the friendship network representsmore individual choice and initiative. People have more discretion in thechoice of friends than they have in the choice of with whom to interactto accomplish work. Achieving a structurally advantageous position ineither the more formal workflow network or the more informal friend-ship network can bring benefits to the individual in terms of diverseinformation and other resources.

Friendship NetworkRespondents were asked to look down an alphabetical list of fellowemployees and place checks next to the names of those individuals theyconsidered “especially good friends.” Friends were defined as “peoplewith whom you like to spend your free time, people you have been with


most often for informal social activities, such as visiting each other’shomes, attending concerts or other public performances.”

Workflow NetworkThe workflow network was modeled after Brass (1981: 332), who arguedthat “task positions and the workers occupying these positions [can be]viewed as interrelated on the basis of the flow of work through the orga-nization.” Respondents were asked to place a check next to the names oftheir workflow contacts. We combined workflow inputs and workflowoutputs to make the questionnaire more manageable and because Brass(1984) found no differences between the predictive power of input andoutput contacts. Workflow contacts were defined as the “set of peoplethat provide you with your workflow inputs taken together with the set ofpeople to whom you provide your workflow output.” We defined work-flow inputs as “any materials, information, clients, etc., that you mustacquire in order to do your job.” Workflow output was defined as “thework that you send to someone else when your job is complete.” Thisnetwork was, therefore, anchored in the actual work processes of theorganization rather than in the more discretionary task advice networksstudied by others (e.g., Podolny and Baron, 1997).

Network Size and StructureA large network, one with many contacts, can enable the individual toaccess numerous others for information and other resources. But the ben-efits of a large network may be offset by the costs involved in maintaininga large number of relationships (Rook, 1984). People who interact withnumerous others in organizations run the risk of running short of timeand other resources necessary for work performance. Thus, people withlarge networks within the organization may not necessarily achieve thehighest performance ratings. They may be so busy maintaining ties atwork that their work performance suffers (see Burt and Ronchi, 1990,for a case study). In considering how network position relates to workperformance, it is therefore important to examine simultaneously the rela-tionships between network size and performance and between between-ness centrality and performance. One of the questions that our researchattempts to answer is, controlling for the size of the individual’s network,does the extent to which the individual’s network spans social divides pre-dict workplace performance? By looking at both network measures simul-taneously, we can assess whether network size and network betweennesshave independent relationships with work performance.

Betweenness CentralityAs a measure of the extent to which each individual occupied a struc-turally advantageous position, connecting otherwise unconnected others


in the friendship and workflow networks, we used betweenness centrality(Freeman, 1979). We chose this measure rather than a more local measureof autonomy, such as constraint (Burt, 1992), because betweenness cen-trality takes both direct and indirect ties into account (Brass, 1984; Brassand Burkhardt, 1993; Krackhardt, 1990), whereas constraint focuses pri-marily on the direct ties in ego’s immediate circle of contacts. More localmeasures of the extent to which individuals span structural holes are use-ful when sampling from large populations for which whole network dataare unavailable (e.g, Burt, 1992).

The (102 × 102) friendship matrix and the (102 × 102) workflowmatrix were each submitted to the betweenness procedure in the networkprogram UCINET IV (Borgatti et al., 2002; see Freeman, 1979, for theformula). The higher the betweenness score of an actor, the greater theextent to which that actor serves as a structural conduit connecting oth-ers in the network. More formally, betweenness centrality measures thefrequency with which an actor falls between other pairs of actors on theshortest or geodesic paths connecting them (Freeman, 1979: 221).

Because it is difficult to interpret measures of betweenness centralityfor nonsymmetric data, we symmetrized the friendship and workflowmatrices using the rule that if either member of a pair nominated theother, the pair was considered to have a tie. To check whether the resultswere affected by this definition, we also symmetrized each matrix usingthe rule that there was a link between two people only if each member ofthe pair nominated the other. The pattern of results remained unchanged.

Network SizeNetwork size was measured as the total number of each individual’sdirect links with other actors in the network, a measure also known asdegree centrality (Scott, 1991: 86–7). To be compatible with measuresof betweenness centrality, we calculated size on friendship and workflowmatrices symmetrized according to the rule that if either member of a pairnominated the other, the pair was considered to have a tie.

PerformanceOur theory of job performance emphasizes the extent to which individ-uals succeed (in the eyes of management) in contributing to organiza-tional ends. In the absence of objective measures of performance acrossjob types in this organization, we relied on supervisory ratings. Using asix-item scale arranged in a five-point Likert format, supervisors ratedthe performance of those subordinates who reported directly to them.As researchers have noted, in work organizations, “the vast majority ofperformance ratings come directly from the immediate supervisor” (Bretzet al., 1992: 331; see also Scullen, Mount, and Goff, 2000). A recent com-prehensive review of performance evaluation in work settings concluded


that supervisory ratings “are most likely valid reflections of true perfor-mance” (Arvey and Murphy, 1998: 163).

We informed supervisors that performance ratings would be confiden-tial and used only for research purposes. Performance ratings obtained forresearch purposes tend to be more reliable and valid than those obtainedfor administrative purposes (Wherry and Bartlett, 1982). The six per-formance items were selected after extensive discussions with the firm’shuman resource director and a group of four employees representing arange of job types at the firm.

Supervisors first evaluated subordinates’ performance on these threeitems: (1) “the overall job performance of the individual” (1 = poor,5 = excellent); (2) the likelihood that the subordinate would “achievefuture career related success (such as promotions, awards, bonuses, andinvolvement in high-profile projects)” at Bayou (1 = very unlikely, 5 =very likely); and (3) “the likelihood that you would pick [the subordinate]to succeed you in your job” (1 = very unlikely, 5 = very likely).

Given the strong emphasis placed on innovation at Bayou and thegrowing recognition among researchers of the importance of contextualaspects of job performance (e.g., Arvey and Murphy, 1998; Borman andMotowidlo, 1993), we also included three items, taken from Scott andBruce (1994), to capture employees’ workplace innovativeness. Supervi-sors rated subordinates’ innovativeness (using five-point scales) on thesethree items: (1) the degree to which the subordinate generated creativework-related ideas; (2) the degree to which the subordinate promoted andchampioned work-related ideas to others; and (3) the degree to which thesubordinate searched out new technologies, processes, techniques, and/orproduct-related ideas.

The reliability of the six-item scale, as measured by Cronbach’s (1951)alpha, was .90. The results of a component analysis showed all six itemsloaded on the same component (eigenvalue = 4.06; all loadings wereabove .76) that explained 68 percent of the overall variance. To checkwhether our results were an artifact of the composition of our perfor-mance measure, for all analyses that included performance, we ran sep-arate tests using (1) the final six-item measure of performance, (2) athree-item measure that excluded the three innovativeness items, and (3)a three-item measure that included only the innovativeness items. The pat-tern of results was unchanged irrespective of the performance measureused.

Self-MonitoringSelf-monitoring was measured with the eighteen-item true-false versionof the Self-Monitoring Scale (Snyder and Gangestad, 1986). Items include“I would probably make a good actor,” and “In different situations and


with different people, I often act like very different persons.” The self-monitoring score, used as a continuous variable, indicates the probabilitythat an individual is a high or low self-monitor (Gangestad and Snyder,1985).

Control Variables

RankDifferences in formal rank are likely to influence patterns of interaction inorganizations. For example, high-ranking individuals, by virtue of theircontrol over resources and their decision-making authority, may be bet-ter positioned to emerge as central actors in social networks (e.g., Ibarra,1992; Lincoln and Miller, 1979). There were three levels of hierarchy inthe company. From company records, we coded rank as 0 for nonsuper-visors, 1 for supervisors, and 2 for top management team members.

TenureThe length of time a person has been with the company is also likelyto affect the pattern of participation in social networks. For example,individuals who have been with the company longer may be more likelyto occupy central positions in social networks. Using company records,we coded tenure as the number of months that the company had employedan individual.

SexWe controlled for sex in each of the regression models because of itspossible impact on network configuration (Brass, 1985; Ibarra, 1993b)and performance evaluation (Burt, 1992). Sex was coded as 0 for womenand 1 for men.

Analysis

Our approach to testing the mediation, moderation, and additive modelsfollows standard statistical procedures (detailed in Baron and Kenny,1986). We controlled for rank, tenure, and sex in each test. To assesssupport for mediation, we conducted three statistical tests to see whetherany significant relation between self-monitoring and performance waseliminated or significantly reduced once network position was controlledfor. First, we used ordinary-least-squares (OLS) regression to examine therelationship between self-monitoring and performance. Second, we usedMANOVA to examine whether self-monitoring significantly predictedthe four network variables taken as a set. Finally, to evaluate support forthe overall mediation model, we used hierarchical regression analysis to


examine whether the inclusion of the four network variables significantlyaffected the relationship between self-monitoring and performance. Ifa significant relationship between self-monitoring and performance iseliminated or significantly reduced as a result of controlling for the fournetwork variables, then this would indicate support for mediation.

We used hierarchical regression analysis to test the interaction model.To correct for the multicollinearity that arises when testing moderatedrelationships among continuous variables, we centered self-monitoringand the centrality variables before generating interaction terms (Aikenand West, 1996; Cohen and Cohen, 1983). Centering consists of sub-tracting the sample mean from each independent variable. The adjustedvariables each have a mean of zero, but their sample distribution remainsunchanged. We computed four interaction terms by multiplying the cen-tered self-monitoring score with each of the four centered centralityscores. Interaction terms were entered in a separate step after the mainterms had already been entered. If the addition of the interaction termsresults in a statistically significant improvement over the regression modelcontaining the main terms, then this would indicate support for the inter-action model.

Testing the additive model was straightforward: Self-monitoring andthe four network variables were included simultaneously as independentvariables. If self-monitoring and the centrality variables were significantlyrelated to performance, then the additive model would be supported.

Size and Betweenness Centrality CollinearityDespite the clear conceptual distinction between the size of the individ-ual’s network and the extent to which the individual’s network links oth-erwise disconnected employees, size and betweenness centrality are oftenhighly correlated (Bonacich, Oliver, and Snijders, 1998: 135). Popularindividuals tend to have high-betweenness centrality scores. Based on ourtheoretical arguments, we were interested in examining how betweennesscentrality relates to dependent variables while controlling for networksize.

Collinearity between variables such as size and betweenness centralitytends to inflate the standard errors of their regression coefficients, makingit more difficult to obtain significant values, but the inflation of standarderrors does not affect the validity of any significant results that are found.As one regression expert explained, a significant value for the beta coeffi-cient in a regression “is just as conclusive when collinearity is present aswhen it is absent” (Darlington, 1990: 130).

To check on the severity of the multicollinearity between size andbetweenness centrality, we examined the conditioning index and vari-ance proportions associated with each independent and control variable(see Belsley, Kuh, and Welsch, 1980, for a discussion). According to


Table 7.1. Means, Standard Deviations, and Correlationsa

Variable Mean SD 1 2 3 4 5 6 7 8

1. Rank 0.31 0.612. Tenure (months) 53.95 39.03 .163. Sex 0.62 0.49 .14 −.064. Self-monitoring 7.12 3.93 .14 .07 .12

Workflow Network5. Betweenness

centrality63.90 77.19 .18∗ .04 .10 .12

6. Size 49.27 19.50 .24∗∗ .04 .15 .24∗∗ .87∗∗∗∗

Friendship Network7. Betweenness

centrality146.63 243.61 −.07 .33∗∗∗ .01 .18∗∗ .07 .14

8. Size 7.24 5.50 .03 .36∗∗ .01 .04 .13 .21∗∗ .80∗∗∗∗9. Performance 20.25 5.08 .36∗∗∗ −.26∗∗ .02 .23∗∗ .26∗∗ .17 .04 −.10

Notes:a N = 93, except performance (N = 92).∗ p < .10.∗∗ p < .05.∗∗∗ p < .01.∗∗∗∗ p < .001.

Tabachnik and Fidell (1996: 86–7), a conditioning index greater than 30and at least two variance proportions greater than .50 indicates seriousmulticollinearity. None of our independent variables violated this crite-rion; multicollinearity thus posed no serious threats to the validity of ouranalyses.

Results

Table 7.1 presents means, standard deviations, and zero-order correla-tions among the variables. The typical employee had been with the firmfor fifty-four months. Men made up 62 percent of the sample. Individu-als who were higher in rank, self-monitoring, and betweenness centralitytended to have higher job performance ratings in these univariate tests.The density of the workflow network, as measured by the average cellvalue in the 102-by-102 binary workflow matrix, was .34. The friendshipnetwork was considerably sparser, with a mean density of .04.

The Mediation Model

According to the mediation model, the success of high self-monitors inoutperforming low self-monitors is due to the greater success of the highself-monitors in occupying strategically advantageous positions in social


Table 7.2. Standardized Regression Coefficients from AnalysesPredicting Performance (N = 92)

Model

Independent Variable 1 2 3∗ 4 5

Rank .40∗∗∗∗ .38∗∗∗∗ .42∗∗∗∗ .40∗∗∗∗ .40∗∗∗∗Tenure −.31∗∗∗ −.32∗∗∗ −.36∗∗∗∗ −.37∗∗∗∗ −.39∗∗∗∗Sex −.06 −.08 −.07 −.08 −.09Self-monitoring (SM) .21∗∗ .19∗∗ .20∗∗Workflow Network

Betweenness centrality .53∗∗∗ .59∗∗∗ .67∗∗∗Size −.37∗∗ −.47∗∗ −.51∗∗∗

Friendship NetworkBetweenness centrality .41∗∗∗ .32∗∗ .28∗Size −.29∗∗ −.22∗ −.19

SM × Workflow betweenness −.11SM × Size of workflow network .11SM × Friendship betweenness .11SM × Size of friendship network −.09Model F 8.51∗∗∗∗ 7.89∗∗∗∗ 6.39∗∗∗∗ 6.33∗∗∗∗ 4.10∗∗∗∗�F 4.29∗∗ 3.95∗∗∗ 4.19∗∗ 0.26R2 .23 .27 .35 .38 .38�R2 .04 .12 .03 .00Adjusted R2 .20 .23 .29 .32 .29

Notes:a �F and �R2 report changes from previous model, except for model 3, which reports change statistics

from model 1 to 3.∗ p < .10.∗∗ p < .05.∗∗∗ p < .01.∗∗∗∗ p < .001.

networks in organizations. To test this model, we first examined therelationship between self-monitoring and performance. The regressionresults presented in model 2 of Table 7.2 show that high self-monitors, asexpected, tended to outperform low self-monitors. When we controlledfor rank, tenure, and sex, self-monitoring significantly predicted perfor-mance (β = 0.21, p < .05), explaining an additional 4 percent of thevariance over the baseline model.

Although high self-monitors may achieve higher job performance thanlow self-monitors, we still need to know whether they also tend to occupystructurally advantageous positions in social networks. The MANOVAresults presented in the last three columns of Table 7.3 show that whenwe control for rank, tenure, and sex, self-monitoring significantly pre-dicted the four network variables taken as a set (F = 3.40, p < .05),explaining an additional 14 percent of the variance over the baselinemodel. Table 7.3 also shows that higher self-monitoring scores predictedboth higher betweenness centrality in the friendship network and larger


Table 7.3. Standardized Regression and MANOVA Coefficients fromAnalyses Predicting Structural Position and Size (N = 93)

Friendship Network Workflow Network MANOVA

Betweenness Betweenness Wilk’s Eta-Variable Centrality Size Centrality Size Lambda Squared F

Rank −.15 −.03 .15 .20∗ .92 .09 1.96Tenure .34∗∗∗∗ .37∗∗∗∗ .01 .00 .86 .15 3.59∗∗∗Sex .03 .04 .07 .10 .99 .01 0.26Self-monitoring .17∗∗ .02 .09 .20∗∗ .86 .14 3.40∗∗Model F 3.92∗∗∗∗ 3.35∗∗ 1.06 2.76∗∗R2 .15 .13 .05 .11Adjusted R2 .11 .09 .003 .07

Notes:∗ p < .10.∗∗ p < .05.∗∗∗ p < .01.∗∗∗∗ p < .001.

size in the workflow network. Thus, high self-monitors, relative to lowself-monitors, did tend to occupy strategically advantageous positions inthe friendship network and to have larger workflow networks.

To evaluate support for the overall mediation model, we examinedwhether the relationship between self-monitoring and performance wasdue to the significant relationship between self-monitoring and the net-work variables. Including the four network variables in the regressionequation, however, did not significantly affect the relationship betweenself-monitoring and performance. The results presented in model 4 ofTable 7.2 show that even though the high self-monitors tended to occupyhigh-betweenness positions in friendship networks, and even though theoccupants of these positions tended to have higher performance, thehigher performance of high self-monitors was not explained by theirdifferential success in occupying high-betweenness positions. After wecontrolled for the significant relationships between the four network vari-ables and performance, self-monitoring continued to explain significantvariance in performance. The full set of results indicates that althoughself-monitoring explains significant variance in performance and in the setof structural variables, and the structural variables predict performance,the mediation model is not supported. There is no evidence of either fullmediation or partial mediation.

To understand these results more fully, we looked at the differing rela-tionships between self-monitoring and the structural variables. Table 7.3shows that higher self-monitoring scores predict higher betweennesscentrality in the friendship network but also larger size in the workflownetwork. High self-monitors, relative to low self-monitors, not onlyoccupy strategically advantageous positions in the friendship network,


they also find themselves busier than the low self-monitors providingwork outputs and receiving work inputs from more people. Table 7.2suggests that the advantages the high self-monitors may gain from occu-pying central positions in the friendship network may be counterbalancedby the disadvantages of having to maintain large workflow networks.Whereas betweenness centrality in the friendship network has a positiverelationship with individual performance, size in the workflow networkhas a negative relationship with performance.

These counterbalanced results suggest that the performance of highself-monitors, relative to low self-monitors, is simultaneously increasedand decreased by the structure of social networks. The high self-monitors’success in spanning structural holes in the friendship network may helpthem increase their performance, but their acceptance of large workflownetworks may make successfully accomplishing tasks more difficult.

The Interaction Model

The interaction model suggests that the relationship between networkposition and performance depends on the self-monitoring orientation ofthe person occupying the network position: High self-monitors (relativeto low self-monitors) should be able to exploit high-betweenness posi-tions more effectively. We found no support for this prediction. Model5 in Table 7.2 shows that high self-monitors were no more likely thanlow self-monitors to benefit from occupying high-betweenness positions.Adding the four interaction terms as a set failed to significantly improvevariance explained over the direct-effects model 4. There was, therefore,no support for the interaction model.

The Additive Model

According to the additive model, self-monitoring and structural positionshould independently predict performance in organizations. To test thismodel, we included self-monitoring and the four network variables in thesame regression equation. In support of the additive model, the resultsshow that high self-monitors tended to outperform low self-monitors,and those occupying high-betweenness centrality positions tended to out-perform those occupying low-betweenness centrality positions: model4 in Table 7.2 shows that (controlling for rank, sex, and tenure) self-monitoring and each of the four network variables explained significantvariance in performance. The full model explained significantly morevariance in performance than model 2, which contained only the controlsand the self-monitoring variable, and model 3, which contained only thecontrols and the four network variables.


Self-Monitoring

Structural Position

Performance

Figure 7.2. Emergent model of self-monitoring and structural positioneffects on individuals’ work performance.

An Emergent Model

Of the three proposed models, the additive model best explains the data,but the overall results suggest a more complex relationship among self-monitoring, structural position and performance than anticipated by anyof the three proposed models. High self-monitors tended to achieve higherperformance, as did individuals who occupied high-betweenness central-ity positions in the friendship and workflow networks. Consistent withthe additive model, self-monitoring and structural position were rela-tively independent predictors of performance. But we also found thatself-monitoring explained significant variance in the set of structural vari-ables: High self-monitors (compared with low self-monitors) tended tooccupy high-betweenness positions in the friendship network and tendedto interact with more people to get their work done. These results indicatethat the variance shared between self-monitoring and the set of structuralvariables did not overlap with the variance that either of these variablesshared with performance, which leads us to the emergent model summa-rized in Figure 7.2.

Network Differences over Time

To further explore the relationship between self-monitoring and socialnetwork position, we looked closely at the network that was mostamenable to individual preferences: the friendship network. Accordingto self-monitoring theory, high self-monitors should move over timeinto positions in the friendship network that link different social worlds,whereas low self-monitors should remain in homogeneous social worlds.

In the absence of longitudinal data, we tested this argument by lookingat whether the interaction of self-monitoring and organizational tenurepredicted betweenness centrality in the friendship network. We first


Table 7.4. Standardized Regression Coefficients PredictingBetweenness Centrality in the Friendship Network (N = 93)

Model

Variable 1 2 3

Rank −.07 −.15 −.15Sex .02 .03 .06Tenure .34∗∗∗ .27∗∗∗Self-monitoring .17∗∗ .11Self-monitoring × tenure .35∗∗∗∗Model F 21 3.92∗∗∗ 6.24∗∗∗∗�F 7.59∗∗∗ 13.33∗∗∗∗R2 .01 .15 .26�R2 .14 .11Adjusted R2 .00 .11 .22

Notes:∗ p < .10.∗∗ p < .05.∗∗∗ p < .01.∗∗∗∗ p < .001.

centered self-monitoring and tenure and then added the interactionbetween these centered variables. The results shown in model 3 inTable 7.4 suggest that the longer the tenure, the more likely werehigh self-monitors to occupy high-betweenness positions, but length oftenure made no apparent difference to the likelihood that low self-monitors would occupy high-betweenness positions. The interaction termexplained an additional 11 percent of the variance in betweenness cen-trality in the friendship network, a statistically significant improvement(p < .001) over model 2, which assessed the direct relationships betweenself-monitoring and betweenness centrality (controlling for tenure, rank,and sex).

To chart this significant interaction, we partitioned the sample so thatindividuals with scores of 11 or greater were classified as high self-monitors (e.g., Gangestad and Snyder, 1985; Kilduff, 1992). Figure 7.3shows that longer-serving high self-monitors tended to have higherbetweenness-centrality scores, whereas length of time in the organizationmade little difference to the betweenness centrality of low self-monitors.These results are compatible with the idea that high and low self-monitorstend to develop different social network structures over time.

Discussion

This research represents a theory-driven examination of how personal-ity relates to social structure and how social structure and personality


700

600

500

400

300

200

100

010 20 40 60 80 100 120 140 160

Bet

wee

nnes

s C

entra

lity

in th

e Fr

iend

ship

Net

wor

k

Tenure (in months)

------- Low Self-Monitors –––––– High Self-Monitors

Figure 7.3. Relationship between tenure and betweenness centrality forhigh and low self-monitors.

combine to predict work performance. Consistent with self-monitoringtheory, we found that high self-monitors tend to occupy positions of high-betweenness centrality. Further, we found that the relation between self-monitoring orientation and performance in the organization remainedsignificant despite controlling for several other significant variables,including four measures of network structure. Although strong claims ofcausality would require studying the effects of self-monitoring on socialstructure over time, we did find that for high self-monitors (but not for lowself-monitors), longer service in the organization predicted the occupancyof strategically advantageous network positions. Our research thereforesuggests three important conclusions. First, personality predicts socialstructure: The high self-monitors tended to occupy central positions insocial networks. Second, personality affects the way individuals buildfriendship networks over time: The high self-monitors (but not the lowself-monitors) became more central the longer they stayed in the orga-nization. Third, self-monitoring and centrality in social networks inde-pendently predict individuals’ workplace performance. The results painta picture of individuals shaping the networks that constrain and enableperformance. It appears that high and low self-monitors pursue differentnetwork strategies, with high self-monitors tending to occupy positionsthat span social divides, whereas low self-monitors remain tied to morehomogenous social worlds. High and low self-monitors, therefore, appearto be active agents in the structuring of distinctive social worlds at work.

In formulating three models of how self-monitoring and network posi-tion together might affect work performance, we have emphasized the


importance of considering alternative linkages between our constructs(cf. Elder, 1973). The particular site that we examined consisted of a rela-tively small, cohesive organization in which there were relatively few highself-monitors. We need further research in other organizational settingsto understand more fully how self-monitoring orientation and networkposition might combine to affect workplace performance.

Future research could also examine different types of performance out-comes to supplement our reliance on supervisory ratings. We are reas-sured by considerable research evidence that, even when supervisors andratees are members of the same network (as in our sample), supervi-sors tend to like subordinates who prove themselves as high performers(Robbins and DeNisi, 1994). It is unlikely, therefore, that ratings werebiased by liking, given that “affect is likely to be a function of how well orpoorly a person performs his or her job” (Arvey and Murphy, 1998: 151).

The picture we present in this chapter of people taking advantage oftheir personality orientations to forge different types of network struc-tures offers a new direction for social network analysis. In the past, net-work research focused almost exclusively on “the overall structure of net-work ties” (Emirbayer and Goodwin, 1994: 1415), neglecting or omittingindividual-level variables (see, for example, Mayhew’s 1980 manifesto).Individual dispositions, to the extent that they have been discussed at allin recent network research, have tended to be dismissed as “the spuriouslysignificant attributes of people temporarily occupying particular positionsin social structure” (Burt, 1986: 106). In this chapter, we demonstratethat self-monitoring theory can enrich our understanding of such vitalnetwork topics as who is likely to bridge structural holes and the connec-tion between structural position and work performance. We encouragefurther examination of the ways in which different types of people forgedistinctively different patterns of social ties in the workplace.

One of the major unanswered questions concerning self-monitoringand social networks is, What motivates high and low self-monitors tobuild such different social worlds? A recent review of the self-monitoringliterature suggested that high and low self-monitors might have differ-ent orientations toward status enhancement. High self-monitors mightseek, above all, to “create public images . . . , that connote social sta-tus.” Low self-monitors, by contrast, may be more interested in investingin “close social relationships in which they and their partners can betrusted” (Gangestad and Snyder, 2000: 547). High and low self-monitorsmay be building different types of social capital, with high self-monitorsfocused on constructing social worlds that function as “effective instru-ments of status enhancement” and low self-monitors focusing on con-structing social worlds that support their reputations as “genuine and sin-cere people” (Gangestad and Snyder, 2000: 547). Future research could


investigate whether high and low self-monitors are differentially awareof the structural holes in social networks and whether they consider thecareer consequences of different social network strategies.

The theory and results that we present in this chapter suggest that highself-monitors, the chameleons of the social world, resemble the proto-typical person featured in sociological research on social networks. Insociological research, individuals tend to take on the attributes and ideasof their associates rather than relying on their own inner beliefs and val-ues (e.g., Carley, 1991). According to sociologists, people strive to occupycentral positions in social networks in order to advance their careers (Burt,1992). Low self-monitors, despite making up approximately 60 percentof the population (Snyder, 1987), seem strangely absent from the socio-logical literature. In our research as well, low self-monitors have featuredmainly as the background against which we have discussed the contri-butions and outcomes of the more visible high self-monitors. For futureresearch, the question remains, How do the organizational networks oflow self-monitors affect contributions and outcomes?

Self-monitoring theory suggests that the social networks of low self-monitors may help enhance several aspects of organizational effective-ness. Low self-monitors’ tendency to forge deep emotional attachments,for example, may facilitate the development of strong network ties usefulin crisis situations (see Chapter 10) and in the transfer of tacit knowledge(Hansen, 1999). The networks of low self-monitoring individuals, there-fore, may help organizations respond to unexpected jolts and to transmitexpertise. Further, low self-monitors’ greater commitment to work rela-tionships may lead to greater commitment to the organization (Jenkins,1993). But if a low self-monitor does leave an organization, there maybe a larger impact on coworkers (in terms of turnover, for example) thanwhen a high self-monitor leaves (see the discussion of turnover effects oncoworkers in Chapter 9).

Self-monitoring orientation is a stable component of the individual’spersonality, but a stable personality trait can be expressed through a rangeof possible behaviors. The practical implications of our findings, there-fore, can involve individuals changing behaviors even if they are unableto change self-monitoring orientations. High self-monitors, for exam-ple, are more other-directed than low self-monitors, meaning that highself-monitors tend to be more susceptible to pressure from other people(Kilduff, 1992; Snyder and Gangestad, 1986). In our results, this other-directedness shows up as an increased workload for high self-monitors interms of a larger number of connections in the workflow network. Thechallenge for the high self-monitor is how to avoid accepting too manydifferent work responsibilities while maintaining friendship ties that spansocial divides. The challenge for low self-monitors is to build on their


ability to say no to an overload of work responsibilities by spending moretime overcoming their marked inclination to retreat into stable friendshipcliques.

Structural analysis offers tools for identifying role structures withingroups (e.g., White, Boorman, and Breiger, 1976) and dynamics betweengroups (e.g., McPherson et al., 1992). Adding personality theory to struc-tural analysis can help forge a powerful approach to understanding indi-vidual behavior in the context of social structure. Rather than accept-ing an inevitable duality between those interested in the psychologicaldeterminants of behavior and those interested in how network struc-ture affects social processes, we need more interdisciplinary research thatdraws from different perspectives and contributes to an enhanced pictureof how action affects outcomes in organizations. In the next chapter,we pursue this interdisciplinary agenda with respect to how personalityvariables (including self-monitoring) predict position in a social networkthat has not previously been studied: the emotion helping network in anorganization.

8

Centrality in the Emotion HelpingNetwork: An Interactionist Approach

An organization can be considered a socio-emotional system in whichenergy must be continually expended in order to keep the system oncourse (cf. Katz and Kahn, 1966). Among the many threats to systemfunctioning are negative emotions. The workplace is a site where peo-ple often experience negative emotions associated with stress, anxiety,tension, and emotional pain (Basch and Fisher, 2000). Two classic exper-iments (Latane and Arrowood, 1963; Schachter, Willerman, Hyman, andFestinger, 1961) established that workers involved in all but the most rou-tinized tasks who are subject to negative emotions tend to suffer decre-ments in the quality and quantity of their production. Further, thesenegative emotions correlate with individuals’ negative work-related atti-tudes (Weiss and Cropanzano, 1996) and health problems (Frost, 2003:3), and can prove contagious (Hatfield, Cacioppo, and Rapson, 1994)with deleterious effects for other employees’ levels of cooperation andperformance (Barsade, 2002). We know that some people become cen-tral actors in dealing with the negative emotions of colleagues (Frost,2003; Frost and Robinson, 1999), but there is still little understanding ofwho these unusual people are. In this chapter, we spotlight the emotionhelping network and its central players from a personality interactionperspective.

Research concerning social support in general has examined the bene-ficial effect of such support on health and well-being (for reviews, seeUchino, Capioppo, and Kiecolt-Glaser, 1996; Viswesvaran, Sanchez,and Fisher, 1999), the features of the support recipient (e.g. Collinsand Feeney, 2004), the antecedents of social support (e.g., Zellars andPerrewe, 2001), and the role of cultural norms (Taylor et al., 2004). But“almost no attention has been paid to social integration, networks, orsupports as dependent variables” (House, Umberson, and Landis, 1988:308). We remedy this omission in focusing on centrality in the emotion

157


helping network. We build on the finding in the previous chapter thatpersonality predicts individuals’ social network positions.

In general, pro-social behavior (such as emotion help) results froma combination of dispositional and status-based characteristics (Batson,1991). We anticipate that people central in the emotion helping networkare likely to have the dispositional skills necessary to notice and alle-viate emotional suffering and the managerial responsibility necessary totake action. We draw from personality interaction theory to suggest thatas managerial responsibility increases, the effects of relevant personal-ity variables – self-monitoring and positive affectivity – are magnified.We also anticipate and control for the role of friendship and coworkersocial networks in ameliorating individuals’ negative emotions. We con-tribute to the ongoing effort to understand how individual personalityinteracts with organizational variables to affect important organizationaloutcomes (cf. Barrick, Parks, and Mount, 2005; Flynn and Ames, 2006).

Theory and Hypotheses

Personality Dispositions and Emotion Helping

Of the many personality variables that might conceivably affect individ-uals’ tendency to become central in the emotion helping network, wefocused on two: positive affectivity and self-monitoring. These two vari-ables emerge from strong theoretical traditions and have shown theirimportance in major studies of organizational behavior relevant to ourresearch topic.

Positive AffectivityThose who engage in the role of alleviating others’ negative emotionson a regular basis are exposed to considerable pain and suffering thathas the potential to lead to burnout and stress (cf. Frost, 2003). There isevidence, however, that some individuals relative to others may enjoy theexperience of helping those in distress deal with their problems and maybe protected from the contagious effects of negative emotions. People highin positive affectivity, relative to those low in positive affectivity, are lesslikely to suffer the harmful consequences of helping others experiencingemotional difficulty (Zellars, Perrewe, Hochwarter, and Anderson, 2006).Meta-analytic results show that high positive affectivity is significantlyrelated to organizational commitment, job satisfaction, lower emotionalexhaustion, and less depersonalization of others (Thoresen et al., 2003).An individual’s positive affectivity disposition tends to “permeate all ofan individual’s experiences” (Barsade, Ward, Turner, and Sonnenfeld,2000: 803). Positive affectivity is a classic personality trait in the sense

Centrality in the Emotion Helping Network 159

of being a consistent disposition over time (Staw et al., 1986; Watson,Clark, and Tellegen, 1988).

People high in positive affectivity, relative to those low in positiveaffectivity, tend to experience a “pleasurable engagement with the envi-ronment” (Watson, 1988: 128). One of the most robust findings in thisresearch literature is that people high in positive affectivity tend to “enjoy,feel more confident in, and even attract social contact” (Lucas and Diener,2003: 48). High positive affectivity people are more sensitive and atten-tive to those with whom they interact (Isen, 1970). Subjects in whom apositive mood has been induced show increased helping behavior relatedto the maintenance of the positive state of the beneficiary (Cunningham,Steinberg, and Grev, 1980) and exhibit more overt friendliness with oth-ers (Cunningham, 1988). Extensive evidence suggests that those high inpositive affectivity (relative to those low in positive affectivity) tend to bemore willing to perform organizational citizenship behaviors (Williamsand Shiaw, 1999) and more willing to engage in altruistic behaviors(Diener, Lyubomirsky, and King, 2001).

High positive affect may well be reinforced by the positive feelings thatpeople experience when they perform altruistic and helpful behaviors(Carlson, Charlin, and Miller, 1988). People with low positive affectivityexperience good mood and happy feelings to a lesser extent than thosewith high positive affectivity (Lyubomirsky, King, and Diener, 2005) andas a result may be less prone to helping others spontaneously (Georgeand Brief, 1992). High positive affectivity individuals – who tend to beenergetic, enthusiastic, and upbeat – are likely to be viewed by employeesas open to helping others regulate their negative emotions; these peopleare also likely to enjoy the role of ameliorating others’ distress and solvingorganizational problems.

Hypothesis 1: Positive affectivity will be positively related to the extentto which individuals occupy central positions in the emotion helpingnetwork.

Self-MonitoringSelf-monitoring theory concerns the monitoring and control of expres-sive behavior (Snyder, 1974), including emotional display (Graziano andBryant, 1998). Individuals high in self-monitoring monitor and controlhow they present themselves to others in response to social cues con-cerning what is appropriate and expected in specific social situations,whereas individuals low in self-monitoring pay less attention to the socialappropriateness of self-presentations and more attention to inner affec-tive states and attitudes (Snyder, 1974). As the theory has developed, two


prototypical types of people have been described: the high self-monitorwho strives to generate affective states and behaviors appropriate tospecific well-defined situations, and the low self-monitor who generatesexpressive behavior from inner affective states and attitudes (Snyder,1979). Compared to high self-monitors, “low self-monitors express it asthey feel it” (Graziano and Bryant, 1998: 251). High self-monitors, bycontrast, are skilled at communicating socially appropriate impressionsboth vocally and nonverbally (Snyder, 1974). With respect to emotions,high self-monitors, because of their greater attentiveness to the social sit-uation, tend to be better at adjusting their emotional displays to the needsof others, but are also more alert to the emotions being experienced byother people (Ickes, Stinson, Bissonette, and Garcia, 1990).

According to recent thinking and research, the high self-monitoringadvantage in expressing a range of socially appropriate emotions and inreading the emotions of others requires considerable effort and representsthe strong motive of high self-monitors to produce successful social inter-actions (Ickes, Holloway, Stinson, and Hoodenpyle, 2006). This greatereffort on the part of high self-monitors, relative to low self-monitors, atmaking social interactions go well is illustrated by research showing thatin unstructured social interactions between strangers, high self-monitorstend to speak first and to use conversational overtures to break periodsof silence (Ickes and Barnes, 1977). A review of this early research sum-marized it as showing that high self-monitors “were concerned that theirinteraction would go well, they acted to ensure that it would, and theyreported being increasingly self-conscious to the extent that it didn’t”(Ickes et al., 2006: 662).

Not only do the high self-monitors, relative to the low self-monitors,try to make social interactions work by taking the initiative in terms ofconversational openings, they also try to inject positive affect into socialinteractions through the use of humor to lift the spirits of others (Turner,1980). Other research shows that high self-monitors express more posi-tive affect in their self-presentations than do low self-monitors (Levine andFeldman, 1997). A recent summary of the evidence concerning the moti-vations of high and low self-monitors concludes that high self-monitors,more than low self-monitors, derive positive self-affect from successfulself-presentations to others: “if the self-presentation of a high self-monitorappears to have the desired effect on his or her interaction partner, thehigh self-monitor would experience a sense of acceptance and validationin the form of positive self-affect” (Ickes et al., 2006: 682).

Thus, the evidence suggests that high self-monitors, relative to low self-monitors, strive to pay attention to the emotions of others and to provideappropriate emotional displays that will produce successful social inter-actions. As part of this effort to make social interactions work, high


self-monitors tend to talk about the other person instead of talking aboutthemselves (Ickes, Reidhead, and Patterson, 1986), to pace conversationsappropriately (Dabbs et al., 1980), and to reciprocate self-disclosures(Shaffer et al., 1982). Not surprisingly, given the substantial cognitiveand emotional resources that high self-monitors bring to their social per-formances (Ickes et al., 2006: 681), high self-monitors tend to emergeas leaders in a variety of situations (Garland and Beard, 1979; Zaccaroet al., 1991). Further, a series of studies of self-monitoring and helpingbehavior suggests that high self-monitors are more accurate in perceivingthe status dynamics of exchange relationships, and elevate their socialstatus by establishing reputations as generous exchange partners willingto help others but reluctant to ask for help in return (Flynn et al., 2006).We build on this cumulating theory and evidence to suggest that it will behigh self-monitors who will tend to monitor and ameliorate the negativeemotions of others in the workplace.

Hypothesis 2: Self-monitoring will be positively related to the extentto which individuals occupy central positions in the emotion helpingnetwork.

Managerial Responsibility and Emotion Helping

Managerial responsibility for the maintenance of morale in organizationshas long been a part of organization theory. Henry Fayol (1916) empha-sized the responsibility of managers for the promotion of harmony andunion among organizational personnel, whereas Chester Barnard (1938)included among the functions of the executive responsibility for man-aging informal communication and morale in organizations. FrederickTaylor’s (1911: 74) principles of scientific management included the needfor “harmony, not discord.”

These classic recognitions of the importance of managerial responsibil-ity for employees’ emotional welfare have been extended by examinationsof emotion management in contemporary firms. The founder and chiefexecutive officer of a cosmetics company was quoted as being “mystifiedby the fact that the business world is apparently proud to be seen ashard and uncaring and detached from human values . . . the word ‘love’was as threatening in business as talking about a loss on the balancesheet” (Martin, Knopoff, and Beckman, 1998: 447). One research reportof how middle managers coped with radical change in a large informa-tion technology firm summarized the findings as follows: “Managers’emotion-attending behaviors reduced a potentially higher state of angerand fear among the employees driven by emotional contagion” (Huy,2002: 60). Managers of many organizations tend to spend considerable


time and attention dealing with the negative emotions prevalent in theworkplace (Frost, 2003: 157–8).

Thus, the processing of employee emotions is a crucial, if overlooked,organizational task (Frost and Robinson, 1999; Maitlis and Ozcelik,2004). Managers, compared to other employees, are expected to helpsolve problems in the workplace, including employees’ emotional prob-lems (Ostell, 1996). To the extent that negative emotions threaten theproper functioning of the organization, it is management’s duty to helpprocess these emotions, and return the workplace to a state of emotionalbalance (Huy, 2002). The managerial role, relative to non-managerialroles, involves less programmed work, more discretion, and a greaterlicense to intervene in others’ affairs to solve problems (Mintzberg, 1973).People with more managerial responsibility (compared to those withlower managerial responsibility) have more decision-making authorityand greater control over resources (e.g., Ibarra, 1992). Therefore, thosewith managerial authority, compared to other employees, are expected tohelp solve problems in the workplace, to intervene in troublesome situa-tions and to make use of the discretion and resources at their disposal toalleviate emotional problems of employees. We summarize this discussionin the following hypothesis.

Hypothesis 3: Managerial responsibility will be positively related to theextent to which individuals occupy central positions in the emotion help-ing network.

An Interactionist Approach

From an interactionist perspective, the effects of personality traits onoutcomes are likely to be magnified to the extent that constraints on theexpression of personality are weakened (Barrick et al., 2005). Particularpersonality characteristics tend not to be evident indiscriminately, butrather to appear in the specific circumstances that make those charac-teristics salient (Reis, 2001: 69). For example, personality predicts jobperformance for occupants of jobs with high as opposed to low auton-omy (Barrick and Mount, 1993). One general indication of autonomy inorganizations is the extent of managerial responsibility: The higher themanagerial responsibility, the more discretion people have in the solutionof workplace problems (Mintzberg, 1973). Building on research suggest-ing that situations that provide individuals with considerable discretionare likely to facilitate the effects of personality (see Weiss and Adler,1984, for a review), we suggest that the more managerial responsibilityindividuals possess, the more likely it is that personality traits will relateto the provision of emotional help to organizational members.


Thus, as an individual accepts more managerial responsibility, self-monitoring and positive affectivity are likely, we suggest, to become morepredictive of discretionary, problem-solving behaviors such as helpingothers with their emotional problems. The acceptance of more managerialresponsibility allows people more discretion, a greater license to intervenein others’ affairs to solve problems (Mintzberg, 1973) and greater powerand status (Hambrick and Finkelstein, 1987). Managerial responsibilityallows people to express their personality dispositions in the workplacein ways that are not available to those with low power or status (Ander-son and Thompson, 2004). For example, the high self-monitoring ten-dency to take the initiative in speaking first in social interactions maywell be suppressed among those with little or no managerial responsibil-ity to intervene in cases where others appear to be suffering emotionaldistress.

Hypothesis 4: The higher an individual’s managerial responsibility, thegreater the effect of positive affectivity on the extent to which the indi-vidual occupies a central position in the emotion helping network.

Hypothesis 5: The higher an individual’s managerial responsibility, thegreater the effect of self-monitoring on the extent to which the individualoccupies a central position in the emotion helping network.

Alternative Explanations

Social Network CovariatesPeople suffering anxiety seek to affiliate with others (Schachter, 1959).More generally, as one review summarized, “social relationships are amajor source of happiness, relief from distress, and health” (Argyle, 1987:31). Therefore, any discussion of who is likely to be prominent in helpingothers with emotions in organizations must control for prominence in theinformal social networks that provide a primary context for seeking andreceiving help. The inclusion of network centrality variables as predictorsseems to be relevant and necessary to show the added explanatory powerof the personality interactions above and beyond the effects of friendshipand work-related interaction.

