asymmetric effects of fashions on the formation and

OrganizationScienceVol. 23, No. 4, July–August 2012, pp. 1114–1134ISSN 1047-7039 (print) � ISSN 1526-5455 (online) http://dx.doi.org/10.1287/orsc.1110.0683

© 2012 INFORMS

Asymmetric Effects of Fashions on the Formation andDissolution of Networks: Board Interlocks with

Internet Companies, 1996–2006

Lori Qingyuan YueMarshall School of Business, University of Southern California, Los Angeles, California 90089,

[email protected]

This paper extends the contextual perspective of network evolution to account for a more complete process of networkevolution by showing that the impacts of fads and fashions on the formation and dissolution of interorganizational

networks are asymmetric. Building on contact theory, this paper proposes that direct contact affords a flow of knowledgethat counters tendencies to social conformity. Network dissolution differs from network formation in that partners havealready obtained direct information. As a result, network dissolution is not as responsive to fads and fashions as networkformation, and network structures induced by fads and fashions often survive beyond the life cycle of a fashion. An analysisof the interlocking ties of S&P 1500 firms with Internet companies from 1996 to 2006 supports the view that fads andfashions have asymmetric effects on the evolution of networks and also shows that (1) fads and fashions have a strongimpact on the formation of networks but not on their dissolution, (2) the networking behaviors of organizations with directcontact are less induced by fads and fashions, and (3) the networks formed by organizations with direct contact during theheyday of a fashion survive longer.

Key words : network evolution; board interlocks; fashion; Internet firmsHistory : Published online in Articles in Advance August 10, 2011.

IntroductionResearchers who study the evolution of interorganiza-tional networks have started to pay more attention tosocial contexts as antecedents of the formation of ties.This shift is based on the increasingly prevalent recog-nition that although mechanisms like homophily, reci-procity, and transitivity can explain substantial variancein network formation, they are more successful in illus-trating the creation of local clusters than the genesis ofties that cut across sector lines (Podolny and 998,Baum et al. 2005). Sorenson and Stuart (2008) devel-oped a contextual perspective on network formationand proposed an additional mechanism—that fads andfashions1 encourage unconnected actors to overcomestructural barriers and meet in a common setting. A log-ical prediction along this line of reasoning is that if anorganization of a particular type suddenly experiences asurge in popularity, then it should be more attractive ineliciting network partners.

Fads and fashions have such power because they rep-resent instances of strong social conformity in deci-sion making. Often emerging around innovations, fadsand fashions are surrounded by a sense of newnessand uncertainties associated with this newness. Whendecision makers are unclear about the potential benefitsand costs of an innovation, they are likely to rely onsocial information and adopt a choice that is perceived

as desirable by others (Banerjee 1992, Abrahamson andRosenkopf 1993, Strang and Macy 2001, Rogers 2003).This is especially the case when social information con-veys the message that an innovation is successful, pro-gressive, and has enormous growth potential. As a result,social conformity induces adoption and generates herd-ing behaviors.

Because fads and fashions frequently arise from socialconformity rather than rational demonstration, they tendto be transitory and fragile. Moreover, the decline offashions can be accelerated by a counter-bandwagon thatis self-reinforcing in a manner similar to the bandwagoninvolved in the rise of a fashion, except in the oppo-site direction (Abrahamson and Rosenkopf 1993). If acounter-bandwagon is operating, decision makers wholook for social information to evaluate their adoptedpractices are likely to abandon these practices in thedownswing of the fashion. Thus, given the transience offads and fashions and the social influence they exert, onequestion is whether the interorganizational ties formedunder the impulse of fashions will last. This is an impor-tant question in the literature because it testifies to howsignificant fads and fashions are as a driving force in theevolution of networks. If the network structures formedunder the impulse of fashions dissolve as soon as enthu-siasm wanes, then we may be at risk of overemphasiz-ing the role of fads-and-fashions-induced networks in

1114

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.

Yue: Asymmetric Effects of Fashions on the Formation and Dissolution of NetworksOrganization Science 23(4), pp. 1114–1134, © 2012 INFORMS 1115

bridging fragmented clusters. Thus, it is important toinvestigate whether the dissolution of networks is as sen-sitive to the rise and fall of fashions as is their formation.In addition, what are the factors that may immunize net-work structures from the transience of fashions?