Two networks in organizational settings provide most of the basis forsocial interaction. One is the affective network of friendship relations, andthe other is the instrumental network of work relationships. The friend-ship network is based on choices concerning whom to like and socializewith, whereas the workflow network reflects choices concerning workcollaboration. These two networks – friendship and workflow – have


long been emphasized as constituting bases of interaction in organizations(e.g., Brass, 1985; Lincoln and Miller, 1979; Roethlisberger and Dickson,1939). Two people in an organizational setting joined by a friendship tieare likely to seek each other out frequently for social exchange, whereastwo people joined by a workflow tie are likely to interact on a regularbasis as part of their work. Note that the workflow network may differfrom the formally designated set of task interdependencies in an orga-nization given that, for example, people may choose not to cooperatewith formally designated work partners (to the detriment of work perfor-mance – cf. Gargiulo and Benassi, 1999) and may provide informationand other resources to people who are not formally recognized as workpartners. Note also that the dependent variable – emotion helping – isdifferent from these network variables: Helping lots of people with theirnegative emotions is not the same as enjoying friendship relations withlots of people, nor is it the same as having lots of work partners. Buta person who is central in either the friendship or the workflow net-work has many opportunities to interact with others and is thereforelikely to be called upon by friends or by work partners to solve prob-lems (Sparrowe et al., 2001), including emotion problems. Therefore,we controlled for the degree of centrality in these two networks in ouranalyses.

Gender and TenureEvidence suggests that women relative to men are more likely to respondempathetically to the distress of others, show greater concern, and providemore comfort (see the review in Baron-Cohen, 2003). Providing sympathymay be more socially acceptable for women compared to men (Eagly andCrowley, 1986) and women tend to be more effective in roles that aredefined in terms of interpersonal ability (Eagly, Karau, and Makhijani,1995). We used tenure as a control variable because it seems likely thatlonger-serving employees might be relied upon more than shorter-servingemployees for emotion support.

Method

Site

We studied a recruiting agency that provided managerial staff for retailoutlets such as supermarkets and grocery stores. The organization had104 employees and was structured as a professional bureaucracy in thattrained professionals in the operating core of the business worked closelywith clients (Mintzberg, 1983a). The company received about five hun-dred original resumes each week and had developed a reputation forprofessionalism in the education and training of its employees.


Procedure

We collected personality, emotion helping, and network data by meansof a questionnaire sent by e-mail attachment to all 104 employees (62women and 42 men). We sent a reminder after two weeks and a secondreminder after a further week. The questionnaire was sent directly by theresearch team to employees without any pressure from the company onemployees to respond, although approval for this research project wasprovided to the research team in advance by the CEO. All respondentswere assured of confidentiality in the handling of their responses, which,given the nature of the social network questions, were not anonymous.Respondents were promised and provided with debriefing sessions withthe research team following data collection. These debriefing workshopsproved popular, with approximately 60 percent of respondents showingup to learn more about how to interpret their individualized reports. Ingeneral, a high level of trust was developed and maintained throughoutthe data collection and debriefing process.

Information concerning managerial responsibility scores, gender, andtenure were derived from company records. The response rate was 92 per-cent for the questionnaire. Nonrespondents did not differ significantlyfrom respondents with regard to managerial responsibility, tenure, gen-der, or age. Missing data on self-monitoring reduced the usable samplefrom ninety-six to ninety-four respondents.

Measures

We were able to test our models with one independent variable (manage-rial responsibility) derived from company records whereas other indepen-dent variables (self-monitoring and positive affectivity) were derived fromself-reports. The network control variables (centrality in friendship andworkflow networks) and the dependent variable (extent of emotion help-ing) were derived not from the self-reports of the focal individual but fromthe reports of others concerning friendship relations, workflow interac-tion, and the tendency to go to the focal individual for emotion help.Thus, we avoided some potential common method variance problemsby using different sources for our independent and dependent variables,consistent with the conceptual framework of our study (cf. Sackett andLarson, 1990: 474).

Independent Variables

Positive AffectivityRespondents rated on a five-point scale the extent to which they gen-erally and on average experienced each of the ten relevant adjectives


(active, alert, attentive, determined, enthusiastic, excited, inspired, inter-ested, proud, and strong) that comprise the positive affectivity trait por-tion of the Positive Affectivity-Negative Affectivity Scale (Watson et al.,1988). For a review of the scale’s predictive and construct validity, seeWatson et al. (1988). Cronbach α for the 10-item scale was .80.

Self-MonitoringThis variable was measured using the eighteen-item true-false Self-Monitoring Scale (Snyder and Gangestad, 1986). As a continuous vari-able, the self-monitoring score indicates the probability that an individualis a high or low self-monitor (Gangestad and Snyder, 1985): The higherthe score, the higher the probability of a high self-monitoring orientation.Representative items include “In different situations and with differentpeople, I often act like very different persons,” and “I may deceive peopleby being friendly when I really dislike them.” The Kuder-Richardson reli-ability for the eighteen-item scale in the present research was .67. For arecent review of the scale’s predictive and construct validity see Gangestadand Snyder (2000).

Managerial ResponsibilityWe measured this with a “job points” score used within the companyto determine the salary and compensation for each position. The scoresranged from 40 for entry-level positions to 100 for that of the CEO.The scoring system shared the Hay Method’s principles of job evaluation(Milkovich and Newman, 1990): (1) More complex or more responsiblework should receive greater compensation than less complex or respon-sible work, and (2) there should be like pay for like work within theorganization. A follow-up interview with the CEO confirmed that higherscores indicated more complex business and supervisory responsibility,less programmed work, and use of greater discretion, consistent with howmanagerial jobs are defined (e.g., Mintzberg, 1973).

Control Variables

Network CentralityWe followed the roster method for capturing network data (cf. Mehraet al., 2001). Respondents were asked to look down alphabetical lists of allemployees and check the names of those they considered to be friends orwork partners. The exact instructions concerning the friendship networkwere as follows:

Whom would you consider to be your especially good friends?Friends are people with whom you like to spend your free time,


people you have been with most often for informal social activ-ities, such as visiting each others’ homes, having lunch togetheroften, attending concerts or other public performances, going outto pubs and clubs, etc.

The exact instructions concerning the workflow network (followingBrass, 1981,1984) were as follows:

Whom would you consider to be your most important workpartners? Workflow contacts are a set of people who provideyou with your workflow inputs, as well as the set of people towhom you provide your workflow output. Workflow inputs areany materials, information, clients, etc., that you might acquirein order to do your job. Workflow output is the work that yousend to someone else when your job is complete.

From these data, we constructed (following the example of previousresearch, i.e. Mehra et al., 2001), for both the friendship network andthe workflow network, a (94 × 94) matrix with (ij) cell entries equal to1 if i indicated a relationship to j and equal to 0 otherwise.

We used the most basic measure of centrality – degree – in calculatingthe exact number of direct links between the individual and others inthe network (Freeman, 1979). The pattern of results did not change ifwe defined a tie as existing only if each member of the pair nominatedthe other. Degree centrality captures the popularity of the individualin the network. The higher an individual’s degree centrality score, thegreater the number of friends or workflow partners the individual has. The(94 × 94) friendship and workflow matrices were submitted to the degreecentrality procedure in the network program UCINET VI (Borgatti et al.,2002).

GenderUsing company records, we coded gender as 0 for men and 1 for women.

TenureThis was measured, using company records, as the number of years thatan individual had been employed by the company.

Dependent Variable: Indegree Centrality in the EmotionHelping Network

To assess the extent to which each individual engaged in providing emo-tion help to colleagues, we collected information from respondents con-cerning who helped them deal with their negative emotions. Note that


our measure recognizes that emotion helping may be unreciprocated inthe sense that an individual can provide emotion help to others withoutseeking emotional help from any of those others. Thus, the dependentvariable – emotion help – was measured as a count of how many othersin the organization nominated the focal individual as someone to whomthey turned to for emotional help. Whereas intra-organizational networkresearch often considers a general form of “advice” network (e.g., Krack-hardt, 1990), we measured the extent to which people helped others dealwith a specific type of problem – stressful and negative emotions. Weprovided organizational members with an alphabetical list of colleaguesand the following instructions:

Whom do you go to when you experience anxiety, tension oremotional pain? Please look down the alphabetical list of yourfellow employees and place a check mark to indicate all the namesof those people who you think help you when you need supportin times of trouble to cope with your personal problems andyour negative emotions. Some people may go to several peoplefor help and support. Some may only go to one person. Somemay not go to anyone within the organization, in which case donot check anyone’s name under that question.

For each individual in our sample, we calculated indegree centrality – thenumber of nominations received from others.

Individuals who received many nominations were considered to beactively taking on the role of regulating others’ negative emotions,whereas individuals who received few or no nominations were consid-ered to be relatively inactive in this role.

Results

Our dependent variable, centrality in the emotion helping network, ismeasured by the number of nominations received from others. Therefore,it is a count variable that takes only non-negative integer values. We ana-lyzed the data using both standard linear regression and Poisson regres-sion recommended for count variables (cf. Long, 1997). Because therewere no significant differences between the results of the two regressionprocedures, we report the data from the more familiar linear regressionmodels. For all tests involving interaction, we corrected for potential mul-ticollinearity among continuous variables by centering self-monitoring,positive affectivity, and managerial responsibility (Aiken and West, 1996;Cohen and Cohen, 1983).


Table 8.1. Means, Standard Deviations, and Correlations

Variable M SD 1 2 3 4 5 6 7 8

1. Emotion Help 3.79 4.09 –2. Gender 0.61 0.49 .19 –3. Tenure 2.58 3.00 .47∗∗∗ −.05 –4. Friendship

centrality6.74 4.34 .25∗ −.14 .10 –

5. Workflowcentrality

32.84 16.26 .43∗∗∗ −.17 .21∗ .19 –

6. Self-monitoring 11.31 3.21 .18 −.24∗ .07 .25∗ .14 –7. Positive

affectivity36.73 5.01 .26∗ −.08 .03 .01 .29∗∗ .11 –

8. Managerialresponsibility

53.51 15.40 .60∗∗ .03 .59∗∗∗ .01 .31∗∗ .21∗ .20 –

Notes: N = 94.∗ p < .05.∗∗ p < 0.01.∗∗∗ p < 0.001.

Table 8.1 presents means, standard deviations, and zero-order correla-tions among the variables. Individuals averaged seven friends and thirty-three workflow relationships, which compares to seven friends and forty-nine workflow relationships reported in previous research on a similarlysized company (Mehra et al., 2001). Importantly for the independence ofour hypotheses, self-monitoring and positive affectivity represented twodifferent constructs – the scores were not significantly correlated (r = .11,ns). Women tended to have lower self-monitoring scores than men(r = −.24, p < .05). Both self-monitoring (r = .18) and positive affectiv-ity (r = .26) were positively correlated with the extent of emotion help,but only the positive affectivity correlation with emotion help reachedconventional levels of significance (p < .05). Individuals active in provid-ing emotional help to many others also tended to have a longer tenure(r = .47, p < .001), and to be higher in workflow centrality (r = .43,p < .001), friendship centrality (r = .25, p < .05), and managerial respon-sibility (r = .60, p < .001).

Table 8.2 shows the results of regression analyses in which the depen-dent variable is, for each individual in the organization, the number ofpeople who nominated that individual as someone they turned to foremotional help. Model 1 shows that the control variables explained 34percent of the adjusted variance. Recall that hypotheses 1 and 2 suggestedthat people with higher scores on positive affectivity and self-monitoringwould tend to help more people with their negative emotions. Model 2 inTable 8.2 shows no main effects of positive affectivity or self-monitoringon the dependent variable. The introduction of these personality variables


Table 8.2. Linear Regression Predicting Centrality in the EmotionHelping Network

Variables Model 1 Model 2 Model 3 Model 4a

Gender (0 for men; 1 for women) 0.09 0.11 0.10 0.16∗Tenure 0.40∗∗∗ 0.40∗∗∗ 0.18 0.02Friendship centrality 0.16 0.15 0.20∗ 0.17∗Workflow centrality 0.33∗∗∗ 0.27∗∗ 0.21∗ 0.06Positive affectivity (PA) 0.16 0.11 0.18∗∗Self-monitoring (SM) 0.09 0.01 0.11Managerial responsibility (MR) 0.41∗∗∗ 0.40∗∗∗Managerial responsibility∗PA 0.44∗∗∗Managerial responsibility∗SM 0.24∗∗F 12.90∗∗∗ 9.63∗∗∗ 12.03∗∗∗ 18.78∗∗∗R2 0.37 0.40 0.50 0.67Adjusted R2 0.34 0.36 0.45 0.63R2 Change 0.37∗∗∗ 0.03 0.10∗∗∗ 0.17∗∗∗

Notes: N = 94; standardized coefficients are presented.a Positive affectivity, self-monitoring, and managerial responsibility have been centered.∗ p < .05.∗∗ p < .01.∗∗∗ p < .001.

did not significantly improve variance explained, so hypotheses 1 and 2were not supported (although the positive affectivity coefficient of 0.16was close to conventional levels of significance at p < .07).

Model 3 in Table 8.2, however, presents evidence to support hypothesis3’s suggestion that the extent of each person’s managerial responsibilitywould relate to how much emotional help each person provided. Theaddition of the managerial responsibility variable (β = .41, p < .001)significantly increased explained variance by 9 percent to an adjustedtotal of 45 percent.

But is it the case that increasing managerial responsibility brings intoplay personality differences with respect to the provision of emotionalhelp in the workplace? Specifically, in line with hypothesis 4, did increasedmanagerial responsibility interact with positive affectivity to predict theextent of emotion helping in the organization? The answer is yes: Asmodel 4 in Table 8.2 shows, the interaction between managerial respon-sibility and positive affectivity was significant (β = 0.44, p < .001). Toillustrate the significant interaction between positive affectivity and man-agerial responsibility, we grouped the employees into three categories(low, medium, and high) on the managerial responsibility score thatranged from 40 to 100. As Figure 8.1 shows, at low levels of managerialresponsibility, there is no discernible difference between those low and


20

16

12

8

4

0

-12 -9 -6 -3 0 3 6 9

High

Medium

Low

Emotion Help

Figure 8.1. Relationship between high, medium, and low managerialresponsibility, positive affectivity, and emotion help.

high on positive affectivity on the extent to which they provided emo-tional help, but positive affectivity differences are discernible for thosewith medium levels of managerial responsibility, especially for those withhigh levels of managerial responsibility.

Thus, individuals high in positive affectivity, to the extent that theyhave the discretion that comes with managerial responsibility, are active inusing their social skills to help ameliorate the effects of negative emotionsin the workplace. Was there also support for hypothesis 5’s prediction thatincreased managerial responsibility would interact with self-monitoringto predict the extent of emotion helping in the organization? The answeris yes. As model 4 in Table 8.2 shows, the interaction between managerialresponsibility and self-monitoring was significant (β = 0.24, p = .01). Theinteraction plots in Figure 8.2 resemble those in Figure 8.1 in showing anincreasing effect of self-monitoring on the extent of emotion helping asmanagerial responsibility increases.

In support of the interactionist approach, the addition of the interactionterms involving personality traits and managerial responsibility (model 4in Table 8.2) significantly improved explained variance over the main-effects model – by 18 percent to an adjusted total of 63 percent. Becauseself-monitoring is a complex construct, it is sometimes useful to exam-ine whether subscales (representing underlying factors) contribute to thestatistical effects. Therefore, we conducted a post hoc analysis to explorethe effects of two subscales (Public Performing and Other-Directedness)


15

12

9

6

3

0

-7 -5 -3 -1 1 3 5 7

High

Medium

Low

Emotion Help

Self-Monitoring (centered)

Figure 8.2. Relationship between high, medium, and low managerialresponsibility, self-monitoring, and emotion help.

identified in previous studies (Briggs and Cheek, 1988; Kilduff, 1992;Miller and Thayer, 1989) on the dependent variable. The analysis sug-gested that, as far as the interaction with managerial responsibility wasconcerned, it was the public performing aspect of self-monitoring that wasassociated with the significant tendency to provide emotional help (β =0.28, p = .001, model adjusted R2 = .66), whereas the interaction ofmanagerial responsibility with other-directedness was not significant (β =.54, ns, model adjusted R2 = .59).

We were intrigued by the suggestion of two anonymous reviewers thatthe higher the individual’s centrality in the friendship and workflow net-works, the more personality effects would be unlocked with respect to theextent of emotion helping. Indeed, post hoc results show that workflowcentrality significantly (p < .05) interacted with both positive affectivityand with self-monitoring to predict the extent of centrality in the emotionhelping network (even when the significant effects of managerial respon-sibility interactions with the personality variables were included). Therewas no significant effect of the interaction between friendship centralityand the personality variables.

In summary, the full model (model 4) suggests that those active inproviding emotional help to others in the workplace tended to be women,tended to have ties of friendship to many people, and tended to possessa combination of managerial responsibility and a high self-monitoring orhigh positive affectivity disposition.


Discussion

The results support an interaction approach to the understanding of cen-trality in emotion helping in organizations. As managerial responsibilityincreased, self-monitoring and positive affectivity effects on the extentof emotion helping also tended to increase. The significant interactionsbetween managerial responsibility and the personality variables providesupport for the idea that the increasing discretion that accompanies man-agerial responsibility unlocks discretionary behavior related to personal-ity differences (cf. Anderson and Thompson, 2004; Barrick and Mount,1993).

There were no significant main effects for self-monitoring and positiveaffectivity on centrality in the emotion helping network in the absenceof the interactions. This should not definitively rule out the main-effecthypotheses, however, given evidence that positive affectivity is generallyrelated to helping behavior (see Brief and Weiss, 2002, for a review)and given emerging theory and evidence that high self-monitors, rela-tive to low self-monitors, strive to provide more help than they receive(Flynn et al., 2006). The zero-order correlations in Table 8.1 betweenpositive affectivity and emotion helping centrality (r = .26) and betweenself-monitoring and emotion helping centrality (r = .18) are similar toor exceed the meta-analytic estimates of correlations between Big Fivepersonality dimensions and measures of work performance (Barrick andMount, 1993; Hurtz and Donovan, 2000).

The finding that the high self-monitoring managers are more cen-tral in the emotion helping network than low self-monitoring managersenhances our theoretical understanding of self-monitoring and contrastswith recent claims that high self-monitors resemble sociopaths in theirgeneral unfitness for managerial responsibility (Bedeian and Day, 2004:689). We show that it is precisely managerial responsibility that revealsthe distinctive advantages high self-monitors bring to the task of pro-viding emotional help. To the extent that the managerial role involvesa pragmatic willingness to intervene to solve organizational problems,it is the high self-monitors who are likely to develop a repertoire ofwell-honed problem-solving scripts (Dabbs et al., 1980; Douglas, 1983)involving directive attention to employees’ behavioral symptoms. Highself-monitoring and high positive affectivity managers may actively reachout to those suffering negative emotions rather than wait for colleaguesto bring these problems to their attention. In our post hoc analysisof self-monitoring subscales, we found that the significant interactionbetween self-monitoring and managerial responsibility tended to derivefrom the public performing aspect of self-monitoring rather than theother-directedness aspect. This suggests that high self-monitors may have


been actively engaged in taking the initiative to help others rather thanpassively waiting to be asked to help.

Future research can investigate the intriguing possibility that high self-monitoring managers’ proclivity for extending help to others is part of anoverall high self-monitoring pursuit of organizational status, a possibilitysuggested in a recent review article (Gangestad and Snyder, 2000) and inrecent empirical research (Flynn et al., 2006). Whether managers high inself-monitoring try to enhance their own status by helping others deal withemotional problems is still unknown. It is possible that employees tendto seek out managers with the appropriate personalities for emotionalsupport and help. Further, it is not entirely clear that an individual’sstatus in an organization is enhanced by being perceived to be a solver ofothers’ emotional problems.

The significant effect of gender, a control variable in our study, is inter-esting in its own right. We found that women, relative to men, tended to bemore likely to enact the role of emotion helper. This finding is compatiblewith research showing that in contrast to men, women tend to be moreempathetic, more loving, and more able to perceive as well as expressnegative emotions (see the review in Baron-Cohen, 2003). There is somesuggestive evidence that women may be slightly better than men at per-ceiving emotion (Mayer, Caruso, and Salovey, 2000) and may, therefore,be more predisposed to offering emotional support. Providing sympathymay be more socially acceptable for women compared to men (Eaglyand Crowley, 1986). Furthermore, women are more likely than men todiscuss personal issues and to self-disclose their emotions (Caldwell andO’Reilly, 1982a), which in turn might encourage reciprocity.

Women, in our results, tended to have lower self-monitoring scoresthan men. This pattern is consistent with evidence from a recent meta-analysis that suggested the sex-related effects for self-monitoring to bepartially responsible for the persistence of the glass ceiling effect in termsof noted disparities between men and women at higher organizationallevels (Day, Schleicher, Unckless, and Hiller, 2002). Clearly, we needmore research on the differing effects of self-monitoring for men andwomen, particularly in light of new research showing that a high self-monitoring orientation provides women (but not men) with more influ-ence in workgroup contexts (Flynn and Ames, 2006). If women are morelikely than men to provide emotional help to fellow employees and if lowself-monitoring is an obstacle to promotability (Kilduff and Day, 1994)then an interesting implication emerges: Emotion helpers at the lowerlevels of the organization may tend to be women, whereas those at higherlevels may tend to be high self-monitoring men.

The results for the friendship centrality control variable are also inter-esting and contribute to social network research a new focus on how


individuals with central network positions are actively engaged in themaintenance of others’ emotional health. This focus adds to the robustfinding from the research literature that individuals with central networkpositions in terms of the diversity of ties tend to be healthier in terms ofmortality (Berkman and Syme, 1979) and resistance to infection (Cohenet al., 1997). Central network positions are typically associated in theresearch literature with personal advancement of individuals in terms offaster promotions (Burt, 1992; Podolny and Baron, 1997) and higherperformance ratings (Mehra et al., 2001). Our research adds to thesefindings a new emphasis on how centrality in the friendship networkinvolves responsibilities for others.

Limitations

A limitation of our study is that, because of the relatively flat organizationstudied, we were unable to investigate how far down the hierarchy man-agers’ involvement in emotional support tends to reach in organizationswith tall hierarchies. Also, we did not examine the possible consequencesfor individuals of emotion helping behavior. Handling the negative emo-tions of others can cause negative outcomes such as frustration, burnout,illness, and failed personal and professional relationships (Frost, 2003;Kahn, 1993; Meyerson, 2000). Finally, we did not examine the types ofnegative emotions for which individuals sought help. Our results suggestthat there might be two types of support systems at work: one that isinformal and based on social network relations, and another activatedby those with both higher formal responsibility and relevant dispositions.Future research efforts could examine the relative effectiveness of thesetwo support systems and whether these two systems provide differentforms of emotional help.

Practical Implications

We draw practical implications for the management of organizationsfrom the personality and social network results. With respect to person-ality interaction effects, the results indicate one important way in whichorganizational systems may be able to cope with the inevitable produc-tion of negative emotions in the workplace. Going beyond previous workshowing the general helpfulness of those high in positive affectivity (Carl-son et al., 1988) and the superior work performance of those high inself-monitoring (Mehra et al., 2001), our results suggest the onerous taskof dealing with upset employees may require employees who demonstrateboth an appropriate disposition and the requisite managerial discretion.A positive affectivity disposition provides many benefits to an individual


across domains as varied as work performance and health (Lyubomirskyet al., 2005). Similarly, high self-monitors tend to outcompete low self-monitors on a range of organizational outcomes, including faster promo-tions (Kilduff and Day, 1994). Our research suggests ways in which thesebenefits can be extended to the improvement of the well-being of othersin organizational settings.

However, there are likely to be different perceptions across and withinorganizations concerning whether the psychological contract betweenemployees and management anticipates the provision of emotional help(cf. Lester, Turnley, Bloodgood, and Bolino, 2002). To the extent thatemployees perceive the psychological contract to include the provisionof emotional help, those with managerial responsibility who are low onpositive affectivity or low on self-monitoring may be seen as less effectivein their managerial roles, and this could affect promotion possibilities tothe extent that employee expectations are endorsed by top management.In such organizations, ambitious people who are either low self-monitorsor low in positive affectivity might be advised to engage in other types ofextra-role behavior particularly valued by top management to the extentthat such behavior is compatible with their personality orientations. Forexample, to the extent that low self-monitors are “independent noncon-formists who think for themselves and stick to their beliefs” (Krosnickand Sedikides, 1990: 724), they might take on the valuable managerialrole of devil’s advocate, putting forward ideas that run counter to organi-zational norms (cf. Premeaux and Bedeian, 2003) and providing accuratepersonnel assessments in the face of pressure to be lenient (cf. Jawahar,2001).

The question has also been raised as to whether individuals occu-pying central positions in social networks benefit only themselves orwhether their activities also benefit the collectivity (Ibarra et al., 2005).Our research shows that for at least one measure of collective benefit –the number of others helped with emotional problems – individuals cen-trally located in the friendship network do contribute to the collectivegood. Thus, in organizations in which the emotional welfare of employ-ees is important, the encouragement of social networking clubs amongemployees (cf. Friedman, 1996) may take on added strategic importance.Indeed, meta-analytic results show that the density of informal expres-sive ties in team social networks predicts team performance and teamcohesion (Balkundi and Harrison, 2006).

Negative emotions are likely to affect work performance across a rangeof organizations from the industrial (e.g., Schachter et al., 1961) to theartistic (e.g., Maitlis and Ozcelik, 2004), and there have been calls forthose with managerial responsibility to take on the “obvious but impor-tant task of . . . handling interpersonal and personal problems of staff ”


(Ostell, 1996: 525). However, there are probably boundary conditionsaffecting the extent of emotion helping and, by extension, the role of per-sonality interactions affecting such emotion helping. We would expect tofind emotion helping in contexts in which emotional labor is an intrinsicpart of the work and in contexts in which there is a necessary coordina-tion across tasks. Thus, it is no surprise to learn that, for example, flightattendants try to improve the morale of depressed coworkers prior toflights (Hochschild, 1983) given the importance of emotional labor in theservice sector and given the importance of coordination between flightpersonnel. In contexts in which individuals pursue their own goals in theabsence of client interaction (e.g., the famous Lincoln Electric plant; cf.Handlin, 1992), a managerial emphasis on emotion helping is likely tobe less important.

Conclusion

The question of who deals with the negative emotions that can threatento overwhelm organizational initiatives (cf. Huy, 2002) and contributeto lower performance (Staw and Barsade, 1993) is an important one thatrelates to the health and well-being of organizational members. In thischapter, we have made a theoretical and empirical attempt at identifyingthose central in the emotion helping network in one organization. To theextent that organizations are arenas in which all human emotions arelikely to emerge, it is vital to understand the actions and motivations ofthose who help others manage the inevitable frustrations and stresses thatarise.

III

Network Dynamics andOrganizational Culture

9

Network Perceptions and Turnoverin Three Organizations

We have emphasized in previous chapters the importance of individuals’perceptions of networks within which they are embedded. In this chapter,we continue this theme, looking this time at the question of organizationalturnover. Do people who perceive each other as playing similar roles inthe organization tend to affect each other’s turnover decisions? And whatabout the attitudes of the people left behind when somebody leaves – howdo people react to the departure of those perceived to be their friends?These are the issues we address in two studies of turnover across threefast food restaurants.

Several reviews of turnover research (e.g., Griffeth, Horn, and Gaert-ner, 2000; Horn and Griffeth, 1995) have underscored the continuinginterest in this area. Models of turnover have become complex (e.g.,Steel, 2002), incorporating in excess of forty organizational, individual,and societal variables in at least one case (Mobley, Griffith, Hand, andMeglino, 1979). This complexity suggests the value of exploring newkinds of variables rather than clouding the picture with more variablesof the same nature. Research on the dynamics of voluntary groups indi-cates the general importance of considering social ties inside and outsidegroups in order to understand rates of turnover (McPherson et al., 1992).A case study of an organization in crisis illustrated the potentially dev-astating effects of turnover on the attitudes of those left behind (Burtand Ronchi, 1990), but without specifying the social psychology of theattitude formation process.

In general, psychological models of the turnover process tend to assumethat turnover occurs atomistically within a workgroup. Each person’sbehavior in a workgroup is considered a stochastic function determinedby various personal and situational characteristics attributed to the per-son. Once those attributes are known, a regression model predicts anindependent probability of each person leaving. In contrast to this work,we present two studies examining (1) whether there is a snowball effect

181

182 Network Dynamics and Organizational Culture

such that turnover itself causes more turnover (cf. Mowday, Steers, andPorter, 1982) and (2) what the effect of friends’ turnover is on the atti-tudes of those who remain behind.

Study 1

The network approach to the question of whether turnover itself cancause more turnover is illustrated by a snowball metaphor. A snowballdoes not randomly accumulate snowflakes in the area. Rather, snowadheres to the snowball in a discernible path. Similarly, the patterns ofturnover will not be independently distributed across any workgroup.People are not independent actors. They affect each other in their behav-ior. Moreover, the degree to which they affect each other depends onthe relationship between them. Social network analysis provides a frame-work for assessing these relationships and for predicting their effects onindividual members.

Role Equivalence in Informal Networks

A defining contribution to network analysis was the concept of struc-tural equivalence (Lorrain and White, 1971). In their landmark article,Lorrain and White proposed that actors could be grouped into similarcategories based on their patterns of interactions in a social system. Twopeople would be considered to have equivalent roles (or to be structurallyequivalent) if they talked to exactly the same other people (although notnecessarily to each other). To the extent that they talked to mostly thesame people, they would occupy similar roles to each other. If they talkedto no one in common, they would occupy very dissimilar roles. Breigerand his colleagues (Breiger et al., 1975) developed an algorithm for oper-ationalizing Lorrain and White’s theory. Since then, many studies haveused this algorithm (Arabie and Boorman, 1982) to identify and interpretinformal groups in social systems. This idea has been generalized to aconcept of role that is more directly relevant to organizations.

Sailer (1978) has argued that two people are equivalent in their roles ifthey communicate with equivalent others. Thus, two supervisors wouldbe equivalent to each other because they each communicate with a groupof equivalent linespeople and to equivalent middle managers. To theextent that they communicate with people who are in different roles,then these two supervisors would be less similar to each other. This differsfrom Lorrain and White’s concept of structural equivalence because Sailerdoes not require that two people talk to the same others in order to beequivalent in their roles. Just as supervisors do not have to supervise the

Network Perceptions and Turnover 183

A

B

D E F G

C

Figure 9.1. Hypothetical advice network of workgroups (B → A indi-cates that person B goes to person A for help and advice).

same exact people to be in equivalent role patterns, the informal rolestructure of any group in an organization is most appropriately assessedusing this Sailer modification of the original Lorrain and White definition.(Sailer’s modification has been formalized and expanded by White andReitz, 1983, 1985, who have coined the term regular equivalence todifferentiate it from Lorrain and White’s structural equivalence.)

To illustrate this Sailer concept, consider a group of seven organiza-tional participants who work together. Assume that over the years theyhave developed a pattern of whom they go to when there is a problemor when they have a question about work-related matters. This patternwill be termed the advice network. For this illustration, assume personsB and C go to A for help and advice, D and E go to B, and F and G go toC. This hypothetical advice network is depicted in Figure 9.1.

Note that this network parallels a typical formal organizational chart.This coincidence is intentional for demonstrative purposes, but oneshould not infer from this example that the informal organization usuallymirrors the organizational chart. Seldom does it do so.

Using the Sailer concept of equivalence (White and Reitz, 1985), onecan measure the equivalence, or at least the similarity, of each pair ofpeople in the hypothetical network. In this example, D, E, F, and G areequivalent in their roles. This is so because they go to the same others(B and C) and no one goes to them. On the other hand, B and C areequivalent because (1) they both go to A and (2) all those (D, E, F, G)who go to them are equivalent to each other.

The matrix in Figure 9.2 shows the results of the algorithm as appliedto this simple example in Figure 9.1. Scores of 100 (e.g., for the pair B, C),indicate that each member of the pair has a role equivalent to the other.A score of 0 indicates that the members of the pair have maximally


A

40 40 0 0 0 0

40

100 40 40 40 40

40 100

40 40 40 40

0 4040

100 100 100

0 4040

100

100 100

0 4040

100 100

100

0 4040

100 100 100

A B C D E F G

B C D E F G

Figure 9.2. Similarity scores between pairs in hypothetical networkdepicted in Figure 9.1.

dissimilar roles (such as the pair A, D, one member of which no onegoes to, the other of which goes to no one). Intermediate scores indicatevarying degrees of similarity in their patterns of communication. Forexample, B and A both have people coming to them. However, D and E(who go to B) are not equivalent to B and C (who go to A). Therefore,A and B will not be equivalent either; their similarity is attenuated by thedegree to which those going to them are dissimilar to each other. Thus,A and B are only moderately similar to each other (score = 40).

To emphasize a point, equivalence between two people is not basedon whether they go to each other. For example, D and G do not go toeach other; they do not even go to the same people. Yet they are perfectlyequivalent in roles (similarity score = 100). Conversely, D goes to B andyet is not very similar to B (similarity score = 40).

Importance of Perceived Similarity

Thus, each pair of actors in a workgroup can be evaluated as to howsimilar they are in their patterns of communication with fellow work-ers. In keeping with the theme of this book, we also emphasize that inpredicting the effect that a network of interactions has on any given indi-vidual’s attitudes, it is important to differentiate between the actual andperceived networks (cf. Burt, 1982). It is the network that is perceived bythe individual that enables that individual to evaluate whether he or sheis similar to a coworker. To the extent that they perceive each other tobe similar, then they are more likely to affect each other’s behavior.

For example, suppose that two supervisors viewed each other as equiv-alent. If one were to leave, the second is likely to view that leaving asrelevant information for himself or herself. The reevaluation process thatMowday and his colleagues (1982) propose would be activated in such a


situation. On the other hand, suppose that the two were quite dissimilarto each other in their informal communication patterns (e.g., one may bean isolated custodian while the second is an active middle manager). Inthis case, one’s leaving may be viewed as irrelevant. The two have littlein common, and as such little dissonance is created when one quits. Inthis way, role equivalence creates a bond between participants in a work-group. The role similarity is determined by the informal communicationpatterns. The resulting bond increases the effect that one’s behavior hason another.

The purpose of study 1 is to test the proposition that such similarityin informal communication patterns can increase the effect of turnoverbehavior on a coworker. Specifically, the more similar coworkers are toeach other, the more likely it is they will leave together (or stay together).If this is the case, then turnover should occur in clusters, which can bepredicted by the informal communication channels.

Hypothesis: Turnover will occur in clusters as defined by the perceivedsocial network, such that those who are perceived similar in position toeach other in the communication network will either stay together orleave together.

Methods

A questionnaire was administered to employees in three fast food restau-rants (N = 16, 27, and 20, respectively, in restaurants A, B, and C). Theaverage age of the employees was nineteen, with 73 percent being eigh-teen years old or less. Forty-eight percent of the employees were female.The only full-time (forty hours or more per week) employees were thestore managers; however, all the people in the sample worked at leasttwenty hours per week. The average job longevity for the employees wasless than eight months, although this number is skewed because turnoveraverages 200 percent per year.

The questionnaire asked each person in the workgroup to list his orher perception of whom people go to for help and advice at their restau-rant. The responses were used to assess how similar people are in termsof the role they had in giving and receiving advice from others. Thesedata enabled us to identify an informal leadership hierarchy beyond thatdescribed by the organizational chart. The directions in part for this sec-tion were as follows:

In this section, you will find several lists of people who workwith you. Each list is started with the question, “Who would thisperson go to for help and advice at work?” That is if this person


had a question or ran into a problem at work, who would theylikely go to ask for advice or help? Please answer the question byplacing a check next to the names of all the people the person islikely to go to.

The network data collected in this study were perceptual (following thesame methods used in Chapters 3 and 4). Each individual provided anentire picture of his or her perception of the social network in whichhe or she is embedded. This permits the testing of the perception-basedhypothesis directly.

The turnover data were collected during the one-month period follow-ing the questionnaire administration. All turnover was counted, includingone transfer and one involuntary turnover.

The reasons for including both of these events in the turnover countwere twofold. First, both of these turnover events resolved a personnelproblem in the organization. Had the turnover not occurred, a voluntaryseparation was likely anyway. Thus, the processes that led to the transferand to the involuntary termination were similar to voluntary termina-tions, leading one reasonably to expect that the effects they would haveon coworkers would be similar. The second reason that these events werenot excluded was that the disruption in the social network was just assevere. A person leaving creates a hole in the network, no matter whatthe reason. It is this disruption that, it is argued here, snowballs.

The hypothesis was tested using the Quadratic Assignment Procedure(QAP), developed by Hubert and his colleagues (Hubert and Schultz,1976). (See Chapter 3 for more details.)


Turnover in the three restaurants combined to 25 percent in the one-month time period under study (range: 20 percent to 38 percent in thethree sites). The differences in turnover rates, χ2 (2, N = 63) = 1.69,and questionnaire response rates (range 81 percent to 85 percent), χ2 (2,N = 63) = .13, were not significant among the three restaurants.

The hypothesis predicted that employees would leave in clusters thatwould map onto perceived role similarity clusters. This hypothesis wastested using three separate approaches with different assumptions under-lying each approach. The first used all the information available: It testedthe hypothesis for each individual employee’s perception of the social net-work. That is, each person had a perception of the entire network. Eachperceived network is translated into a role similarity matrix using Sailer’sdefinition. Each similarity matrix can then be compared to the turnovermatrix and tested to see whether the two are significantly related.


Although this test took advantage of all the information, it was sub-ject to methodological criticism, because the meta-analysis was based oncorrelated observations (i.e., each individual was observing the “same”network). The second test eliminated the statistical independence prob-lem, but in the process destroyed the perceptual map information. Thistest used one representation of the network for each site by aggregatingperceptions of whether person i goes to person j. This single representa-tion then was tested against the turnover matrix.

The third test represented a compromise, restoring much of the percep-tual information and retaining statistical integrity for the significance test.First, individual perceptual maps are translated into similarity measures,as done in the first test. Then these similarity measures are aggregatedinto one similarity matrix, as in test 2. This summary matrix for each siteis tested against the turnover matrix for significance.

Before we proceed to the specifics of the testing procedures, we shouldnote that the aggregations for tests 2 and 3 do not alter the basic dyadicnature of this study. The aggregations are of people’s perspectives. Thedyadic relation between each (i, j) pair is still the unit of analysis thatforms the basis for the significance tests that follow. As we will show, therobustness of the findings is emphasized by the similarity in the results ofeach test.

Test 1: Individual Perceived NetworksThe first step was to determine the perceived similarity between pairs ofcoworkers. Let k represent the respondent who filled out the question-naire, i represent the coworker who seeks advice at work, and j representthe coworker who potentially could be approached by i for advice. Then,let A(i, j, k) (referring to the raw advice matrix) be a matrix of dimensionN × N × K (that is, K is the number of respondents, N is the number ofcoworkers, including K), such that A(i, j, k) = 1 if k perceives that i goesto j for help or advice at work, and A(i, j, k) = 0 otherwise. This matrixcan be separated into k adjacency matrices, A(i, j), each representing whogoes to whom for advice as perceived by k. Using the algorithm proposedby White and Reitz (1985), each A(i, j) is transformed into a RegularSimilarity matrix, RS(i, j), representing k’s perception of how similar (inthe Sailer sense) i is to j. If the hypothesis is correct, then each individ-ual’s map of the role similarity should closely correspond to the turnoverclusters. This was tested by creating a Turnover matrix (N × N) whosecells T(i, j) = 1 if i and j both either left or stayed (that is, both i and jbehaved in similar ways), and whose cells T(i, j) = 0 if either i or j (butnot both) left (that is, i and j behaved dissimilarly). This matrix was thencompared to the RS(i, j) matrix for Person k to see whether, in general,their similar behavior in turnover (the 1s in the T[i, j] matrix) matched


the perceptions that k had of how similar the two people (i and j) werein their roles in the advice network. QAP yielded a normalized statistic(expressed in Z scores), which would be large to the extent that this matchup was greater than would occur by chance reassignment of turnovers(i.e., if a different group, but the same number, of people had left). Inaddition to the significance test, a gamma (Goodman and Kruskal, 1963)was calculated between the matched cells of the two matrices, RS(i, j)and T(i, j), where i does not = j. Gamma, a nonparametric correlationmeasure, is particularly appropriate here, because one of the matrices iscomposed of a dichotomous variable turnover.

The results of these tests are summarized in Table 9.1. Each person (k)was tested against the T(i, j) matrix, producing a Z score and a gamma.For each site, the gammas and the Z scores were averaged. Because thestandard error was estimated, a t test was used to test the null hypothesisthat, on the average, no relation exists between turnover clusters andperceived similarity. Each site was considered an independent test. Theoverall relation was assessed using Rosenthal’s (1978) suggested meta-analysis approach. Each of the three tests was transformed into a Z scorethat corresponds to the same significance level. These three scores werethen summed and divided by the square root of N (N = 3 sites), yieldinga joint Z. The overall significance level is determined by this joint Z valuein the normal distribution.

The overall gamma is the simple average of the three sample gammas.It should be noted that correlation statistics, such as gamma or Pearson’sr, are not normally distributed, but rather are skewed toward 0. Thisleads to conservative estimates of the population parameter when simpleaverages are calculated, as is done in Table 9.1. Also, in the interest ofconservatism, the correlations are not weighted by the sample N sizes. Thelargest sample had the strongest correlation (as is true in practically allthe analyses reported in this work), which reduces the magnitude of whatmight be considered the appropriate overall strength of association. Thesize of both of these biases is not substantial, however, and the resultingoverall average can be considered a reasonable, if somewhat conservative,estimate.

Given these caveats, a discernable trend is observed in Table 9.1. Two ofthe three sites show significant relations between the pattern of perceivedrole similarities and turnover. Combining these results, the significancelevel is persuasive (less than .0005). The strength of these results, onthe other hand, is modest. Ranging from 0 to .23 by group, the averagegamma is only .10.

Thus, using individual maps of the perceived role similarities in thetest of the hypothesis, it is concluded that the null hypothesis of norelation is rejected in favor of a positive relation. The strength of thatrelation is questionable. One may wonder how an impressive significance


Table 9.1. Quadratic AssignmentProcedure Test of Association betweenIndividual Perceptions of RoleSimilarity and Turnover

Group and Measure Result

Group AZ .055N 13SD .9629SE .2780t .1993p .NSγ .00

Group BZ 1.907N 23SD 1.9725SE .4205t 4.536p <.0001γ .23

Group CZ .515N 17SD 1.0228SE .2557t 2.013p <.03γ .07

Meta-AnalysisN 3Joint Z 3.37p <.0005γ .10

level can be associated with such a modest correlation and N size. Theanswer lies in the fact that the standard error of the average associationsbetween individuals’ perceived maps and turnover is quite small. That is,the association may not be strong, but practically all the subjects “agree”it was there. This tight standard deviation leads to a small standard errorand thus highly significant mean.

Test 2: Local Aggregated Networks and Turnover ClustersIt may be argued, and justifiably so, that averaging individuals’ rela-tionships between perceived networks and turnover clusters is aninappropriate test of the hypothesis, because these observations are not


independent of one another (i.e., each person is presumably perceivingthe “same” social network). A more appropriate test, this logic contin-ues, would be to summarize the perceived network in one matrix andtest that one matrix against the turnover matrix for the particular site. Asan alternative test of the hypothesis, then, the following procedure wasundertaken. The transformation of information for the Role Similaritymatrix will be described first.

Recall that A(i, j, k) is the original data matrix containing all of k’sperceptions of whether i goes to j for advice, for all i and j (i not = j). Letthe Local Aggregated Advice matrix, or LAA, entries LAA(i, j) = 1 if andonly if both A(i, j, i) = 1 and A(i, j, j) = 1. Let LAA(i, j) = 0 otherwise.In other words, if i and j both agree about the fact that i goes to j for helpand advice at work, then in the summary matrix LAA cell, (i, j) = 1. Ifthey disagree, or if they agree that i does not go to j, the cell is set equal to0. This strict intersection rule is consistent with the conservative stancerepeatedly adopted in this research effort. If one person claims to begoing to another for advice, but the second person denies it, such a claimis considered unreliable. When both agree that the first goes to the second,it is reasonable to assume that the connection actually does take place.

With the same algorithm used to create the RS(i, j) matrix, this sum-mary matrix LAA(i, j) is transformed into a Local Aggregated RegularSimilarity matrix, LARS(i, j;). Again, the cells take on continuous valuesfrom 0 to 100, where 0 indicates that i and j are totally dissimilar intheir roles in the advice network, and 100 indicates that the two haveidentical roles. We test the hypothesis by comparing this matrix with theT(i, j) matrix. Each of the three sites yields one summary test (instead ofaveraging the tests of each individual’s matrix). The QAP test results aregiven in Table 9.2.

Groups B and C show significant, if again somewhat weak, associationsbetween the role similarity matrix and the turnover similarity matrix.Overall, the relation is reconfirmed as before: significant (p < .05) butnot strong (unweighted average gamma = .16).

Test 3: Average Perceived SimilaritiesAlthough the preceding computations of local-aggregated networksaddressed the problem of independent observations, they did so at con-siderable cost. It was argued earlier that it is the individual’s perceptionof the network that affects him or her, and not links in the network asdetermined or perceived by others. The use of traditional local-aggregatednetworks abolishes the perceived network, beyond the individual’s localinput.

To take advantage of all the perceptual information focused on anindividual, we performed a third test. Instead of aggregating the raw data


Table 9.2. Quadratic AssignmentProcedure Test Association betweenLocal Aggregated Network of RoleSimilarity and Turnover


Group AZ −1.23γ −.11p ns

Group BZ 2.98γ .34p <.002

Group CZ 1.70γ .24p <.05

Meta-AnalysisZ (Joint) 1.99γ .16p <.05

and then transforming to similarity (or distance) matrices, we performedthe aggregation this time just prior to the QAP test.

Recall that RS(i, j, k) is the matrix of role similarity scores perceivedby k between i and j. The Average Perceived Role Similarity, or APRS,is the average of i and j’s overall perception of how similar i and jare. Then let APRS(i, j) = (RS(i, j, i) + RS(i, j, j))/2. This summarymatrix, then, retains all the perceptual information calculated in A(i, j,k), distorted only to the extent that i and j disagree on their mutualsimilarity.

Table 9.3 displays the results of the QAP test between the summarymatrix APRS(i, j) and T(i, j) for each site. One of the three sites (group B)is significant. However, once again, the overall results were convincinglysignificant (p < .004) but not strong (average gamma = .15).

Study 1 Discussion

Three separate analyses pointed to the same conclusion: Turnover doesnot occur randomly throughout a workgroup. Rather, it is concentratedin patterns that can be delineated by role similarities in a communica-tion network. Although these are not the only analyses possible, the fact


Table 9.3. QAP Test of Associationbetween Average Perceived RoleSimilarity and Turnover


Group AZ .83γ .11P ns

Group BZ 4.66γ .42P <.000l

Group CZ −.79γ −.09p ns

Meta-AnalysisZ (Joint) 2.71γ .15p <.004

that this snowball pattern was so consistent across the three analytictechniques presented here emphasizes the robustness of this finding.

Of interest to the manager in these results are the implications they holdfor dealing with turnover phenomena in the organization. Frequently,interventions to reduce turnover are spread out over organizational levels(see, for example, Krackhardt, McKenna, Porter, and Steers, 1981). Amore cost-efficient approach might be to concentrate resources on thosewho are observing similar coworkers leave. In doing so, expenses ofturnover reduction programs could be minimized.

Before projected savings of such a strategy are calculated, however, it isimportant to recall the relative strength of the findings reported here. Thesnowball effects were attenuated by two major factors. First, methodolog-ical issues must be considered. The single-item measure of connection –which is the building block of our role equivalence assessment – maybe unreliable. Unsystematic error may be present in the data, reducingany observed correlations. Given the lack of power gained from usingholistic perceptions, it may be preferable in future research to invest inmultiple indicators of local connections rather than attempting to captureeach individual’s perception of the entire network. In a similar vein, thecorrelation statistic used here, the gamma, does not have the usual upperbound of +1 in such matrices. Thus the apparent low values may bemisleading. At the least, they are conservative.


The second significant factor contributing to lower-than-expected cor-relations is inherent in the nature of the problem being studied. Turnoverdecisions are determined by a complex set of variables (Mobley, 1982;Steel, 2002), not simply whether a certain coworker leaves. The datareported here suggest that turnover of role-equivalent coworkers con-tributes only a part of the picture.

Left unstudied is the effect that turnover has on people left behind inthe organizations, and it is to this that we turn next.

Study 2

Does an individual who decides to continue working for an organizationafter a coworker leaves become more negative or more positive toward theorganization? Prior research suggests, on the positive side, that turnovercreates internal promotion opportunities for those who remain (Daltonand Todor, 1979; Staw, 1980a). Further, employees may increase theirsatisfaction with the job and the organization to justify their own decisionto stay (as suggested by Mowday, 1981). Also, if those who left were poorperformers, those who stay on are likely to benefit and be more satisfiedwith their jobs (cf. Mowday et al., 1982).

Conversely, turnover could leave behind more discouraged, less satis-fied coworkers. Each of the reasons for positive consequences previouslymentioned could be turned around to predict negative consequences. Forexample, the termination of a coworker could require more work of thosewho remain to make up for the work not being accomplished by the per-son who left (cf. Mowday et al., 1982). This would be particularly trueif the person who left was a valued employee.

Particularly critical may be the social relationship the stayer has to theleaver: When the person leaving is a close friend, the effect on the stayer“may be particularly traumatic” ( Mowday et al., 1982: 148; cf. Burt andRonchi, 1990). Perhaps the most useful model to organize the possibleoutcomes of this dynamic social interaction process is Heider’s (1958)balance theory. In this model, a triangle of relationships is describedbetween an observer (self), another person, and an object of commoninterest. In this case, the observer (stayer) is faced with a coworker (whois a friend) and the job (see Figure 9.3). For the purpose of exposition,it is assumed that the link between each pair of vertices is positive priorto the departure of the coworker. That is, the triangle is balanced: Theobserver has positive affect for the job, the observer has positive affectfor the coworker, and the coworker has positive affect toward the job.

How this triangle might change (or not change) as a result of the termi-nation of the coworker is depicted in Figure 9.3 (effects A, B, and C). In


AFTER TURNOVER

Effect A: External Attribution

Friend

Effect B: Dissonance Reduction

Friend

BEFORE TURNOVER Friend

Effect C: Insufficient Justification

Friend

Self

Self

Self

Self

Job

Job

Job

Job

No Change

Negative Attitude Change

Positive Attitude Change

Figure 9.3. Possible effects of turnover of friend on stayer.

each of these predictions, it is assumed that positive attitudes held towardthe friends remain, or at least do not become negative. This assumptionis supported in friendship studies, where such links are generally stableover long periods of time (e.g., Newcomb et al., 1967).

The first prediction is that no change in attitude toward the job wouldoccur. This could happen if the employee attributed exogenous reasonsto the friend’s departure (effect A). In this way, an attribution of job


satisfaction to the friend can be maintained in the face of the friend’sleaving (e.g., “My friend liked the job, but she had to leave because ofschool”). Mowday (1981) proposed a similiar argument to explain hisresults, referring to such external attributions as the “pull” forces ofturnover.

Effect B in Figure 9.3 depicts the possibility of a negative change inattitude resulting from a friend’s leaving. In this scenario, the employeeattributes dissatisfaction to the friend who left. This creates dissonance,which is resolved in the triangle by the stayer becoming more dissatisfiedwith his or her job.

Effect C represents a possibility not predicted directly from balancetheory but that has some support in dissonance studies. If the personobserves a coworker leaving and attributes dissatisfaction to the leaver,then the person’s decision to stay may require more justification (Staw,1976; 1980b). One way this justification could occur is for the stayer todevelop more and stronger positive attitudes toward the workplace.

Turnover Embedded in Network Structures

These scenarios represent the possibilities at the micro level between twopeople and their job. However, the workplace is seldom restricted totwo people in their organization. Instead, each of N employees mustbalance N-1 such triangles in his or her head. Few probably actually doso, but it is likely that such forces on a person’s psychology are to someextent additive, at least figuratively. That is, if many of a person’s friendsleave, then the effects described in Figure 9.3 are likely to be strongerthan if only one friend leaves. Moreover, the closer the friends are tothe person, the stronger the effect is likely to be. Viewed from a moremacro perspective, this phenomenon dictates that effects of turnover onstayers will not be uniformly nor randomly distributed among the stayersin the organization. Rather, these effects will be localized and focusedon those stayers who are closest to those who left. The social network,then, describes the topology of forces that reverberate throughout anorganization when someone leaves (Burt, 1977; Lewin, 1936).

The friendship network in Figure 9.4 illustrates this proposed effect.Each letter represents an employee; a line connecting two employees indi-cates that the two employees are friends. Thus, A is a friend of B and Cbut not a friend of the remaining employees (D through H). If A were toleave, it is proposed that B and C would be most strongly affected.

A person who is not a friend but is seen as a friend of a friend is moreapt to have more influence than someone who is not seen as a friend ofa friend. By extension, one is more affected by a friend of a friend of afriend than by someone further out in the friendship chain. Thus, it is


A

B

C

D E F

G

H

Figure 9.4. Hypothetical friendship network.

proposed that A’s termination would affect D more than E and that Hand G would be least affected.

Another contextual effect must be considered when moving from sim-ple dyads to the entire network. An individual is influenced by thosewho stay as well as by those who leave. That is, in Figure 9.3, if theperson’s friend does not leave, then the triangle in “Before Turnover” isreinforced. If many of the coworkers who remain are friends and onlyone friend leaves, then the impact that this termination will have on theindividual will be attenuated.

This balancing effect of leavers versus stayers is depicted in Fig-ure 9.5. Four extreme scenarios are represented. In each case, personA has eight coworkers, four of whom leave. Scenario 1 (in the upper-leftcorner of Figure 9.5) predicts the maximum impact on person A of thefour turnovers. That is, because A is close to all four leavers and not closeto any of the four stayers, whatever impact the turnover will have wouldbe relatively large. At the other extreme (scenario 4), when A is close tothe stayers and not close to the leavers, the impact of the turnovers wouldbe less. Scenarios 2 and 3 represent two more moderate effects; however,they represent moderate positions for different reasons. In scenario 2,the impact is neutral because each of the actors is not connected (eitherdirectly or indirectly) to A; thus, there is little impact from either stayersor leavers. In scenario 3, the relatively strong impact of those who left isbalanced by the impact of an equal number of coworkers who stayed.

To be consistent with the psychological foundation of this book, how-ever, we must make one final modification to the preceding structuralarguments. Person A’s leaving will affect person B, assuming that personB perceives that person A is a friend. The effect is attenuated if personB perceives that person A is only a friend of a friend, and so on. Forexample, in Figure 9.4, if person D does not perceive person A to be afriend of B or C, and thus person C sees no connection at all between

Clo

se to

Lea

vers

Not

Clo

se

to S

taye

rs

Clo

se to

S

taye

rs

Not

Clo

se to

Lea

vers

2. N

eutr

al Im

pact

on

A

4. L

owes

t Im

pact

on

A

1. H

ighe

st Im

pact

on

A

3. N

eutr

al (

Bal

ance

d)

Impa

ct o

n A

A AAA

Circ

led

dots

repr

esen

t Lea

vers

. U

ncirc

led

dots

repr

esen

t Sta

yers

. A

line

con

nect

ing

a do

t (co

wor

ker)

to A

indi

cate

s th

at a

per

ceiv

es th

e co

wor

ker t

o be

a c

lose

frie

nd.

Figu

re9.

5.Fo

urex

trem

esc

enar

ios

depi

ctin

gva

riou

sde

gree

sof

impa

ctfr

omle

aver

s.

197


herself and A, then the effect of A leaving will not be felt by D, eventhough in “reality” A is connected indirectly to D. Specifically, the effectof turnover on coworkers will depend, it is hypothesized, on how closein the friendship network the leaver was to the stayer as perceived by thestayer. We examined the relationship between turnover and subsequentorganizational attitudes of those who remained and, in particular, howthis relationship was moderated by the perceived position of leavers inthe friendship network.

Methods

The three-restaurant sample was identical to that described in study 1. Apre-post natural quasi-experimental design was used. At time 1, a ques-tionnaire was administered that included network questions and attitudeitems. One month later, at time 2, a second questionnaire with the atti-tude items was administered. The major treatment variable, turnover,was recorded during the interval between time 1 and time 2 at each ofthe sites. Using this design, we could determine the relationship eachrespondent had to each of the coworkers who left, and we could assessthe degree of change in stayers’ attitudes subsequent to the turnover oftheir coworkers.

Our analysis this time focused on perceived friendship rather thanadvice relations. Thus, we were interested in each person’s perceptions ofwho was a friend of whom, using the same question outlined in Chap-ter 3. Each person provided an entire picture of his or her perception of thefriendship network in the restaurant in which he or she was embedded.These data allowed us to construct, for example, Henry’s perception ofthe entire network in the group, Rita’s perception of this network, andso on.

Independent VariableFrom these data, we calculated how close in the perceived friendshipnetwork the respondent perceived himself or herself to be to eachother coworker. These friendship links were combined with subsequentturnover data to create the independent variable in this study – theIMPACT index. This variable is a summary indication of how muchpotential influence there is on an individual stayer (k in the following for-mula) from friends who terminated, relative to those friends who stayed(see Figures 9.4 and 9.5).

IMPACTk =N∑

j=1

[FDk−1 ( j) × T ( j)].


for all j not = k; where

IMPACTk = the potential influence of the leavers relative to

the stayers

F Dk−1 ( j) = person k’s perceived friendship closeness to

coworker j (closeness = 1/distance)

T( j) = turnover of coworker j ( = 1 if j left; = −1 if

j stayed)

Dependent VariablesOrganizational CommitmentThis was measured by the fifteen-item Likert-scale Organizational Com-mitment Questionnaire (Mowday, Steers, and Porter, 1979) that assessesthe degree to which the employee is committed to the firm for which he orshe works. It has been shown to predict turnover reliably and consistently(Mowday, Steers, and Porter, 1979).

Relative Job Satisfaction (Self)One section of the questionnaire assessed perceptions of how satisfiedemployees were. The respondent was instructed to place beside each indi-vidual’s name a number that indicated the relative amount of job satisfac-tion that individual had (e.g., a “1” beside the most satisfied coworker, a“2” beside the next most satisfied, etc.). Where the respondent placed himor herself in this ranking provided an indication of his or her satisfactionrelative to the other coworkers. This self-ranking score was then trans-formed into a percentile by reverse scoring and normalizing. In assigninga rank to those people who had already left, the respondent was to recall“how satisfied they were just before they left this job.”


A rank-order correlation was used (Goodman and Kruskal’s gamma) toassess the degree to which individuals attributed relatively more dissatis-faction to those who left than they did to those who stayed. This com-parison was done within individuals. That is, each respondent receiveda score equal to his or her gamma, indicating his or her associationbetween coworkers leaving and the relative job satisfaction that he orshe attributed to those coworkers. These gammas were then averaged foreach group.

Because, following Figure 9.5, three different predictions were madeabout the effect that turnover of friends might have on those who


Table 9.4. Associations within Individuals between Turnover ofOthers and Attributed Job Satisfaction

Attributed Attributed � AttributedSample Satisfaction T1 Satisfaction T2 Satisfaction

Group A (N = 16) γ = −.38 γ = −.45 γ = −.21N = 13 N = 8 N = 8SD = .201 SD = .290 SD = .317

Group B (N = 27) γ = −.34 γ = −.51 γ = −.18N = 23 N = 15 N = 15SD = .392 SD = .365 SD = .364

Group C (N = 20) γ = .01 γ = −.10 γ = −.15N = 17 N = 12 N = 12SD = .282 SD = .363 SD = .365

Weighted average γ = −.24 γ = −.36 γ = −.18association SD = .310 SD = .347 SD = .354

remained, two-tailed tests were used to evaluate the significance of therelationship observed in these links. Overall correlations were calculatedbased on the pooling of the information from all three sites, rather thanthe simple averaging of the three individual correlations (cf. Hunter,Schmidt, and Jackson, 1982), for two reasons. First, the dependent vari-ables – organizational commitment and satisfaction – were all normalizedfor group size and may be considered continuous variables. Second, theindependent variable, IMPACT, is theoretically meaningful in its abso-lute form. If a clique of close-knit friends exists and all members butone leave the clique, the effect on the remaining group member shouldbe most pronounced, even in comparison with employees in other workgroups. To average the three within-site correlations would destroy thisinformation.

The Organizational Commitment Questionnaire (OCQ) ranged in reli-ability (Cronbach’s alpha) from .84 to .90 across the three sites in bothadministrations. In addition, both the job satisfaction (self-ranked) mea-sure and the OCQ at time 1 predicted subsequent turnover. Correlationswith turnover ranged for the three sites from −.16 to −.52 for the twomeasures. Thus, these instruments displayed both reliability and predic-tive validity properties that were acceptable and consistent with priorassessments of similar measures (Mobley, 1982; Mowday et al., 1979).

The results, shown in Table 9.4, indicate that people do attributemore dissatisfaction to those who leave. The overall gamma indicatesthat stayers rank those people who left lower in satisfaction after theyleave (gamma = −.18). The lack of a particularly strong relationshipis explainable by referring to the first two columns of Table 9.4. It is


Table 9.5. Correlation of IMPACT with OrganizationalCommitment

Sample at T1 at T2 � (T2 – T1)

Group A r = .21 r = .35 r = .34N = 8 N = 8 N = 8p = NS p = NS p = NS

Group B r = .49 r = .16 r = −.17N = 21 N = 15 N = 15p < .05 p = NS p = NS

Group C r = .41 r = 39 r = .00N = 13 N = 12 N = 12p = NS p = NS p = NS

Combined r = .51 r = .34 r = −.04N = 42 N = 32 N = 35p < .001 p < .05 p = NS

Note: Probability levels are all based on two-tailed tests.

clear that at time 1 employees were able to predict who was dissatisfiedenough to leave (average gamma = −.24). At time 2, this predictabilitywas somewhat stronger (average gamma = −.36). That is, employeeswere attributing dissatisfaction to those who left; however, the changein dissatisfaction ratings was attenuated by the fact that by time 1 theyalready had been able to anticipate who would leave. (Although the sig-nificance of these scores could be tested, this would be inappropriate,because the errors are not independent in this case [cf. Box, Hunter, andHunter, 1978: 78–82]; therefore, the data are provided for descriptivepurposes only.)

IMPACT was a summary index of the relative closeness of the individ-ual to those who left. A relationship between this index and organizationalcommitment would indicate that commitment was differentially affectedby leavers depending on how close the leavers, relative to stayers, wereperceived to be to the employee. As with attributed satisfaction, reportedpreviously, the relationship between IMPACT and commitment is shownfor time 1, time 2, and changes between time 1 and time 2 in Table 9.5.The changes in commitment were inconsistently related to IMPACT, as isevident from the last column in Table 9.4. In group A, the correlation wasmoderate (.34), although it was insignificant. Groups B and C showedweaker correlations. The combined correlation was practically zero.

On the other hand, the correlations between IMPACT and commitmentat time 1 and time 2 were significant and relatively strong (.51 and .34).Thus, commitment does seem to be related to the degree to which friendsleave. The direction of this relationship is particularly interesting: The


Table 9.6. Correlation of IMPACT with Job Satisfaction(Self-Ranking)

Sample at T1 at T2 � (T2 – T1)

Group A t = .32 r = .34 r = .02N = 8 N = 8 N = 8p = NS p = NS p = NS

Group B r = .08 r = .36 r = .40N = 21 N = 15 N = 15p = NS p = NS p = NS

Group C r = .04 r = .42 r = .50N = 13 N = 11 N = 11p = NS p = NS p = NS

Combined r = .11 r = .35 r = .38N = 42 N = 34 N = 34p = NS p < .05 p < .05

Note: Probability levels are all based on two-tailed tests.

closer in friendship distance the leavers were to the respondent, the higherthe degree of respondent commitment, both at time 1 and time 2. Itis difficult to ascertain a causal direction in this relationship, however,because the change in commitment score was negligible.

It was predicted that the turnover of close friends would result indifferential satisfaction in stayers. In other words, IMPACT would berelated to the change in satisfaction. As can be seen in Table 9.6, in eachof the three groups a positive correlation existed between IMPACT andthe change in self-ranked satisfaction (although none reaches significancebecause of small N sizes). The overall relationship is .38 and is significant.Thus, it would appear that when closer friends leave, the person who staysbecomes even more satisfied relative to the other stayers.

The relationship at time 2 between satisfaction and IMPACT is alsopositive (.35). At time 1, the relationship is even weaker (.11) and notsignificant. Because the change in satisfaction is more strongly correlatedwith IMPACT than satisfaction at either time 1 or time 2, it wouldappear plausible that the turnover of friends may contribute to the jobsatisfaction of stayers.

Our primary concern here was to assess the net effect turnover had onthose who remained. To this end, we presented the relationship betweenjob attitudes and IMPACT, a summary index representing the effectsillustrated in Figures 9.4 and 9.5. Thus far, we have ignored in this analy-sis the role of attributed satisfaction. The following results explore morefully the extent to which attributed satisfaction moderates or explains theobserved relationship between IMPACT and the dependent variables.


One question is whether the correlation between IMPACT and atti-tudes is spurious due to attributed satisfaction. As shown in Table 9.4,attributed satisfaction is a reasonably good predictor of turnover. Thus,it is possible that IMPACT is a surrogate for the influence from those whowere dissatisfied. To test this, hierarchical regressions were performed onthe set of six dependent variables (self-ranked satisfaction and organi-zational commitment at time 1, time 2, and the change from time 1 totime 2) on the data combined across the three sites (Cohen and Cohen,1983). Forced to enter at step 1 of the regression was a new compos-ite variable, FD∗AttrSat, which was similiar to IMPACT except that thebinary turnover component was replaced with the extent to which theattributed satisfaction of the coworker was above or below the median(50 percent). In other words, this index was strongly positive when theclose coworkers were relatively satisfied; it was strongly negative when theclose coworkers were relatively dissatisfied. (See Krackhardt and Porter,1985, for formulae for the additional attributed satisfaction measuresdiscussed here.)

At step 2 of the regression, IMPACT was added to the equation. IfFD∗AttrSat is a source of spuriousness, then it would be significant atstep 1 of the hierarchical regression, and the addition of IMPACT wouldnot improve the R square significantly.

Of the six hierarchical regressions, only one (where commitment at time1 was the dependent variable) was significant at either step 1 or step 2.The fact that commitment at time 1 emerged as the most significant find-ing is not surprising given the results reported previously in Tables 9.4and 9.5. The N size was larger at time 1, reducing the standard error ofthe estimates in the regression. Moreover, at time 1, self-ranked satisfac-tion was not related to IMPACT, whereas commitment showed a strongcorrelation (.51). Although the pattern of coefficients was similar acrossall six regressions, only the significant regression is detailed here.

The results, shown in Table 9.7, indicate strong support for theIMPACT index over and above attributed satisfaction as a correlate ofcommitment. FD∗AttrSat is not significantly related to commitment ateither step 1 or step 2. The amount of variance explained by addingIMPACT, however, is substantial (�R2 = .251, p = .001).

A second question of interest is whether attributed satisfaction inter-acts with turnover to influence attitudes and whether IMPACT explainsany variance over and above the satisfaction-turnover interaction. Onecould reasonably expect that satisfaction would moderate the effect thatturnover has on stayers. We could observe this interaction by separat-ing the attributed satisfaction of stayers from the attributed satisfac-tion of leavers (see Table 9.8). Again, six hierarchical regressions were


Table 9.7. Hierarchical Regression (Dependent Variable: Commitmentat Time 1)

OverallIndependent StandardizedVariable Beta t p R2 F (df) p

Step lFD∗AttrSat .08 .52 NS .007 .274 (1,40) NS

Step 2FD∗AttrSat .03 .18 NSIMPACT .50 3.63 .001 .258 6.76 (2,39) .005

Test of Increment�R2 = .251 F = 13.17 df = 1,39 p < .001

Table 9.8. Hierarchical Regression (Dependent Variable: Commitmentat Time 1)

OverallIndependent StandardizedVariable Beta t p R2 F (df) p

Step 1FD∗AttrSat (stayers) .24 1.52 NSFD∗AttrSat(leavers) −.16 −1.02 NS .103 2.34 (2,39) NS

Step 2FD∗AttrSat (stayers) .14 .90 NSFD∗AttrSat (leavers) −.21 −1.50 NSIMPACT .49 3.62 .001 .333 6.34 (3,38) .005

Test of increment�R2 = .231 F = 11.73 df = 1,38 p < .001

performed. Step 1 forced in both these interaction variables; step 2 addedIMPACT. As in the preceding case, five of the six regressions were insignif-icant. Only commitment at time 1 resulted in a significant equation.

The results for this regression are in Table 9.8. Again, the interactionterms for attributed satisfaction of stayers and leavers were not significantin step 1 or step 2. With the addition of IMPACT at step 2, however, theregression became significant, and the increment in explained variancewas also significant (�R2 = .231, p = .001). Attributed satisfaction thusadded little to our ability to understand the process that enables IMPACTto predict attitudes.

Study 2 Discussion

It appears that the insufficient justification model in Figure 9.3 receivedthe strongest support from the data. In general, dissatisfaction was


attributed to those who left. This suggests that external attributions, ifthere were any, were not strong enough to justify the coworker’s depar-ture. Thus the model of effect A depicted in Figure 9.3 is not supported.Effect B in Figure 9.3, although predicting correctly the attributed nega-tive link between coworker and job, was incorrect in its prediction of thesubsequent dissatisfaction of the stayer. Effect C correctly identified boththe negative link between coworker and job and the positive subsequentchange in stayers’ attitudes.

It is worth noting that the two dependent variables, organizationalcommitment and job satisfaction, did not respond in identical patterns.Although commitment was correlated with IMPACT at time 1 and time2, the change in commitment was not. It is reasonable to expect this,given the strength of the correlation at time 1. If the employee knows thata close coworker is about to leave, and at the same time knows that hehimself or she herself is going to stay, then the insufficient justificationprocess proposed earlier is likely to be operating at time 1. Given this,one would expect that little change would be observed.

This anticipatory effect does not explain the satisfaction pattern. Cor-relations with the change in satisfaction subsequent to the turnover ofcoworkers indicate that employees were affected by the turnover itself. Ifthe insufficient justification was enough to force stayers to be positivelydisposed toward the organization at time 1, why did the same forces notwork to improve their satisfaction at time 1, also? There are two possi-ble explanations of this inconsistency, one based on methods, the secondbased on theory.

The inconsistency may be a result of the satisfaction measures, whichare partially ipsative (Smith, 1967). That is, they measure satisfactiononly in a relative sense. Consequently, an increase in a satisfaction scorecould result from a person becoming more satisfied or from a personperceiving that others are less satisfied. This makes the interpretationof these change scores somewhat tentative as compared, for example,to Likert scales. In contrast, the Organizational Commitment Question-naire is a standardized instrument whose psychometric properties are wellestablished (Mowday et al., 1979). The OCQ score provides an absoluteindication of commitment. As such, increases or decreases in commitmentare readily interpretable.

The problem lies not in the advantages or disadvantages of eitheripsative or nonipsative scales (cf. Kerlinger, 1973; Smith, 1967). Rather,of concern here is that the difference in results between the satisfactionand commitment measures could be partly a function of how they weremeasured. Although this is a possibility, it should be remembered thatthe results in Table 9.3 are based on correlation, not on absolute differ-ences. The fact that IMPACT is positively related to satisfaction at time 2


can be interpreted as meaning that those with high IMPACT scoresreported themselves to be relatively more satisfied than their cowork-ers. This interpretation is not substantially different from that given tothe similar positive correlation to commitment in Table 9.2: Those withhigh IMPACT scores report themselves to be relatively more committedthan their coworkers. The similarity in interpretation can be extended tothe change scores in both satisfaction and commitment. Thus, althoughthe two measures do exhibit different psychometric properties, there isno reason to assume these differences would lead to the observed discrep-ancies between the satisfaction and commitment results.

A more interesting and theoretically based explanation lies in anothermodel of work attitude formation, social information processing theory(Salancik and Pfeffer, 1978). Suppose that employee A is about to leave,and B is A’s good friend. A’s behavior during these last weeks may includeproviding B with an earful of why it is that A is leaving (complainingabout the work, the supervisor, etc.). B’s evaluation of the work duringthis time is influenced by A on two counts. First, because A is a friend,the frequency of interaction will be higher, allowing A more opportunityto provide negative social cues. Second, and equally important, because Bperceives A to be a friend, B may take cues coming from A more seriouslythan cues coming from a stranger. Thus, not only are the social cues fromA more frequent, but also B’s receptivity to such cues is enhanced by thefriendship link. Once A has terminated, this source of negative informa-tion about the workplace also diminishes, resulting in a higher percentageof positive social cues about the work. Hence, B’s job attitude toward thejob itself improves. Moreover, because the job is more immediate to theemployees’ experience of work than is an evaluation of the organization,it would seem reasonable that shared communications would focus morefrequently on the job than on the organization. Organizational commit-ment, then, was probably not as susceptible to social information cues;thus, changes in this attitude were less likely to be affected by the turnoveritself and more likely to be governed by the anticipation of turnover asdescribed previously.

Conclusion

Calls for more studies of the effects of network change have increasedin recent years, with the realization that “enormous research remains tobe done in the dynamics of social networks” (Degenne and Forse, 1999:159). However, one of the limitations with respect to formal organiza-tions is that networks tend to be relatively inertial. It is difficult to studychange processes in interpersonal social networks if ties between people


remained stubbornly fixed (Mollica, Gray, and Trevino, 2003). Thus,the advantage of studying network change and its effects in fast foodrestaurants is that change is ever present in the fluctuating population ofemployees joining and leaving.

It would be helpful to have further studies of network dynamics inother types of organizations given that most of the employees in thefast food restaurants studied here were adolescents less concerned withcareers than with high school and social relationships. These adolescentswere not trapped in this type of work or these organizations, but werefreer to quit than would be the case for more career-oriented employees.Indeed, with turnover running at about 200 percent annually in theserestaurants, employees may become inured to role equivalents or friendsquitting. Thus, we would expect the effects in more typical career-orientedlocations to be stronger than the ones observed in this study.

Individual actors behave in organizations in ways that are influencedby the larger context in which they find themselves. Researchers, how-ever, have a tendency to focus on either micro or macro factors, perhapsbecause of training as psychologists or sociologists. Organizational psy-chology and social psychology have explored individuals’ values, beliefs,perceptions, and motives, which can lead to their observed behavior.And organizational sociology has focused on the structural constraintsto such behavior. The purpose of the current chapter, following upfrom the previous section, is to demonstrate that the combination ofboth orientations can lead to new insights into organizational dynam-ics. This demonstration employs a distinctly macro, structural lens tolook at micro-organizational research questions concerning the effects ofnetwork turnover in organizations. The results confirm the importanceof structural approaches, but at the same time reaffirm the richness ofpsychological explanations. We continue our examination of networkdynamics in the next chapter, which focuses on organizational crisis.

10

Organizational Crises

The previous chapter discussed the ways in which social networks sys-tematically affect patterns of turnover in organizations and attitudes ofpeople left behind. The current chapter takes a closer look at how suchpatterns of cross-unit network links help organizations deal with crises.It is inevitable that people cross organizational boundaries in pursuit ofcareers. But as they move – from one department to another, for example –they also connect units that might not have been linked previously. Thus,a somewhat haphazard network of informal relations is likely to char-acterize any system of organizational units within an overall umbrellaorganization.

Recently, there has been considerable attention given to the ways inwhich patterns of clustering and connectivity develop in networks (Doro-govtsev and Mendes, 2003), but little of this work has focused on thedesign of organizational networks. In general, the new science of “smallworlds” has been content to assume optimal designs emerge throughrelatively serendipitous processes, although the possibility of more goal-directed network design has been discussed (e.g., Kilduff and Tsai, 2003).In this chapter, the emphasis is on comparing informal patterns of orga-nization across two types of structure: an optimal structure, constructedon the basis of theory; and a “natural” structure that emerges on the basisof social interaction.

It is argued here that emergent networks, left to themselves withoutthe aid of conscious design, will form naturally in ways that are subop-timal, even dysfunctional, for the organization. Moreover, we posit thatthe degree to which the informal organization is designed optimally ismeasurable. The argument behind this theory will be built up from a setof seven assumptions presented and defended in the sections that follow;the assumptions will culminate in a set of propositions. The first of thesepropositions is tested in a set of organizational simulations. The resultsof these tests provide strong support for this primary proposition.

208

Organizational Crises 209

The Structuring of Organizations

Differentiation produces work units at many levels in an organization(Lawrence and Lorsch, 1967; Miller, 1958). We will refer to formal sub-divisions within organizations as “subunits”; these units may vary widelyin size. Subunits usually are considered in terms such as department, unit,center, or workgroup. The characteristic that is important to the follow-ing discussion is that these units are recognized formally as differentiatedfrom other units.

Friendship Patterns

Assumption 1: Organizations tend to evolve into friendship cliques (densefriendship networks) primarily within subunits.

As we have shown in Chapter 6, similarity is a basis for identificationand friendship formation. One important type of similarity is physicalpropinquity: similarity in geographic location. It is well known that suchgeographic closeness induces frequent interaction and leads to friend-ship. Festinger, Schachter, and Back (1950) studied a group of studentsto determine whether a causal link between propinquity and friendshipcould be inferred reasonably. Specifically, married students were assignedto housing units according to when they applied to school. This order ofapplication was independent of other factors that may have led to friend-ship, such as similarity of interests or college majors. Yet those living nextdoor to each other were far more likely to become friends with each otherthan were other pairs of couples. Those who lived at the physical ends ofthe buildings were more likely not to have any friends in the complex.

Other researchers have found similar relationships. Several studiesshow that students who sat close to each other in classes in a board-ing school were more likely to like each other (Byrne and Buehler, 1955;Maissoneuve, Palmade, and Fourment, 1952). Sykes, Larntz, and Fox(1976) found that those who bunked next to each other interacted witheach other and liked each other more. In summarizing this research, Shaw(1981: 84–5) notes that propinquity creates at least an opportunity forfriendship formation: “Clearly, persons who are physically close to eachother are more likely to form affiliative relationships than those who aremore distant from each other.”

Subunits usually are organized physically so that members are proxi-mate to each other. More interaction during the workday occurs withinsubunits than between subunits; as a result, friendships are more likely tooccur within than across subunits. Miller (1958) describes how subunits,


particularly those based on territory and time divisions, develop groupsolidarities and inward connectedness. Obvious examples include indi-viduals working in the same department or the manner in which workerson the same shift are likely to form relationships.

Such structurally constrained interaction also is likely to lead to somecommon perceptions of the units and the individuals in the organization.Subunits divided by technology often derive solidarity from what Selznick(1957) termed “distinctive competence.” The arrangements of the tasknot only make them specialists but also give them a particular vantagepoint and perspective on the organization. To the extent that individualsare grouped by occupation, common points of view may derive from thecreation or maintenance of occupational communities.

The creation of common points of view also is strengthened by anotherforce at work in the friendship network. Allocation of resources usuallyis made to subunits in order to facilitate coordination and control ofresources within the organization. Monitoring and budgeting funds tosubunits is much easier than monitoring and budgeting individuals. As aresult, those in a subunit are viewed as allies in the battle of the budget,and those within a unit may see other units as competitors for resourcesin organizational decision processes. Such a view is explicit in the workof both Allison (1971) on decision making in the government and Pfeffer(1981) on organizational power.

In summary, organizations are structured in subunits that have suffi-cient boundaries to be structurally conducive to the formation of friend-ships with units. Although friendship links between individuals occuracross subunits, these ties occur at a lower rate than within-unit ties.

We now digress to introduce the idea of organizational crisis. It will beargued subsequently that crises create conditions in which the arrange-ment of friends becomes especially critical.

Organizational Crises

Organizational crisis has been the subject of a variety of analyses thatdebate the definition of crisis itself (e.g., Billings, Milburn, and Schaal-man, 1980; Hermann, 1969; Milburn, Schuler, and Watman, 1983; Smartand Vertinsky, 1978). Most discussions use Hermann’s (1963) definition:A crisis is a stimulus (situation) that consists of a threat to desired orga-nizational goals, in which decision time is short and surprise has founddecision makers unprepared to act.

Billings and his colleagues (1980) find this concept incomplete, andsuggest the additional importance of the triggering event and the mannerin which it is sensed. The stimulus needs to be considered in comparisonwith current standards of performance outcomes; otherwise, there will


be no apparent threat. A discrepancy would produce an evaluation of theprobability that the organization would face a loss, and of the magnitudeof that loss. Crises are most likely when required responses are uncertainand must be original (Billings et al., 1980: 302–5).

In building a general model of organizational crisis, Milburn and hiscolleagues (1983) add that a crisis may occur not only because of theopportunity to achieve desired goals but also because of a threat to thosegoals. They emphasize the organization’s need to resolve the crisis andthe likelihood that the “resolution strategy is uncertain” (Milburn et al.,1983: 1144).

From the crisis research, we can extract a working definition for thisanalysis. “Crisis” refers to a situation facing an organization that requiresthe organization to engage, under time constraint, in new, untested,unlearned behaviors to obtain or maintain the organization’s desired goalstates. A crisis could result from events within an organization, such as aplanned effort for change or the sudden loss of critical personnel, as wellas from external sources.

In the simplest terms, a crisis requires uncertain action under timepressure. When uncertain action is required without time pressure, thesituation may be viewed as a problem rather than a crisis. When requiredactions and outcomes are known but when time pressure exists, organi-zations engage standard, albeit critical, procedures or routines.

The importance of friendship networks for organizational crises isrevealed in the conditions that exist within an organization during acrisis and in the behaviors that are required to manage the situation.

Assumption 2: Crisis leads to a perception of uncertainty and threat ofchange.

This assumption follows closely from the definition of crisis itself. Athreat or opportunity requires uncertain response, the outcomes of whichare also uncertain. Billings and his colleagues (1980) discuss the perceivedprobability of loss. Milburn and his colleagues (1983) examine the stressthat results from this uncertainty. Meyer (1982) provides one of the clear-est statements in discussing environmental jolts: “When they are labeledcrises, jolts infuse organizations with energy, legitimize unorthodox actsand destabilize power structures” (Meyer, 1982: 553). The behaviorsrequired are “unorthodox,” unlearned, and untried; especially to thepoint, they have unpredictable consequences. Thus the power structures,which enforce order and stability, are challenged directly.

A wide range of threats to the current order must be acknowledged.Resources may be reallocated or changed in absolute availability. Per-ceived resource scarcity may increase. Power may be redistributed and


day-to-day organizational procedures may shift. Expectations of behav-ior and overall performance may change. Threats of change in the nor-mative, political, and physical environments can arrive in widely rangingforms.

Assumption 3: Such perceived uncertainty and threat of change will resultin conflict among subunits.

The perceived threat will lead units to defend their resource base andpatterns of action. Hermann (1963) suggests that preexisting conflictswill be aggravated during crises. The internal allocation of resources willseem more like a zero-sum game. Milburn and his colleagues (1983)argue that strategies of centralization tend to be applied in crises. Theybelieve that such strategies reduce the cohesiveness of the organization asa system. There will be a greater tendency for conflict to develop amongunits (Milburn et al., 1983: 1171) and for cooperation among units todecline (Schein, 1967). Milburn and his colleagues (1983) posit a cycleof deepening crisis as a result. This argument leads to further elaborationof the nature of the conflict.

Assumption 4: Conflict will lead to two separate but commensurate out-comes: (1) increased commitment to the home subunit and (2) reducedcooperation with other subunits in the organization.

Theory suggests that intergroup conflict leads to increased group cohe-siveness, a stereotyped image of other groups, an inability to cooperate,hostility, and clear formulation of group beliefs (Coser, 1956; Simmel,1955). Most of these hypotheses have been verified in experiments sincethe work of Sherif, Harvey, White, Hood, and Sherif (1961) (see alsoSchein, 1967). Wheaton (1974) also shows that conflict can lead toincreased group cohesion when members are forced to rally around acommon set of principles, as one might observe within an organizationalsubunit.

Cooperation, in this case, entails the notion that people are willing towork with others, even though some of the behavior is not likely to benefittheir unit to the maximum degree. In addition to the preceding discussion,Hermann’s (1963) analysis of crises includes a similar observation. Thecentralizing tendency of crisis management reduces the use of normalchannels to collect and disseminate information. As a result, cooperationand coordinating in general become more difficult. Centralization is anattempt to force integrated action by the subunits in order to managethe crisis. The ordinary level of differentiation of the subunits must beovercome temporarily.


Assumption 5: Adaptation to crisis requires increased cooperation.

Crises, by their nature, have uncertain outcomes; usually it is not clearwho will benefit from the new required behaviors. Some of these behaviorsrequire organization and coordination across subunits; thus some of thesebehaviors must be cooperative.

Khandwalla (1978) discusses the response phase of crisis as one thatinvolves increased collaborative relations and the establishment of inte-grative mechanisms. Solutions to major crises described by Starbuck,Greve, and Hedberg (1978) show the need for increased connectednessbetween units that previously were unconnected.

Assumption 6:Trust enhances cooperation.

In a trusting relationship, one imputes “honorable” motives to another.That is, if person A trusts person B, by implication A expects that B willnot intentionally use information or engage in behavior at A’s expense.To violate this expectation is to violate trust. Once trust has been violated,cooperation is diminished greatly; cooperation without initial trust is verydifficult to implement.

Forced compliance to cooperative behaviors is not an efficient answerto the problem. As Kelman (1958) points out, forced compliance requiresexpensive and constant surveillance; it engenders distrust and negativeaffect. People are tremendously creative at undermining systems of con-trol; the popular press often has highlighted examples of resistance, suchas the now-classic case of the auto workers at Lordstown, Ohio, in theearly 1970s (Lee, 1983).

Assumption 7: Strong friendship includes trust.

Although not isomorphic with trust, friendship implies trust. Withouttrust, friendship does not exist. As Bell (1981) observes in his researchon friendship, “When we asked people to describe what was importantto friendship, their most common answer was ‘trust’. This was becauseclose friendships are possible only if certain barriers are eliminated andthe two people can come to an understanding. This further means thatwhat they do and get from each other is based on trust” (Bell, 1981: 16).

These arguments suggest the reasonableness of the following proposi-tion:

Proposition 1: In times of crisis, more effective organizations will be thosewith friendship links between subunits (as opposed to strong friendshiplinks within subunits).


Insofar as strong friendship ties exist across subunit boundaries, moretrust and cooperation will be possible. Uncertain solutions are more likelyboth to be suggested and to be implemented. More cooperation willresult in greater ability to adapt to the new situation created by crisis.The importance of this proposition lies not only in its prediction thatthe tendencies toward differentiation in organizations work against theexistence of optimal conditions, but also in its contradiction of the generalprinciple of impersonality as found in the traditional bureaucratic model,which advocates eliminating friendship ties as a strategy for efficiency.