In this paper, I build on the intergroup contact the-ory (Allport 1954, Pettigrew 1998), which posits thatdirect contact with an out-group offers learning opportu-nities and thus reduces a focal actor’s tendency to referto stereotypes. Contact not only allows for the flow ofdirect knowledge that reduces a focal actor’s relianceon social wisdom, but it also supplements that knowl-edge with vivid self-experience that can correct inac-curate social evaluations. Extending the contact theory,I argue that, just as individuals can overcome stereo-types through their contact with an out-group, organi-zations should be less influenced by exaggerated socialinformation if they have direct connections to infor-mation sources. Fads and fashions create opportuni-ties for distant actors to meet and form relationships.Once formed, these relationships offer an opportunityto acquire direct knowledge and experience, which inturn reduces the reliance of network partners on socialinformation and promotes independent decision making.Direct knowledge, especially if it disconfirms exagger-ated or inaccurate social evaluations, should effectivelyimpede organizations from engaging in social compli-ance. Thus, the dissolution of networks should be lessresponsive to external social information, so that the net-work architecture formed during the heyday of a fad orfashion can live beyond the fashion’s life cycle.

An extended argument about direct knowledge is that,even in the context of network formation, the effectof fads and fashions should be attenuated among firmsthat have direct knowledge and experience. Organiza-tions are less likely to conform to social influence iftheir direct knowledge discounts the value presentedby social information. Organizations’ individual assess-ments should both reduce their conformity to fashionsin picking up network partners and improve the qualityof their partner choice decisions.

I tested these propositions by analyzing the formationand dissolution of board interlocks established betweenInternet companies and non-Internet firms (henceforth,“conventional firms”) from 1996 to 2006, which coversthe Internet bubble period. I found that the enthusiasmsurrounding Internet companies significantly increasedthe chance of tie formation between conventional firmsand the Internet sector. However, once these ties wereestablished, they were resilient to the external rise andfall of fashions. In addition, knowledge of Internet-basedbusiness models flowing from direct contact attenuatedconventional firms’ networking behaviors in two ways.One is that conventional firms with previous contactwere less likely to be induced by fads and fashions toform ties with the Internet sector. The other is that ties

that were formed by conventional firms with direct con-tact during the heyday of a fad or fashion lasted longerthan those formed by conventional firms without directcontact during the heyday of a fad or fashion.

This paper makes four major contributions. First,I extend prior work on mechanisms affecting networkevolution by showing how fads and fashions help explainnot just the formation of networks but also a more com-plete process of network evolution. In particular, I pro-pose a theory that combines both micro- and macroforcesby emphasizing the importance of both embedding orga-nizations’ networking actions into social contexts andtaking into account microstrategic reactions that gener-ate heterogeneity rather than uniformity. By extending thecontextual perspective, my study not only answers thecall to expand research on network evolution (Brass et al.2004, Zaheer and Soda 2009) but also joins a growingbody of literature exploring factors that drive organiza-tions to seek network partners beyond their local cliques(Beckman et al. 2004, Baum et al. 2005, Vissa 2012).Second, I show that fads and fashions have strong impactson network formation but not on dissolution. This asym-metric relationship suggests that transitory forces such asfads and fashions can generate long-lasting effects on net-work evolution. Third, this paper simultaneously investi-gates the formation and dissolution of ties and contributesto remedying the lack of investigation on network dissolu-tion. Moreover, by revealing that network formation anddissolution are asymmetric, this study argues that networkdissolution is a more complicated process and deservesmore research attention. Finally, whereas the prior inter-lock literature assumes that board interlocks tend to berelatively stable (Beckman and Haunschild 2002), thispaper suggests that significant changes in interlock net-works are possible under certain conditions and showsthat fads and fashions are another set of antecedents in theevolution of board interlocks.

TheoryThe contextual perspective has been developed toaddress the shortcoming in the network literature con-stituted by the lack of investigation of network evo-lution (Brass et al. 2004); especially lacking is theorythat can explain the evolution of interorganizational net-works that cut across sectional lines (Podolny and Page1998, Baum et al. 2005). The contextual perspectiveposits that the social contexts in which potential net-work partners are embedded affect their chance of beingconnected. The change in social contexts can providean impetus for organizations to move out of their localzones to seek network partners unconnected via conve-nient means such as homophily, reciprocity, transitivity,and repeated ties. The theoretical origin of the contextualperspective rests on the social foci proposition advancedby Feld (1981). Feld observed that personal relation-ships are often developed in a certain type of context

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.