Caution should be exercised in interpreting this proposition. It appliesto crisis situations rather than to normal, routine operations. Althoughlinks among units still are important even in ordinary organizationalfunctioning, ties within subunits improve normal performance by per-mitting coordination and cooperation in unit work tasks. Intraunit tiespromote a more positive social environment for unit members who mustspend most of their workday within the unit. This proposition is notintended to disregard the importance of internal group ties to ordinarygroup functioning. Rather, it emphasizes the extraordinary circumstancescreated by crisis conditions, which make an abundance of external tiesmore effective than internal ties.

Dimensions of Network Characteristics

In the model of crisis management proposed in this chapter, the focus is onfriendship as the link among members of an organization. Friendship ties(links) have been identified as both external and internal to organizationalsubunits. External links are friendship links between members of differentsubunits; internal links are friendships between members of the samesubunit. According to proposition 1, for any given density of friendshiprelations, external links are more important for the management of acrisis. According to assumption 1, however, internal links are more likelyto form. An index of the relationship between external and internal links isrequired to evaluate the proposition. Therefore the E-I index is proposedas follows:

E-I index = EL − ILEL + IL

where:

EL = the number of external friendship linksIL = the number of internal friendship links


The possible scores for this index range from –1.0 to +1.0. As the E-Iindex approaches +1, all the links would be external to the subunits. Ascore of –1 would indicate that all the links are internal. If the links aredivided equally, the index will equal zero.

Several facets of this index are worth exploring. One may note, forexample, that the index is a measure of dominance of external overinternal ties, not simply a measure of external links. Thus the index notonly decreases with a decrease in external ties, as can be deduced directlyfrom the theory developed here, but also can be decreased by increasingthe internal ties.

The rationale for calculating the index as a ratio instead of as a simplecount of external ties is threefold. First, on the theoretical side, the higherthe density of internal ties, the greater the identification that members willmake with the subunit per se. Such an identification will exacerbate theproblem of increased commitment to the home unit (see assumption 4).The lower the density of friends, the easier it will be to induce the organi-zationwide identification and commitment necessary for the cooperationrequired to face the crisis successfully.

Second, one may assume that individuals have a limited amount oftime, energy, and need for the social interaction and intimacy that aredemanded in maintaining friendships. Given this assumption, one willfind, on the average, a trade-off between the number of friends that onecan maintain outside the subunit and the number one can maintain insidethe subunit. In this sense, the more internal links one has, the fewerlinks one can foster outside the subunit. Thus internal links represent an“opportunity cost” to the subunit.

Third, there is a practical, methodological reason for including internallinks in the E-I measure. The concept of “friend” is somewhat elusive;some people may respond to a specific question about what constitutesfriendship (and therefore who their friends are) differently from otherpeople. That is, some may have a high threshold of friendship (and thusreport few such friends), whereas others may have a low threshold (andthus report many such friends). By comparing the external to the internallinks, we are controlling automatically for this source of variation in themeasure. Those with low thresholds will contribute correspondingly toboth internal and external links; those with high thresholds will deprive,in a sense, both external and internal links. On average, then, in a ratiomeasure such as the E-I index the “overestimates” of externals wouldbe balanced by “overestimates” of internals. Thus this measure is not sosensitive to such kinds of measurement error.

Other facets of this index are worth noting as well. As the organizationgrows larger, for example, the potential number of external links grows


much faster than the potential number of internal links. This point canbe demonstrated easily, as follows:

The maximum possible number of external links (E∗) is a function ofthe size of each subgroup (Si) and the number of subgroups (N). Thisrelationship is computed readily as:

E∗ =N−1∑i=1

N∑j=i+1

Si S j

If we assume that the size of all the subgroups is equal (i.e., Si = Sj forall i, j), this formula reduces to:

E∗ =(

N2

)S2

i = N (N − 1)2

S2i

The maximum possible number of internal links (I∗) also is a functionof N and of the size of each Si:

I∗ = 12

N∑i=1

Si (Si − 1)

Again, if we assume that the size of all subgroups is equal, this formulareduces to:

I∗ = NSi (Si − 1)2

The external possibilities are approximately proportional to the squareof N times the square of Si. The internals are approximately proportionalto N times the square of Si. In other words, the potential number ofexternal links usually will be greater – by a factor of about N – thanthe potential number of internals. Moreover, when the groups are of thesame size, the potential number of external links exceeds the numberof internal links approximately by a factor of N. In an organization ofany reasonable size, it would seem difficult to find an E-I index of lessthan zero because E∗ outnumbers I∗ so strongly. Of course, assumption 1argues that most links in organizations will tend naturally to be internal;thus it is our conjecture that negative E-I indices would be common,despite the handicap that I∗ imposes relative to E∗.

Mariolis (1985) points to this problem and argues that such indicesshould be normalized against the maximum possible to give a true indi-cation of the propensity to be dominated by external or internal relations.We have chosen not to do this for three reasons. First, calculating the num-ber of possible external and internal ties is cumbersome, relative to the E-Iindex calculation. Second, if assumption 1 is correct, it would be difficult


to find examples of positive E-I values. That is, external links are moredifficult to form and are costly to maintain. To normalize this alreadysmall number by dividing by the large number of potential external linkswould place too much constraint on the index. The observed index val-ues would be very small (i.e., all the values would approach –1.0) anddifficult to interpret. Third, and most important, the theory rests in parton the degree to which members of subunits are oriented inwardly (insidethe subunit) or outwardly (to the organization as whole). An individualwho has many ties to other parts of the organization, even though thoseties represent a tiny fraction of the maximum “possible” ties, will sharea wider, more organizational view of the world. He or she need not betied to everyone else (an impossible task anyway, in most cases) in orderto share that view. As long as most of those ties are oriented outward,the individual will be influenced toward cooperation. If all members ofthe organization tend generally to be tied to more members outside theirsubunit than inside, the E-I index will summarize that trend with a pos-itive value. Because we seek to measure this trend, we prefer the simple,unstandardized E-1 index to the more complex, more sophisticated alter-natives proposed by Mariolis (1985).

This index allows the formulation of a specific hypothesis deduciblefrom proposition 1:

Hypothesis 1: Organizations with a high (positive) E-I index will be moreeffective in the face of crises than organizations with a low (negative) E-Iindex.

Further Propositions

The dominant theme in this argument is that the relative density of exter-nal friendship links is the critical determinant of effectiveness in facingan organizational crisis. It would be misleading, however, to suggest thatthis is the only contributing factor. As suggested elsewhere (Tichy, 1981),one must pay attention to several contingencies in the design of organi-zations. Indeed, the theoretical discussion that led to proposition 1 canbe used to develop additional propositions regarding other such contin-gencies and design factors. (These propositions will not be tested in thischapter.) Propositions 2 through 4, which follow, address some of thesedesign issues. They are phrased specifically to retain the preeminence ofthe first proposition.

Proposition 2: If the E-I index is held constant, the number of subunits inan organization is correlated inversely with organizational effectivenessin the face of a crisis.


We stated earlier that subunits are any number of types of organi-zational units. They are distinct and usually there is consensus in theorganization about their distinctness. From the earlier discussion, it isclear that the fewer subunits exist, the fewer problems will be created bydifferentiation. In effect, the greater the level of organizational complex-ity based on the number of subunits, the greater the potential problemof insufficient ratios of external to internal links. In addition to the dom-inance of external ties and the number of subunits, the distribution ofexternal ties may be important, as suggested in the following proposi-tion:

Proposition 3: If the E-I index is held constant, the variance of the dis-tribution of links between pairs of subunits is correlated inversely withorganizational effectiveness in the face of a crisis.

Variance is defined here as follows:

Var =∑Ns−1

i=1

∑Ns−1j=i L2

i j12 Ns (Ns − 1)

−(∑Ns−1

i=1

∑Ns−1j=i Li j

12 Ns (Ns − 1)

)2

where

Ns = number of subunits in organizationLi j = number of external links between subunit i and subunit j

This measure is simply the variance in the number of links (Lij) betweenall pairs of subunits.

Given a fixed number of subunits and a fixed number of externallinks, we can arrange those links in several different ways, but not allsuch arrangements would be equally effective in promoting the necessarycooperation. For example, if all the links were concentrated between twosubunits and if no links were present between any of the other subgroups(i.e., if variance were high), one would predict trouble in cooperativebehaviors among those other subunits. On the other hand, if the links weredistributed equally across all pairs of subunits (variance = 0), potentialfor cooperation and cooperative attitudes would be distributed equallyacross the organization. Thus a high-variance condition (in which thelinks are concentrated in a few pairs of subunits) will not be as robust oras suitable to crisis situations as a low-variance condition (in which thelinks are parceled out among pairs of subunits).

Proposition 4: If the E-I index and the variance in external links areheld constant, the match between density of external links and needs forcoordination of those pairs of subunits will be correlated positively withorganizational effectiveness in the face of crisis.


One additional highly probable contingency follows from the precedingdiscussion: Subunits will vary in the amounts of coordinative activity theyrequire (Thompson, 1967). Pairs of subunits that require such coordina-tion will be sensitive to the extent to which external links exist to handlethem. The creation of a high density of links is imperative to high needsfor coordination (e.g., Lawrence and Lorsch, 1967). When two subunitshave little need for coordination, the corresponding need for externallinks is attenuated. If the linkage pattern and the need for coordinationoverlap, greater human resources are available for crisis management.

We are not arguing that external links between pairs are unnecessarywhere coordination needs are low. Indeed, such links can prevent prob-lems that might arise independently from technological or other demands.We simply are arguing that they are needed more where coordinationneeds are higher.

We turn now to a description of the empirical support found in anorganizational experiment that we conducted to test this theory. Ourintention was to test only hypothesis 1 (and, by inference, only proposi-tion 1). Testing other parts of this model for the remaining propositionsis left to future research.

Method

To test hypothesis 1, we made six trials of an experiment. In each trial, twoorganizations were designed to be comparable in every way except for thearrangement of their friendship links. In one organization, the number offriendship links between subunits of the organization was maximized atthe expense of the number of links within any one subunit. In the secondorganization, the number of links within the subunits was maximized sothat there was a minimum number of links between subunits.

We simulated the “organizations” using Miles and Randolph’s (1979)Organization Game. Other researchers also have found the game a usefuldevice for simulating the complexities and demands found in real orga-nizations (Cameron and Whetten, 1981). It provides a fertile ground fortesting the hypothesis, for two reasons. First, it allows for experimentalcontrol of the design variable of interest – namely, the pattern of friend-ship links.

Second, and equally important, the Organization Game is the epitomeof crisis as we defined it previously. This definition has several compo-nents, including the requirement that organizations, “under time con-straints, engage in new, untested, unlearned behaviors. . .”; the Organiza-tion Game meets these requirements. Consider the first condition, that of“time constraints.” The game is divided into several periods, each lastingbetween 30 and 75 minutes, with the longer periods at the beginning of


the game. Within these strictly enforced time limits, each participant mustfigure out what to do and then do it. With the exception of a few peoplewho find out that they have nothing to contribute, all the “employees”discover that they are under pressure to complete certain tasks (gatherand disseminate information, fill out forms, allocate resources) before theend of the session. The penalty for not completing these tasks varies butfrequently affects the entire organization. Invariably, some of these tasksare not completed on time.

The second condition in our definition of crisis – that the organizationis required to “engage in new, untested, unlearned behaviors” – also isendemic to the Organization Game. Wealth is accumulated or lost overtime through a complex formula that incorporates productivity measuresand the quality of decisions about resource allocation (see this chapter’sappendix for more details). The formula was not given directly to the par-ticipants; most players learn it through trial and error and through thehints given in the participant’s manual. Moreover, only a cursory skeletonof an organizational structure is provided. Even communication betweenunits is restricted through a formal mechanism involving a limited numberof passes. Some people are assigned to positions in the organization, butthey must determine for themselves their formal and informal responsibil-ities in that position. If the participants do not organize themselves, allo-cate jobs, and begin to produce outputs, the indicators of performance willdrop and the organization will fail. The ultimate creation of a hierarchic orcentralized control structure also is a function of participants’ decisions.

In reference to the definition given previously, the initial crisis is a threatto the organization’s goals of survival and performance; decision makersare surprised by the organization’s complexity and operating rules, whichdictate behaviors that are verv different from the accustomed behaviors.The required responses are uncertain and original in their configuration.Some known and some unfamiliar behaviors must be organized in newways to stabilize the organization. This initial crisis approximates theconcept as developed by Billings and his colleagues (1980), Hermann(1963), and Milburn and his colleagues (1983).

Little stability or reduction in confusion occurs for at least three roundsof play (Stern, 1974). After some stability has developed, periodic spe-cial events are introduced simultaneously into the paired games to ini-tiate a further crisis condition. Such events have included simulationsof disasters such as fires or earthquakes, the introduction of designatedminority players who must be given positions of power within a certainperiod (thus current power holders must be replaced), competitors inthe product market and takeover attempts by outside corporations, andgovernment regulatory rulings. These crises, usually introduced duringsession 5, approximate the environmental jolts regarded commonly asorganizational crisis situations.


Thus the experience for each participant is one of uncertainty aboutwhat to do or how to do it. Roles develop through time in the game,through negotiations, interaction, and agreements. Yet these roles changefrequently because of political infighting or renegotiations, or simply be-cause organizational goals were not being met. The Organization Game,then, amounts to a continuous, interunit crisis environment, where new,untested, unlearned behaviors emerge and then frequently are discardedin favor of new behaviors.

Each simulation was conducted as part of a class at one of two univer-sities. Four of the trials used undergraduates (trials 2, 4, 5, and 6) and twoused graduate (MBA) students (trials 1 and 3). The procedure for design-ing the organizations was as follows. Everyone in the class completed aquestionnaire that asked them to rate every other person in the class asto how close a friend he or she was. The directions for this questionnaireincluded the following:

Please place a check in the space that best describes your rela-tionship with each person on the list.

The names of everyone participating in the game were listed below,with five categories from which the respondent could choose: “trust as afriend,” “know well,” “acquaintance,” “associate name with face,” and“do not know.” Only the first category, “trust as a friend,” was used toassess the friendship links. The pattern of these links was used to assignindividuals to positions in the organization.

Two organizations were created from each class. In the first organi-zation, called here the “optimal” organization, we broke up friendshipclusters by assigning the members to different organizational divisionsand taking care not to put friends within the same division (there arefour divisions in the Organization Game). In the second organization,called the “natural” organization, we maintained the cluster of friends byassigning as many friends as possible within the same division and as fewas possible in different divisions. This organization was termed “natural”because it reflected how organizations tend to form in the working world(see assumption 1).

The actual procedure for assigning subjects to groups was constrainedby the need to maintain some friendship cliques and to break up oth-ers. The friendship checkoff procedure was used to identify cliques; thesecliques were assigned randomly to the “optimal” and the “natural” simu-lations. Individuals then were assigned to one of the four units, dependingon whether the clique was to be preserved within a unit or spread outacross units. We began to place the members of the largest cliques first,then members of smaller cliques, then dyads; finally we assigned isolatesto even out the sizes of the four units and the two simulations. This pro-cedure did not guarantee that no friendships existed within units in the


“optimal” organization or that no friendships existed between units in the“natural” organization, but it did result in organizations that approachedthese ideal conditions. In all cases, the “optimal” organizations had pos-itive E-I indices and the “natural” organizations had negative E-I indices(see Table 10.1).

The two organizations were run simultaneously in separate areas ofthe building. We took care to ensure that each organization was exposedto the same resources, enforcement of rules, time deadlines, and othervariables that can affect performance of the organization. Each trial lastedfrom five to seven rounds, depending on the time constraints placed bythe host institution. Total playing time for the trials ranged from sevento nine hours. Although individuals were not provided with any externalincentives to motivate them in the simulation, they were all asked tospecify individual goals at the outset and to evaluate periodically howwell they thought their organization was doing. Further, participants wereaware that another organizational simulation involving their classmateswas going on at the same time.

The dependent variables in these trials were the four OrganizationGame performance indicators, determined by the rules given in the gamemanual (Miles and Randolph, 1979). Each indicator summarizes howwell the organization is doing on one of the four dimensions: (1) resourcebase (RB), or how effectively the organization is replenishing the resourcesit consumes; 2) total output (TO), or the effectiveness of the organizationat producing goods and services; (3) internal cohesion (IC), or the state ofcollaboration between individuals and groups in the organizations; and(4) member commitment (MC), or how many members are satisfied withthe organization’s functioning, structure, and values. We determined thetotal performance of the organization by averaging the four indicators.The rules stipulate that if any organization drops to zero on any indicator,that organization is declared bankrupt (see the appendix at the end of thischapter for details on calculating the four indicators).

Results

The means, standard deviations, and intercorrelations among the fourperformance measures are provided in Table 10.2. These measures werecalculated in each session, but for reasons of clarity and brevity, we reportthe data only for the first and last sessions. Our primary interest here isin the scores for the last session because these represent how well theorganization coped with the various problems encountered during theentire game.

Tab

le10

.1.

E-I

Inde

xof

Org

aniz

atio

nsin

Eac

hT

rial

Tri

al1

Tri

al2

Tri

al3

Tri

al4

Tri

al5

Tri

al6

Spri

ng19

83Fa

ll19

83Sp

ring

1984

Fall

1984

Spri

ng19

85Sp

ring

1986

Lin

ksN

atur

alO

ptim

alN

atur

alO

ptim

alN

atur

alO

ptim

alN

atur

alO

ptim

alN

atur

alO

ptim

alN

atur

alO

ptim

al

E30

361

254

490

2512

7016

38I

557

501

323

567

2010

254

Inde

x−0

.294

0.67

4−0

.961

0.92

3−0

.778

0.88

5−1

.000

0.56

2−0

.250

0.75

0−0

.220

0.81

0

Num

ber

of26

2531

3229

2936

3531

3133

33pa

rtic

ipan

ts

223


Table 10.2. Means, Standard Deviations, and Intercorrelations of FourPerformance Measures after First Session and after Last Sessiona

RB1 TO1 IC1 MC1 RB7 TO7 IC7 MC7

Mean 76.25 72.0 67.333 73.0 58.25 110.333 64.417 90.667SD 14.771 10.804 8.793 13.300 52.127 55.516 41.816 41.121

CorrelationsRBI 1.0000 0.3121 0.4513 0.4053 0.5112 0.5389 0.5665 0.6210TO1 1.0000 0.9300 0.9805 0.4174 0.1499 0.2603 0.3607IC1 1.0000 0.9171 0.4817 0.2781 0.4117 0.4983MCI 1.0000 0.4320 0.2254 0.3255 0.4238RB7 1.0000 0.8971 0.8970 0.8875TO7 1.0000 0.9331 0.9049IC7 1.0000 0.9640MC7 1.0000

Notes:a A “1” following the performance abbreviation indicates that the score was measured after the first session.

A “7” indicates that the score was measured after the last session.

The design of the study allows direct comparison between the twopaired organizations when we control for many extraneous factors thatfrequently affect the performance of such simulations. It is worth empha-sizing here that the trials are comparable only in such a pairwise manner.In any given pair of games, the researcher would insert a “crisis” simul-taneously into both organizations. Resources were dispensed or (morefrequently) withdrawn at arbitrary times. Players were reassigned arbi-trarily (through fake “affirmative action” dicta). We made every effort tominimize any differences in such resources or rule enforcements betweenorganizations within an experimental pair. It was impossible, however,to control such differences from one administration of the experiment tothe next.

Thus the appropriate analysis is a nested multivariate analysis of vari-ance with four dependent variables, one blocking variable (experimentaltrials), and one treatment variable (the “optimal” versus the “natural”organizational design). We performed a MANOVA on these data, using atest of Pillai’s trace as a test of significance of the independent variable onthe set of final performance indicators. Pillai’s trace was equal to 3.018,and the approximate F24,20 = 2.56 (p <.018). Thus the treatment vari-able (friendship link patterns) affected the set of performance indicatorssignificantly.

We conducted the MANOVA because we expected the four dependentvariables to be intercorrelated (their determinants shared common ele-ments – see the appendix). Indeed, as Table 10.2 shows, the four variablesare related strongly in the last session of the simulation (generally they arerelated less strongly after session 1). The strength of these correlations


Table 10.3. Results of Organization Game Experiment

First Session Last Session

Natural Optimal Session # Natural Optimal

Trial 1 78 83 7 127 150Trial 2 49.25 67 6 11 77.5Trial 3 63 62 7 50 62Trial 4 74.75 72.75 7 79.5 161.25Trial 5 82.75 78.75 6 34 45Trial 6 72.5 81.5 5 75 98Significance test:

paired ta 1.16 2.94df 5 51-tailed p-level NS p <.02

Note:a A paired t-test was used here because the organizations were matched in each trial

on several important dimensions. These dimensions included size, type of participant(undergraduate, MBA), degree of enforcement, and interpretation of ambiguous rules,number and length of game sessions, and, perhaps most important, the precise timingand character of the artificial “crises” that were introduced into the organizations.

provided an opportunity to perform a simpler, more straightforwardanalysis of the data. We combined the four dependent variables intoone average performance score (Cronbach’s alpha = .89 for the aver-age scores calculated after session 1, and Cronbach’s alpha = .98 forthe average after the last session). With a single dependent variable, thedata lend themselves to a simple paired t-test. The results are shown inTable 10.3.

A graphic representation of these results appears in Figure 10.1. Thegraph shows the difference in average performance scores between thepaired optimal and natural organizations. Points below zero representmore effective performance by the natural organization; those above zerorepresent more effective performance by the optimally designed organi-zation. Scores were plotted at the completion of each simulation session.

In reviewing Table 10.3 and Figure 10.1, one can see that neither thenatural nor the optimal organizations have a significant edge over theircounterparts after the first round. That situation would be expected atthe start because everyone is still learning the rules and no one has figuredout yet what is required to make the organization succeed.

By the last round (indicated by a circle in Figure 10.1), however, eachof the optimal groups outperformed its natural counterpart. The pairedt-test indicates that those organizations designed with a high density offriendship links across subunits did significantly better than the organiza-tions in which most friendships are within the subunits. These significancelevels were attained despite the small number of observations (six trial

85

75

65

55

45

35

25

15 5

–5

–15

–25

12

34

56

7

Indi

cate

s fin

al s

essi

on

Tria

l 1Tr

ial 2

Tria

l 3Tr

ial 4

Tria

l 5Tr

ial 6

Figu

re10

.1.

Dif

fere

nce

betw

een

opti

mal

and

natu

ralp

erfo

rman

cein

dica

tors

for

each

sess

ion

inea

chtr

ial.

226


pairs). In addition, an acute observer will note the frequent recurrenceof inflection points at session 5, when the secondary, external crisis wasintroduced into the organization. At this point in trials 1, 2, 3, and 5, theoptimal organizations begin to demonstrate substantially better perfor-mance than the natural groups.

Cooperation and Conflict Processes

The results described in the preceding section are reasonably strong andconvincing when we consider that a minimum number of experimentswas conducted and that the focus was on only the first hypothesis in thetheory. The size of the differences, on the whole, was substantial andrelatively consistent. The conclusion one draws from these results is thatthe structure of friendship patterns in such situations was an importantcontributor to organizational success.

It is important to note, however, that the theory as described in theintroduction to this chapter has not been tested in full. This experimentis only a surrogate for organizational phenomena; the simulation lacksthe history and the resultant culture that characterize most organizations.Such cultures could act to moderate the effects of friendship links (or thelack thereof). That history and culture also might include greater central-ized authority than participants create in the organizational simulation;the experiment uses a relatively young organization.

In addition, the performance indicators give little insight into the pro-cesses that produce cooperation, trust, and success or their opposites inthe game. Even so, observing the participants’ activity provides a meansfor interpreting the levels of internal cohesion and resource accumulationthat develop. Observation of participants also produces support for theprocess implied by the assumptions in the model.

During the experimental trials, participants’ patterns of activity wereobserved and each participant kept a diary describing events and reactionsto those events. Entries were made after each session, and the diaries weresubmitted to the instructor as part of the class assignment. Frequently,those in the optimal organizations were seen to cooperate with each otherin the face of the dilemmas that they encountered. Divisions in the natu-ral organizations, however, frequently directed participants’ attention toprotecting or enhancing the resources of their own division rather thanthose of the entire group.

Some examples will help to illustrate this difference in cooperation.At the close of the first session in trial 1, the Red Division in (coinci-dentally) both organizations failed to turn in some necessary forms. Thepenalty for failing to do so was reasonably severe for each organization,


although those in the Red Division would bear the brunt of the penalty.In both cases, the members of the Red Division felt disillusioned, embar-rassed, and angry, but the responses of those in the other divisions dif-fered markedly between the two organizations. In the natural organiza-tion, the people in the Red Division were blamed for the oversight; theywere labeled as incompetents who were going to ruin the organizationand were isolated from the rest of the organization. Future attempts bythose in the Red Division to help the organization were met with sus-picion.

In the optimal organization, a delegation of representatives from theother divisions (who were friends of those in the Red Division) ap-proached those in Red to ask what had happened. Those in the Red Divi-sion replied that they were simply unaware of the rule that had requiredthe forms to be turned in. (The same lack of awareness was responsiblefor Red’s mistake in the natural organization.) The group asked whetherthere was anything it could do to help Red at this point. It was decidedjointly to spread out the penalty in such a way as to minimize the impacton the organization as a whole rather than to let Red suffer the penaltyalone. By the middle of session 2, those in the Red Division were inte-grated into useful roles throughout the organization.

The fact that the optimal organization outperformed the natural onein this trial is particularly interesting because, as was discovered duringthe postgame debriefing, one person in the natural organization had hada copy of the solutions to production problems in the game. This personwas part of the Yellow Division, which was responsible for producingsolutions to puzzles (for which the organization received profits). Theability to solve these puzzles was one of the primary ways in which theorganization made money and increased its various performance indica-tors, especially resource base and total output. Having these solutionsput the organization at a tremendous advantage. The advantage was notsufficient, however, to overcome the disadvantage that the natural orga-nization suffered because of its suboptimal informal design.

Trial 2, which ended after six rounds because of the bankruptcy of thenatural organization, illustrates the process involved in cooperation andspeaks to the strength of the assumptions of our model. The descriptionfocuses first on the action of the production units, which were solvingword puzzles as their production activity, and second on the way in whichthese units interacted with the other units in their respective simulations.

Production in the optimal organization was not confined solely toefforts of the assigned production units. An integrative form of subcon-tracting developed in which the Red Division, job- and resource-poor, wasgiven puzzles to solve on commission. Red borrowed funds from otherunits to supplement what little money it owned and had the production


units buy it puzzles. The return on solved puzzles was split among theproduction units and the Red Division. In the fourth round of this trial,several diaries show that staff units loaned funds to the production unitsto purchase raw materials. Units in the optimal game also began to poolresources for reinvestment in the firm as early as round 2. They cooper-ated in an effort to deal with the unexpected absence of a unit head, whichcould have cost the entire group available resources and opportunities.

This cooperative effort contrasts sharply with events in the parallelnatural organization. Two production units that must cooperate weremerged through the actions of one of those units and then changed theirname to Supercomtin. One member’s diary for session 3 says, “I really dothink we are the most important (unit). We are really tight now. We havedecided to stay together no matter what. It is also evident that the restof the organization is against us.” The head of the super production unitwas quite explicit about these views. In session 3, a member of anotherunit came to ask whether the production group would contribute toreinvestment in the organization. The head of Supercomtin wrote in hisdiary:

They never invested any money in us to buy puzzles and theykept giving money to Routin [the other production unit] whichwas less successful than we were at production. We just felt thatwe were supposed to support ourselves.

This theme continues in the next session (session 4):

Although we did well, there was a lot of inter-group tension. Ithink people were jealous of us and were p . . . off at our attitude.We felt we didn’t need the rest of the organization and showed itin our relationship with the other group. Because of this, we hada meeting between the leaders of the groups. I, however, chosenot to attend.

During this trial, both groups were presented with a high-risk oppor-tunity to expand their markets and to make a substantial improvementin performance indicators. Chance of success was only 50 percent, how-ever, and failure meant a major decline in the indicators. In the optimalorganization, the unit heads sent delegates to a meeting arranged by thecommunications unit and decided that they were doing well enough andshould not take the risk. The natural organization proceeded in a dra-matically different manner. The unit with information on the currentindicator levels decided that for the “good of the organization,” the mar-ket expansion should be attempted. The members of the unit drafted astatement that they read to other units, telling them that the group couldafford to take the risk and would be all right even if the attempt failed.


To persuade the other units, the communication group presented falseinformation to the others regarding current indicator levels.

One diary describes this episode by saying, “Crunode [communication]was lying about some of the indicators in order to get the other units tocontribute to the investment.” Then she points out that there was “com-plete lack of trust and sympathy between Emrel [personnel], Crunode andComtin [production] which I understand since Crunode has been totallydishonest with us all until now.”

The dynamic described here points out the individual-unit–centeredactivity in the natural organization and the integrative intergroup activityin the optimal structure. One member of the natural organization providesa succinct summary for sessions 1 and 2: “There was a lot of cooperationwithin the Blue Division. However, when members of other divisionscame into the room, there was deception and non-cooperation.” Fig-ure 10.1 shows dramatically the effects of those differences. Session 5 oftrial 2 shows the adverse effects of the efforts devoted to the decision onthe high-risk option. The difference in session 6 demonstrates the clearsuperiority of the optimal organization’s treatment of this issue.

One additional example from trial 4 illustrates the differing dynamicsof the games. Several members of both organizations became idle perma-nently because they had no job or income for two consecutive sessions.These members were sent back into the games as “government observers”in round 4. In round 5, they regained active status as “minorities” whohad to be given positions of responsibility. In the optimal organization,the Board of Directors met and decided on a position for each “minority”member in the organization. The person currently occupying that positionwas given compensation including a vacation, a permanent salary, and abonus. In the natural game, the “minority” member went from divisionto division asking for work (the group received a penalty if the person didnot find a job). After a substantial effort, the “minority” member madean arrangement by written contract with one unit. When the next roundof play began, however, that unit said that the deal was a lie, made uponly to avoid the penalty. Substantial conflict erupted immediately. Oneof the “minority” members who moved from an ordinary player in onegame to a “minority” in the other characterized the two situations interms of his own enjoyment. “In Ingot [natural] it was fun screwing overthe environment and other people. In Extol [optimal] it was fun makingdecisions in a corporate manner with results that reinforce your belief inyourself.”

Differences between the organizations also are reflected in the personalassessment forms collected in trials 1 and 3 (these forms were not collectedfor the other four trials). One question on these forms asked how wellparticipants thought the organization was doing. The response options


consisted of a three-point scale: 1 = the organization is not doing well;2 = the organization is doing fairly well; and 3 = the organization is doingvery well. After round 1, the average responses in the natural organiza-tions were 2.25 and 1.87 respectively for trials 1 and 3. These averageswere slightly better than in the optimal organizations, whose correspond-ing scores were 2.11 and 1.61 respectively. By the end of the last round,however, the self-evaluations were decidedly different. The naturals haddropped their average evaluations to 2.06 and 1.47 respectively in trials1 and 3. The optimals, in contrast, raised their evaluations to 2.71 and2.44.

Discussion

Although descriptions of game events and diary entries tend to support thecontentions of the major proposition of the model, they also contribute tothe viability of the set of assumptions underlying the model. In the model,assumptions such as conflict leading to commitment to the home unit(assumption 4) were taken as axiomatic, and we did not undertake a testof such statements. The participants’ diaries, however, give evidence thatthe players saw the work in ways that were consistent with assumptions3 through 7.

Resource scarcity and uncertainty about game rules led initially to con-flict over proper strategy. Groups sought to ensure the security of theirhome units and attended to attainment of unit goals in the initial sessions.The learning of roles clarified the need for distribution of resources, andunits began to approach one another. Efforts at interunit cooperationresulted where friendship links extended across groups, but concentra-tions of within-group friendships produced cooperation only within sin-gle units. The outcome was differential performance on the overall levelsof performance that required organizationwide cooperation.

Several particularly observant participants noted aspects of this effectin their diary entries:

[Trial 2, session 5, optimal] We are all participating in invest-ments and sinking money into buying puzzles. At the start ofthe game, this never happened. Every department was concernedwith their [sic] own success and didn’t discuss matters of an orga-nizational nature. Now it seems like everyone is out for the goodof the organization. I think we have established a trust based oncompetency and effectiveness.

[Trial 2, postgame, natural] Together with friends the membersdidn’t need to trust those they didn’t know because they already


had a network of trust built in. They could communicate with anddepend on each other so there was no need to reach out beyondtheir original circle of friends. Since people tend to choose friendswho are similar to themselves, those in the same group thoughtalike and would go along with any ideas that came up – evenif it should have been discussed more thoroughly and subjectedto differing opinions. Each group thought only of how they [sic]could benefit themselves and their friends.

These two participants illustrate the importance of preexisting friend-ship ties, the formation of cohesive, within-group interests, and the dif-ficulty of cooperating across units. The latter example focuses on thedifficulties created by strong within-group ties; the former exemplifiesthe benefits of ties between groups. The importance of external links isdemonstrated by the superior speed with which the optimal groups raisedthe performance indicators and solved unexpected problems presentedduring the simulations (e.g., Figure 10.1, trials 2, 3, and 5). The qualita-tive observations and diary records enhanced our faith in the validity ofthe model by illuminating the process that led to these results.

Future Considerations

We do not expect that this theory will be universally applicable. Not allorganizations will benefit from friendships that cross division boundaries;such friendships may incur costs (cronyism, for example). Another issue iswhether all organizations require the same amount of cooperation, evenduring a crisis. When the organization, such as a bank, is facing pooledinterdependence, very little coordination is required across subunits tobe efficient. That is, some crises may be restricted to single units or mayrequire much less coordinated effort because of the technology employedby the organization or the limited scope of the crisis itself. In such cases,the cooperation that results from friendship links between subunits maybe less useful. In any event, finding the boundary conditions and the asso-ciated moderating variables will be an important part of the developmentof this model.

Let us ask a related question: Is there a linear relationship between theE-I index in the organization and the cooperation between subunit pairs?Perhaps the cooperation reaches a maximum asymptotically with rela-tively few links between the pairs of subunits. Answers, even speculation,await further research.

Issues of External Validity

As with any laboratory experiment, questions arise as to how far theresults can be generalized to “real-life” situations. We have tried to


emulate an organization in chaos and crisis, but there are significantdifferences between these simulated organizations and typical businessorganizations. In our organizations, for example, no hierarchy of divi-sions or centralized authorities was imposed on the players. Part of theirtask was to create their own hierarchy or centralization, if they chose todo so. It is possible that a centralized, hierarchic structure would respondto crises differently from organizations without such central authority.Another similar argument could be raised about the existence of formalcommunication links that cut across divisional boundaries. These linksmay not be based on friendships, but it could be argued that such formalchannels might facilitate a cooperative effort in the face of a crisis.

We have no formal basis for discounting such objections. Indeed, orga-nizations often have more centralized authority and more formal commu-nication channels than those in our experiment. Yet, as mentioned earlier,the literature on crises indicates that these formal, routinized links oftenare abandoned during crises and may be ineffective. Informal, nonhier-archic ties become more prominent. Nonetheless, it would be interestingto test whether the density of friendship ties across work-unit boundarieshas as great an effect on performance in highly centralized firms as indecentralized firms.

This simulation also has other limitations that prevent us from gener-alizing too quickly into the field. The performance indicators of neces-sity are objectively quantifiable – and consequently somewhat arbitrary.Although arbitrary indicators of performance often exist as well in the realworld, they do not necessarily take the same form as in our simulation.Absences, for example, carry some cost to an organization, but to statethat they deplete “total output” by precisely two points each may notreflect most organizations’ assessment of such costs. In fact, if the rightpeople were absent, the organization might enjoy increased efficiency andoutput.

As arbitrary as these indicators are, they are also known to the playersof the Organization Game. Our intent here was not to mimic exactly theperformance and reward system of the “real” world. Rather, we createda complex system of performance indicators that the players would haveto work to fathom, unravel, and deal with – just as they would in thereal world. In our current design, however, we cannot attest whether theresults reported here are sensitive to these arbitrary reward structures.Again, this is an area for future research.

Conclusion

Organizations are cooperative systems by nature. We have suggested herethat the friendships that exist in all organizations can either hinder or


facilitate that cooperation in times of crisis, depending on whether thosefriendships cut across subunit boundaries. In particular, we argue herethat organizations, left to their natural progression, will develop dysfunc-tional structures that concentrate friendships within subunits. Effectivefriendship structures, then, must be designed consciously.

The theory would suggest that more attention be given to organizinginformal relationships and nurturing their development across organiza-tional subunits. Organizations that must deal with crises should encour-age personal relationships outside their immediate work unit. The exis-tence of a dense network of cross-unit ties may result in the resolution ofcomplex situations before they escalate to the point where organizationmembers define them as crises. That is, external ties may affect the mag-nitude and frequency of crises in the organization. In contrast to the viewthat impersonality is preferred in organizations and that personal rela-tionships are dysfunctional (Merton, 1957), we believe that personalizedties are a reserve resource that provides the potential for the coordinationneeded to meet rapidly changing circumstances.

The challenge in organizational design, therefore, is to take the naturaltendency for people to cluster together in friendship groups and use thisto help bridge across administrative units. New research suggests that thischallenge is not to be underestimated. Members of organizations prefer tosee informal organizational structure in terms of clusters of friends, withclusters connected within organizational units (Kilduff, Crossland, Tsai,and Krackhardt, forthcoming). The task of changing this set of cognitiveexpectations to incorporate the different design principles articulated inthis chapter would seem to require intensive training (cf. Janicik andLarrick, 2005). And, as we discuss in the next chapter, friendship clusterstend to enforce their own normative interpretations on their membersand therefore may be difficult for management to influence.

Appendix to Chapter 10: Detail on the Calculationof the Performance Indicators

The performance indicators are functions of actions taken by game par-ticipants. All four indicators are set initially at 100. The following Orga-nization Game rules stipulate how each indicator is calculated after eachsession:

� Resource base. RB is a function of natural decline (–10 percent ofRB in the prior session); investments in organizational improve-ments (+20 percent of the amount invested); production rawmaterials purchased (–1 or –2 per unit, depending on type); new


job assignments (+1 per assignee); absences (–2 per absence);job quits and firings (–3 per action); vacations (–1 per vaca-tion); temporary unemployment (–1 per person); and producingan accounting report (+2 if produced, –3 if not).

� Total output. TO is a function of natural decline (–10 percent ofTO in prior session); investment in organizational improvementprograms (+10 percent of amount invested) and human resourcedevelopment programs (+10 percent of amount invested); pro-duction (+3 or +4, depending on type); new job assignments (–1each); absences (–2 each); firings and quits (–3 each); vacations(–1 each); temporary unemployment (–1 per person); and pro-duction of management consulting report (+2 if produced, –3 ifnot).

� Internal cohesion. IC is a function of natural decline (–10 per-cent of IC in the prior session); investment in human resourcedevelopment programs (+20 percent of the amount invested);interunit cooperation in production (+2 per unit produced);new job assignments (–1 each); absences, firings, and quits (−3each); vacations (–1 each); permanent unemployment (–3 perperson); temporary unemployment (–1 per person); and produc-ing a report on organizational communication (+ 2 if produced,–3 if not).

� Member commitment. MC is a function of natural decline(−10 percent of MC in the prior session); investment in humanresource development programs (+20 percent of the amountinvested); overall production (+2 per unit); absences (–2 each);firings and quits (–3 each); permanent unemployment (–5 each);temporary unemployment (–1 each); and production of a man-agement consulting report (+2 if produced, –3 if not).

These adjustments to the indicators are applied at the end of each ses-sion of the simulation in which they occur, except for certain investmentsthat are delayed one round pending actual implementation of the appro-priate program. Finally, players in certain positions are given a “salary,”the size of which depends on how well the organization is doing (asdefined by the preceding performance indicators). These funds can beused to invest in programs, buy raw materials, or pay for other items;such decisions affect the performance scores in succeeding rounds.

11

The Control of OrganizationalDiversity

In the previous two chapters, we showed the relevance of social networksfor the understanding of turnover and crises in organizations. But whatenables organizations to promote coordination and collectivity? How dopeople with diverse backgrounds, goals, and values successfully coordi-nate their activities in organizations? The usual answer to these questionsis that organizational culture provides the glue that keeps the organiza-tion together. But the organizational culture literature has neglected theimportance of social connections in producing shared systems of meaning.In this chapter, we begin the process of remedying this oversight throughan emphasis on how people who are tied to each other create locallyshared cognitive understandings. First, we provide an in-depth analysisof how the diversity of cultural interpretations within one organization iscontrolled through the friendship network. Second, we extend the discus-sion to include how network embeddedness affects agreement concerningthe structuring of networks across three different organizations.

Previous organizational culture research has tended to treat the cultureof an organization as an independent variable that can be manipulatedto control deviant behavior (e.g., Ouchi, 1980). From this culture-as-a-managerial-tool perspective, an effective organization is like a clan, in thatit relies on mechanical solidarity – a religious adherence to common beliefsand practices – to ensure cooperation (Durkheim, 1933: 175–8). The clancannot tolerate any divergence from the “totality of belief and sentimentscommon to all members of the group” ( Durkheim, 1933: 129).

In contrast to this managerial emphasis, anthropological research hasshown that organizational culture is an emergent property of informalrelationships within workgroups (for reviews, see Baba, 1986; Holzbergand Giovannini, 1981; Trice 1985). Researchers within this traditionhave investigated how norms, beliefs, attributions, behaviors, and otheraspects of organizational culture are controlled through the informal net-works of coworkers. Each culture develops its own system of knowledge

236

The Control of Organizational Diversity 237

(Romney and D’Andrade, 1964) and this knowledge is dispersed bothamong experts and novices (Romney, Weller, and Batchelder, 1986).Interaction between group members results in knowledge diffusion(Carley, 1991) concerning important aspects of the culture, such as thedistribution of roles and relations (D’Andrade, 1984: 110). Effectiveaction within a specific culture requires an understanding of how thatparticular world is organized (D’Andrade, 1995: 182). That is, an impor-tant part of cultural knowledge is the knowledge of how to operate inthis complex web of relations and dependencies. This knowledge in turndepends in part on knowing who is related to whom in important ways (cf.Chapter 5).

Thus we see two different views on organizational culture in the lit-erature. From the culture-as-a-managerial-tool perspective, culture is aunifying force that binds people together (Siehl, 1985). Culture as anemergent property of personal relationships suggests a more fragmentedview of culture, with the possibility of competing subcultures existingwithin the same organization (cf. Gregory, 1983). This chapter builds onboth of these perspectives to suggest that institutionalized traditions, setin place by the organization’s founders, shape and are shaped by emergentbeliefs and actions. Organizational culture, at any point in time, can beexpressed as a set of social constructs negotiated between organizationalmembers to anticipate and control the motivational and cognitive diver-sity in the organization (cf. Wallace, 1970: 36). These shared constructsallow organizational members to make sense of ongoing organizationalactivities.

In this chapter, we treat culture as a cognitive system (as defined byKeesing, 1974) that is negotiated between interacting individuals whocreate what Geertz (1973) has referred to as locally shared systems ofmeaning. In study 1, we describe a method for eliciting the overarchingcultural constructs utilized by people in organizations, and we look athow the network of informal relationships in the organization controlsthe way people use culturally defined constructs. In study 2, we investigatehow clusters of individuals reinforce idiosyncratic understandings, witha specific focus on locally reinforced perceptions of network relations.

Study 1

The organization selected as the research site was a small regional dis-tributor called Pacific Distributors Incorporated (PacDis) that employeda total of 162 people at its headquarters and four branch offices. Thecompany had been founded by its current president, John Briggs, but wasrun on a day-to-day basis by Bob Jamison, who had been with the firm


since its inception. Jamison had worked his way up from salesperson tochief operating officer.

According to consultants’ reports, there was an ongoing ideologicalstruggle between two main groups in the organization. On the one hand,there were those like Jamison, who believed in the primary importanceof maintaining good social relations within the organization. On theother hand, there were those like Ralph Gibson, the chief financial offi-cer, who believed that financial control and the bottom line were ofparamount importance. Compared to Jamison and Gibson, PresidentBriggs was removed from the everyday running of the organization, butas the founder of PacDis he had been instrumental in establishing thecultural and expressive components of the company (cf. Pettigrew 1979:574). He was a strong believer in the importance of a friendly, open styleof management that placed a great deal of trust in each employee.

We conducted a series of structured interviews with a sample of PacDisemployees to uncover the cultural dimensions that characterize the work-place. This phase was designed to elicit a set of constructs used by theseemployees to organize the diversity of styles and approaches that we hadobserved and to anticipate each other’s behavior. Based on the results ofthis first phase, we developed a questionnaire to examine how the net-work of relationships influenced the applications and interpretations ofthese cultural constructs.

Phase 1: Eliciting the Cultural Constructs

Phase 1 consisted of eliciting the cultural constructs from our structuredinterviews of PacDis personnel.

MethodSubjectsWe interviewed six men and four women, chosen from the full sample ofkey management and operational personnel whom we planned to includein the second phase (see the “Method” section of phase 2). Previousresearch has indicated that the majority of all constructs can be generatedby a relatively small sample within a population (Dunn, Cahill, Dukes,and Ginsberg, 1986: 372). We wanted to capture the diversity of per-spectives that existed in the organization. To this end, we interviewedthe three top executives who epitomized the cultural tensions in the firm(the owner, the chief operating officer, and the chief financial officer)and seven other employees selected to represent four diverse functionalareas and various levels in the firm (two department heads, three supervi-sors, and two nonsupervisory employees). Each individual in the samplewas promised and provided with personalized feedback concerning theconstructs elicited in the interviews.


ProcedureGiven our cognitive approach to organizational culture, we turned topersonal construct theory (Kelly, 1955) for a technique designed specifi-cally to elicit the cognitive constructs that individuals use to anticipate thebehavior of others. George Kelly’s repertory grid technique has been usedin a wide variety of settings to enable individuals to verbalize the cogni-tive constructs they use to organize and anticipate events (e.g., Romneyand D’Andrade, 1964: Wexler and Romney, 1972).

Each of the ten employees was interviewed at the work site by aresearcher for up to ninety minutes using the structured but informalgrid technique outlined by Eden, Jones, and Sims (1983). The interview-ers presented the participants with three names at a time, asking, “In whatimportant way are two of these people alike but different from the third?”and, “How is this person different?” Nine names of PacDis employeeswere utilized, and twenty-four triads were presented to each participantso that each pair of names occurred twice (Burton and Nerlove, 1976).Research has shown that each individual has only a limited number ofconstructs relevant to any particular domain and that few new constructsare elicited after twenty or so triads have been presented (Hunt, 1951).

For each triad, a similarity and a difference were elicited to form the ver-bal labels of two poles of a bipolar construct. The interviewers followedKelly’s method of encouraging participants to articulate the distinctionsand similarities suggested by the triads, to elaborate spontaneously onthe bases for their discriminations. Facile similarities, such as, “They’reboth in marketing,” were not ignored, but following Kelly (1955: 222),participants were encouraged to keep talking so that important psycho-logical similarities and differences would emerge. As verbal labels forconstruct poles were elicited, they were written down by the researchersand confirmed by the participants. The informality of this technique wasdesigned to encourage “thinking aloud,” the verbalization of unconsciousand taken-for-granted constructs. This flexible form of the repertory gridtechnique provides much more information about the subjects’ constructsthan paper and pencil tests (Kelly, 1955: 224).

Results of Phase IOn average, each subject used twenty-one constructs (standard devia-tion = 4.5), with the number ranging from thirteen to twenty-nine. Weexamined the ten lists of elicited constructs to see whether any commonconstructs were present. According to Kelly, verbal labels are not theconstructs themselves but merely signify processes that may or may nothave been previously verbalized. Therefore, we looked for similarities inideas rather than in exact wording. For example, “Follows proceduresvs. More freewheeling” and “Likely to go by the book vs. Likely to breakrules” were counted as representing the same basic construct. Seven such


Table 11.1. The Seven Elicited Constructs and the Number of People WhoUsed Each Construct

Number of PeopleWho Used Construct(Maximum = 10) Constructs

10 Inflexible, critical ←→ Flexible, tolerant8 Does the job and nothing more ←→ Eats, sleeps, and breathes PacDis9 Goes by the book ←→ Prepared to cut corners6 Lets things slide ←→ Efficient, organized9 Easy-going, relaxed ←→ Aggressive, competitive6 Tactful, diplomatic ←→ Straightforward, blunt

10 People-oriented ←→ Task-oriented

constructs were identified, each of which had been spontaneously gener-ated by at least six out of the ten participants. Verbal labels for the polesof the seven constructs were selected from individual protocols to accu-rately reflect their consensual ideas. The seven constructs are presentedin Table 11.1.

These constructs summarize the major contrasts in behavioral stylesexperienced by organizational members. According to Wallace’s (1970)view of culture, these constructs allow organizational members to antici-pate the diversity of behaviors in the organization. The constructs capturevarious aspects of the organization’s main ideological struggle as delin-eated by the consultants’ reports and supported by our own observations.This struggle was between those who, like Bob Jamison, favored a flexi-ble, easy-going company, and those who, like Ralph Gibson, preferred acritical, rule-based approach.

In summary, a set of seven constructs was elicited from a subsampleof ten people using Kelly’s repertory grid technique. By eliciting the cul-tural constructs from organizational members, we were able to approachculture from the participants’ rather than the survey researchers’ point ofview. The seven constructs were assumed to express vital aspects of theorganization’s culture and to possess psychological resonance for eachindividual in the organization.

Phase 2: Network Relations and Cultural Attributions

Our view of organizational culture as a cognitive system negotiatedbetween interacting individuals suggests that people use the social net-work to find support for their own interpretations of experience. Weexpected that PacDis employees would tend to agree with their friendson how flexible or inflexible other employees were, how people-orientedversus task-oriented they were, and so on. Through processes of social


comparison (Festinger, 1954), the attributions people make about othersin the organization tend to be influenced by and aligned with the attri-butions of their friends. From the social comparison perspective, peopleevaluate beliefs (such as whether Smith is efficient) by comparing theiruncertain opinions with others in their social network (Festinger et al.,1950; Kilduff, 1990).

Based on the preceding discussion, we hypothesized that, relative tononfriends, friends would construe fellow workers similarly on each ofthe seven cultural dimensions (hypothesis 1).

The social network operates not only to support idiosyncratic patternsof attributions but also to control the diversity of possible attributions.People who can find little support for their opinions among their friendsare likely to be in a state of discomfort or cognitive dissonance (Festinger,1957) because they hold two beliefs that are incongruous with each other,namely, “I like my friends,” and, “My friends dislike my opinions.” Thisdiscomfort is likely to manifest itself in a number of ways, including areduction in overall satisfaction with work. Thus, we predicted a positivecorrelation between how closely individuals agree with their friends andhow satisfied they would be with their jobs (hypothesis 2).

MethodSubjectsForty-seven of the 162 PacDis employees (twenty-four men and twenty-three women) were paid $10 each to complete and return a lengthy ques-tionnaire. (Although the questionnaire contained forty-eight names, onlythe responses from the forty-seven people who completed the instrumentwere included in the subsequent analysis.) We used two criteria for selec-tion in the sample. First, we included all the supervisors and managementpersonnel at headquarters and at each of the four branches. Second, weasked the chief operating officer, Mr. Jamison, to assist us in identifyingkey operating personnel in accounting, purchasing, and manufacturingin the organization. We added these employees to our sample to ensurethat we captured the entire operational core of the organization. Eachsubject was thoroughly briefed concerning the aims and outcomes of theresearch after the study was completed.

MeasuresTo measure friendship choices, each person was provided with a list of allforty-eight people in the sample and asked to check the names of his orher personal friends. On a separate form, each respondent was also askedto check the names of individuals whom the respondent thought wouldconsider the respondent a personal friend. These data were aggregatedinto one N × N matrix using the following rule: If persons i and j both


agree that person i considers person j to be a personal friend, then entry(i, j) in the matrix equals 1. Otherwise, entry (i, j) equals 0. The resultingadjacency matrix was labeled the Friendship Matrix.

Attributions about fellow workers were measured using the sevenelicited constructs from phase 1. For each of the seven constructs, eachperson rated every other person on a seven-point Likert scale (where1 = one end of the bipolar dimension, and 7 = the opposing end of thesame dimension). For example, Bob Jamison rated Ralph Gibson on howflexible and tolerant he was (as opposed to inflexible and critical), howtask-oriented he was (as opposed to people-oriented), and so on. Jamisonthen rated each of the other forty-seven people (including himself) on thesame scales.

From these data, a coefficient of similarity for each pair of individ-uals for each construct was created. We accomplished this by calculat-ing the Pearson correlation between their vectors of ratings. For exam-ple, the coefficient of similarity on the flexibility construct for Jamisonand Gibson was the correlation between Jamison’s vector of “flexibility”ratings for each of the forty-eight employees and Gibson’s correspond-ing “flexibility” ratings for those same forty-eight employees. Repeatingthis procedure for each pair of individuals permitted the creation of anN × N similarity matrix of such scores for each of the seven constructs.These seven Attribution Similarity matrices were hypothesized to maponto the friendship social network.

Overall job satisfaction was measured using the five items from theMichigan Organizational Assessment Questionnaire (Cammann, Fich-man, Jenkins, and Klesh, 1983). These items consisted of seven-pointLikert scales with end points labeled “strongly disagree” and “stronglyagree.”

AnalysesTo test hypothesis 1 (relative to nonfriends, friends would construe fellowworkers similarly), the Friendship Matrix was correlated with each of theseven Attribution Similarity Matrices. To the extent that the hypothesisis true, a positive correlation should be observed between these matrices(i.e., the 1s in the Friendship Matrix should match up with high similarityscores in the Attribution Similarity Matrix). Because the unit of analysisfor this correlation was the dyad, the test for this correlation was based onN × (N – 1) nonindependent observations. Thus, a nonparametric test, theQuadratic Assignment Procedure (QAP), was used to test the significanceof the correlation (Baker and Hubert, 1981; Hubert, 1987; Hubert andSchultz, 1976). The QAP is a permutation-based test of significance forinterdependent data of the sort encountered here (Krackhardt, 1988; seeChapter 3 for more details).


Although the QAP procedure provides a significance test (expressed asa Z score), it does not reveal the strength of the relationship betweentwo matrices. To measure the strength of the correlations, we calculatedGoodman and Kruskal’s (1963) gamma, a nonparametric correlationcoefficient that is a more appropriate descriptive measure than Pearson’sr for skewed and binary data such as are contained in the FriendshipMatrix.

We tested the second hypothesis – that relative to those who agreed withtheir friends, those who disagreed would be less satisfied – by creating anindex of average agreement with friends for each individual. This indexwas the arithmetic mean of the attributional similarity scores that werecalculated between each person and his or her friends. The agreementindex was correlated with satisfaction scores for each person. Pearsoncorrelations and t-tests were used instead of the gammas and QAP testsused for the first hypothesis, because hypothesis 2 involved predictions atthe level of the individual.

Results of Phase 2A map of the friendship links in the organization (see Figure 11.1) showsa center-periphery structure in which there are no obvious cliques. At thecenter of the network, with many friends, is the chief operating officer,Bob Jamison (21). Close to Jamison, in terms of his role in the infor-mal network, is the president, John Briggs (13), but far removed fromthe center of the network is the chief financial officer, Ralph Gibson (41).Consistent with our informal observations, the map shows that both Gib-son and Jamison were friends with the president, but not with each other.There were five individuals who had no friendship links with anyone.These individuals were either from the computer group or from outlyingbranches. Their contact with other organizational members was minimaland mainly involved questions of technical support.

The first question to ask is, How much diversity was there concern-ing attributions about fellow workers in this organization? Table 11.2indicates that the diversity of evaluations on the shared constructs wasextreme: Some pairs of individuals agreed completely on how they viewedothers, whereas other pairs disagreed completely (correlations rangedfrom – 1 to + 1).

The diversity of opinions that existed in this organization is also indi-cated by the magnitude of the standard deviations of the average cor-relations between attributional vectors. For example, on the construct“Prepared to cut corners vs. Goes by the book,” Table 11.2 showsthat the average Pearson correlation between the 1,081 possible pairsof individuals’ vectors was .21, with a standard deviation of .24. On thisdimension, 17 percent of the correlations were actually less than zero,


Table 11.2. Distributions of Attribution Similarity Scores for Each Construct

MeanAttributionSimilarity Standard % of Scores

Construct Score Deviation Minimum Maximum below .0

Flexible, tolerant .33 .21 −.50 1.00 6.3Eats, sleeps, and breathes PacDis .44 .21 −1.00 1.00 1.7Prepared to cut corners .21 .24 −.98 1.00 17.3Efficient, organized .30 .22 −.78 1.00 8.9Aggressive, competitive .37 .21 −.58 .93 3.8Straightforward, blunt .32 .24 −.98 .96 8.8Task-oriented .26 .22 −.81 1.00 11.5

10

28

22

1

155

13

12

25

3531

40

18

21

1714

8

199

45

4

37

11

48

20

29

24

44

42

383

33

39

23

4634

26

43

27

7

36

32

1641

Figure 11.1. Friendship sociogram on multidimensional scaling repre-sentation of path distances.

indicating considerable differences in how individuals construed theirfellow workers.

Another descriptive issue is whether the seven dimensions can bereduced to fewer dimensions for analysis (i.e., did people think of “flexi-bility” and “prepared to cut corners” as the same thing?). To shed lighton this issue, the correlations among the seven dimensions are reported


Table 11.3. Correlations among the Seven Dimensions

Dim1 Dim2 Dim3 Dim4 Dim5 Dim6 Dim7

Diml Flexible, tolerant 1.000 .091 .149 .126 .275 .220 .224Dim2 Eats, sleeps, and breathes PacDis .091 1.000 −.012 .330 .167 .111 .113Dim3 Prepared to cut corners .149 −.012 1.000 .061 .083 .117 .095Dim4 Efficient, organized .126 .330 .061 1.000 .069 .023 .088Dim5 Aggressive, competitive .275 .167 .083 .069 1.000 .181 .181Dim6 Straightforward, blunt .220 .111 .117 .023 .181 1.000 .249Dim7 Task-oriented .224 .113 .095 .088 .181 .249 1.000

Table 11.4. The Relationship between Friendship Links andSimilarity of Cultural Attributions

Construct Gamma Z (QAP) p-Level

Flexible, tolerant .33 3.960 .0001Eats, sleeps, and breathes PacDis .28 3.330 .0005Prepared to cut corners .31 3.854 .0001Efficient, organized .33 4.276 .0001Aggressive, competitive .24 2.866 .005Straightforward, blunt .27 3.205 .001Task-oriented .32 4.342 .0001

in Table 11.3. Most of the correlations were small. Of twenty-onepairs of dimensions, only one pair was correlated higher than .3: “Eats,sleeps, and breathes PacDis” and “Efficient, organized” were correlatedat .33. Rather than collapsing these dimensions into subscales, we consid-ered these dimensions to be sufficiently independent to warrant separateanalyses.

The first hypothesis asks, Was the diversity of attributions random orwas the diversity patterned by the friendship network? The answer isgiven in Table 11.4, which shows that the attributions of friends weresignificantly more similar than those of nonfriends for each of the sevenconstructs (p ≤.005 for each construct). The gammas, ranging from .24to .33, indicate a moderately strong relationship between friendship andattribution similarity.

There are at least two alternative explanations for this relationship.From a demographic perspective, those who join a firm around the sametime form a cohort within which attitudes are likely to be similar (becauseof similar experiences) and friendships are likely to develop (Pfeffer, 1983;Wagner, Pfeffer, and O’Reilly, 1984). From this perspective, we wouldexpect the observed relationship between friendship and attributionalsimilarity to disappear when we control for tenure in the organization.


Table 11.5. Partial Correlations between the Friendship Network andAgreement on Each Construct Controlling for Alternative Explanations

Construct Controlling for: Gamma Z (QAP) p-Level

Flexible, tolerant Tenure .29 3.961 .0001Formal organization .26 3.735 .0001

Eats, sleeps, and breathes PacDis Tenure .27 3.327 .0005Formal organization .25 3.302 .0005

Prepared to cut corners Tenure .27 3.855 .0001Formal organization .25 3.777 .0001

Efficient, organized Tenure .31 4.277 .0001Formal organization .27 4.081 .0001

Aggressive, competitive Tenure .22 2.872 .005Formal organization .18 2.619 .005

Straightforward, blunt Tenure .26 3.203 .001Formal organization .20 2.911 .005

Task-oriented Tenure .27 4.355 .0001Formal organization .25 4.098 .0001

To test for this alternative explanation, we used a multiple regressionextension of QAP (Krackhardt 1987b, 1988). To partial out the effectsof tenure, we created an N × N matrix whose (i, j) cell was set equalto one if i and j arrived the same year at PacDis (i.e., were cohorts inthe same entering “class”); otherwise the cell (i, j) was set equal to zero.Table 11.5 shows that the hypothesized relationship remained strong andsignificant, even controlling for tenure. The p-values ranged from .005 to.0001 and the gammas ranged from .22 to .31 (compared to .24 to .33 iftenure is not controlled for).

The second alternative explanation that we considered derives from theidea that people in similar organizational positions make similar kindsof judgments (Walker, 1985). Perhaps people make friendship choicesfrom among those in similar roles, and thus the observed correlationbetween friendship and attribution similarity is spurious. To test thisalternative explanation, we controlled for formal organizational position.An N × N matrix describing the formal organization was created suchthat the (i, j) cell was set equal to one if i reported to j in the formalorganizational chart; the (i, j) cell was set to zero otherwise. As the resultsin Table 11.5 show, the hypothesized relation between friendship andattributional similarity was still highly significant (p-values ranging from.005 to .0001), and the gammas, ranging from .18 to .27, continued toindicate a moderately strong correlation.

Thus the hypothesis that pairs of friends, compared to pairs of non-friends, would be more similar in how they construed their fellow work-ers received strong support. The relationship between friendship and


Table 11.6. Correlations between Attitudes and AverageAgreement with Friends

Construct r t-Value p-Level

Flexible, tolerant .48 3.627 .0005Eats, sleeps, and breathes PacDis .47 3.615 .0005Prepared to cut corners .69 6.355 .0001Efficient, organized .47 3.540 .0005Aggressive, competitive .22 1.484 NSStraightforward, blunt .45 3.384 .001Task-oriented .23 1.620 NS

attributional similarity remained significant even controlling for eithercohort or organizational structure effects.

Friends, then, tend to see the world similarly. But what if individualsdisagree with their friends? The results in Table 11.6 suggest that suchdisagreement reduces job satisfaction, as predicted in hypothesis 2. Forfive of the seven dimensions, the Pearson correlations between agreementand satisfaction ranged from .45 to .69 (p-values from .001 to .0001).These high correlations indicate that whether individuals agreed or dis-agreed with how their friends viewed others in the organization had apowerful influence on their levels of job satisfaction.

Study 1 Discussion

We have described a method for uncovering the cultural constructs thatpeople use to make sense of their interpersonal experiences in organi-zations. In the first phase of the research, we found that the repertorygrid technique captured the ongoing tension in the PacDis organizationbetween established and emergent norms. The original values of flexibil-ity and people-orientation stressed by the organization’s founders werebeing challenged by a more rule-bounded and task-oriented approach.

In the second phase of the research, we confirmed that interpersonalnetworks support individual interpretations of experience, and that thesenetworks help control the diversity of possible interpretations. Interper-sonal networks are one of the media through which organizational cul-ture is maintained and challenged. Those who find support among theirfriends for idiosyncratic interpretations of the culture are more satisfiedwith their jobs than those whose interpretations run counter to friends’views.

The two-phase research design allowed us to capture some of the rich-ness of a particular organizational setting in the actual questionnaire usedto test theoretically derived hypotheses. But the present study lacks much


of the thick description and longitudinal analysis that characterizes ear-lier ethnographic studies of informal relationships at work (e.g., Dalton,1959; Whyte, 1948). In addition, the question arises as to whether per-ceptions of the structure of an organizational network are themselvesaffected by people’s structural positioning in the network. It is to thislatter question that we turn next.

Study 2

Anthropological studies of culture have emphasized the degree towhich consensus concerning kinship and other social relations servesto define different cultures (Romney and D’Andrade, 1964; Romney,Batchelder, and Weller, 1987; Romney, Weller, and Batchelder, 1986).Network structures in traditional societies determine “most of one’spositions . . . and most of what one will be expected to do” (D’Andrade,1995: 19).

An important question that any approach to culture must address isthe relationship between culture and social structure (see discussion inD’Andrade, 1984). Cultural knowledge is clustered in the minds of inter-acting individuals. The organization resembles a magnetic field “of per-sonal forces” (Barnard, 1938: 75) in which individuals and groups attractand repel each other, developing idiosyncratic interpretations of the cul-ture that are reinforced through social interactions (as shown in study1). Respondents “may give strikingly different descriptions” of the net-work relations within a particular group (Geertz and Geertz, 1975: 1).To understand the culture is to understand how the network ties betweenindividuals shape their perceptions of the social world.

The Structure of Cultural Agreement

Individuals who interact with each other are likely to have a higheragreement concerning the culture than non-interacting individuals (aswe demonstrated in study 1). Further, some relations (strong ties, forexample) are likely to produce more cultural agreement than others.

Simmel (1950) moved beyond the distinction between strong and weakties by examining the special nature of dyadic ties embedded within tri-ads. He suggested that relations embedded in a triad are stronger, moredurable, and in particular more able to produce agreement between actorsthan relations that are not so embedded. Research confirms that dyadicrelations embedded in triads (relative to dyadic relations in general) aremore stable over time (Krackhardt, 1998) and exert more pressure onpeople to conform to clique norms and behavior (Krackhardt, 1999).


Due to Simmel’s pioneering work in this area, we refer to dyadic tiesembedded in three-person cliques as Simmelian ties.

The culture of the organization is communicated through social net-works (Krackhardt and Kilduff, 1990). But part of what is communicatedis information about the social network itself. The social network is, there-fore, both the vehicle through which cultural meaning is communicatedand an important topic of cultural communication (see the related dis-cussion in D’Andrade, 1984). Clusters of individuals reinforce potentiallyidiosyncratic understandings of many aspects of organizational culture,including the structure of roles and relationships.

Advice and Friendship Relations in OrganizationsThus, cognitions about social relations are an important aspect of cul-ture. In modern organizations, informal advice and friendship relationsare critical for decision making and resource allocation (see the discussionin Krackhardt and Hanson, 1993). The structure of these networks residesas tacit knowledge in the minds of organizational members in the form ofcognitive maps (Krackhardt and Kilduff, 1999; Kumbasar et al., 1994).To the extent that people agree about the structure of advice and friend-ship relations in organizations, they share an understanding of importantaspects of the culture of the organization.

Our focus on knowledge concerning advice and friendship networksenables us to examine both instrumental and expressive domains (cf.Lincoln and Miller, 1979). Knowledge about advice relations is instru-mental in the sense that such knowledge is the key to understanding howwork gets done, how daily routine exceptions are handled, and who theexperts are in the organization. Knowledge of who goes to whom foradvice can be advantageous in short-circuiting long indirect chains ofinformation gathering in the firm. Knowledge about friendship relations,on the other hand, is useful in determining who can trust whom, who ismore likely to cooperate with whom, and who is likely to go to whosedefense in a political scrap (Krackhardt, 1992).

Dyadic and Simmelian TiesAs advice and friendship networks develop in a firm, how does culturalagreement emerge? Certainly, agreement could follow the structure ofthe ties themselves. As two people interact in an advice relationship, forexample, they are likely to share information about who else advisesothers. Dyadic ties are, therefore, likely to induce similarity in beliefsabout the advice network. Similarly, friends may influence each other intheir beliefs about who is a friend of whom in the firm.

But, if Simmel is to be believed, these similarities in perceptions shouldbe enhanced through the agreement-creating force of Simmelian triads


(see the discussion in Krackhardt, 1999). The likelihood that two friendsor advice partners will reach agreement concerning the structure of socialnetworks should increase if these two people are members of the samestrong clique. Disagreements within the clique are more likely to be medi-ated by a third party friendly to two antagonists in a three-person cliquethan in a dyad. Further, sense-making processes within such cliques arelikely to be particularly effective in providing individuals with opportu-nities to compare beliefs with similar others (cf. Festinger, 1954), thusfacilitating the process of clarification concerning many aspects of orga-nizational culture, including information about who the informal leadersare and who is connected to whom. In summary, we predict that relativeto dyads in general, dyads embedded in Simmelian triads are likely tohave higher agreement concerning who is tied to whom in the organiza-tion (hypothesis 3).

Dyads embedded in Simmelian triads (relative to dyads in general) arelikely to exhibit agreement on many other aspects of the structure ofthe social worlds in which individuals’ careers are formed. The socialstructure of organizations is likely to be opaque and subject to discussionand interpretation. For example, an important aspect of social structureis the organization of the network into cliques. From the perspective ofcoalition formation, knowledge about cliques is likely to be useful inpredicting where alliances might form (see discussion in Murnighan andBrass, 1991). The members of Simmelian triads are, we argue, likely toshare understandings concerning who is Simmelian-tied to whom. Ofcourse, dyads in general will tend to share beliefs about who is in whichinformal group, but dyads embedded in Simmelian triads are likely tohave higher agreement concerning who are embedded together in triadsin the organization (hypothesis 4).

Method

To test the general proposition that Simmelian ties produce more confor-mity in cultural beliefs than raw dyadic ties, we examined the structuralrelations and beliefs about these structural relations among employeesin three entrepreneurial firms. These firms were all small (less than twohundred employees) and involved in state-of-the-art technologies in eachof their areas. They all faced stiff competition from much larger playersin their industries but were doing well within their particular niches.

At each of the three organizations (Silicon Systems, Pacific Distributors,and High-Tech Managers – all described in Chapter 3), participants werepromised and given an overview of the findings. At all three sites, the samequestionnaire was used as described in Chapter 4. The high response rates


(varying from 92 percent to 100 percent) reduced problems associatedwith nonresponse bias.

Each respondent gave us a complete cognitive map of his or her percep-tions concerning who were friends with whom in the organization, andwho went to whom for advice about work-related matters (as describedin Chapter 3).

Dyadic TiesThe raw dyadic ties (Ri j ) were created from the locally aggregated struc-ture (Krackhardt, 1987a). The procedure was the same for both thefriendship and advice networks. A tie existed from person i to j onlyif person i claimed that i was a friend of (or asked advice from) j andperson j agreed that person i was a friend of (or asked advice from) j.Thus, a friendship or advice link from i to j was defined as existing whenboth parties agreed that it existed. If both respondents did not confirm theexistence of this relationship, then the tie was considered not to exist. Ifeither person did not fill out the questionnaire, then the other’s responsewas taken as a valid indication of the relationship. If neither of the twopeople filled out the questionnaire, then the relationship was deemed toexist if and only if the majority of others in the sample said that theparticular relationship existed.

Simmelian Ties and HypergraphsSimmelian ties are dyadic in nature (they occur between pairs of people)but they require more than dyadic information to ascertain. To generatethe S matrix of Simmelian ties from R, the matrix of raw dyadic ties,the hypergraphH matrix was first created (see Berge, 1989; Wassermanand Faust, 1994, for more information on hypergraphs). This hyper-graph recorded every instance in which an actor belonged to a completetriad (defined as a triad in which each actor was tied to every otheractor).

LetH represent the hypergraph of all N actors mapped onto the set ofcomplete triads,Hi j = 1 if and only if actor i is a member of the triad j;otherwise,Hi j = 0. We can use this representation to uncover Simmelianties by multiplying the matrix form ofH by its transpose and then takingthe boolean of that matrix:

S = bool [HH]

S will be an N×N matrix such thatSi j = 1, if actors i and j are Simmelian-tied to each other, andSi j = 0 otherwise. One implication of this S matrixis that it not only reveals who are in the same connected triple but also,by implication, who are in the same strongly connected informal group


A B

C D

Person 2’s Perception of i’s

Relation to j i tied to j i not tied

Person 1’s Perception of i’s

Relation to j

to j

to j

i tied to j

i not tied

Figure 11.2. Comparison of network perceptions of two people.

or clique in the organization (Krackhardt, 1999). That two actors areSimmelian-tied implies that they are comembers of the same clique andvice versa; or, in other words, S is a dichotomized clique comembershipmatrix.

Cultural Agreement within DyadsWe assessed the degree to which dyads reached an agreement concerningthe structure of social networks as follows. Each person’s cognitive slice,B, of the structure of raw ties in the organization was taken directly fromthe individual’s responses to the questionnaire (see Krackhardt, 1987a).Thus, for the advice network for respondent k,B i j(k) = 1 if and only ifperson k checked person j’s name in response to the question “Who doesi go to help or advice?”; otherwise,B i j(k) = 0. Thus, B is a matrix of whatperson k perceives the network to be.

We created a cultural similarity matrix, C, by calculating an agreementmeasure (Pearson’s r) between each pair of individuals’ perceptions asgiven in B. As illustrated in Figure 11.2, the 2×2 table reflected two indi-viduals’ perceptions of the set of dyads among all the actors in the firm.Methods of calculating Pearson’s r for such a table are numerous, usingdifferent nomenclatures (e.g., ϕ, point biserial correlation, Spearman’srho, or S14), but all yield identical values (for a discussion, see Harris,1975: 226). We used the S14 formula to calculate r (Gower and Legendre,1986; see Krackhardt, 1990, for an example).

The four cells in Figure 11.2 contain frequency counts. For example,A is the number of dyads where both persons 1 and 2 agree a tie goesbetween two other actors in the system (from i to j); B is the numberof dyads where person 1 claimed a tie existed from i to j and person 2claimed that no tie existed from i to j, and so on. C, then, is the matrixof these measures of agreement, whereCi j = r for the corresponding cellvalues in B(i) (B as perceived by respondent i) and B(j) (B as perceived byrespondent j).


To calculate the extent of dyadic agreement concerning Simmelian ties,we had to convert each person k’s belief matrix B(k) to a correspondingperceived Simmelian tie matrix, S(k). Each S(k) was created in the samemanner as before. A hypergraph matrixH(k) was created based on B(k)and then converted to a boolean S(k). We created a matrix of agreementconcerning these Simmelian ties, K, by calculating r for each pair of S(k).Thus,K i j is the Pearson correlation of the corresponding cell values in S(i)

and S(j).

Data AnalysisTo test the hypotheses, we assessed the overall relationship between thecultural agreement matrices (C and K) and the structural matrices (R andS) using the Goodman and Kruskal (1963) gamma. We chose gamma as ameasure of association because of its direct interpretation in this context:It reveals the proportional reduction in error in “guessing” whether onepair of people will be in more agreement than a second pair given thatthe first pair of people is related (directly or Simmelian-tied) and thesecond pair is not. In other words, gamma tells us the extent to whichour theory is making correct predictions (those people linked together aremore similar in their perceptions than those who are not linked together).

The other important methodological issue to be raised here is thatC, S, and K are all symmetric matrices by construction. However, R(the matrix of raw dyadic ties) is not symmetric; indeed, many of theadvice ties themselves are asymmetric. It is much more difficult for twomatrices, one being symmetric and the other being nonsymmetric, tobe strongly correlated with each other than two symmetric matrices.Thus, our predictions that Simmelian tie structures (which are symmetric)predict cultural agreement (also symmetric) better than raw dyadic ties(which are nonsymmetric) would become artificially supported.

To eliminate this source of substantial bias, we temporarily sym-metrized R (using a union rule for symmetry) before calculating the cor-relation (gamma) between R and the two cultural agreement matricesC and K. This union rule (as opposed to intersection rule) was chosenbecause it more closely reflects a solid theoretical interpretation of thesocial phenomenon under scrutiny. To see this, consider the two casesseparately, one with a union rule and one with an intersection rule. Forthe advice network (the more asymmetric of the two relations under con-sideration), a tie is retained in the symmetrized version if either person igoes to person j or vice versa. In either of these cases, interaction occursbetween i and j, and this interaction (no matter who initiates it) can leadto the exchange of information and influence in assessing what the rest ofthe network looks like. If we were to restrict the symmetrized network toan intersection case, then those cases where one person goes to another


Table 11.7. Gamma Correlations Showing the Extent to Which (forAdvice Networks in Three Organizations) Dyads Linked by Raw orSimmelian Ties Exhibited Agreement Concerning Organizationwide Rawand Simmelian Ties

Dyadic Structure

Dyadic Agreement about Ties Raw Tie Simmelian Tie

Raw tie SilSys = 0. 06 SilSys = 0.20PacDis = 0.19 PacDis = 0.24HiTecMgrs = −0.10 HiTecMgrs = 0.17

Simmelian tie SilSys = 0.07 SilSys = 0.37PacDis = 0.34 PacDis = 0.49HiTecMgrs = 0.06 HiTecMgrs = 0.57

Note: The highest correlations for each site are bolded.

for advice (but not vice versa) would be ignored (set to 0). Thus, theywould be considered the same as two people who do not interact at all.It is the interaction of these people, and not just who initiates it, thatcreates the opportunity for influence and exchange of information. Thus,the union rule for symmetry makes more theoretical sense in this contextthan the intersection rule.

This temporary symmetry adjustment permitted us to interpret thegamma correlation between the raw network R and C (or K) as theextent to which a tie from either i to j or from j to i, through interactionand the exchange of information, predicted cultural agreement between iand j.

Results

Our hypotheses, derived from cultural agreement theory, were that dyadsembedded in Simmelian triads (relative to dyads in general) would exhibitgreater agreement in the social structure of the organization. Table 11.7presents the results of tests of the hypotheses for the network of advicerelations.

In general, the results supported the hypotheses. Specifically, whentwo people joined by an advice relation were embedded in a three-personadvice clique, then those two people were more likely to have higher agree-ment concerning which other people in the organization were (a) joinedby an advice relation and (b) joined by an advice relation that was embed-ded in a clique. The gamma correlations on the right side of Table 11.7are larger than those on the left side, showing that dyads embedded in Sim-melian triads had higher agreement than ordinary dyads. The strongest


Table 11.8. Gamma Correlations Showing the Extent to Which (forFriendship Networks in Three Organizations) Dyads Linked by Raw orSimmelian Ties Exhibited Agreement Concerning OrganizationwideRaw Ties and Simmelian Ties

Dyadic Structure

Dyadic Agreement about Ties Raw Tie Simmelian Tie

Raw tie SilSys = 0.42 SilSys = 0.65PacDis = 0.50 PacDis = 0.64HiTecMgrs = 0.49 HiTecMgrs = 0.67

Simmelian tie Sil Sys = 0.33 SilSys = 0.44PacDis = 0.31 PacDis = 0.53HiTecMgrs = 0.56 HiTecMgrs = 0.57

Note: The highest correlations for each site are bolded.

result was found in the bottom-right quadrant of the table, where the gam-mas for the three organizations were .37, .49, and .57. Being a member ofa Simmelian advice clique appeared to predict particularly high agreementconcerning which other people were embedded in similar cliques.

Dyads embedded in Simmelian advice cliques may be prone to reachingagreement concerning the structure of social worlds, but is the sametrue for the friendship network? The answer is yes. Table 11.8 showsthat, relative to ordinary friendship dyads, dyads embedded in Simmelianfriendship cliques tended to have high agreement concerning both aspectsof social structure that we investigated. All of the correlations on theright side of Table 11.8 are higher than the corresponding correlationson the left side of the table, indicating that Simmelian-tied dyads tended tohave higher agreement concerning (a) who was friends with whom in theorganization and (b) which friendship pairs were embedded in Simmeliantriads.

Comparing Table 11.8 to Table 11.7, we see that the correlationsfor friendship agreement were consistently higher than those for adviceagreement. Further, whereas dyads embedded in Simmelian advice triadstended to reach the highest agreement on the question of which advicepairs were similarly embedded in Simmelian triads, the pattern was dif-ferent for the friendship network. The highest friendship correlations arein the top-right quadrant of Table 11.8: Dyads embedded in Simmeliantriads tended to be most in agreement concerning which others in theorganization formed friendship pairs (irrespective of whether the pairswere Simmelian-tied). The correlations for this quadrant across the threeorganizations were .64, .65, and .67.


Study 2 Discussion

The results support the idea that Simmelian-tied dyads (relative to dyadsin general) reach a higher agreement concerning the informal social struc-ture of organizations. The degree of agreement appears to vary dependingon the type of structure, the type of network, and the particular organi-zation.

In testing the effects of Simmelian ties on cultural agreement, we builton study 1’s investigation of the ways in which social structures con-strain the expression and interpretation of culture. We articulated theidea that the structure of an organization consists of the relationshipsamong the actors of that organization. In network terms, the structure isa set of dyadic statements describing who is related to whom on particulardimensions, such as friendship and advice.

The results are compatible with the Simmelian argument that we havepresented but leave room for alternative explanations. For example, thedata are binary and provide no information concerning tie strength. Analternative explanation might be that Simmelian ties are stronger ties,and that if one were to measure the strength of actors’ relations and notsimply whether actors are members of the same clique, one might findthat stronger ties predict more agreement. Although this “strength of ties”argument is plausible, such an explanation is certainly consistent with theSimmelian argument. Simmel would argue that co-cliqued relations willbe stronger relations. But if stronger ties lead to more cliquing (ratherthan the other way around), then Simmelian ties are spuriously related toagreement. Our guess is that both explanations are true: Cliques lead tostronger ties and stronger ties lead to cliques in a reciprocating processthat reinforces the relationship between Simmelian ties and agreement. Itwould be useful to have better access to “strength of tie” data to be ableto explore this alternative explanation in more detail.

Despite this possible tweaking of the underlying explanation of theseresults, we find support consistent with the theory that the social struc-ture influences cultural understandings. The relation between social struc-ture and culture appears much stronger for friendship structures than foradvice structures. It is possible that friendship structures, with their impli-cations of trust and cooperation, are more critical to the dynamic oper-ation of work organizations. More energy may be spent on monitoringand sharing information about friendships than about advice relations.

Simmelian ties predict higher levels of cultural agreement than raw ties.This appears to be true, independent of firm or of cultural domain (rawties or Simmelian ties; advice or friendship). That such group-based tiesare sources of powerful conformities speaks to the wisdom of Simmel’soriginal thesis. As Romney and his associates (1986) discovered in a


similar argument, the agreement takes on a group form, and differentgroups can create their own cultural definitions. Their insight was animportant first step. We have gone one step further in suggesting thatdyadic processes of agreement formation become particularly powerfulin the context of a specific type of group – the Simmelian triad.

General Discussion

In these two studies, we argued that cultural beliefs emerged through anegotiated process. Cultural agreements are far from uniform across theorganization but rather occur between sets of actors. Subcultures evolveas one group within the system forges agreement on one set of beliefswhile other groups emphasize different cultural truths. Culture itself,then, becomes structured to the extent that different actors agree withother specific actors within the system. At the micro level, each dyad canbe characterized by the extent to which the two individuals in the dyadagree on a particular cultural domain; this level of agreement betweenthe actors constitutes a belief relationship between those two actors. At amore macro level, the aggregate set of dyadic belief relations among theactors of an organization can be considered one aspect of the structure ofculture.

One implication of these findings is that organizational culture can beonly an imperfect management control device. To the extent that informalnetworks transmit and transmute the culture of the organization, cultureis clearly outside the control of the formal organizational socialization andreward system. A subculture can flourish among a group of friends whouse the same constructs as everyone else but interpret them differently.For example, everyone in the organization may believe in the virtues ofboth honesty and initiative, but people may differ as to how a specificbehavior such as insider trading should be interpreted. Should one viewthose engaging in insider trading with admiration, because they displaygreat initiative? Or should one condemn these traders because they aredishonest? The present research suggests that within any organizationalculture the same set of cultural values can lead to discrepant attributionsabout the same people.

In conclusion, we have found that friends significantly affect eachother’s evaluations of fellow employees on culturally relevant criteria;and dyads embedded within three-person cliques reach higher agreementconcerning who is tied to whom and who are embedded together in tri-ads in organizations. People’s attributions are to some extent controlledby the need to be in harmony with others in their networks. These net-works are likely to resist management attempts to initiate discrepant


cultural values or interpretations (cf. Siehl, 1985). The organization canbe depicted as a magnetic field in which individual components attractand repel each other (Nord, 1985). Within this fragmented universe, pairsof people establish mutually reinforcing interpretive systems. The controlof organizational diversity, therefore, may be as much an interpersonalinitiative as it is a prerogative of management manipulation.

12

Future Directions

In this book, we have emphasized the distinctiveness of the individualin the context of the structuring of organizational social networks. Thisrelationship between the micro and macro has proved elusive for networkresearch. Thus, we have renewed the call to “bring the individual backin” when conducting structural analysis (Kilduff and Krackhardt, 1994).Our objective includes helping the next generation of network researchersunderstand the benefits of simultaneously considering individuals andsocial structures.

In this last chapter, we anticipate future directions for the researchprogram described in this book. In looking to the future, we try to adoptsome of the advantages and overcome some of the limitations of existingapproaches to social network research. The influential structural holeperspective and similar work focused on actor centrality have brought awelcome focus on the agency of central individuals, but have tended todeliberately neglect the cognitions and personalities of actors in favor ofan assumption of rational pursuit of personal advantage (Burt, 1992). Bycontrast, the new surge of work focused on small worlds is welcome inbringing an emphasis on dynamics to the network field, but too often thiswork tends to treat actors as pawns subject to all-powerful system forces(e.g., Dorogovtsev and Mendes, 2003). In looking to the future, we firstreview possible extensions of cognitive social network research and thenexplore topics related to the dynamic interplay of distinctive individualsin complex social networks in organizations.

Network Structure Affects Cognition

Granovetter (1973) famously described how individuals could experi-ence cohesion within cliquelike groups that were disconnected from eachother in social space. This paradox of perceived local cohesion within

259


overall fragmentation can be fruitfully revisited by cognitive networkresearchers. Individuals’ positions in friendship networks can bias per-ceptions of the environment to the relative exclusion of more objectiveoutside views, potentially reinforcing similar views within friendship clus-ters (cf. Krackhardt and Kilduff, 1990; Chapter 7 this book). CEOs ofpoorly performing firms tend to seek advice from within their network offriends concerning perceived market opportunities (McDonald and West-phal, 2003), but the extent to which top management team perceptionsof the environment coalesce over time tends to predict firm performance(Kilduff et al., 2000). Thus, there are considerable opportunities forresearch that specifically investigates how the network of ties within andacross management teams affects the cognitive construction of strategicopportunities and how these cognitive constructions differ across com-petitive landscapes.

Future research can examine the importance of network effects on indi-vidual perceptions at the organizational level. We know that managerswhose previous organizations featured structural holes tend to be bet-ter able to see such holes in new organizational settings and are therebymore likely to forge viable top management team coalitions (Janicik andLarrick, 2005). Going further with such research may require the exam-ination of the links between network structure, perceptions, and actionsin a dynamic field of interaction. For example, it would be interesting toinvestigate the extent to which individuals occupying brokerage positionscan profit from such brokerage if the two parties connected by the brokerthemselves perceive the network opportunity and view the broker to beself-interested. Actors who span across structural holes in networks maybe able to exploit advantages only if they are seen by others to be notopenly pursuing their own agendas (Fernandez and Gould, 1994). Peopleoccupying brokerage positions in organizations are reported to benefit inmany ways (including higher salaries and faster promotions; Burt, 2004).However, we do not know the extent to which the benefits flowing tobrokers depend on the misperceptions of other (more marginally located)actors concerning their own potential for activating potential links insteadof depending on brokers. What might be the implications for membersof two different cliques of their absence of knowledge concerning theextent to which the two cliques constitute an overlapping social circle (cf.Kadushin, 1966)?