Yue: Asymmetric Effects of Fashions on the Formation and Dissolution of Networks1116 Organization Science 23(4), pp. 1114–1134, © 2012 INFORMS

and argued that contexts (i.e., social foci) matter becausethey appeal to similar people and result in homophilousrelations.

Sorenson and Stuart (2008) extended the social focitheory from interpersonal networks to interorganiza-tional relations and built a contextual perspective ofnetwork evolution. In particular, they argued that thepopularity of a setting attracts dissimilar actors by pro-viding a point of first contact. Despite its novelty, thecontextual perspective, however, has fallen short of pro-viding a theoretical account for the evolution of net-works beyond first contact. We still know relatively littleabout how stable these context-induced relations wouldbe and whether organizations are equally susceptible tothe influence of contexts. These are important questionsbecause the promise of contextual variables as a driv-ing force of network evolution lies in change. Changingenvironments legitimize different types of organizationsas promising and validate different kinds of minglingopportunities as desirable. Consequently, the changedattitudes and beliefs shape the perceived value of addingorganizations of a certain type as network partners. How-ever, change, being either exogenous or endogenous,also sows the seeds of instability. If social contexts arechanging, then network structures induced by these con-textual variables are inevitably fragile. Extending thecontextual perspective beyond the initial point of contactdirectly indicates how significant contextual factors actas a driving force in the evolution of networks. There-fore, it is critical to simultaneously test the impacts ofcontextual factors on network formation and dissolution.

Network dissolution is yet another topic that hasreceived little research attention (Broschak 2004;Sullivan et al. 2007; Vissa 2011, 2012). One importantreason for the lack of investigation of network disso-lution is that it is often assumed to be the inverse offormation (Broschak 2004). If the factors that lead tonetwork formation automatically prevent network disso-lution, then the lack of investigation on network dissolu-tion will not be problematic. However, if the processes ofnetwork formation and dissolution are symmetric, thenthe network structure induced by a changing contextmay be only transitory. Thus, it is important to under-stand network dissolution and to investigate what makesit different from network formation.

Admittedly, organizational theorists have long recog-nized the importance of social contexts for interorga-nizational networks. Researchers, especially those whoinvestigate consequences of networks, have found that theefficacy of interorganizational diffusions depends onsocial contexts. For example, Davis and Greve (1997)found that interlocks were effective in diffusing poi-son pills but not golden parachutes, and they attributedthe divergent effects to the institutional environment thataffected the perceived legitimacy of different practices.Haunschild and Beckman (1998) investigated when board

interlocks affected corporate behaviors and found that theavailability of alternate information sources that eithersubstituted for or complemented interlocks affected theefficacy of interlocks. Mizruchi et al. (2006) studied thecontingency of the impact of board interlocks on firms’financial strategies and found that the influence of boardinterlocks declined over time with the changes in insti-tutional environments. In addition, researchers have alsofound that the benefits of advantageous network positionsare contingent on contextual factors such as national cul-ture (Xiao and Tsui 2007). Together, these studies havefocused on the consequences of networks and examinedthe extent to which these consequences vary across socialcontexts. Despite the contributions of these studies toestablishing the importance of contextual factors, theyhave not yet treated networks per se as the subject of inter-est nor examined the impact of contextual factors on theevolution of networks.

Fads and Fashions as Antecedents ofNetwork FormationFads and fashions are a powerful contextual force thataffects organizational decision making through a pro-cess of social conformity. Researchers in psychology,economics, and sociology have offered a range of expla-nations for their influence. Psychologists, focusing onthe motivations of individuals, have developed the con-cept of social validation: decision makers are more likelyto follow ideas that are favored by many other peoplebecause they perceive such ideas to be more correct andvalid. As Cialdini (1993, p. 131) pointed out, the reason-ing underlying social validation is that “if a lot of peopleare doing the same thing, they must know something wedon’t. Especially when we are uncertain, we are willingto place an enormous amount of trust in the collectiveknowledge of the crowd.” Extending individuals’ ratio-nales into a population-level phenomenon, the economicherding model suggests that rational decision makerswho infer from others’ actions their private evaluationsand assume the superiority of the choice preferred by themajority will engage in accelerating cascades of choiceconvergence (Banerjee 1992, Bikhchandani et al. 1992).Similarly, focusing on the behaviors of organizations,Strang and Macy (2001) proposed a theory of fashionsbuilt on the thesis of the “search for excellence,” wherethe selection pressure pushes organizations to imitateothers who have been successful, resulting in a conver-gence in organizational practices.