Although researchers have identified many discrete structures in orga-nizational social networks, including dyads, triads, cliques, and socialcircles, the extent to which individuals automatically encode and there-fore perceive these structures as entities in themselves remains unknown.Research suggests that there may be important differences flowing fromthe tendency to recognize certain group structures as entities (e.g.,

Future Directions 261

Campbell, 1958; McConnell, Sherman, and Hamilton, 1997; see alsoChapter 11). We also know little about people’s ability to record changesin an organizational social structure. We know, for example, that humanshave difficulty keeping track of the movements of more than four units ata time (Dehaene, 1997), but we have little research on whether membersof even relatively small organizational networks are able to accuratelyrecord changes in connections. To the extent that organizational mem-bers remain unaware of their structural constraints and opportunities,many of the purported benefits that can flow from network embedded-ness and connectedness may fail to materialize.

The organization can be understood as a marketplace of perceptionsin which different schemas compete for adoption, alerting people to dif-ferent signaling options. According to signaling theory (Spence, 1973),for a signal to be convincing, it must be difficult or expensive to produce(e.g., a Harvard diploma). How might this be relevant to network ties?High-status partners, with whom it is difficult to form ties, may serveas signals of an individual’s or an organization’s quality (see Chapter 3).The extent to which individual actors are perceived to have a high reputa-tion may depend on which perceptual framings currently dominate socialconstructions, and these perceptual framings may vary between groupsand subcultures. There may be a cognitive tipping point, such that percep-tions, shared among a few key players, may create consensus in the wholenetwork. Central actors tend to persist in seeing expected patterns, ignor-ing potentially important but fleeting information discrepant with theirexpectations (Freeman et al., 1987). The social construction of reputationcan therefore be a fragile undertaking, subject to sudden disconfirmation.Such social constructions can extend not only to individual people, butalso to the creation of “celebrity firms” (Rindova, Pollock, and Hayward,2006).

Cognition Affects Network Structure

We know that cognitive biases affect perceptions of social structure.Experimental evidence (De Soto, 1960; Freeman, 1992) suggests thatpeople think of friendship relations in terms of reciprocated ties (if Johnlikes Alan, Alan will like John) and in terms of transitive ties (if Johnhas two friends, the two friends will be friends of each other) in sup-port of Heider’s (1958) notion of a strain toward balance in relationsinvolving sentiment. People tend to bias their own friendship relations infavor of balance (thus preserving their own emotional tranquility), andthey tend to bias in favor of balance their perceptions of the relations ofcomparative strangers far removed from them in the organization (thus


economizing on the necessity of keeping track of partially learned rela-tionships). As part of this biased set of perceptions concerning friendship,people are also likely to see themselves as closer to the center of friendshipnetworks in organizations than do the other members of the organization(Kumbasar et al., 1994). Thus, people may strive to preserve a perspec-tive of a just world in which relations are ordered appropriately andin which they perceive themselves as more important (as measured bycentrality in the network) than they are regarded by others. Actual datafrom organizations tend to show surprisingly low average levels of per-ceived reciprocity and transitivity in friendship networks (see Chapter 2).Perceptions of the balanced world may, therefore, be surprisingly fragileand subject to recurrent disconfirmation, perhaps motivating individu-als to try to repair gaps in the network or to try to impose inaccurateperceptions on recalcitrant structures.

As we consider how perception structures networks, a host of impor-tant questions emerge concerning accuracy, schemas, and cognitive tiesbetween actors. Under which circumstances does it matter whether indi-viduals have accurate cognitions regarding who is connected to whom?If individuals’ accurate perceptions of advice networks (but not friend-ship networks) lead to positions of power (as cross-sectional work hasimplied; see Chapter 5), do individual differences with respect to socialintelligence predict who in the network is likely to be most accurate?High self-monitoring individuals (acutely aware of the demands of socialsituations; Snyder, 1974, 1979) tend to occupy more central positions innetworks (Chapter 7; Oh and Kilduff, forthcoming), perhaps because oftheir greater accuracy in attending to such relevant signals as nonverbalbehavior (Mill, 1984) and others’ emotions (Geizer, Rarick, and Soldow,1977; Chapter 8 of this book). Given the importance of cognitive heuris-tics in the structuring of network relations, can more accurate individualspotentially take advantage of others’ biased perceptions to promote theirown agendas?

There are many different types of network relations, but research oncognitive schemas has tended to focus almost exclusively on friendshipand influence networks (e.g., De Soto, 1960; Chapter 4 of this book).Network relations range from the primal (such as kinship, which remainsan important determinant of outcomes in the many large and small family-run firms) to the fleeting (such as homophily that can change dependingon the specific mix of people in a social context; see Chapter 6). Differentcognitive schemas may help structure different types of networks (seethe discussion of communication rules in Monge and Contractor, 2003:88). Evidence suggests that people in organizations differ in the extent towhich they develop new schemas to codify their perceptions of recurringnetwork patterns (such as structural holes; Janicik and Larrick, 2005).


To what extent is behavior a function of competition between activatednetwork schemas (cf. Macrae and Bodenhausen, 2000)?

Some schemas – such as the balance schema – tend to be chronicallyaccessible for most individuals as default options in the perception ofsocial relationships. Most of us, for example, tend to perceive friendshipas a reciprocal relationship. Evidence suggests that people can be taughtnew schemas through exposure to patterns of social relationships and thatsuch schemas can provide advantages in the structuring of relationships inorganizations (Janicik and Larrick, 2005). To the extent that schemas ingeneral are slow to change and represent generic expectations about theworld, we need to know more about how slow (i.e., schematic) learningof social network connections combines with the fast learning of novelconnections to produce cognitive maps and social consensus (cf. March,1991).

As cognitions, many disparate organizational elements can be includedin the same analysis, thus fulfilling one of the aims articulated by theactor-network theory research program (e.g., Law and Hassard, 1999)that has proved relatively intractable for the more quantitatively ori-ented social network research perspective. Building on the traditionalassumption that humans are the nodes of the network, researchers canexplore how novel kinds of ties between these nodes (including similarityof cognitions concerning technology) structure patterns of interaction.Block-modeling analysis can incorporate different kinds of cognitive tiesbetween the same set of nodes in the search for underlying structure, butit may also be possible to discover alternative structural configurationsderiving from cognitions relative to more “concrete” kinds of relations.For example, two people may be said to have a tie between them in thatthey have the same perception of the importance of the organizationaldatabase, or, alternatively, the same two people may have a tie to theextent that they both routinely input information (or extract informa-tion) from the database. The perceptual and the behavioral networks areunlikely to be identical and may differ in the extent to which they pre-dict outcomes such as the extent to which people rely on technology tomediate workflow (rather than using human beings).

The Dynamic Interplay between Distinctive Individualsand Complex Social Networks

Building on the ideas suggested so far in this chapter, we see possibil-ities for extending research in terms of the dynamic interplay betweenthe psychology of individuals and the complexity of social networkswithin which they interact. Networks in which people (as organizational


members or as representatives of organizations) constitute the nodes areunusual in that each node is itself a complex adaptive system. The nodesare constituted in part through their relationships with others in the net-work, but they also bring to any particular network idiosyncratic networkexpectations and perceptions. Thus, network stability and change involveboth the patterns of interactions within the overall network system andthe idiosyncrasies of the network actors in terms of their cognitions, per-sonalities, and expectations regarding the social network.

Therefore, social networks as complex systems are constituted by andhelp constitute the complexity of the nodes making up the system. Thisrecursiveness between complex nodes linked together in complex systemsmight be the focus of future research. Actual networks are reflected in,constituted by, and sometimes discrepant with the perceptions of indi-viduals. Both actual network patterns and perceived patterns can beapproached in terms of underlying structures.

We suggest that organizational network research could move forwardby incorporating actors’ memories and desires, their bounded rationalityand structural biases, and their creation and re-creation of structures thatexhibit both stability and change. Organizational networks change con-stantly in some respects and yet remain stable in other respects, just asorganizations can be considered both rapidly changing engines of creativ-ity (Burns and Stalker, 1961) and stable bundles of routines (Nelson andWinter, 1982). At the perceptual level, perceptions of network structuresevolve as individuals learn to perceive structural holes and other unusualfeatures of the interpersonal landscape. Perceptions tend to be stablegiven that people rely on default schemas such as the balance schema (seeChapter 4). But people can learn to change their perceptions to includenew types of schematic knowledge, such as the expectation that friendshipnetworks will exhibit surprising gaps (Janicik and Larrick, 2005). Differ-ences in perceptions can lead to differences in how people try to changethe network. Small changes of network relations at the local level – oneindividual adding a single friendship tie, for example – can have globalimplications for change for the whole network – by bringing disconnectedclusters of people much closer together, for instance.

Conventional wisdom suggests that networks tend to be relatively sta-ble, but this apparent stability can mask many types of change. For exam-ple, reciprocated relationships among strangers brought together in acollege dormitory tend to stabilize over a period of about three weeks(Newcomb, 1961). But a closer examination of these data suggests that“reciprocity never converges in any meaningful sense, but instead fluc-tuates substantially over the entire observation period” of fifteen weeks(Moody et al., 2005: 1227). Some actors form stable relations, but others


“dance between friends throughout the observation period” (Moodyet al., 2005: 1229). This combination of stability and change offers con-siderable opportunities for organizational network research.

The complexity of individuals is immense, with the human brain com-monly understood to be the most complex object in the universe. Socialnetworks of relations are also complex; even a small social network offifty individuals on one dimension such as friendship involves the presenceor absence of 2,450 ties. Further, each individual is distinctive in terms ofsocial attributes, personality, and membership in associations, whereaseach network, comprising the interactions of individuals, is distinctivein terms of size, dynamics, and structural attributes. We suggest thatorganizational network research can advance by seeking to capture thecomplexity and distinctiveness of both individuals and social networks interms of mutual constitution and change.

Structuration theory is one approach to interaction between agents andsystems that has sensitized researchers to the duality of social structure –the ways in which knowledgeable agents draw on rules and resources inconstituting and reconstituting the social structures that both enable andconstrain (Giddens, 1984). The system properties of networks are nonre-ducible to the properties or the motivations of individuals. The structuresthat evolve from the interactions of individuals take on system-specificcharacteristics (Barley, 1986) in terms of centralization and density, forexample. Organizational network research can enhance the structurationapproach by investigating the dynamic interplay between the psychologyof individuals and the complexity of social networks within which theyinteract; and by investigating how perceived and actual network systemsmutually constitute each other.

Networks are constituted in the minds of individuals as memories,thoughts, and desires. Network change can be traced in the changingperceptions of individuals concerning the creation and disappearance ofties between actors. Networks undergo constant change as actors repeland attract each other like components in a magnetic field (Barnard,1938: 75). Thus, network change tends to be messy, with links appearingand disappearing in different places rather than the whole system shiftingfrom one steady state to another. Locally shared systems of meaningare created when friendship groups form. The stability of such friendshipgroups depends on the continual efforts by members to engage in a mutualadjustment concerning how the world is perceived (see Chapter 11).

Beyond these overall research directions, we can draw specific insightsand research ideas from considering how the core constructs at the heartof network research can be reinterpreted to emphasize the dynamic inter-play of distinctive individuals in complex social networks. In the spirit


of Lakatos (1970), we offer emergent research directions regarding rela-tional ties, embeddedness, the social utility of connections, and structuralpatterning. The goal is not to delineate small puzzles whose outcomes arealready predetermined (as advocated from a paradigm perspective on sci-entific progress; Kuhn, 1962), but to signal new ideas about phenomenaunanticipated by conventional wisdom.

Primacy of Relations

Social networks can be understood as sets of interlinked actors continu-ally forming and reforming, continually in the process of becoming. Socialnetworks, as outcomes of human agency, carry with them in the cogni-tions of their members, memories of their past states, as well as hopesof their future states (Emirbayer and Mische, 1998: 963). At the levelof the individual actors, the social network includes absent actors in theform of their memories in the minds of actors currently present in the net-work. Like the legendary Japanese soldier who retained his organizationalloyalty during decades of hiding in the Philippine jungle, organizationssurvive in the memories and purposes of their actors. Some organizationsprocess people through their cultures and then return them to the exter-nal world. Examples include universities and military organizations. Wesuggest that organizational network research can incorporate a focus onthe neglected phenomenon of the influence of exiles on the organizationsthey have left behind.

Thus, the tension between stability and change is affected by connec-tions (both cognitive and actual) to absent nodes that continue as anactive force within the network. People who remain in the organizationselectively remember the history of who was in which office, who usedto say what at meetings, and who could be relied upon to help whentimes were difficult. Similarly, the continuing activities of some of thoseno longer formally part of the current network continue to be impor-tant to the network. Prominent exiles continue to influence the networkfrom afar through their examples of what can be achieved by networkmembers. Their successes are envied, their failures commented on, theirups and downs serving as important social comparison indicators. Thenetwork may be considered a virtual set of nodes that stretches backwardin time, forward to include those anticipated to join, and is dispersed spa-tially to include those whose continuing histories are vividly present (asexiles) even though formally they have no official links. When a personleaves an organization, this signals the appropriateness of exit for all thoseindividuals who play similar roles in the social network (cf. Chapter 9).


These signals affect not just interpersonal relations but also interorga-nizational networks. One recent example, in the realm of intercollegiateathletics, involved three members of the Big East athletic network suc-cessively yielding to the temptation to switch allegiance to the AtlanticCoast Conference (Spirer, 2003).

The social network, therefore, exists as layer upon layer of relationsbuilt up over time and space in the cognitions of members. The past,for some members, may be more vividly present than the present. Old-timers may wander the halls carrying on imaginary conversations withcolleagues long departed (cf. Berendt, 1994). Relative newcomers maymourn the departure of high-flying friends to other organizations, andmay benchmark their own ideas and progress against those of people withwhom they rarely share a conversation. Current members of governingcoalitions in organizations are likely to be influenced by those temporar-ily out of power. Deal makers may operate behind the scenes to influenceappointments and policies. For each individual in the network, its reachmay be idiosyncratically defined, shaped by memory and desire, reach-ing outside the set of obvious colleagues to include those forgotten byothers.

An organizational example of the importance of nostalgia for vanishedtimes discussed the case of a university faculty who mourned the van-ished past in which standards were higher, purposes were clearer, andcohesion in the network of faculty and students was greater (Brown andHumphreys, 2002). Such nostalgia can constitute a powerful integrat-ing narrative of resistance to change and can propel actors to organizeagainst apparently overwhelming forces for modernization (Welcomer,Gioia, and Kilduff, 2000). Nostalgia provides a framework for interpret-ing knowledge flows, a framework that may differ sharply from othersense-making recipes.

To summarize this section, therefore, is to emphasize that apparentstability masks continual adjustment and negotiation. The core conceptof the primacy of relations must be understood to include virtual actorswhose “ghost” ties may constrain and enable network member actions.(See the related discussion of how past relationships continue to affectcurrent relationships in Moody et al., 2005.) An important new researchdirection involves, therefore, a focus on how the history of the network,retained in the selective memories and interactions of its members, influ-ences network change (Soda, Usai, and Zaheer, 2004). We propose thatto the extent that members of the social network retain and rememberghost ties to former members of the network, this ongoing strengtheningof relationships with exiles will restrict the extent to which the internalorganizational network can shape cognition and behavior.


Dynamic Embeddedness

We conceive networks as evolving through a dynamic embeddednessprocess that takes into account both individual network positions aswell as system-level network change. We build on recent work showingthat individuals in the same network may be embedded in idiosyncraticpositions that subject them to unique constraints and opportunities.

An individual who is a member of many different cliques is potentiallysubject to the distinctive norms and values of all of those cliques. If theindividual shares membership in many cliques with one or more otherindividuals, these co-clique members may function as watchdogs, alert tosigns that the individual is violating in one clique the norms and valuesimportant to members in one of the other cliques to which they bothbelong. (See our discussion of Simmelian ties in Chapter 11; Krackhardt,1998, 1999.) Building on this work, we emphasize how actors in theselocal structures can experience significant network change even thoughtheir direct ties remain the same. We propose that, as friends’ friendschange, actors may find themselves serving as central conduits for theexchange of knowledge and other resources between distant actors towhom they are tied only indirectly. Because of changes in ties over whichthey have no control, their structural positions shift. (For a visualizationof how actors’ positions can change even though their own ties do notchange, see Moody et al., 2005.) Changes in network ties far from theindividual can, therefore, affect individual outcomes in ways not currentlyincorporated in research that emphasizes the importance of the direct tiesof actors.

Individuals are likely to differ in their ability to notice and respond tothese changes in embeddedness in the larger social environment. Someindividuals (high self-monitors) are acutely aware of and responsive tothe modulations of the interlinked system that creates the roles they areable to play (Snyder, 1987). These individuals scan the system for cuesas to how to behave in ways familiar to sociological thinking concern-ing the responsiveness of individuals to the ideas and attributes of theirassociates (Kilduff, 1992). Other individuals (low self-monitors) look toa subset of the system for support for the roles they have decided to play,roles that may or may not find support or encouragement in the larger sys-tem where ideas and actions are traded and careers are traversed (Kilduffand Day, 1994; Chapter 7). New research investigating network structur-ing among a community of expatriate Korean small-business owners (Ohand Kilduff, forthcoming) shows that low self-monitors, relative to highself-monitors, are more embedded in social networks in terms of transitiv-ity (their acquaintances tend to be mutual acquaintances of each other),betweenness centrality (the acquaintances of their acquaintances tend to


be acquainted), and diversity of ties outside the community. The indi-vidual’s personality appears to have ripple effects across social structure.A fruitful research direction involves investigating how self-monitoringorientation contributes to the ongoing re-creation of the network sys-tem of constraint and facilitation in ways unanticipated by sociologicalapproaches that treat structural positions as system constraints ratherthan as emergent properties of the interactions of distinctive individuals.

Each actor, occupying a distinctly different position in the network,possesses a cognitive map of all the connections between all of the actorsin the network (Krackhardt, 1987a). Each actor sees the network differ-ently. Thus, if the perspectives of all the different actors in a forty-eight-person network are collected, it may appear that there are forty-eightdifferent networks. Some of these cognitive maps are more accurate thanother maps. Accuracy refers to the degree to which an actor’s cogni-tive map of ties overlaps with a consensus map (e.g., Chapter 5). Someactors will have only confused perceptions of the quality of relationshipsbetween network members whereas other actors will be able to describesuch relationships with great clarity in terms of their strength, frequency,existence of mutual admiration, and so on. Accuracy of perceptions ofnetworks is likely to predict the skill with which actors engage in socialinteraction.

Occupants of the same social space may anticipate very different ver-sions of the social network to which they both belong. Actors, embeddedcognitively in their own perceptions of social networks, and drawing fromtheir biased perceptions of social ties, may, we suggest, attempt to enactidiosyncratic structures of constraint and opportunity at the local level.Changes to local level structures can drastically affect global propertiesof networks (Robins, Pattison, and Woolcock, 2005).

Meanings and other resource flows may tend to move through rathernarrow conduits that can compress knowledge, distort it, or excludeimportant parts of it. We know that certain types of network connec-tions can handle richer streams of knowledge than other types (Hansen,1999; Tsai, 2002). But there is little research on how the embedded-ness of individual actors can interrupt or supplement flows of knowledgeacross networks. We suggest that actors embedded in relatively openstructures, with ties to several clusters, may become experienced facilita-tors of new knowledge flow, whereas actors in relatively closed structuresmay block incoming knowledge flow discrepant with taken-for-grantedassumptions. Further, we think it likely that certain signals, because ofthe asymmetric nature of network ties, may fail to be amplified above thethreshold necessary to move beyond a certain status level in the organi-zational network. There is a rich set of research opportunities relating tohow embeddedness restricts or facilitates rumor transmission (see Burt,


2001, for one review), and how such transmission across asymmetricboundaries in organizations alters social network structures.

To summarize, changes in the larger context within which local net-works are embedded can affect flows of knowledge and other resourcesthrough those local clusters (within which ties may remain stable). Theextent to which people can track changes in such global and local con-nections is likely to correlate with self-monitoring orientation and is alsolikely to promote purposive social action. Actors who change ties at thelocal level may affect overall network functioning by, for example, bridg-ing across clusters of hitherto isolated actors.

Social Utility of Network Connections

In this section, we focus on how perceptions of centrality and centraliza-tion affect the utility of network connections. There are two questionshere, one related to how actors perceive themselves and the other relatedto how actors are perceived by others.

People tend to overestimate the number of friends they actually have inorganizations (Kumbasar et al., 1994) and they may, therefore, anticipatethat they have more social capital to support initiatives than is actuallythe case. The extent of this popularity illusion is likely to differ acrossindividuals, perhaps as a function of self-monitoring orientation, giventhe greater social acuity of high self-monitors (Berscheid et al., 1976;Hosch et al., 1984; Jones and Baumeister, 1976). There may be penaltiesattached to miscalculating the extent of personal popularity in organi-zational contexts in which, for example, people jockey for support forleadership positions. A chairperson of a department who miscalculateshow much support exists for the renewal of his or her tenure may suffera damaging blow if the majority of the department members vote fornonrenewal. On the other hand, the illusion of popularity may facilitatea self-fulfilling prophecy: Those who think others like them may recipro-cate this perceived liking and thereby create the very friendship links thatinitially did not exist.

To the extent that actors in a network are connected to each other, eachactor is exposed to perceptions from both proximate and distant actors.Perceptions concerning network structures (such as who is connected towhom) are an important aspect of the shared knowledge in the mindsof organizational members that constitutes organizational culture (seeChapter 11). We suspect that slight, initial differences with respect toperceptions of popularity may be transmitted to different parts of thenetwork and can lead to accumulating advantages in actual networks(cf. the “popularity is attractive” principle – Dorogovtsev and Mendes,2003). Social networks may take on different social capital characteristics


depending on the characteristics of the central actors who emerge fromthis process.

Order – as represented by the emergence of consensus concerning whois central in the network – is possible because each actor’s perceptionsare appraised in the eyes of proximate actors. As Adam Smith famouslyobserved, “the countenance and behavior of those [we live] with . . . is theonly looking glass by which we can, in some measure, with the eyes ofother people, scrutinize the propriety of our conduct” (quoted in Bryson,1945: 161). Thus, perceptions and behavior are subject to the scrutinyand appraisal of neighboring actors, whose perceptions also flow throughthe network establishing reputations through a collective process of allthe actors in the network (see Chapter 3).

As part of this system by which centralization emerges, there maybe a tendency to perceive popular actors as being even more popularthan they really are. Given that humans, as “cognitive misers,” tend tosimplify complex social network information (see Chapter 4), people maytend to perceive networks as dominated by a few central actors ratherthan spending the cognitive energy to keep track of the fine gradationsof popularity. If there is a tendency to cognitively enhance the popularityof central actors (cf. Kilduff et al., forthcoming), this attributional bias islikely to affect important outcomes, including the extent to which peopleare perceived to be performing well in their jobs (see Chapter 3). Further,it is possible that a misattribution of popularity can enhance the possibilityof the actor actually becoming popular. For example, a researcher whosework is assumed to be highly cited is likely to receive more citations, thuspropelling the researcher further into the center of the relevant citationnetwork.

On the negative side of the ledger, it is possible that being falselyperceived to be connected to many others may increase others’ expecta-tions concerning the focal actor’s performance. Higher standards may beapplied to those perceived to be part of a central elite. These issues maytake on particular salience when the network is perceived to be central-ized around a few central actors. Those who perceive themselves to be onthe margins may self-select not to attempt to pursue options that appearto be controlled by a self-perpetuating elite.

Of course, the question of who is central and who is marginal is affectedby the structural configuration of the network itself, and by the type ofcentrality being discussed. An individual who is influential within one partof the social structure may be revealed as relatively marginal in the largercontext of the whole social system. Conversely, people perceived by localgroup members to be insignificant may, like Swann in Marcel Proust’s(2003) masterpiece, maintain close connections with (absent) kings andprinces. Further, an actor who is popular may be less influential than anactor with fewer ties who bridges between disconnected others (Brass,


1992). The meaning and relevance of network ties are likely to vary fromone social context to another even when the structural form is identical(Gould and Fernandez, 1989).

Perceptions of centrality and of centralization are, therefore, fluid inter-pretations subject to change depending upon social context. Given thewell-known asymmetry of relations within organizational networks, cen-tral players, relative to less central players, are likely to be able to commu-nicate their own reputational messages with high fidelity. Thus, networkflows are likely to be nonlinear. Most actors will be able to initiate rel-atively inconsequential network flows, but a few actors will be able toexploit the clustering and connectivity of the network to influence a largeproportion of the network members.

Structural Patterning

The small world effect (Watts, 1999) originally investigated in the 1960s(Milgram, 1967) has grown into a dominant force in structural configu-ration research (e.g., Kogut and Walker, 2001; Uzzi and Spiro, 2005). Asmall world network structure is unusual in that the network exhibits twonetwork characteristics – high local clustering and short average paths –that are normally divergent (Watts and Strogatz, 1998). Local clusteringmeans that actors in the network tend to link together in several clusters,whereas short average path length means that any actor in the networkhas a good chance of reaching any other actor through a small number ofintermediaries. Thus, the hub-and-spoke U.S. airline system is an exampleof a small world network, whereas the interstate highway system is not.There are many questions yet unasked that could be opened up from adynamic stability perspective. Structural configuration researchers neglectthe question of how actors discover the shortest paths connecting them toothers in organizational small world networks. For individual actors, thediscovery of short paths is critical to their occupation of and exploitationof strategically central positions. But such discovery may prove difficultbecause network paths are properties of the whole global network systemwhereas actors are likely to have information heavily biased toward theirown local network position (Watts, 2003). Actors who already occupycentral positions may have advantages in terms of gaining diverse infor-mation about the structure of the network through short paths. Actorsmay gain these central positions in part through small initial advantagesthat translate into accumulating network ties as the network changes andgrows over time.

From this perspective, growing networks tend to produce a surpris-ingly robust topology, with distinct regions. (See Simon, 1996, on the


evolutionary advantages of decomposable complex systems.) Such self-organized networks may prove highly resilient to disruption and highlyefficient in the transmission of information across large distances. Re-search questions might include the following: What is the effect of differ-ences in early structuring on the likelihood that small worlds will emergein particular organizational arenas? What are the consequences, in termsof the social utility at the system level, of differing network structures?

There has long been interest in the topology of human cognition (e.g.,Lewin, 1936). We know that each individual in an organization has acognitive map of the relations between all individuals (cf. Krackhardt,1990). To what extent does the system of cognitions concerning networkrelations tend to organize according to small world principles? Organizingand keeping track of organizational relationships is likely to be especiallychallenging for a difficult-to-discern relationship such as friendship. Wepropose that boundedly rational people keep track of friendship relationsin organizational settings by using a simple set of cognitive small worldrules that can be summarized as follows: Arrange people in dense clustersand connect the clusters with short paths. Cognitive small-worldednesscan, we argue, facilitate the systemwide organization of perceptions andreduce the cognitive burden of trying to keep track of hundreds of discreterelationships (cf. Kilduff et al., forthcoming).

Each person within a network might exhibit a different level of relianceon small world perceptual organization with respect to network percep-tions. Of course, researchers could compare individuals’ perceptions withactual maps of the “real” relations existing between actors. But it may bepossible to discover benefits to individuals, irrespective of accuracy, con-sequent upon the organization of perceptions according to small worldprinciples. Potential research questions could be the following: To whatextent does the organization of individuals’ cognitive maps in terms ofsmall worlds facilitate sense making and action by the individuals? Arethere cognitive biases evident in the way people organize their perceptionsin terms of small world structures, and if so, what are the advantages anddisadvantages of such biases? Is it possible to “rewire” individuals’ cogni-tions concerning organizational networks without damaging the efficacyof their social cognition as long as small world structure is preserved?

Conclusion

In this concluding chapter, we have drawn attention to the ongoingmutual constitution of complexity and distinctiveness by both networksand actors. The activities of the social actor cannot be understoodexcept in terms of the network of relationships within which the actor


is embedded; and the emergence of system-level properties cannot beunderstood except in terms of the relationships forged by individualactors. Actors, as complex systems themselves, bring distinctive quali-ties to the network that can provide initial advantages or drawbacks inthe relationship-forging process. Small initial advantages can lead to long-term structural advantages for the actor, and small changes at the levelof local networks surrounding particular actors can have large effects onsystem-level properties such as average path length. Apparent stability ofnetworks can mask many types of change, and the network system, at anypoint in time, carries memories of its past states and anticipation of itsfuture states distributed in the minds of actors. We offered several futureresearch ideas relating to, among other topics, the importance of ghostties, the likelihood that people’s cognitions of networks are organizedas small worlds, and the likelihood that individual dispositions predictembeddedness in personal, organizational, and extra-organizational net-works.

It is through systems of relationships that people are able to enacttheir desires, pursue their affections, and get work done. In studyingthe evolution of social network relationships as the reciprocally emer-gent and re-created outcomes of purposive action, we need to discoverwhy network connections bypass or avoid crossing certain territories.As one organizational theorist demanded to know some years ago, weneed to discover not just the effects of structural holes but the reasonwhy they exist in the first place (Salancik, 1995). What is the nature ofthe prohibition that prevents connections crossing between clusters? Theidiosyncrasies of social actors and the flows of meanings between themhelp sustain the social structures that guide action.

References

Agger, B. 1991. Critical theory, poststructuralism, postmodernism: Their socio-logical relevance. In W. R. Scott (ed.), Annual Review of Sociology, vol. 17:105–31. Palo Alto: Annual Reviews.

Ahn, W. K., Brewer, W. F., and Mooney, R. J. 1992. Schema acquisition from asingle example. Journal of Experimental Psychology: Learning, Memory, andCognition, 18: 391–412.

Aiken, L., and West, S. 1996. Multiple regression: Testing and interpreting inter-actions. Newbury Park: Sage.

Allison, G. T. 1971. Essence of decision. Boston: Little, Brown.Anderson, C., and Thompson, L. L. 2004. Affect from the top down: How pow-

erful individuals’ positive affect shapes negotiations. Organizational Behaviorand Human Decision Processes, 95: 125–39.

Anderson, L. R., and Tolson, J. 1989. Group members’ self-monitoring as apossible neutralizer of leadership. Small Group Behavior, 20: 24–36.

Arabie, P., and Boorman, S. 1982. Blockmodels: Developments and prospects. InH. Hudson (ed.), Classifying social data: New applications of analytic methodsfor social science research: 177–98. San Francisco: Jossey-Bass.

Argyle, M. 1987. The psychology of happiness. London: Methuen.Argyle, M., and Henderson, M. 1985. The anatomy of relationships and the rules

and skills needed to manage them successfully. London: Heinemann.Arvey, R. D., and Murphy, K. R. 1998. Performance evaluation in work settings.

Annual Review of Psychology, 49: 141–68.Auletta, K. 1986. Greed and glory on Wall Street: The fall of the house of Lehman.

New York: Random House.Baba, M. L. 1986. Business and industrial anthropology: An overview. Washing-

ton: American Anthropological Association.Bachrach, P., and Baratz, M. S. 1962. The two faces of power. American Political

Science Review, 56: 947–52.Bailey, F. G. 1969. Stratagems and spoils: A social anthropology of politics.

Oxford: Blackwell.Baker, F. B., and Hubert, L. J. 1981. The analysis of social interaction data: A

non-parametric technique. Sociological Methods and Research, 9: 339–61.

275

276 References

Balkundi, P., and Harrison, D. A. 2006. Ties, leaders, and time in teams: Stronginference about network structure’s effects on team viability and performance.Academy of Management Journal, 49: 49–68.

Balkundi, P., Kilduff, M., Barsness, Z., and Michael, J. H. 2007. Demographicantecedents and performance consequences of structural holes in work teams.Journal of Organizational Behavior, 28: 241–60.

Barley, S. R. 1986. Technology as an occasion for structuring: Evidence fromobservations of CAT scanners and the social order of radiology departments.Administrative Science Quarterly, 31: 78–108.

Barnard, C. 1938. The functions of the executive. Cambridge: Harvard UniversityPress.

Baron, R. A. 1989. Personality and organizational conflict: Effects of the typeA behavior pattern and self-monitoring. Organizational Behavior and HumanDecision Processes, 44: 196–281.

Baron, R. A., and Markman, G. D. 2000. Beyond social capital: How social skillscan enhance entrepreneurs’ success. Academy of Management Executive, 14:106–16.

Baron, R. M., and Kenny, D. A. 1986. The moderator mediator variable dis-tinction in social psychological research: Conceptual, strategic and statisticalconsiderations. Journal of Personality and Social Psychology, 51: 173–1182.

Baron-Cohen, S. 2003. The essential difference: Men, women and the extrememale brain. London: Penguin.

Barrick, M. R., and Mount, M. K. 1993. Autonomy as a moderator of therelationships between the big five personality dimensions and job performance.Journal of Applied Psychology, 78: 111–18.

Barrick, M. R., Parks, L., and Mount, M. K. 2005. Self-monitoring as a modera-tor of the relationships between personality traits and performance. PersonnelPsychology, 58: 745–67.

Barsade, S. G. 2002. The ripple effect: Emotional contagion and its influence ongroup behavior. Administrative Science Quarterly, 47: 644–75.

Barsade, S. G., Ward, A., Turner, J., and Sonnenfeld, J. 2000. To your heart’scontent: The influence of affective diversity in top management teams. Admin-istrative Science Quarterly, 45: 802–36.

Basch, J., and Fisher, C. D. 2000. Affective events-emotions matrix: A classifica-tion of work events and associated emotions. In N. M. Ashkanasy, C. Hartel,and W. Zerbe (eds.), Emotions in the workplace: 36–48. Westport/London:Quorum.

Bass, B. M. 1990. Bass & Stogdill’s handbook of leadership: Theory, researchand managerial application (3rd ed.). New York: Free Press.

Batson, C. D. 1991. The altruism question: Toward a social-psychological answer.Hillsdale: Erlbaum.

Baum, J., and Oliver, C. 1992. Institutional embeddedness and the dynamics oforganizational population. American Sociological Review, 57: 540–59.

Becker, G. 1976. The economic approach to human behavior. Chicago: Universityof Chicago Press.

References 277

Bedeian, A. G., and Day, D. V. 2004. Can chameleons lead? Leadership Quar-terly, 15: 687–718.

Bell, R. R. 1981. Worlds of friendship. Beverly Hills: Sage.Belsley, D. A., Kuh, E., and Welsch, R. E. 1980. Regression diagnostics: Identi-

fying influential data and sources of collinearity. New York: Wiley.Berendt, J. 1994. Midnight in the garden of good and evil: A Savannah story.

New York: Random House.Berge, C. 1989. Hypergraphs: Combinatorics of finite sets. North-Holland:

Amsterdam.Berkman, L. F., and Syme, S. L. 1979. Social networks, host resistance and

mortality: A nine-year follow-up study of Alameda County residents. AmericanJournal of Epidemiology, 109: 186–204.

Bernard, H. R., Killworth, P., Kronenfeld, D., and Sailer, L. 1984. The problemof informant accuracy: The validity of retrospective data. Annual Review ofAnthropology, 13: 495–517.

Berscheid, E., Graziano, W. G., Monson, T., and Dermer, M. 1976. Outcomedependency: Attention, attribution, and attraction. Journal of Personality andSocial Psychology, 34: 978–89.

Bierstadt, R. 1950. An analysis of social power. American Sociological Review,15: 730–8.

Billings, R. S., Milburn, T. W., and Schaalman, M. L. 1980. A model of crisis per-ception: A theoretical and empirical analysis. Administrative Science Quarterly,25: 300–16.

Blau, P. M. 1977. Inequality and heterogeneity: A primitive theory of socialstructure. New York: Free Press.

———. 1982. Structural sociology and network analysis: An overview. InP. V. Marsden and N. Lin (eds.), Social structure and network analysis: 273–9.Beverly Hills: Sage.

———. 1993. Multilevel structural analysis. Social Networks, 15: 201–15.Bonacich, P. 1972. Factoring and weighting approaches to status scores and clique

identification. Journal of Mathematical Sociology, 2: 113–20.Bonacich, P., Oliver, A., and Snijders, T. A. B. 1998. Controlling for size in

centrality scores. Social Networks, 20: 135–41.Borgatti, S. P., Everett, M. G., and Freeman, L. C. 2002. UCINET for Windows:

Software for social network analysis. Harvard: Analytic Technologies.Borman, W. C., and Motowidlo, S. J. 1993. Expanding the criterion domain to

include elements of contextual performance. In N. Schmitt and W. C. Borman(eds.), Personnel selection in organizations: 71–98. San Francisco: Jossey-Bass.

Bossard, J. H. S. 1945. The law of family interaction. American Journal of Soci-ology, 50: 292–4.

Bougon, M. G., Weick, K. E., and Binkhorst, D. 1977. Cognition in organizations:An analysis of the Utrecht Jazz Orchestra. Administrative Science Quarterly,22: 606–39.

Box, G. E. P., Hunter, W. G., and Hunter, J. S. 1978. Statistics for experimenters:An introduction to design, data analysis, and model building. New York: Wiley.

278 References

Boyd, J. P. 1991. Social semigroups: A unified theory of scaling and blockmodelingas applied to social networks. Fairfax: George Mason University Press.

Bradley, R. T. 1987. Charisma and social structure: A study of love and power,wholeness and transformation. New York: Paragon.

Brass, D. J. 1981. Structural relationships, job characteristics, and worker satis-faction and performance. Administrative Science Quarterly, 26: 331–48.

———. 1984. Being in the right place: A structural analysis of individual influencein an organization. Administrative Science Quarterly, 29: 518–39.

———. 1985. Men’s and women’s networks: A study of interaction patterns andinfluence in an organization. Academy of Management Journal, 28: 327–43.

———. 1992. Power in organizations: A social network perspective. In G. Mooreand J. A. Whitt (eds.), Research in politics and society: 295–323. Greenwich:JAI Press.

Brass, D. J., and Burkhardt, M. E. 1992. Centrality and power in organizations.In N. Nohria and R. G. Eccles (eds.), Networks and organizations: structure,form, and action: 191–215. Boston: Harvard Business School Press.

———. 1993. Potential power and power use: An investigation of structure andbehavior. Academy of Management Journal, 36: 441–70.

Brass, D. J., and Krackhardt, D. 1999. The social capital of twenty-first centuryleaders. In L. G. Hunt, G. E. Dodge, and L. Wong (eds.), Out-of-the-box lead-ership: Transforming the twenty-first-century army and other top-performingorganizations: 179–94. Stamford: JAI Press.

Breiger, R. L., and Ennis, J. G. 1979. Personae and social roles: The networkstructure of personality types in small groups. Social Psychology Quarterly,42: 262–70.

Breiger, R. L., Boorman, S. A., and Arabie, P. 1975. An algorithm for clusteringrelational data with application to social network analysis and comparison withmultidimensional scaling. Journal of Mathematical Psychology, 12: 328–83.

Bretz, R. D., Jr., Milkovich, G. T., and Read, W. 1992. The current state of perfor-mance appraisal research and practice: Concerns, directions, and implications.Journal of Management, 18: 321–52.

Brief, A. P., and Weiss, H. M. 2002. Organizational behavior: Affect in theworkplace. Annual Review of Psychology, 53: 279–307.

Briggs, S. R., and Cheek, J. M. 1988. On the nature of self-monitoring: Problemswith assessment, problems with validity. Journal of Personality and SocialPsychology, 54: 663–78.

Brock, T. C. 1965. Communicator-recipient similarity and decision change. Jour-nal of Personality and Social Psychology, 1: 650–4.

Brown, A. D., and Humphreys, M. 2002. Nostalgia and the narrativization ofidentity: A Turkish case study. British Journal of Management, 13: 141–59.

Bryson, G. 1945. Man and society: The Scottish inquiry of the eighteenth century.Princeton: Princeton University Press.

Burkhardt, M. E., and Brass, D. J. 1990. Changing patterns or patterns of change:The effect of a change in technology on social network structure and power.Administrative Science Quarterly, 35: 104–27.

References 279

Burns, M. O., and Seligman, M. E. P. 1989. Explanatory style across the life-span: Evidence for stability over 52 years. Journal of Personality and SocialPsychology, 56: 471–7.

Burns, T., and Stalker, G. M. 1961. The management of innovation. London:Tavistock.

———. 1994. The management of innovation (3rd ed.). Oxford: Oxford Univer-sity Press.

Burt, R. S. 1976. Positions in networks. Social Forces, 55: 93–122.———. 1977. Positions in multiple network systems, part one: A general concep-

tion of stratification and prestige in a system of actors cast as a social typology.Social Forces, 56: 106–31.

———. 1982. Toward a structural theory of action: Network models of socialstructure, perception, and action. New York: Academic Press.

———. 1983. Cohesion versus structural equivalence as a basis for networksubgroups. In R. S. Burt and M. J. Minor (eds.), Applied network analysis: Amethodological introduction: 262–82. Beverly Hills: Sage.

———. 1986. Comment. In S. Lindberg, J. S. Coleman, and S. Novak (eds.),Approaches to social theory: 105–7. New York: Russell Sage.

———. 1987. Social contagion and innovation: Cohesion versus structural equiv-alence. American Journal of Sociology, 92: 1287–1335.

———. 1992. Structural holes: The social structure of competition. Cambridge:Harvard University Press.

———. 1997. The contingent value of social capital. Administrative ScienceQuarterly, 42: 339–65.

———. 2000. The network structure of social capital. Research in OrganizationalBehavior, 22: 345–423.

———. 2001. Bandwidth and echo: Trust, information, and gossip in socialnetworks. In J. E. Rauch and A. Casella (eds.), Networks and markets: 30–74.New York: Russell Sage Foundation.

———. 2004. Structural holes and good ideas. American Journal of Sociology,110: 349–99.

———. 2005. Brokerage and closure: An introduction to social capital. Oxford:Oxford University Press.

———. 2007. Second-hand brokerage: Evidence on the importance of local struc-ture for managers, bankers, and analysts. Academy of Management Journal,50: 119–48.

Burt, R. S., and Ronchi, D. 1990. Contested control in a large manufacturingplant. In J. Weesie and H. D. Flap (eds.), Social networks through time: 121–57. Utrecht: Isor.

Burt, R. S., Jannotta, J. E., and Mahoney, J. T. 1998. Personality correlates ofstructural holes. Social Networks, 20: 63–87.

Burton, M. L., and Nerlove, S. B. 1976. Balanced designs for triads tests. SocialScience Research, 5: 247–67.

Byrne, D., and Buehler, J. A. 1955. A note on the influence of propinquity uponacquaintanceships. Journal of Abnormal and Social Psychology, 51: 147–8.

280 References

Caldwell, D. F., and O’Reilly, C. A. 1982a. Responses to failure: The effects ofchoice and responsibility on impression management. Academy of ManagementJournal, 25: 121–136.

———. 1982b. Boundary spanning and individual performance: The impact ofself-monitoring. Journal of Applied Psychology, 67: 124–77.

Cameron, K. S. and Whetten, D. A. 1981. Perceptions of organizational effec-tiveness over organizational life cycles. Administrative Science Quarterly, 26:525–44.