Recently, network scholars have proposed that fadsand fashions can explain the formation of nonlocal tiesbecause the pressure of social conformity drives sociallydistant actors to participate in a common, popular set-ting and form relationships. For example, Sorenson andStuart (2008) found that two distant venture capital firmsare more likely to form syndicate relationships with

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


each other when they are both attracted by a fashion-able investment setting. The fads and fashions that theseauthors point to are a third common “setting,” the popu-larity of which attracts actors of the same type but whoare socially distant from one another. An extension oftheir argument is that the popularity of certain partnersitself should afford more networking opportunities aswell. Thus, when the social information conveys a strongmessage that a certain type of organization is new andprogressive, and has enormous growth potential (regard-less of its technical merits), other organizations shouldbe more likely to establish networks with organizationsof this type.

Hypothesis 1 (H1). If a certain type of organiza-tion becomes fashionable, then conventional organiza-tions are more likely to form ties with organizations ofthis type.

Fads and Fashions as Antecedents ofNetwork Dissolution

Transience of Fads and Fashions. Although the mech-anisms through which fads and fashions influence orga-nizations’ behaviors are similar to those that result ininstitutionalization, the hallmark of fashions lies in theirconstant transience, which results in unstabilized prac-tices (Zucker 1988, Abrahamson and Fairchild 1999).Fads and fashions have short life cycles, and they aresustained by exaggerated zeal for a limited period fol-lowed by a decline. Fashion scholars depict the canoni-cal life cycle of a fashion as “a short-lived, bell-shaped,symmetric popularity curve” (Abrahamson and Fairchild1999, p. 711).

The transience of fashions can be attributed to at leastthree factors. First, systematic errors can be generatedwith the diffusion of social information (Gilovich 1987).Information is not only “sharpened” when moving fromone cascade to the next, but it also intensifies as moreand more people come to hold the same attitude whilethe actual value remains unchanged. Second, fashionscan surge in a way that is disproportionate or evenirrelevant to the underlying technical merits, and con-sequently, they are inevitably fragile (Abrahamson andRosenkopf 1993). Third, the process of adoption is oftenan emotional one in which organizations may overreactto fads and fashions (Abrahamson and Fairchild 1999).Disruptive innovations can easily provoke a sense of cri-sis and trigger anxieties leading to efforts to alleviate thesituation. Organizations desperately jump on bandwag-ons that are not necessarily in their best interests. Oncethe crisis seems to have been alleviated, the fashionablepractice is likely to be abandoned.

Because fashions are transitory and fragile, there areample reasons to suspect that network structures maydissolve as soon as enthusiasm wanes. Just as socialactors infer the value of adopting an innovation from

media reports and the actions of peers, their decisionto abandon an innovation is also likely to be affectedby these sources of social information. A lower rateof adoption or substantial shrinkage in media reportsis an indicator of discounted value. Abrahamson andRosenkopf (1993) developed the concept of a counter-bandwagon to describe a negative feedback loop in thedownswing of a fashion, where decreases in the num-ber of adopters cause a drop in the bandwagon pres-sure, which in turn triggers more abandonments andfurther decreases in the number of adopters. As thisnegative feedback loop repeats itself, more and morerejections occur. A counter-bandwagon is more likelyto form when uncertainty about the value of innova-tions persists even after adoption (Abrahamson 1991,Rao et al. 2001). In this circumstance, decision mak-ers use social information to evaluate the innovationsthey have adopted. For example, Greve (1995) reportedthat radio stations that faced uncertainties regarding thefuture performance of current and alternative strategiesexamined their peers’ actions to decide whether to dis-continue the practices of their organizations, causingcontagious abandonments.