Cammann, C., Fichman, M., Jenkins, G. D., Jr., and Klesh, J. R. 1983. Assessingthe attitudes and perceptions of organizational members. In S. E. Seashore, E.E. Lawler III, P. H. Mirvis, and C. Cammann (eds.), Assessing organizationalchange: 71–138. New York: Wiley.

Campbell, D. T. 1958. Common fate, similarity, and other indices of the statusof aggregates of persons as social entities. Behavioral Science, 3: 14–25.

Campbell, D. T., and Fiske, D. W. 1959. Convergent and discriminant validationby the multitrait-multimethod matrix. Psychological Bulletin, 56: 81–105.

Carley, K. 1991. A theory of group stability. American Sociological Review, 56:331–54.

Carlson, M., Charlin, V., and Miller, N. 1988. Positive mood and helping behav-ior: A test of six hypotheses. Journal of Personality and Social Psychology, 55:211–29.

Carmines, E. G., and Zeller, R. A. 1979. Reliability and validity assessment.Beverly Hills: Sage.

Cartwright, D., and Harary, F. 1956. Structural balance: A generalization ofHeider’s theory. Psychological Review, 63: 277–93.

Cialdini, R. B. 1989. Indirect tactics in impression management: Beyond basking.In R. A. Giacalone and P. Rosenfeld (eds.), Impression management in theorganization: 45–56. Hillsdale: Erlbaum.

Cialdini, R. B., and Richardson, K. D. 1980. Two indirect tactics of impressionmanagement: Basking and blasting. Journal of Personality and Social Psychol-ogy, 39: 406–15.

Cialdini, R. B., Borden, R. J., Thorne, A., Walker, M. R., Freeman, S., and Sloan,L. R. 1976. Basking in reflected glory: Three (football) field studies. Journal ofPersonality and Social Psychology, 34: 366–75.

Cohen, J., and Cohen, P. 1983. Applied multiple regression: Correlation analysisfor the behavioral sciences (2nd ed.). Hillside: Erlbaum.

Cohen, S., Doyle, W. J., Skoner, D. P., Rabin, B. S., and Gwaltney, J. M. 1997.Social ties and susceptibility to the common cold. Journal of the AmericanMedical Association, 277: 1940–4.

Coleman, J. S. 1988. Social capital in the creation of human capital. AmericanJournal of Sociology, 94: S95–S120.

———. 1990. Foundations of social theory. Cambridge: Harvard UniversityPress.

Coleman, J. S., Katz, E., and Menzel, H. 1966. Medical innovation. New York:Bobbs-Merrill.

References 281

Collins, N. L., and Feeney, B. C. 2004. Working models of attachment shapeperceptions of social support: Evidence from experimental and observationalstudies. Journal of Personality and Social Psychology, 87: 363–3.

Cook, K. S., and Emerson, R. M. 1978. Power, equity, and commitment inexchange networks. American Sociological Review, 43: 712–39.

Cook, K. S., Emerson, R. M., Gillmore, M. R., and Yamagishi, T. 1983. Thedistribution of power in exchange networks: Theory and experimental results.American Journal of Sociology, 89: 275–305.

Coser, L. A. 1956. The functions of social conflict. Glencoe: Free Press.Crockett, W. H. 1982. Balance, agreement, and positivity in the cognition of

small social structures. In L. Berkowitz (ed.), Advances in Experimental SocialPsychology, vol. 15: 1–57. New York: Academic Press.

Cronbach, L. J. 1951. Coefficient alpha and the internal structure of tests. Psy-chometrika, 16: 297–334.

Cross, R., Parker, A., and Sasson, L. (eds.). 2003. Networks in the knowledgeeconomy. New York: Oxford University Press.

Cunningham, M. R. 1988. Does happiness mean friendliness? Induced mood andheterosexual self-disclosure. Personality and Social Psychology Bulletin, 14:283–97.

Cunningham, M. R., Steinberg, J., and Grev, R. 1980. Wanting to and havingto help: Separate motivations for positive mood and guilt-induced helping.Journal of Personality and Social Psychology, 38: 181–92.

Cyert, R. M., and March, J. G. 1963. A behavioral theory of the firm. EnglewoodCliffs: Prentice Hall.

Dabbs, J. M., Evans, M. S., Hopper, C. H., and Purvis, J. A. 1980. Self-monitorsin conversation: What do they monitor? Journal of Personality and SocialPsychology, 39: 278–84.

Dalton, D. R., and Todor, W. D. 1979. Turnover turned over: An expanded andpositive perspective. Academy of Management Review, 4: 225–35.

Dalton, M. 1959. Men who manage. New York: Wiley.D’Andrade, R. 1984. Cultural meaning systems. In R. A. Shweder and R. A.

LeVine (eds.), Culture theory: Essays on mind, self, and emotion: 88–119.Cambridge: Cambridge University Press.

———. 1992. Schemas and motivation. In R. D’Andrade and C. Strauss (eds.),Human motives and cultural models: 23–44. Cambridge: Cambridge UniversityPress.

———. 1995. The development of cognitive anthropology. Cambridge: Cam-bridge University Press.

Dansereau, F. 1995. A dyadic approach to leadership: Creating and nurturingthis approach under fire. Leadership Quarterly, 6: 479–90.

Dansereau, F., Graen, G., and Haga, W. J. 1975. A vertical dyad linkage approachto leadership within formal organizations: A longitudinal investigation of therole making process. Organizational Behavior and Human Performance, 13:46–78.

Darlington, R. B. 1990. Regression and linear models. New York: McGraw-Hill.

282 References

Davis, J. A. 1979. The Davis/Holland/Leinhardt studies: An overview. In P. W.Holland and S. Leinhardt (eds.), Perspectives on social network research: 51–62. New York: Academic Press.

Dawes, R. 1976. Shallow psychology. In J. Carroll and J. Payne (eds.), Cognitionand social behavior: 3–12. Hillsdale: Erlbaum.

Day, D. V., and Kilduff, M. 2003. Self-monitoring personality and work rela-tionships: Individual differences in social networks. In A. M. Ryan andM. R. Barrick (eds.), Personality and work: 205–28. San Francisco: Jossey-Bass.

Day, D. V., Schleicher, D. J., Unckless, A. L., and Hiller, N. J. 2002. Self-monitoring personality at work: A meta-analytic investigation of constructvalidity. Journal of Applied Psychology, 87: 390–401.

De Soto, C. B. 1960. Learning a social structure. Journal of Abnormal and SocialPsychology, 60: 417–21.

Degenne, A., and Forse, M. 1999. Introducing social networks. London: Sage.Dehaene, S. 1997. The number sense: How the mind creates mathematics. Oxford

University Press, New York.Dekker, D., Snijders, T. A. B., and Krackhardt, D. Forthcoming. Sensitivity of

MRQAP tests to collinearity and autocorrelation conditions. Psychometrika.Despartes, J. P., and Lemaine, J. M. 1986. Les groupes humains dans la nature:

Nouvelle analyse des distributions de leurs effectifs [Naturally occurring humangroups: A new analysis of size distributions]. Annee Psychologique, 87: 567–80.

Diener, E., Lyubomirsky, S., and King, L. A. 2001. Is happiness a good thing?The benefits of long-term positive affect. Manuscript in preparation, Universityof Illinois.

DiMaggio, P. 1991. The macro–micro dilemma in organizational research: Impli-cations of role-system theory. In J. Huber (ed.), Macro–micro linkages in soci-ology: 76–98. Newbury Park: Sage.

———. 1997. Culture and cognition. Annual Review of Sociology, 23: 263–87.Doreian, P., Kapuscinski, R., Krackhardt, D., and Szczypula, J. 1996. A brief

history of balance through time. Journal of Mathematical Sociology, 21: 113–31.

Dorogovtsev, S., and Mendes, J. 2003. Evolution of networks: From biologicalnets to the internet and WWW. Oxford: Oxford University Press.

Douglas, W. 1983. Scripts and self-monitoring: When does being a high self-monitor really make a difference? Human Communication Research, 10: 81–96.

Drazin, R., and Auster, E. R. 1987. Wage differences between men and women:Performance appraisal ratings vs. salary allocation as the locus of bias. HumanResource Management, 26: 157–68.

Dubin, R. 1957. Power and union management relations. Administrative ScienceQuarterly, 2: 60–81.

Duck, S. W. 1973. Personal relationships and personal constructs: A study offriendship formation. London: Wiley.

Dunn, W. N., Cahill, A. G., Dukes, M. J., and Ginsberg, A. 1986. The policy grid:A cognitive methodology for assessing policy dynamics. In W. N. Dunn (ed.),

References 283

Policy analysis: Perspectives, concepts, and methods: 355–75. Greenwich: JAIPress.

Durkheim, E. 1933. The division of labor in society. Glencoe: Free Press.Eagly, A. H., and Crowley, M. 1986. Gender and helping behavior: A meta-

analytic review of the social psychological literature. Psychological Bulletin,100: 283–308.

Eagly, A. H., Karau, S., and Makhijani, M. 1995. Gender and the effectivenessof leaders: A metaanalysis. Psychological Bulletin, 117: 125–45.

Eden, C., Jones, S., and Sims, D. 1983. Messing about in problems. Oxford:Pergamon Press.

Eden, D., and Leviatan, U. 1975. Implicit leadership theory as a determinant ofthe factor structure underlying supervisory behavior scales. Journal of AppliedPsychology, 60: 736–41.

Ehrlich, D., Guttman, I., Schonbach, P., and Mills, J. 1957. Post-decision exposureto relevant information. Journal of Abnormal and Social Psychology, 54: 98–102.

Elder, G. H., Jr. 1973. On linking social structure and personality. In G. H. Elder,Jr. (ed.), Linking social structure and personality: 7–22. Beverly Hills: Sage.

Emerson, R. M. 1962. Power-dependence relations. American SociologicalReview, 27: 31–41.

Emirbayer, M., and Goodwin, J. 1994. Network analysis, culture, and the prob-lem of agency. American Journal of Sociology, 99: 1411–54.

Emirbayer, M., and Mische, A. 1998. What is agency? American Journal ofSociology, 103: 962–1023.

Fayol, H. 1916. Administration industrielle et generale. Paris: Dunod.Feld, S. L., and Grofman, B. 1989. Toward a sociometric theory of representation:

Representing individuals enmeshed in a social network. In M. Kochen (ed.),The small world: 100–7. Norwood: Ablex.

Fernandez, R. M., and Gould, R. V. 1994. A dilemma of state power: Broker-age and influence in the national health policy domain. American Journal ofSociology, 99: 1455–91.

Fernandez, R. M., and Weinberg, N. 1997. Sifting and sorting: Personal contactsand hiring in a retail bank. American Sociological Review, 62: 883–902.

Fernandez, R. M., Castilla, E. J., and Moore, P. 2000. Social capital at work:Network and employment at a phone center. American Journal of Sociology,105: 1288–1356.

Festinger, L. 1954. A theory of social comparison processes. Human Relations,7: 117–40.

———. 1957. A theory of cognitive dissonance. New York: Row, Peterson.Festinger, L., and Hutte, H. A. 1954. An experimental investigation of the effect

of unstable interpersonal relations in a group. Journal of Abnormal and SocialPsychology, 49: 513–23.

Festinger, L., Schachter, S., and Back, K. W. 1950. Social pressures in informalgroups. Stanford: Stanford University Press.

Fiedler, F. E. 1971. Validation and extension of the contingency model of lead-ership effectiveness: A review of empirical findings. Psychological Bulletin, 76:128–48.

284 References

Fiedler, F. E., and House, R. J. 1988. Leadership theory and research: A reportof progress. In C. L. Cooper and I. T. Robertson (eds.), International Reviewof Industrial and Organizational Psychology: 73–92. New York: Wiley.

Fiske, A. P. 1992. The four elementary forms of sociality: Framework for a unifiedtheory of social relations. Psychological Review, 99: 689–723.

Fiske, S. T., and Taylor, S. E. 1991. Social cognition. New York: McGraw-Hill.Flynn, F. J., and Ames, D. 2006. What’s good for the goose may not be as good

for the gander: The benefits of self-monitoring for men and women in taskgroups and dyadic conflicts. Journal of Applied Psychology, 91: 272–81.

Flynn, F. J., Reagans, R. E., Amanatullah, E. T., and Ames, D. R. 2006. Self-monitoring and the perception and emergence of exchange relations. Journalof Personality and Social Psychology, 91: 1123–37.

Freeman, L. C. 1979. Centrality in social networks: Conceptual clarification.Social Networks, 1: 215–39.

———. 1992. Filling in the blanks: A theory of cognitive categories and thestructure of social affiliation. Social Psychology Quarterly, 55: 118–27.

———. 2004. The development of social network analysis: A study in the sociol-ogy of science. Vancouver: Empirical Press.

Freeman, L. C., and Romney, A. K. 1987. Words, deeds and social structure.Human Organization, 46: 330–4.

Freeman, L. C., Freeman, S. C., and Michaelson, A. G. 1988. On human socialintelligence. Journal of Social and Biological Structures, 11: 415–25.

Freeman, L. C., Romney, A. K., and Freeman, S. C. 1987. Cognitive structureand informant accuracy. American Anthropologist, 89: 310–25.

French, J. R. P., and Raven, B. 1959. The bases of social power. In D. Cartwright.(ed.), Studies in social power: 150–67. Ann Arbor: University of MichiganPress.

Friedkin, N. E. 1998. A structural theory of social influence. New York: Cam-bridge University Press.

Friedman, R. A. 1996. Defining the scope and logic of minority and femalenetwork groups: Can separation enhance integration? In G. R. Ferris (ed.),Research in Personnel and Human Resource Management, vol. 14: 307–49.Greenwich: JAI Press.

Frost, P. 2003. Toxic emotions at work: How compassionate managers handlepain and conflict. Boston: Harvard Business School Press.

Frost, P., and Robinson, S. 1999. The toxic handler: Organizational hero andcasualty. Harvard Business Review, 77: 96–106.

Funder, D. C. 1987. Errors and mistakes: Evaluating the accuracy of social judg-ment. Psychological Bulletin, 101: 75–90.

Furnham, A., and Capon, M. 1983. Social skills and self-monitoring processes.Personality and Individual Differences, 3: 311–20.

Galaskiewicz, J. 1985. Professional networks and the institutionalization of thesingle mind set. American Sociological Review, 50: 639–58.

Gangestad, S. W., and Snyder, M. 1985. To carve nature at its joints: On theexistence of discrete classes in personality. Psychological Review, 92: 317–40.

———. 2000. Self-monitoring: Appraisal and reappraisal. Psychological Bulletin,126: 530–55.

References 285

Gargiulo, M., and Benassi, M. 1999. The dark side of social capital. In S. Gabbayand R. Leenders (eds.), Social capital and liability: 298–322. Norwell: Kluwer.

Garland, J., and Beard, J. F. 1979. Relationship between self-monitoring andleader emergence across two task situations. Journal of Applied Psychology,64: 72–6.

Geertz, C. 1973. Ritual and social change: A Javanese example. In The inter-pretation of cultures: Selected essays by Clifford Geertz: 142–69. New York:Basic.

Geertz, H., and Geertz, C. 1975. Kinship in Bali. New York: Free Press.Geizer, R. S., Rarick, D. L., and Soldow, G. F. 1977. Deception and judgmental

accuracy: A study in person perception. Personality and Social PsychologyBulletin, 3: 446–9.

George, J. M., and Brief, A. P. 1992. Feeling good – doing good: A conceptualanalysis of the mood at work–organizational spontaneity relationship. Psycho-logical Bulletin, 112: 310–29.

Gerhart, B. 1987. How important are dispositional factors as determinants of jobsatisfaction? Implications for job design and other personnel programs. Journalof Applied Psychology, 72: 366–73.

Gibson, C., and Vermeulen, F. 2003. A healthy divide: Subgroups as astimulus for team learning behavior. Administrative Science Quarterly, 48:202–39.

Giddens, A. 1984. The constitution of society. Berkeley: University of CaliforniaPress.

Gnyawali, D. R., and Madhavan, R. 2001. Network structure and competitivedynamics: A structural embeddedness perspective. Academy of ManagementReview, 26: 431–45.

Goethals, G. R., and Darley, J. M. 1987. Social comparison theory: Self-evaluation and group life. In B. Mullen and G. R. Goethals (eds.), Theoriesof group behavior: 21–47. New York: Springer-Verlag.

Goodman, L. A., and Kruskal, W. H. 1963. Measures of association for cross-classifications, III: Approximate sampling theory. Journal of the American Sta-tistical Association, 58: 310–64.

Gould, R. V., and Fernandez, R. M. 1989. Structures of mediation: A formalapproach to brokerage in transaction networks. Sociological Methodology, 19:89–126.

Gouldner, A. W. 1960. The norm of reciprocity: A preliminary statement. Amer-ican Sociological Review, 25: 161–79.

———. 1973. The importance of something for nothing. In For sociology:Renewal and critique in sociology today: 260–99. New York: Basic.

Gower, J. C., and Legendre, P. 1986. Metric and euclidean properties of dissimi-larity coefficients. Journal of Classification, 3: 5–48.

Graen, G. B., and Cashman, J. 1975. A role-making model of leadership in formalorganizations: A development approach. In J. G. Hunt and L. L. Larson (eds.),Leadership frontiers: 143–66. Kent: Kent State University Press.

Graen, G. B., Novak, A. M., and Sommerkamp, P. 1982. The effects of leader-member exchange and job satisfaction design on productivity and satisfaction:

286 References

Testing a dual attachment model. Organizational Behavior and Human Per-formance, 30: 109–31.

Granovetter, M. 1973. The strength of weak ties. American Journal of Sociology,78: 1360–80.

———. 1974. Getting a job: A study of contacts and careers. Cambridge: HarvardUniversity Press.

———. 1985. Economic action and social structure: A theory of embeddedness.American Journal of Sociology, 91: 481–510.

Graziano, W., and Bryant, W. H. 1998. Self-monitoring and the self-attributionof positive emotions. Journal of Personality and Social Psychology, 74: 250–61.

Gregory, K. L. 1983. Native-view paradigms: Multiple cultures and culture con-flicts in organizations. Administrative Science Quarterly, 28: 359–76.

Griffeth, R. W., Horn, P. W., and Gaertner, S. 2000. A meta-analysis ofantecedents and correlates of employee turnover: Update, moderator tests, andresearch implications for the next millennium. Journal of Management, 26:463–88.

Gronn, P. 1999. Substituting for leadership: The neglected role of the leadershipcouple. Leadership Quarterly, 10: 41–62.

Gronn, P. C. 1983. Talk as the work: The accomplishment of school administra-tion. Administrative Science Quarterly, 28: 1–21.

Gulati, R., and Gargiulo, M. 1999. Where do interorganizational networks comefrom? American Journal of Sociology, 104: 1439–93.

Hambrick, D. C., and Finkelstein, S. 1987. Managerial discretion: A bridgebetween polar views of organizational outcomes. Research in OrganizationalBehavior, 9: 369–407.

Hambrick, D. C., and Mason, P. A. 1984. Upper echelons: The organization as areflection of its top managers. Academy of Management Review, 9: 193–206.

Hambrick, D. C., Cho, T., and Chen, M. J. 1996. The influence of top manage-ment team heterogeneity on firms’ competitive moves. Administrative ScienceQuarterly, 41: 659–84.

Handlin, H. 1992. The company built upon the golden rule: Lincoln electric.Journal of Organizational Behavior Management, 12: 151–63.

Hannan, M., and Freeman, J. 1984. Structural inertia and organizational change.American Sociological Review, 49: 149–64.

Hansen, C. H., and Hansen, R. D. 1988. Finding a face in the crowd: An angersuperiority effect. Journal of Personality and Social Psychology, 54: 917–24.

Hansen, M. T. 1999. The search-transfer problem: The role of weak ties in sharingknowledge across organization subunits. Administrative Science Quarterly, 44:82–111.

Harary, F. 1969. Graph theory. Reading: Addison-Wesley.Hargadon, A. B., and Sutton, R. I. 1997. Technology brokering and innovation

in a product development firm. Administrative Science Quarterly, 42: 716–49.Harris, R. J. 1975. A primer of multivariate statistics. New York: Academic Press.Hatfield, E., Cacioppo, J. Y., and Rapson, R. L. 1994. Emotional contagion.

Cambridge: Cambridge University Press.Hedges, L. V., and Olkin, I. 1985. Statistical methods for meta-analysis. Orlando:

Academic Press.

References 287

Heider, F. 1958. The psychology of interpersonal relations. New York: Wiley.Hermann, C. F. 1963. Some consequences of crisis which limit the viability of

organizations. Administrative Science Quarterly, 8: 61–82.Hermann, C. F. 1969. Crises in foreign policy: A simulation analysis. Indianapolis:

Bobbs-Merill.Higgins, M. C. 2001. Changing careers: The effects of social context. Journal of

Organizational Behavior, 22: 595–618.Higgins, M. C., and Kram, K. 2001. Reconceptualizing mentoring at work: A

developmental network perspective. Academy of Management Review, 26:264–88.

Hochschild, A. R. 1983. The managed heart: Commercialization of human feel-ing. Berkeley: University of California Press.

Holland, J. L. 1985. Making vocational choices: A theory of careers. EnglewoodCliffs: Prentice-Hall.

Holland, P. W., and Leinhardt, S. 1973. The structural implications of measure-ment error in sociometry. Journal of Mathematical Sociology, 3: 85–111.

———. 1977. Transitivity in structural models of small groups. In S. Leinhardt(ed.), Social networks: A developing paradigm: 49–66. New York: AcademicPress.

Holzberg, C. S., and Giovannini, M. J. 1981. Anthropology and industry:Reappraisal and new directions. Annual Review of Anthropology, 10(3):17–360.

Hooijberg, R., Hunt, J. G., and Dodge, G. E. 1997. Leadership complexity anddevelopment of the leaderplex model. Journal of Management, 23: 375–408.

Horn, P. W., and Griffeth, R. W. 1995. Employee turnover. Cincinnati: South-western.

Hosch, H. M., Leippe, M. R., Marchioni, P. M., and Cooper, D. S. 1984. Vic-timization, self-monitoring, and eye-witness identification. Journal of AppliedPsychology, 69: 28–288.

House, J. S., Umberson, D., and Landis, K. R. 1988. Structures and processes ofsocial support. Annual Review of Sociology, 14: 293–318.

House, R. J. 1977. A 1976 theory of charismatic leadership. In J. G. Hunt and L.L. Larson (eds.), Leadership: The cutting edge: 189–273. Carbondale: SouthernIllinois University Press.

Hubbell, C. H. 1965. An input-output approach to clique identification. Sociom-etry, 28: 377–99.

Hubert, L. J. 1987. Assignment methods in combinatorial data analysis. NewYork: Marcel Dekker.

Hubert, L. J., and Golledge, R. G. 1981. A heuristic method for the comparisonof related structures. Journal of Mathematical Psychology, 23: 214–26.

Hubert, L. J., and Schultz, L. 1976. Quadratic assignment as a general dataanalysis strategy. British Journal of Mathematical and Statistical Psychology,29: 129–241.

Hughes, E. C. 1946. The knitting of racial groups in industry. American Socio-logical Review, 11: 512–15.

Hunt, D. E. 1951. Studies in role concept repertory: Conceptual consistency.Unpublished master’s thesis, Ohio State University, Columbus.

288 References

Hunter, J. E., Schmidt, F. L., and Jackson, G. B. 1982. Meta-analysis: Cumulatingresearch findings across studies. Beverly Hills: Sage.

Hurtz, G. M., and Donovan, J. J. 2000. Personality and job performance: Thebig five revisited. Journal of Applied Psychology, 85: 869–79.

Huy, Q. N. 2002. Emotional balancing of organizational continuity and radicalchange: The contribution of middle managers. Administrative Science Quar-terly, 47: 31–69.

Ibarra, H. 1992. Homophily and differential returns: Sex differences in networkstructure and access in an advertising firm. Administrative Science Quarterly,37: 422–47.

———. 1993a. Personal networks of women and minorities in management: Aconceptual framework. Academy of Management Review, 18: 56–87.

———. 1993b. Network centrality, power, and innovation involvement: Determi-nants of technical and administrative roles. Academy of Management Journal,36: 471–501.

Ibarra, H., Kilduff, M., and Tsai, W. 2005. Zooming in and out: Connectingindividuals and collectivities at the frontiers of organizational network research.Organization Science, 16: 359–71.

Ickes, W. J., and Barnes, R. D. 1977. The role of sex and self-monitoring inunstructured dyadic interactions. Journal of Personality and Social Psychology,35: 315–30.

Ickes, W. J., Reidhead, S., and Patterson, M. 1986. Machiavellianism and self-monitoring: As different as “me” and “you.” Social Cognition, 4: 58–74.

Ickes, W., Holloway, R., Stinson, L., and Hoodenpyle, T. 2006. Self-monitoringin social interaction: The centrality of self-affect. Journal of Personality, 74:659–84.

Ickes, W., Stinson, L., Bissonette, V., and Garcia, S. 1990. Naturalistic socialcognition: Empathic accuracy in mixed-sex dyads. Journal of Personality andSocial Psychology, 59: 730–42.

Insko, C. A. 1981. Balance theory and phenomenology. In R. E. Petty, T. M.Ostrom, and T. C. Brock (eds.), Cognitive responses in persuasion: 309–38.Hillsdale: Erlbaum.

———. 1999. A balance-logic perspective on Kruglanski and Thompson’s single-route approach to persuasion. Psychological Inquiry, 10: 127–37.

Isen, A. M. 1970. Success, failure, attention, and reaction to others: The warmglow of success. Journal of Personality and Social Psychology, 15: 294–301.

Janicik, G. A., and Larrick, R. P. 2005. Social network schemas and the learningof incomplete networks. Journal of Personality and Social Psychology, 88:348–64.

Jawahar, I. 2001. Attitudes, self-monitoring, and appraisal behaviors. Journal ofApplied Psychology, 86: 875–83.

Jenkins, J. M. 1993. Self-monitoring and turnover: The impact of personality onintent to leave. Journal of Organizational Behavior, 14: 83–9.

Jones, S. R. G. 1990. Worker independence and output: The Hawthorne studiesreevaluated. American Sociological Review, 55: 176–90.

References 289

Jones, E. E., and Baumeister, R. 1976. The self-monitor looks at the ingratiator.Journal of Personality and Social Psychology, 44: 654–74.

Kadushin, C. 1966. The friends and supporters of psychotherapy: On social circlesin urban life. American Sociological Review, 31: 685–99.

Kahn, W. A. 1993. Caring for the caregivers: Patterns of organizational caregiv-ing. Administrative Science Quarterly, 38: 539–63.

Kahneman, D., and Tversky, A. 1973. On the psychology of prediction. Psycho-logical Review, 80: 237–51.

———. 1979. Prospect theory: An analysis of decision under risk. Econometrica,47: 263–91.

Kanter, R. M. 1977. Men and women of the corporation. New York: Basic.———. 1979. Power failure in management circuits. Harvard Business Review,

July/August: 65–75.Katz, D., and Kahn, R. L. 1966. The social psychology of organizations. New

York: Wiley.Keesing, R. M. 1974. Theories of culture. Annual Review of Anthropology, 3:

73–97.Kelly, G. A. 1955. The psychology of personal constructs. New York: Norton.Kelman, H. C. 1958. Compliance, identification and internalization: Three pro-

cesses of attitude change. Journal of Conflict Resolution, 2(5): 1–60.Kenny, D. A., and DePaulo, B. M. 1993. Do people know how others view

them? An empirical and theoretical account. Psychological Bulletin, 114: 145–61.

Kenny, D. A., Bond, C. F., Jr., Mohr, C. D., and Horn, E. M. 1996. Do weknow how much people like one another? Journal of Personality and SocialPsychology, 71: 928–36.

Kenny, D., and Zaccaro, S. 1983. An estimate of variance due to traits in leader-ship. Journal of Applied Psychology, 68: 678–85.

Kephart, W. M. 1950. A quantitative analysis of intragroup relationships. Amer-ican Journal of Sociology, 55: 544–9.

Kerlinger, F. N. 1973. Foundations of behavioral research. New York: Holt,Rinehart & Winston.

Kerr, S., and Jermier, J. M. 1978. Substitutes for leadership: Their meaning andmeasurement. Organizational Behavior and Human Performance, 22: 375–403.

Khandwalla, P. N. 1978. Crisis responses of competing versus noncompetingorganizations. In C. F. Smart and W. T. Stanbury (ed.), Studies on crisis man-agement. Toronto: Butterworth.

Kilduff, M. 1990. The interpersonal structure of decision-making: A socialcomparison approach to organizational choice. Organizational Behavior andHuman Decision Processes, 47: 270–88.

———. 1992. The friendship network as a decision-making resource: Dispo-sitional moderators of social influences on organizational choice. Journal ofPersonality and Social Psychology, 62: 168–80.

Kilduff, M., and Day, D. V. 1994. Do chameleons get ahead? The effects ofself-monitoring on managerial careers. Academy of Management Journal, 37:1047–60.

290 References

Kilduff, M., and Krackhardt, D. 1994. Bringing the individual back in: A struc-tural analysis of the internal market for reputation in organizations. Academyof Management Journal, 37: 87–108.

Kilduff, M., and Oh, H. 2006. Deconstructing diffusion: An ethnostatisticalexamination of Medical Innovation network data reanalyses. OrganizationalResearch Methods, 9: 432–55.

Kilduff, M., and Regan, D. T. 1988. What people say and what they do: The dif-ferential effects of informational cues and task design. Organizational Behaviorand Human Decision Processes, 41: 83–97.

Kilduff, M., and Tsai, W. 2003. Social networks and organizations. London:Sage.

Kilduff, M., Angelmar, R., and Mehra, A. 2000. Top management-team diversityand firm performance: Examining the role of cognitions. Organization Science,11: 21–34.

Kilduff, M., Crossland, C., Tsai, W., and Krackhardt, D. Forthcoming. Networkperceptions versus reality: A small world after all? Organizational Behaviorand Human Decision Processes.

Kilduff, M., Tsai, W., and Hanke, R. 2006. A paradigm too far? A dynamicstability reconsideration of the social network research program. Academy ofManagement Review, 31: 1031–48.

Klein, K. J., Lim, B., Saltz, J. L., and Mayer, D. M. 2004. How do they get there?An examination of the antecedents of centrality in team networks. Academy ofManagement Journal, 47: 952–63.

Knoke, D., and Burt, R. S. 1983. Prominence. In R. S. Burt and M. J. Minor(eds.), Applied network analysis: A methodological introduction: 195–222.Beverly Hills: Sage.

Knoke, D., and Kuklinski, J. H. 1982. Network analysis. Beverly Hills: Sage.Kogut, B., and Walker, G. 2001. The small world of Germany and the durability

of national networks. American Sociological Review, 66: 317–35.Kotter, J. B. 1982. What effective general managers really do. Harvard Business

Review, 60(6): 156–67.Kozlowski, S. W. J., and Kirsch, M. P. 1987. The systematic distortion hypoth-

esis, halo, and accuracy: An individual level of analysis. Journal of AppliedPsychology, 72: 252–61.

Krackhardt, D. 1987a. Cognitive social structures. Social Networks, 9: 109–34.———. 1987b. QAP partialling as a test of spuriousness. Social Networks, 9:

171–86.———. 1988. Predicting with networks: Nonparametric multiple regression anal-

ysis of dyadic data. Social Networks, 10: 359–81.———. 1990. Assessing the political landscape: Structure, cognition, and power

in organizations. Administrative Science Quarterly, 35: 342–69.———. 1992. The strength of strong ties: The importance of philos in orga-

nizations. In N. Nohria and R. Eccles. (eds.), Networks and organizations:Structure, form and action: 216–39. Boston: Harvard University Press.

———. 1993. MRQAP: Analytic versus permutation solutions. Working paper,Carnegie Mellon University, Pittsburgh.

References 291

———. 1998. Simmelian ties: Super, strong and sticky. In R. Kramer and M.Neale (eds.), Power and influence in organizations: 21–38. Thousand Oaks:Sage.

———. 1999. The ties that torture: Simmelian tie analysis in organizations.Research in the Sociology of Organizations, 16: 183–210.

Krackhardt, D., and Hanson, J. 1993. Informal networks: The company behindthe chart. Harvard Business Review, 71: 104–11.

Krackhardt, D., and Kilduff, M. 1990. Friendship patterns and culture: Thecontrol of organizational diversity. American Anthropologist, 92: 142–54.

———. 1999. Whether close or far: Social distance effects on perceived balance infriendship networks. Journal of Personality and Social Psychology, 76: 770–82.

———. 2002. Structure, culture and Simmelian ties in entrepreneurial firms.Social Networks, 24: 279–90.

Krackhardt, D., and Porter, L. T. 1985. When friends leave: A structural analy-sis of the relationship between turnover and stayers’ attitudes. AdministrativeScience Quarterly, 30: 242–61.

———. 1986. The snowball effect: Turnover embedded in communication net-works. Journal of Applied Psychology, 71: 1–6.

Krackhardt, D., and Stern, R. 1988. Informal networks and organizational crises:An experimental simulation. Social Psychology Quarterly, 51: 123–40.

Krackhardt, D., McKenna, J., Porter, L. W., and Steers, R. M. 1981. Supervisorybehavior and employee turnover: A field experiment. Academy of ManagementJournal, 24: 249–59.

Krackhardt, D., Blythe, J., and McGrath, C. 1994. Krackplot 3.0: An improvednetwork drawing program. Connections, 17(2): 53–5.

Krosnick, J. A., and Sedikides, C. 1990. Self-monitoring and self-protective biasesin use of consensus information to predict one’s own behavior. Journal ofPersonality and Social Psychology, 58: 718–28.

Kruskal, W. H. and Wallis, W. A. 1952. Use of ranks in one-criterion varianceanalysis. Journal of the American Statistical Association, 47: 583–621.

Kuethe, J. L. 1962. Social schemas. Journal of Abnormal and Social Psychology,64: 31–8.

Kuhn, T. S. 1962. The structure of scientific revolutions. Chicago: University ofChicago Press.

Kumbasar, E. A., Romney, K., and Batchelder, W. H. 1994. Systematic biases insocial perception. American Journal of Sociology, 100: 477–505.

Kunda, G. 1993. Engineering culture: Control and commitment in a high techcorporation. Philadelphia: Temple University Press.

Kunreuther, H. 1978. Disaster insurance protection: Public policy lessons. NewYork: Wiley.

Lakatos, I. 1970. Falsification and the methodology of scientific research pro-grams. In I. Lakatos and A. Musgrave (eds.), Criticism and the growth ofknowledge: 91–132. New York: Cambridge University Press.

Larson, A. 1992. Network dyads in entrepreneurial settings: A study of thegovernance of exchange processes. Administrative Science Quarterly, 37: 76–104.

292 References

Latane, B., and Arrowood, A. 1963. Emotional arousal and task performance.Journal of Applied Psychology, 47: 324–7.

Laumann, E. O., and Pappi, F. U. 1976. Networks of collective action: A Per-spective on community influence systems. New York: Academic Press.

Law, J., and Hassard, J. (eds.). 1999. Actor network theory and after. Oxford:Blackwell.

Lawler, E. E. III, Kuleck, W. J., Jr., Rhode, J. G., and Sorensen, J. E. 1975.Job choices and post-decision distance. Organizational Behavior and HumanPerformance, 13: 133–45.

Lawrence, P. R., and Lorsch, J. W. 1967. Organization and environment: Man-aging differentiation and integration. Boston: Harvard University Press.

Lee, H. 1983. Lordstown plant of General Motors case. In J. R. Gordon (ed.),A diagnostic approach to organizational behavior: 569–75. Boston: Allyn andBacon.

Lester, S., Turnley, W., Bloodgood, J., and Bolino, M. 2002. Not seeing eye toeye: Differences in supervisor and subordinate perceptions of and attributionsfor psychological contract breach. Journal of Organizational Behavior, 23:39–56.

Levine, S. P., and Feldman, R. S. 1997. Self-presentational goals, self-monitoring,and nonverbal behavior. Basic and Applied Social Psychology, 19: 505–18.

Lewin, K. 1936. Principles of topological psychology. New York: McGraw-Hill.Lewin, K., Lippitt, R., and White, R. 1939. Patterns of aggressive behavior in

experimentally created “social climates.” Journal of Social Psychology, 10:271–99.

Lincoln, J. R., and Miller, J. 1979. Work and friendship ties in organizations: Acomparative analysis of relational networks. Administrative Science Quarterly,24: 181–99.

Long, J. S. 1997. Regression models for categorical and limited dependent vari-ables. Thousand Oaks: Sage.

Longenecker, C. O., Sims, H. P., Jr., and Gioia, D. A. 1987. Behind the mask:The politics of employee appraisal. Academy of Management Executive, 1(3):183–93.

Lord, R. G., and Emrich, C. G. 2001. Thinking outside the box by looking insidethe box: Extending the cognitive revolution in leadership research. LeadershipQuarterly, 11: 551–79.

Lorrain, F. P., and White, H. C. 1971. Structural equivalence of individuals insocial networks. Journal of Mathematical Sociology, 1: 49–80.

Lucas, R. E., and Diener, E. 2003. The happy worker: Hypotheses about the roleof positive affect in worker productivity. In A. M. Ryan and M. Barrick (eds.),Personality and work: 30–59. San Francisco: Jossey-Bass.

Lukes, S. 1974. Power: A radical view. London: MacMillan.Lyubomirsky, S., King, L., and Diener, E. 2005. The benefits of frequent posi-

tive affect: Does happiness lead to success? Psychological Bulletin, 131: 803–55.

Macrae, C. N., and Bodenhausen, G. V. 2000. Social cognition: Thinking cate-gorically about others. Annual Review of Psychology, 51: 93–120.

References 293

Maissoneuve, J., Palmade, G., and Fourment, C. 1952. Selective choices andpropinquity. Sociometry, 15: 135–40.

Maitlis, S., and Ozcelik, H. 2004. Toxic decision processes: A study of emotionand organizational decision making. Organization Science, 15: 375–93.

Mandler, J. M. 1979. Categorical and schematic organization in memory. In C. R.Puff (ed.), Memory organization and structure: 259–99. New York: AcademicPress.

Mantel, N. 1967. The detection of disease clustering and a generalized regressionapproach. Cancer Research, 27: 209–20.

March, J. G. 1991. Exploration and exploitation in organizational learning. Orga-nization Science, 2: 71–87.

Mariolis, Peter. 1985. Concepts, models and measures: Toward an analyticalframework for social network analysis. Paper presented at the fifth annualSunbelt Social Network Conference, Palm Beach.

Markus, H., and Zajonc, R. B. 1985. The cognitive perspective in social psychol-ogy. In G. Lindzey and E. Aronson (eds.), Handbook of social psychology (3rd.ed.): 137–230. New York: Random House.

Martin, J., Knopoff, K., and Beckman, C. 1998. An alternative to bureaucraticimpersonality and emotional labor: Bounded emotionality at The Body Shop.Administrative Science Quarterly, 43: 429–69.

Mayer, J., Caruso, D., and Salovey, P. 2000. Emotional intelligence meets tradi-tional standards for intelligence. Intelligence, 27: 267–98.

Mayhew, B. 1980. Structuralism versus individualism: I – Shadowboxing in thedark. Social Forces, 59: 335–75.

Mayo, M. C., Meindl, J. R., and Pastor, J. C. 2003. Shared leadership in workteams: A social network approach. In C. Pierce and J. Conger. (eds.), Sharedleadership: Reframing the hows and whys of leadership: 193–214. ThousandOaks: Sage.

McClelland, D. C. 1975. Power: The inner experience. New York: Irvington.McConnell, A. R., Sherman, S. J., and Hamilton, D. L. 1997. Target entatitivity:

Implications for information processing about individual and group targets.Journal of Personality and Social Psychology, 72: 750–62.

McDonald, M. L., and Westphal, J. D. 2003. Getting by with the advice oftheir friends: CEOs’ advice networks and firms’ strategic responses to poorperformance. Administrative Science Quarterly, 48: 1–32.

McGuire, W. J. 1968. Personality and susceptibility to social influence. In E.F. Borgatta and W. W. Lambert (eds.), Handbook of personality theory andresearch: 1130–87. Chicago: Rand McNally.

———. 1984. Search for the self: Going beyond self-esteem and the reactiveself. In R. A. Zucker, J. Aronoff, and A. I. Rabin (eds.), Personality and theprediction of behavior: 73—120. New York: Academic Press.

McPherson, J. M., Popielarz, P. A., and Drobnic, S. 1992. Social networks andorganizational dynamics. American Sociological Review, 57: 153–70.

McPherson, M., Smith-Lovin, L., and Cook, J. M. 2001. Birds of a feather:Homophily in social networks. Annual Review of Sociology, 27: 415–44.

294 References

Mechanic, D. 1962. The sources of power for lower participants in complexorganizations. Administrative Science Quarterly, 7: 349–64.

Mehra, A., Kilduff, M., and Brass, D. 1998. At the margins: A distinctivenessapproach to the social identity and social networks of underrepresented groups.Academy of Management Journal, 4: 441–52.

———. 2001. The social networks of high and low self-monitors: Implicationsfor workplace performance. Administrative Science Quarterly, 46: 121–46.

Merton, R. K. 1957. Social theory and social structure. Glencoe: Free Press.Meyer, A. D. 1982. Adapting to environmental jolts. Administrative Science

Quarterly, 27: 515–37.Meyer, J., and Rowan, B. 1977. Institutionalized organizations: Formal structure

as myth and ceremony. American Journal of Sociology, 83: 340–67.Meyerson, D. 2000. If emotions were honored: A cultural analysis. In S. Fineman

(ed.), Emotion in organizations (2nd. ed.): 167–83. London: Sage.Milburn, T. W., Schuler, R. S., and Watman, K. H. 1983. Organizational crisis,

part I: Definition and conceptualization. Human Relations, 36: 1141–60.Miles, R. H., and Randolph, W. A. 1979. The organization game. Glenview:

Scott, Foresman.Milgram, S. 1967. The small world problem. Psychology Today, 1: 61–7.Milkovich, G., and Newman, J. 1990. Compensation. Homewood: Irwin.Mill, J. 1984. High and low self-monitoring individuals: Their decoding skills

and empathic expression. Journal of Personality, 52: 372–88.Miller, E. J. 1958. Technology, territory and time. Human Relations, 12: 243–

72.Miller, M. L., and Thayer, J. F. 1989. On the existence of discrete classes in

personality: Is self monitoring the correct joint to carve? Journal of Personalityand Social Psychology, 57: 143–55.

Mintzberg, H. 1973. The nature of managerial work. New York: Harper & Row.———. 1983a. Structure in fives: Designing effective organizations. Englewood

Cliffs: Prentice Hall.———. 1983b. Power in and around organizations. Englewood Cliffs: Prentice

Hall.Mizruchi, M. S. 1983. Who controls whom? An examination of the relation

between management and boards of directors in large American corporations.Academy of Management Review, 8: 426–35.

Mobley, W. H. 1982. Employee turnover: Causes, consequences and control.Reading: Addison-Wesley.

Mobley, W. H., Griffeth, R. W., and Hand, H. H., and Meglino, B. M. 1979.Review and conceptual analysis of the employee turnover process. Psychologi-cal Bulletin, 86: 493–522.

Mollica, K. A., Gray, B., and Trevino, L. K. 2003. Racial homophily and itspersistence in newcomers’ social networks. Organizational Science, 14: 123–36.

Monge, P. R., and Contractor, N. S. 2003. Theories of communication networks.Oxford: Oxford University Press.