Direct Contact and Knowledge. The fundamental rea-son why bandwagons or counter-bandwagons occur isthat actors rely on social information to resolve uncer-tainties in decision making, and this results in a con-vergence in behaviors. Knowledge flowing from directexperience, especially that which is inconsistent withexaggerated and inaccurate social information, shouldcurb an actor’s tendency to engage in social compliance.This hypothesis is best presented by the contact theory.Building on decades of research in the social psychol-ogy of intergroup relations, scholars have concluded thatcontact allows a flow of direct knowledge and experiencethat corrects stereotypes and diminishes intergroup prej-udice (Allport 1954; Pettigrew 1998, 2008). Althoughmost of these studies were conducted at the individ-ual level and examined how direct contact reduces per-sonal bias based on salient individual characteristics, theinsight that direct knowledge counters the effect of socialcategorization can improve our understanding of theimpact of fashions on the evolution of networks. Orga-nizational decision makers who have direct knowledgeshould rely less on social information when decidingwhether to form or dissolve a tie. Moreover, direct con-tact with firsthand information sources provides expo-sure to a set of situational factors that may underminethe promise of exaggerated social information. Organi-zations should be able to make wiser decisions in pick-ing up network partners because competing evaluationsbetween direct and social information open up possibil-ities for choice.

Organizational scholars have documented numerouscases in which divergent information improves the qual-ity of organizations’ decision making. For example, Rao

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


et al. (2001) found that stock analysts who follow socialclues in forecasting stock prices are able to quicklyabandon their imitation once they obtain disconfirminginformation from their own experience. Beckman andHaunschild (2002) found that organizations that receivediverse information about acquisitions and mergers areable to pay lower premiums for their acquisitions andmergers. Strang and Kim (2004) found that organizationsincorporate their own knowledge of effective actionswhen deciding whether to adopt fashionable manage-ment practices.

The distinct function of fashions in the networkingsetting is to create an opportunity for meeting. Whatmakes the dissolution of networks different from theirformation is that network partners have already met andobtained firsthand knowledge of each other. Uncertain-ties about network partners are greatly reduced after adirect tie has been established. Direct knowledge allowsorganizations to make independent judgments about thequality of their partners, rendering them less subject tothe influence of fashions. Organizations that concludethat their fashionable network partners are of low qual-ity may be more likely to abandon the ties even dur-ing the upswing of a fashion. Similarly, organizationsthat find their network partners to be of high qualitymay be more likely to retain the ties even during thedownswing of a fashion. Thus, organizations that relyon direct knowledge may be less likely to conform toa counter-bandwagon and to abandon networks simplybecause of a drop in the bandwagon pressure.

Moreover, direct contact is especially important in dis-counting the impact of fashions because of the scarcityof information. Because information about an innovationmay not be available from other sources and a set ofstandards may not have been established or widely dis-seminated, knowledge may be held only by those whoare directly involved in the innovation. Direct contact islikely to be a major source of knowledge transfer. Thus,direct contact may lessen the effect of social influenceand result in asymmetric relations between fashions andthe formation and dissolution of networks.

Hypothesis 2 (H2). The dissolution of networks isless sensitive than their formation to the fashions asso-ciated with network partners.

An individual organization’s direct knowledge shouldalso moderate its tendency to form networks with fash-ionable partners. Organizations that have direct con-tact with fashionable network partners should be betterinformed about this type of partner than others. Theirbehavior is likely to diverge from those without directcontact in two ways. First, they should be less respon-sive to external fashions in forming networks with fash-ionable partners if they receive a disconfirming signalthrough their direct contact. In contrast, organizationswithout direct knowledge should be more likely to be

induced by fashions to form networks with fashionablepartners.

Second, when selecting partners, organizations withdirect knowledge should make better decisions when itcomes to selecting fashionable network partners. Thedifference in decision quality should be especially largefor those ties that are formed during the heyday of afashion, when other organizations without direct knowl-edge are more eager to jump on the fashion bandwagon.Thus, because the network formation decisions made byorganizations with direct knowledge are less induced byfashions, their ties with fashionable partners should beformed on a sounder ground and have a lower risk offailure.

Hypothesis 3 (H3). A conventional organization thathas had prior contact with a fashionable type of networkpartner is less likely to form networks with such part-ners during the heyday of a fashion than conventionalorganizations without such prior contact.

Hypothesis 4 (H4). Networks that are formed duringthe heyday of a fashion by conventional organizationsthat have had prior contact with their fashionable type ofnetwork partner are less likely to fail than those formedin the heyday of the fashion by conventional organiza-tions without such prior contact.