Moody, J., McFarland, D. A., and Bender-DeMoll, S. 2005. Dynamic networkvisualization. American Journal of Sociology, 110: 1206–41.

References 295

Moreland, R. L., and Levine, J. M. 1992. The composition of small groups.Advances in Group Processes, 9: 237–80.

Mowday, R. T. 1981. Viewing turnover from the perspective of those who remain:The relationship of job attitudes to attributions of the cause of turnover. Journalof Applied Psychology, 66: 120–23.

Mowday, R. T., Steers, R. M., and Porter, L. W. 1979. The measurement oforganizational commitment. Journal of Vocational Behavior, 14: 224–7.

Mowday, R. T., Porter, L. W., and Steers, R. M. 1982. Employee-Organizationlinkages: The psychology of commitment, absenteeism, and turnover. NewYork: Academic Press.

Murnighan, J. K., and Brass, D. J. 1991. Intraorganizational coalitions. Researchon Negotiations in Organizations, 3: 283–307.

Murray, S. L., Holmes, J. G., and Griffin, D. W. 1996. The benefits of positiveillusions: Idealization and the construction of satisfaction in close relationships.Journal of Personality and Social Psychology, 70: 79–98.

Neisser, U. 1976. Cognition and reality: Principles and implications of cognitivepsychology. New York: Freeman.

Nelson, R. R., and Winter, S. G. 1982. An evolutionary theory of economicchange. Cambridge: Belknap Press of Harvard University Press.

Newcomb, T. M. 1961. The acquaintance process. New York: Holt, Reinhart,and Winston.

Newcomb, T. M., Koenig, K. E., Flacks, R., and Warwick, D. P. 1967. Persistenceand change: Bennington College and its students after twenty-five years. NewYork: Wiley.

Nord, W. R. 1985. Can organizational culture be managed? A Synthesis. InP. Frost, L. F. Moore, M. R. Louis, C. C. Lundberg, and J. Martin (eds.),Organizational culture: 187–96. Beverly Hills: Sage.

Oh, H., Chung, M. H., and Labianca, G. 2004. Group social capital and groupeffectiveness: The role of informal socializing ties. Academy of ManagementJournal, 47: 860–75.

Oh, H., and Kilduff, M. Forthcoming. The ripple effect of personality on socialstructure: Self-monitoring origins of network brokerage. Journal of AppliedPsychology.

Oh, H., Kilduff, M., and Brass, D. J. 2005. The network dilemmas of ethnicentrepreneurs: The case of Koreans in a Canadian city. Working paper, YonseiUniversity, Korea.

O’Reilly, C. A. 1977. Supervisors and peers as information sources, group sup-portiveness, and individual performance. Journal of Applied Psychology, 62:632–5.

O’Reilly, C. A., and Roberts, K. H. 1977a. Communication and performance inorganizations. Proceedings of the Academy of Management: 375–9.

———. 1977b. Task group structure, communication, and effectiveness in threeorganizations. Journal of Applied Psychology, 62: 674–81.

Ostell, A. 1996. Managing dysfunctional emotions in organizations. Journal ofManagement Studies, 33: 525–57.

Ouchi, W. G. 1980. Markets, bureaucracies, and clans. Administrative ScienceQuarterly, 25: 129–41.

296 References

Palmer, D. 1983. Broken ties: Interlocking directorates and intercorporate coor-dination. Administrative Science Quarterly, 28: 40–55.

Pastor, J. C., Meindl, J. R., and Mayo, M. C. 2002. A network effects model ofcharisma attributions. Academy of Management Journal, 45: 410–20.

Pescosolido, A. T. 2001. Informal leaders and the development of group efficacy.Small Group Research, 32: 74–93.

———. 2002. Emergent leaders as managers of group emotion. Leadership Quar-terly, 13: 583–99.

Pettigrew, A. M. 1973. The politics of organizational decision-making. London:Tavistock.

———. 1979. On studying organizational cultures. Administrative Science Quar-terly, 24: 570–81.

Pfeffer, J. 1981. Power in organizations. Boston: Pitman.———. 1983. Organizational demography. Research in Organizational Behavior,

5: 299–357.———. 1991. Organization theory and structural perspectives on management.

Journal of Management, 17: 789–803.Pfeffer, J., and Salancik, G. 1978. The external control of organizations: A

resource dependence perspective. New York: Harper & Row.Pfeffer, J., Salancik, G. R., and Leblebici, H. 1976. The effect of uncertainty on

the use of social influence in organizational decision making. AdministrativeScience Quarterly, 21: 227–45.

Pieters, G. R., Hundert, A. T., and Beer, M. 1968. Predicting organizationalchoice: A post hoc analysis [Summary]. Proceedings of the 76th Annual Con-vention of the American Psychological Association, 3: 573–4.

Podolny, J. M. 1998. Network forms of organization. Annual Review of Sociol-ogy, 24: 57–76.

———. 2001. Networks as the pipes and prisms of the market. American Journalof Sociology, 107: 33–60.

Podolny, J. M., and Baron, J. N. 1997. Resources and relationships: Social net-works and mobility in the workplace. American Sociological Review, 62: 673–93.

Podsakoff, P. M., and MacKenzie, S. B. 1997. Kerr and Jermier’s substitutesfor leadership model: Background, empirical assessment, and suggestions forfuture research. Leadership Quarterly, 8: 117–25.

Podsakoff, P. M., Todor, W. D., and Skov, R. 1982. Effect of leader contin-gent and noncontingent reward and punishment behaviors on subordinateperformance and satisfaction. Academy of Management Journal, 25: 810–21.

Popielarz, P. A. 1999. Organizational constraints on personal network formation.Research in the Sociology of Organizations, 16: 263–81.

Portes, A. 2000. The two meanings of social capital. Sociological Forum, 15:1–11.

Powell, W. W., Koput, K. W., and Smith-Doerr, L. 1996. Interorganizational col-laboration and the locus of innovation: Networks of learning in biotechnology.Administrative Science Quarterly, 41: 116–45.

References 297

Premeaux, S. F., and Bedeian, A. G. 2003. Breaking the silence: The moderatingeffects of self-monitoring in predicting speaking up in the workplace. Journalof Management Studies, 40: 1537–62.

Proust, M. 2003. Swann’s way. New York: Viking.Radloff, R. 1968. Affiliation and social comparison. In E. F. Borgatta and

W. W. Lambert (eds.), Handbook of personality theory and research: 943–58. Chicago: Rand McNally.

Reis, H. 2001. Relationship experiences and emotional well-being. In C. Ryffand B. Singer (eds.), Emotion, social relationships, and health: 57–85. Oxford:Oxford University Press.

Reynolds, L. G. 1951. The structure of labor markets. New York: Harper.Riley, D., and Eckenrode, J. 1986. Social ties subgroup differences in costs and

benefits. Journal of Personality and Social Psychology, 51: 770–8.Rindova, V. P., Pollock, T. G., and Hayward, M. L. A. 2006. Celebrity firms: The

social construction of market popularity. Academy of Management Review, 30:50–71.

Robbins, T. L., and DeNisi, A. S. 1994. A closer look at interpersonal affect as adistinct influence on cognitive processing in performance evaluations. Journalof Applied Psychology, 29: 341–53.

Roberts, K. H., and O’Reilly, C. A., III. 1979. Some correlations of commu-nication roles in organizations. Academy of Management Journal, 22: 42–57.

Roberts, N. 1986. Organizational power styles: Collective and competitive powerunder varying organizational conditions. Journal of Applied Behavioral Sci-ence, 22: 443–58.

Robins, G., Pattison, P., and Woolcock, J. 2005. Small and other worlds: Globalnetwork structures from local processes. American Journal of Sociology, 110:894–936.

Roethlisberger, F., and Dickson, W. J. 1939. Management and the worker: Anaccount of a research program conducted by the Western Electric Company,Chicago. Cambridge: Harvard University Press.

Romney, A. K., and D’Andrade, R. G. 1964. Cognitive aspects of English kinshipterms. American Anthropologist, 66: 146–70.

Romney, A. K., Batchelder, W., and Weller, S. 1987. Recent applications ofcultural consensus theory. American Behavioral Scientist, 31: 163–77.

Romney, A. K., Weller, S. C., and Batchelder, W. H. 1986. Culture as consensus:A theory of culture and informant accuracy. American Anthropologist, 88:313–38.

Rook, K. S. 1984. The negative side of social interaction: Impact on psychologicalwell-being. Journal of Personality and Social Psychology, 46: 1097–1108.

Rosenthal, R. 1978. Combining results of independent studies. PsychologicalBulletin, 85: 185–93.

Roy, D. 1954. Efficiency and “the fix”: Informal intergroup relations in a piece-work machine shop. American Journal of Sociology, 60: 255–66.

Runkel, P. J., and Peizer, D. B. 1968. The two-valued orientation of currentequilibrium theory. Behavioral Science, 13: 56–65.

298 References

Sackett, P. R., and Larson, J. R. 1990. Research strategies and tactics in industrialand organizational psychology. In M. D. Dunnette and L. M. Hough (eds.),Handbook of industrial and organizational psychology (2nd. ed.), vol. 1: 419–89. Palo Alto: Consulting Psychologists Press.

Sailer, L. D. 1978. Structural equivalence: Meaning and definition, computationand application. Social Networks, 1: 73–90.

Salancik, G. R. 1995. Wanted: A good network theory of organization. Admin-istrative Science Quarterly, 40: 345–9.

Salancik, G. R., and Pfeffer, J. 1977. Who gets power and how they hold onto it:A strategic-contingency model of power. Organizational Dynamics, 5: 3–21.

Salancik, G. R., and Pfeffer, J. 1978. A social information processing approach tojob attitudes and task design. Administrative Science Quarterly, 23: 224–53.

Sampson, E. E., and Insko, C. A. 1964. Cognitive consistency and conformityin the autokinetic situation. Journal of Abnormal and Social Psychology, 68:184–92.

Sampson, F. S. 1968. A novitiate in a period of change: An experimental andcase study of social relationships. Unpublished doctoral dissertation, CornellUniversity, Ithaca.

Saxenian, A. 1990. Regional networks and the resurgence of Silicon Valley. Cal-ifornia Management Review, 33: 89–112.

Schachter S. 1959. The psychology of affiliation. Palo Alto: Stanford UniversityPress.

Schachter, S., Willerman, B., Hyman, R., and Festinger, L. 1961. Emotionaldisruption and industrial productivity. Journal of Applied Psychology, 45: 201–13.

Schein, E. H. 1967. Attitude change during management education. Administra-tive Science Quarterly, 11: 601–28.

Schwab, D. P. 1982. Recruiting and organizational participation. In K. Rowlandand G. Ferris (eds.), Personnel management: 103–28. Boston: Allyn & Bacon.

Schwab, D. P., Rynes, S. L., and Aldag, R. J. 1987. Theories and research on jobsearch and choice. Research in Personnel and Human Resource Management,5: 129–66.

Schwarz, N. 1998. Warmer and more social: Recent developments in cognitivesocial psychology. Annual Review of Sociology, 24: 239–64.

Scott, J. 1991. Social network analysis: A handbook. London: Sage.Scott, S. G., and Bruce, R. A. 1994. Determinants of innovative behavior: A path

model of innovation in the workplace. Academy of Management Journal, 37:580–607.

Scullen, S. E., Mount, M. K., and Goff, M. 2000. Understanding the latent struc-ture of job performance ratings. Journal of Applied Psychology, 85: 956–70.

Seidel, M. D. L., Polzer, J. T., and Stewart, K. J. 2000. Friends in high places: Theeffects of social networks on discrimination in salary negotiations. Administra-tive Science Quarterly, 45: 1–24.

Selznick, Philip. 1957. Leadership in administration. Evanston: Row, Peterson.Shaffer, D. R., Smith, J. E., and Tomarelli, M. 1982. Self-monitoring as a deter-

minant of self-disclosure reciprocity during the acquaintance process. Journalof Personality and Social Psychology, 43: 163–75.

References 299

Shaw, M. E. 1981. Group dynamics: The psychology of group behavior (3rd.ed.). New York: McGraw-Hill.

Sherif, M., Harvey, O. J., White, B. J., Hood, W. R., and Sherif, C. 1961. Inter-group conflict and cooperation: The robbers cave experiment. Norman: Uni-versity Book Exchange.

Sherony, K. M., and Green, S. 2002. Coworker exchange: Relationships betweencoworkers, leader-member exchange, and work attitudes. Journal of AppliedPsychology, 87: 542–8.

Siehl, C. 1985. After the founder: An opportunity to manage culture. In P. Frost,L. F. Moore, M. R. Louis, C. C. Lundberg, and J. Martin (eds.), Organizationalculture: 125–40. Beverly Hills: Sage.

Simmel, G. 1950. The isolated individual and the dyad. The sociology of GeorgSimmel: 118–36. Glencoe: Free Press.

———. 1955. Conflict and the web of group-affiliations (K. H. Wolff andR. Bendix, trans.). New York: Free Press.

Simon, H. A. 1996. The sciences of the artificial (3rd. ed.). Cambridge: MIT Press.Smart, C. F., and Vertinsky, I. 1978. Diagnosing corporate effectiveness and

susceptibility to crisis. In C. F. Smart and W. T. Stanbury (eds.), Studies oncrisis management: 53–92. Toronto: Buttenvorth.

Smith, L. H. 1967. Some properties of ipsative, normative, and forced choicenormative measures. Philadelphia: Franklin Institute Research Laboratories.

Snyder, M. 1974. Self-monitoring of expressive behavior. Journal of Personalityand Social Psychology, 30: 526–37.

———. 1979. Self-monitoring processes. Advances in Experimental Social Psy-chology, 12: 85–128.

———.1987. Public appearances, private realities: The psychology of self-monitoring. New York: W. H. Freeman.

Snyder, M., and Cantor, N. 1980. Thinking about ourselves and others: Self-monitoring and social knowledge. Journal of Personality and Social Psychol-ogy, 39: 222–34.

Snyder, M., and DeBono, K. G. 1985. Appeals to image and claims about quality:Understanding the psychology of advertising. Journal of Personality and SocialPsychology, 49: 586–97.

Snyder, M., and Gangestad, S. 1982. Choosing social situations: Two investiga-tions of self-monitoring processes. Journal of Personality and Social Psychol-ogy, 43: 123–35.

———. 1986. On the nature of self-monitoring: Matters of assessment, mattersof validity. Journal of Personality and Social Psychology, 51: 125–39.

Snyder, M., and Monson, T. C. 1975. Persons, situations, and the control ofsocial behaviors. Journal of Personality and Social Psychology, 32: 637–44.

Snyder, M., and Simpson, J. A. 1984. Self-monitoring and dating relationships.Journal of Personality and Social Psychology, 47: 1281–91.

Snyder, M., Gangestad, S., and Simpson, J. A. 1983. Choosing friends as activ-ity partners: The role of self-monitoring. Journal of Personality and SocialPsychology, 51: 181–90.

Snyder, M., Simpson, J. A., and Gangestad, S. 1986. Personality and sexualrelations. Journal of Personality and Social Psychology, 51: 181–190.

300 References

Soda, G., Usai, A., and Zaheer, A. 2004. Network memory: The influence of pastand current networks on performance. Academy of Management Journal, 47:893–906.

Soelberg, P. O. 1967. Unprogrammed decision making. Industrial ManagementReview, 8: 19–29.

Sparrowe, R. T., and Liden, R. C. 2005. Two routes to influence: Integratingleader–member exchange and network perspectives. Administrative ScienceQuarterly, 50: 505–35.

Sparrowe, R. T., Liden, R. C., Wayne, S. J., and Kraimer, M. L. 2001. Socialnetworks and the performance of individuals and groups. Academy of Man-agement Journal, 44: 316–25.

Spence, A. M. 1973. Job market signaling. Quarterly Journal of Economics, 87:355–74.

Spirer, J. 2003. Fall and rise of the Big East. Hoya, November 14.Starbuck, W. H., Greve, A., and Hedberg, B. 1978. Responding to crises: Theory

and the experience of European business. In C. F. Smart and W. T. Stanbury(eds.), Studies on crisis management: 107–34. Toronto: Buttenvorth.

Staw, B. M. 1976. Knee-deep in the big muddy: A study of escalating com-mitment to a chosen course of action. Organizational Behavior and HumanPerformance, 16: 27–44.

———. 1980a. The consequences of turnover. Journal of Occupational Behavior,1: 253–73.

———. 1980b. Rationality and justification in organizational life. In B. M. Stawand L. L. Cummings (eds.), Research in Organizational Behavior, 2: 45–80.Greenwich: JAI Press.

Staw, B., and Barsade, S. 1993. Affect and managerial performance: A test of thesadder-but-wiser vs. happier-and-smarter hypotheses. Administrative ScienceQuarterly, 38: 304–31.

Staw, B., and Ross, J. 1985. Stability in the midst of change: A dispositionalapproach to job attitudes. Journal of Applied Psychology, 70: 469–80.

Staw, B., Bell, N., and Clausen, J. 1986. The dispositional approach to job atti-tudes: A lifetime longitudinal test. Administrative Science Quarterly, 31: 56–77.

Steel, R. P. 2002. Turnover theory at the empirical interface: Problems of fit andfunction. Academy of Management Review, 27: 346–60.

Stern, R. 1974. Market behavior in a simulated society: Illustrating the sociologyof economic phenomena. Simulation and Games, 5: 347–62.

Stevens, S. S. 1957. On the psychophysical law. Psychological Review, 64: 153–81.

———. 1962. The surprising simplicity of sensory metrics. American Psycholo-gist, 17: 29–39.

Stevenson, W. B., Pearce, J. L., and Porter, L. W. 1985. The concept of coalitionin organizational theory and research. Academy of Management Review, 10:256–68.

Swidler, A. 1986. Culture in action. American Sociological Review, 51: 273–86.Sykes, R. E., Larntz, K., and Fox, J. C. 1976. Proximity and similarity effects on

frequency of interaction in a class of naval recruits. Sociometry, 39: 263–9.

References 301

Tabachnik, B. G., and Fidell, L. S. 1996. Using multivariate statistics (3rd. ed.).New York: Harper Collins.

Tashakkori, A., and Insko, C. A. 1979. Interpersonal attraction and the polarityof similar attitudes: A test of three balance models. Journal of Personality andSocial Psychology, 37: 2262–77.

———. 1981. Interpersonal attraction and person perception: Two tests of threebalance models. Journal of Experimental Social Psychology, 17: 266–85.

Taylor, F. 1911. Principles of scientific management. New York: W. W. Norton.Taylor, S. E. 1981. A categorization approach to stereotyping. In D. L. Hamilton

(ed.), Cognitive processes in stereotyping and intergroup behavior: 83–114.Hillsdale: Erlbaum.

———. 1991. The interface of cognitive and social psychology. In J. H. Harvey.(ed.), Cognition, social behavior, and the environment: 189–211. Hillsdale:Erlbaum.

Taylor, S. E., and Brown, J. D. 1988. Illusion and well-being – a social psycho-logical perspective on mental-health. Psychological Bulletin, 103: 193–210.

Taylor, S. E., and Fiske, S. T. 1978. Salience, attention, and attribution: Top ofthe head phenomena. Advances in Experimental Social Psychology, 11: 249–88.

Taylor, S., Sherman, D., Kim, H., Jarcho, J., Takagi, K., and Dunagan, M. 2004.Culture and social support: Who seeks it and why? Journal of Personality andSocial Psychology, 87: 354–62.

Tedeschi, J. T., and Melburg, V. 1984. Impression management and influence inthe organization. In S. B. Bacharach and E. J. Lawlor (eds.), Research in theSociology of Organizations, vol. 3: 31–58. Greenwich: JAI Press.

Thomas, D. A. 1993. The dynamics of managing racial diversity in developmentalrelationships. Administrative Science Quarterly, 38: 169–94.

Thompson, J. D. 1967. Organizations in action. New York: McGraw-Hill.Thoresen, C., Kaplan, S., Barsky, A., Warren, C., and de Chermont, K. 2003.

The affective underpinnings of job perceptions and attitudes: A meta-analyticreview and integration. Psychological Bulletin, 129: 946–72.

Tichy, N. M. 1981. Networks in organizations. In P. C. Nystrom and W. H.Starbuck (eds.), Handbook of organizational design, vol. 2: 225–49. New York:Oxford University Press.

Tichy, N. M., Tushman, M. L., and Fombrum, C. 1979. Social network analysisfor organizations. Academy of Management Review, 4: 507–19

Trice, H. M. 1985. Rites and ceremonials in organizational culture. Research inthe Sociology of Organizations, 4: 271–309.

Tsai, W. 2002. Social structure of “coopetition” within a multiunit organiza-tion: Coordination, competition, and intra-organizational knowledge sharing.Organizational Science, 13: 179–90.

Tsui, A. S., and Barry, B. 1986. Interpersonal affect and rating errors. Academyof Management Journal, 29: 586–99.

Tsui, A. S., and O’Reilly, C. A., III. 1989. Beyond simple demographic effects: Theimportance of relational demography in superior-subordinate dyads. Academyof Management Journal, 32: 402–23.

302 References

Tuma, N. B., and Hallinan, M. T. 1979. The effects of sex, race and achievementon schoolchildren’s friendships. Social Forces, 57: 1286–1309.

Turner, R. G. 1980. Self-monitoring and humor production. Journal of Person-ality, 48: 163–72.

Uchino, B. N., Cacioppo, J. T., and Kiecolt-Glaser, J. K. 1996. The relationshipbetween social support and physiological processes: A review with emphasison underlying mechanisms and implications for health. Psychological Bulletin,119: 488–531.

Uzzi, B. 1996. The sources and consequences of embeddedness for the eco-nomic performance of organizations: The network effect. American Sociologi-cal Review, 61: 674–98.

———. 1997. Social structure and competition in inter-firm networks: The para-dox of embeddedness. Administrative Science Quarterly, 42: 35–67.

Uzzi, B., and Spiro, J. 2005. Collaboration and creativity: The small world prob-lem. American Journal of Sociology, 111: 447–504.

Van Maanen, J. 1983. Golden passports: Managerial socialization and graduateeducation. Review of Higher Education, 6: 435–55.

Vaux, A. 1988. Social support: Theory, research and intervention. New York:Praeger.

Viswesvaran, C., Sanchez, J., and Fisher, J. 1999. The role of social support inthe process of work stress: A meta-analysis. Journal of Vocational Behavior,54: 314–34.

von Hecker, U. 1993. The significance of balance, generalization, and positivityas cognitive rules. Zeitschrift fur Experimentelle und Angewandte Psychologie,40: 548–76.

Vroom, V. H. 1966. Organizational choice: A study of pre- and post-decisionprocesses. Organizational Behavior and Human Performance, 1: 212–25.

Vroom, V. H., and Yetton, P. W. 1973. Leadership and decision-making. Pitts-burgh: University of Pittsburgh Press.

Wagner, W. G., Pfeffer, J., and O’Reilly, C. A., III. 1984. Organizational demog-raphy and turnover in top-management groups. Administrative Science Quar-terly, 29: 74–92.

Walker, G. 1985. Network position and cognition in a computer software firm.Administrative Science Quarterly, 30: 103–30.

Wallace, A. F. C. 1970. Culture and personality. New York: Random House.Wasserman, S., and Faust, K. 1994. Social network analysis: Methods and appli-

cations. Cambridge: Cambridge University Press.Watson, D. 1988. The vicissitudes of mood measurement: The effects of varying

descriptors, time frames, and response formats on measures of positive andnegative affect. Journal of Personality and Social Psychology, 55: 128–41.

Watson, D., Clark, L., and Tellegen, A. 1988. Development and validation ofbrief measures of positive and negative affect: The PANAS scales. Journal ofPersonality and Social Psychology, 54: 1063–70.

Watts, D. J. 1999. Small worlds: The dynamics of networks between order andrandomness. Princeton: Princeton University Press.

———. 2003. Six degrees: The science of a connected age. New York: W. W.Norton.

References 303

Watts, D. J., and Strogatz, S. 1998. Collective dynamics of small world networks.Nature, 393: 440–2.

Weick, K. E. 1995. Sensemaking in organizations. Thousand Oaks: Sage.Weick, K. E., and Bougon, M. G. 1986. Organizations as cognitive maps: Chart-

ing ways to success and failure. In H. P. Sims and D. A. Gioia (eds.), Thethinking organization: Dynamics of organizational social cognition: 102–35.San Francisco: Jossey-Bass.

Weiss, H. H., and Adler, S. 1984. The role of personality in organizational behav-ior. In B. Staw and L. Cummings (eds.), Research in Organizational Behavior,vol. 6: 1–50. Greenwich: JAI Press.

Weiss, H., and Cropanzano, R. 1996. Affective events theory: A theoretical dis-cussion of the structure, causes and consequences of affective experiences atwork. Research in Organizational Behavior, 18: 1–74.

Welcomer, S., Gioia, D. A., and Kilduff, M. 2000. Resisting the discourse ofmodernity: Rationality versus emotion in hazardous waste siting. Human Rela-tions, 53: 1175–1205.

Wellman, B. 1988. Structural analysis: From metaphor to substance. In B. Well-man, and S. D. Berkowitz (eds.), Social structures: A network approach: 19–61.New York: Cambridge University Press.

Westphal, J. D., and Milton, L. P. 2000. How experience and social networksaffect the influence of demographic minorities on corporate boards. Adminis-trative Science Quarterly, 45: 366–98.

Wexler, K., and Romney, A. K. 1972. Individual variations in cognitive structures.In A. K. Romney, R. N. Shepard, and S. B. Nerlove (eds.), Multidimensionalscaling: Theory and applications in the behavioral sciences, vol. 2. New York:Seminar Press.

Wheaton, B. 1974. Interpersonal conflict and cohesiveness in dyadic relationships.Sociometry, 37: 328–48.

Wherry, R. J., and Bartlett, C. J. 1982. The control of bias in ratings. PersonnelPsychology, 35: 521–51.

White, D. R., and Reitz, K. P. 1983. Graph and semigroup homomorphisms ornetworks of relations. Social Networks, 5: 193–234.

———. 1985. Rethinking the role concept: Homomorphisms on social networks.In L. Freeman, D. White, and A. K. Romney (eds.), Research methods in socialnetwork analysis. Unpublished manuscript, University of California, Irvine.

White, H. C. 1970. Chains of opportunity. Cambridge: Harvard University Press.———. 1992. Identity and control: A structural theory of social action. Princeton:

Princeton University Press.White, H. C., Boorman, S. A., and Breiger, R. L. 1976. Social structure from

multiple networks: Blockmodels of roles and positions. American Journal ofSociology, 81: 730–80.

Whitmore, M. D., and Klimoski, R. J. 1984. Leader emergence and self-monitoring behavior under conditions of high and low motivation. Paper pre-sented at the annual meeting of the Midwestern Psychological Association,Chicago.

Whyte, W. F. 1948. Human relations in the restaurant industry. New York:McGraw-Hill.

304 References

Williams, S., and Shiaw, W. T. 1999. Mood and organizational citizenship behav-ior: The effects of positive affect on employee organizational citizenship behav-ior intentions. Journal of Psychology, 133: 656–68.

Wrong, D. H. 1968. Some problems in defining social power. American Journalof Sociology, 73: 673–81.

Zaccaro, S. J., Foti, R. J., and Kenny, D. A. 1991. Self-monitoring and trait-basedvariance in leadership: An investigation of leader flexibility across multiplegroup situations. Journal of Applied Psychology, 76: 308–15.

Zellars, K. L., and Perrewe, P. L. 2001. Affective personality and the content ofemotional support: Coping in organizations. Journal of Applied Psychology,86: 459–67.

Zellars, K. L., Perrewe, P., Hochwarter, W., and Anderson, K. 2006. The inter-active effects of positive affect and conscientiousness on strain. Journal ofOccupational Health Psychology, 11: 281–9.

Zuckerman, E. W. 1999. The categorical imperative: Securities analysts and theillegitimacy discount. American Journal of Sociology, 104: 1398–1438.

Index

Ahn,W. K., 4Amanatullah, E. T., 5Ames, D. R., 5, 158, 174Arabie, P., 103, 182Argyle, M., 59, 163Auletta, K., 19, 27, 35, 57

Back, K. W., 29, 209Baker, F. B., 50, 118, 242balance schema, 6, 24, 26, 39–41,

55–6, 59, 61–5, 82, 193, 195,261, 263–4

Balkundi, P., 34, 176Barley, S. R., 265Barnard, C., 161, 248, 265Baron, J. N., 133, 142, 175Baron, R. A., 134–5Baron, R. M., 145Baron-Cohen, S., 164, 174Barrick, M. R., 158, 162, 173Barsade, S. G., 31, 157–8, 177basking-in-reflected-glory effect,

39–40, 49, 55–7Bass, B. M., 16, 19Batchelder, W. H., 61, 237, 248Baum, J., 35Beard, J. F., 135, 161Becker, G., 114Bedeian, A. G., 173, 176Bell, R. R., 213Benassi, M., 164Berge, C., 251

Berkman, L. F., 175Bernard, H. R., 90Berscheid, E., 138, 270Binkhorst, D., 3Blau, P. M., 4, 56, 131Bonacich, P., 104, 146Boorman, S., 103, 156, 182Borgatti, S. P., 15, 51, 104, 143, 167Bossard, J. H., 63–4Bougon, M. G., 3, 20–1, 41Brass, D. J., 1–2, 4–5, 15, 21, 27, 30,

33, 47, 80, 82, 84, 86–7, 89–90,96, 131, 133, 141–3, 145, 164,167, 250, 271

Breiger, R. L., 103, 132, 156, 182Brief, A. P., 159, 173Briggs, S. R., 172brokerage, 4–5, 19, 30, 33, 260Bryant, W. H., 159–60Burkhardt, M. E., 47, 80, 141, 143Burns, T., 15, 133, 264Burt, R. S., 1, 4, 13, 15–16, 19, 22–3,

28, 29, 31–3, 46, 82, 87, 103,114–15, 121, 131–3, 137, 141–3,145, 154–5, 175, 181, 184, 193,195, 259–60, 269

Caldwell, D. F., 123, 135–6, 138, 174Cameron, K. S., 219Campbell, D. T., 3, 48, 261Carley, K., 155, 237Cartwright, D., 56

305

306 Index

centralitybetweenness, 91, 132, 142–3, 146,

148–53, 268degree, 143, 167eigenvector, 105indegree, 44, 46–8, 50, 52, 54,

167–8Cheek, J. M., 172Chen, M.-J., 30Chesterton, G. K., 34Cialdini, R. B., 2, 40cliques, 30–1, 132, 156, 209, 221,

243, 249–50, 255, 257, 260, 268cognitive approach

to leadership, 23to organizational culture, 239to social networks, 3

cognitive dissonance theory, 241cognitive maps, 4, 25, 41–2, 44, 47–8,

50, 67, 69, 73, 91, 249, 251, 269,273

cognitive miser model, 60, 64–5, 76,78

cognitive network theory, 2, 7, 17, 23Cohen, S., 175Coleman, J. S., 15, 114Contractor, N. S., 5, 262Cook, K. S., 132Coser, L. A., 212Crockett, W. H., 59Cross, R., 6Crossland, C., 4, 24, 234Cunningham, M. R., 159Cyert, R. M., 29

D’Andrade, R., 60, 237, 239, 248–9Dalton, D. R., 193Dalton, M., 248Darley, J. M., 114Davis, J. A., 55, 79, 81Dawes, R., 63Day, D. V., 5, 173–4, 176, 268De Soto, C. B., 62, 79, 261–2Degenne, A., 206Dickson, W. J., 1, 141, 164DiMaggio, P., 3, 59, 82distinctiveness theory, 102, 110

Doreian, P., 81Dorogovtsev, S., 25, 208, 259, 270Duck, S. W., 122Durkheim, E., 236

Eckenrode, J., 129, 131E-I index, 214–17embeddedness, 1, 14, 19, 22, 23, 28,

29, 31, 32, 38, 236, 261, 266,268

Emerson, R. M., 85, 132Emirbayer, M., 131, 154, 266emotion helping network, 8, 156, 157,

159, 162, 163, 167, 172emotional tension model, 60, 62, 65,

76, 77Everett, M. G., 15

Faust, K., 15, 28, 59, 251Fayol, H., 161Feld, S. L., 69Fernandez, R. M., 22–3, 26, 260, 272Festinger, L., 29, 41, 60, 103, 114,

157, 209, 241, 250Fiedler, F. E., 16, 86Finkelstein, S., 163Fiske, A. P., 60Fiske, D. W., 48Fiske, S. T., 4, 63Flynn, F. J., 5, 158, 161, 173–4Fombrum, C., 14Forse, M., 206Freeman, J., 23, 36Freeman, L. C., 14–15, 24, 26, 47,

62–3, 79, 87, 90–2, 94, 141, 143,167, 261

Freeman, S. C., 26, 87French, J. R. P., 86Friedkin, N. E., 5Friedman, R. A., 176Frost, P., 34, 157–8, 162, 175Funder, D. C., 80

Gaertner, S., 181Galaskiewicz, J., 36Gangestad, S. W., 5, 123–5, 134, 136,

144–5, 152, 154–5, 166, 174

Index 307

Gargiulo, M., 20, 35, 37, 164Garland, J., 135, 161Geertz, C., 237, 248George, J. M., 159Gerhart, B., 123ghost ties, 267, 274Gibson, C., 31Giddens, A., 265Gioia, D. A., 49, 267Gnyawali, D. R., 15Goethals, G. R., 114Goodwin, J., 131, 154Gould, R. V., 26, 260, 272Gouldner, A. W., 81Graen, G. B., 16, 20Granovetter, M., 1, 9, 14, 29, 56, 105,

114, 121, 128, 132, 259Gray, B., 207Graziano, W. G., 138, 159–60Gregory, K. L., 237Griffeth, R. W., 181Gronn, P. C., 33, 135Gulati, R., 20, 35, 37

Hambrick, D. C., 29–30, 163Hanke, R., 3Hannan, M., 23, 36Hansen, C. H., 64Hansen, M. T., 155, 269Hansen, R. D., 64Harary, F., 56, 69Hargadon, A. B., 20Harrison, D. A., 176Hayward, M. L. A., 261Hedges, L. V., 71Heider, F., 4, 24, 39, 41, 55–6, 59–61,

64, 68, 81, 193, 261Higgins, M. C., 33, 114Hochschild, A. R., 177Holland, J. L., 123Holland, P. W., 68, 81, 141homophilous networks, 29homophily, 104–5, 108, 262Hooijberg, R., 16, 28Horn, P. W., 181House, R. J., 16, 86Hubbell, C. H., 84

Hubert, L. J., 50, 118, 126, 186, 242Hughes, E. C., 4Huy, Q. N., 161–2, 177

Ibarra, H., 3, 20, 47, 103–4, 113,132, 145, 162, 176

Ickes, W. J., 135, 160–1identity network, 105–7Insko, C. A., 60–1, 82interorganizational networks, 38, 267Isen, A. M., 159

Janicik, G. A., 4, 17, 19, 26, 64, 132,234, 260, 262–4

Kadushin, C., 260Kahn, R. L., 157Kahn, W. A., 175Kahneman, D., 27, 114Kanter, R. M., 85, 103Katz, D., 157Katz, E., 114Keesing, R. M., 237Kelly, G. A., 239–40Kenny, D. A., 16, 28, 81, 134, 145Kephart, W. M., 64Kilduff, M., 1–6, 14, 16–17, 21, 23–5,

30, 34, 36, 49, 56, 59, 66–7, 71,80, 83, 101, 121–2, 129, 134–5,141, 152, 155, 172, 174, 176,208, 234, 241, 249, 259–60, 262,267–8, 271, 273

Killworth, P., 90Klein, K. J., 5Knoke, D., 46, 114, 116Kogut, B., 15, 25, 272Kotter, J. B., 135Krackhardt, D., 1, 3–4, 6, 15, 24, 26,

28, 31, 41, 43–4, 47, 50–1, 53,57, 59, 66–8, 71, 80–1, 83, 86–7,90–1, 95–6, 104, 110, 114–15,118, 122, 141, 143, 168, 192,203, 234, 242, 246, 248–52,259–60, 268–9, 273

Kraimer, M. L., 17Kram, K., 33Kronenfeld, D., 90

308 Index

Kuethe, J. L., 63Kuhn, T. S., 266Kumbasar, E. A., 61, 80, 141, 249,

262, 270Kunda, G., 36

Labianca, G., 34Lakatos, I., 14, 266Larrick, R. P., 4, 17, 19, 26, 64, 132,

234, 260, 262–4Larson, A., 20, 35Laumann, E. O., 84law of family interaction, 63–4Lawrence, P. R., 209, 219leader effectiveness

social network effects on, 19, 28,31, 34

social network measures of, 33social network outcomes of, 20, 33

leadershipaccuracy of network perceptions

and, 21, 270cognitive revolution in research and,

23distributed, 33ego network density and, 28emergence of, 24, 27, 31, 134–5formal, 22implicit, 20, 28informal, 29, 31–2, 185in large networks, 25secondary networks and, 32self-monitoring and, 135

Leinhardt, S., 68, 81, 141Levine, J. M., 80, 160Lewin, K., 16, 195, 273Liden, R. C., 17, 23, 32Lincoln, J. R., 87, 145, 164, 249Lord, R. G., 21, 23, 28Lorrain, F. P., 114, 182–3Lorsch, J. W., 23, 209, 219

Madhavan, R., 15Maitlis, S., 34, 162, 176March, J. G., 29, 37, 263marginality in networks, 19, 101,

103–4, 110

Markus, H., 59, 64Martin, J., 161Mayhew, B., 57, 122, 154Mayo, M. C., 17, 33McDonald, M. L., 30, 260McPherson, M., 29, 63, 131, 156, 181Mehra, A., 1–2, 4, 6, 29–30, 33,

166–7, 169, 175Meindl, J. R., 17, 33Mendes, J., 25, 208, 259, 270Menzel, H., 114Merton, R. K., 234Meyer, A. D., 211Meyer, J., 23Milgram, S., 25, 272Milton, L. P., 30Mintzberg, H., 23, 97, 135, 162–4,

166Mische, A., 266Mizruchi, M. S., 85Mobley, W. H., 181, 193, 200Mollica, K. A., 207Monge, P. R., 5, 262Moody, J., 37, 264–5, 267–8Moreland, R. L., 80Mowday, R. T., 182, 184, 193, 195,

199–200, 205Murnighan, J. K., 250

Neisser, U., 59network cognition

and accuracy, 21, 26and leadership, 20

network research core ideas, 14, 16,21, 28, 38, 265

embeddedness, 268primacy of relations, 266social utility of network

connections, 270structural patterning of social life,

272Newcomb, T. M., 61, 114, 132, 194,

264

Oh, H., 30, 34, 36, 121, 262, 268Oliver, A., 146Oliver, C., 35

Index 309

Olkin, I., 71O’Reilly, C. A., 34, 49, 123, 133,

135–6, 138, 174Ozcelik, H., 34, 162, 176

Palmer, D., 29Pappi, F. U., 84Pastor, J. C., 17, 33Pattison, P., 269perceptions of social networks, 1–3,

6–7, 23–4, 27, 39, 44, 55, 59, 63,65, 67, 79–80, 97, 115, 187, 198,248, 260, 262

personal construct theory, 239personality and social networks

positive affectivity, 158, 162–3, 172self-monitoring, 1, 5, 7, 98, 123,

133, 154–5, 158, 162–3, 172Pfeffer, J., 84–6, 96–7, 128–9, 131,

206, 210, 245Podolny, J. M., 3–4, 19, 21, 26, 133,

142, 175Pollock, T. G., 261Popielarz, P. A., 29, 63Porter, L. T., 25, 87, 114, 182, 192,

199, 203Portes, A., 15Powell, W. W., 20, 35

Raven, B., 86Reagans, R. E., 5reciprocity

actual, 68, 81, 264perceived, 67, 69, 72, 76–7, 79, 262

Regan, D. T., 122regular equivalence, 183Reitz, K. P., 103, 183, 187reputation

and power, 96and self-monitoring, 124, 154, 161individual, 1, 7interorganizational, 21market for, 26, 40organizational, 125perceived, 3, 39, 83performance, 38, 41, 48, 53social construction of, 261, 271

Riley, D., 129, 131Rindova, V. P., 261Roberts, K. H., 34, 133Robins, G., 269Robinson, S., 157, 162Roethlisberger, F., 1, 141, 164Romney, A. K., 26, 61, 87, 94, 236,

237, 239, 248, 256Ronchi, D., 22, 142, 181, 193Rook, K. S., 129, 142Rowan, B., 23Roy, D., 1

Sailer, L., 90, 182–3, 186–7Salancik, G., 85, 96, 128, 206, 274Sampson, F. S., 3, 132Saxenian, A., 36Schachter, S., 29, 157, 163, 176, 209Schein, E. H., 212Schleicher, D. J., 174Seidel, M. D. L., 22Sherif, C., 212Sherif, M., 212Simmel, G., 30, 81, 113, 212, 248–9,

256Simmelian ties, 249–51, 253, 255–6Simon, H. A., 272small worlds, 25, 208, 259, 272–4Smith-Lovin, L., 29Snijders, T. A., 51, 146Snyder, M., 1, 5, 123–5, 134–8, 144,

145, 152, 154–5, 159–60, 166,174, 262, 268

social capital, 2, 15–16, 21, 28–9, 33,38–9, 154, 270

social comparison theory, 41, 103,114, 122, 130, 241

social distance, 62, 65, 69, 76, 78, 81Soda, G., 267Sparrowe, R. T., 17, 23, 32, 164Spence, A. M., 40–1, 261Stalker, G. M., 15, 133, 264Staw, B. M., 123, 159, 177, 193,

195Steers, R. M., 182, 192, 199Stern, R., 86–7, 95, 220Stevenson, W. B., 25

310 Index

structural equivalence, 7, 103,115–16, 120, 182

structural hole theory, 31, 259structural holes, 26, 28, 34, 82–3,

131–2, 141, 143, 150, 154–5,260, 262, 264, 274

structural patterning of social life,14–15, 132, 266

structuration theory, 265Sutton, R. I., 20Swidler, A., 59Syme, S. L., 175

Taylor, S. E., 4, 27, 63, 157Thomas, D. A., 34Thompson, J. D., 219Tichy, N. M., 14, 217transitivity

actual, 69, 81, 268perceived, 68, 72, 77, 79–81, 262

Trevino, L. K., 207Tsai, W., 1, 3–5, 14, 17, 23–4, 36, 49,

208, 234, 269Tsui, A. S., 49, 56Tushman, M. L., 14Tversky, A., 27, 114

underrepresented groups, 33, 101–3,108, 113

Usai, A., 267

Uzzi, B., 14, 17, 20, 23, 25, 28–9, 32,35–6, 132, 272

Van Maanen, J., 102Vermeulen, F., 31Vroom, V. H., 16, 125

Walker, G., 3, 15, 25, 80, 123, 246,272

Wallace, A. F. C., 237, 240Wasserman, S., 15, 28, 59, 251Watts, D. J., 25, 272Wayne, S. J., 17Weick, K. E., 3, 41, 82Wellman, B., 15Westphal, J. D., 30, 260Whetten, D. A., 219White, D. R., 103, 183, 187White, H. C., 40, 57, 114, 134, 156,

182–3Whyte, W. F., 1, 248Woolcock, J., 269workflow network, 90, 133, 141–3,

147, 149, 151, 155, 163–5, 167,172

Zaccaro, S. J., 16, 134, 161Zaheer, A., 267Zajonc, R. B., 59, 64Zuckerman, E. W., 3, 19, 21

Documents

Interpersonal Networks in Organizations: Cognition, Personality, Dynamics, and Culture (Structural Analysis in the Social Sciences)