Empirical Context: The Internet Bubble andDirectorate InterlocksInternet companies represent a new type of businessmodel that is based on the advancement and spread ofInternet technology. The rise of small Internet compa-nies triggered a boom that began in the mid-1990s. Mar-ket confidence in the new technology and related busi-ness models fueled fads and fashions around Internetcompanies as well as a stock bubble. The height of theboom was marked by events like the AOL/Time Warnermerger and the peak of the NASDAQ Composite Indexat 5,132.52 on March 10, 2000. The bubble burst in2001 when stock prices fell, and many Internet compa-nies burnt through their venture capital without turninga profit. The tightening of financial conditions generatedincreasing skepticism about Internet companies’ basicbusiness models. As the growth of many Internet compa-nies proved to be illusory, pessimistic views surged. Thedownswing of the Internet bubble was further compli-cated by legal scandals around some Internet companies’accounting frauds.

For a number of reasons, the Internet bubble pro-vides an excellent context for investigating the impactof fads and fashions on the formation and dissolutionof interorganizational networks. First, the Internet bub-ble presents a typical case in which fads and fashionsemerge around a technical innovation. Social evaluationof Internet-based business models changed a lot withthe rise and fall of the bubble. The boom period belief

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


that the Internet would, through “digital Darwinism”(Schwartz 1999) displace traditional business modelswas quickly replaced by the bust-period view that theInternet would not change business fundamentals andthat the previously perceived gap was superficial. Thischange in social evaluation provides an opportunity totest the impact of fads and fashions on organizations’networking behaviors.

Second, the Internet bubble emerged around a ques-tionable business model. The canonical Internet-basedbusiness model relied on harnessing network effects toobtain market share by operating even at a loss. Thebet on future rather than current profitability resultedin significant ambiguities in evaluating these compa-nies’ qualities. However, during the boom period, thenovelty of this business concept enabled some compa-nies with flawed business plans to acquire substantialresources (Perkins and Perkins 2001). The ambiguitiesand uncertainties surrounding the Internet-based busi-ness models present a good context for investigating therole of social conformity in determining organizations’networking behaviors.

Last, but not least, board interlocks with Internetcompanies are inexpensive, trustworthy, and credibleinformation sources for conventional firms to acquireknowledge about the Internet sector. Directorate inter-locks, which happen when one company’s directors orexecutives sit on the boards of other firms, transmitresources and information between organizations (for areview, see Mizruchi 1996). In particular, informationtransferred through board interlocks directly drives afirm’s competitive strategy. Useem (1984) was amongthe first to point out that interlocks enabled managersto achieve an optical business scan of the latest com-petitive practices and to get a sense of the direc-tion of future business development. Similarly, throughinterviewing corporate directors and executives, Lorschand MacIver (1989) concluded that directors playedan important role in firms’ strategy formulations asadvisors and evaluators. These conclusions have alsobeen supported by numerous quantitative studies thatshow that interlocks affect firms’ acquisition and mergerstrategies (Haunschild 1994, Haunschild and Beckman1998, Beckman and Haunschild 2002, Beckman et al.2004), financial strategies (Mizruchi and Stearns 1994,Mizruchi et al. 2006), and organizational strategies(Palmer et al. 1993).

In the current research context, interlocks are an espe-cially appropriate theme for investigation because mostInternet companies are still in their early stage of devel-opment and other channels of connection may not havebeen developed. The motivation of conventional firmswas more to obtain information because they have not yetfigured out how the Internet might affect them or howthey can incorporate the Internet into their own opera-tions. Amid high levels of uncertainty, board interlocks

are a low-cost and convenient way to obtain informa-tion; as Haunschild and Beckman (1998, p. 817) pointedout, “[D]irectors are required for all public firms, and theinformation that comes from a director is thus an inex-pensive by-product of such mandated relationships.” Inaddition, there is anecdote evidence suggesting that, atthe height of the Internet boom, conventional businessesactively sought Internet interlock partners. Business Weekreported that at the time there were a lot of boards thatsought talent that could help them compete against thenewbies on the block (Byrnes and Judge 1999).

MethodDataThe sample for this study consisted of S&P 1500 firmsfrom 1996 to 2006, covering the boom and bust peri-ods of the Internet bubble. S&P 1500 firms are com-posed of approximately 500 large firms, 600 midsizedfirms, and 400 small firms. The list of S&P 1500 firmschanges slightly from year to year, and the annualaverage addition and drop-off rates during the periodinvolved are each about 5%. For the network formationanalysis, because I studied conventional firms’ construc-tion of new ties (compared with the previous year), thefinal sample excluded observations pertaining to Inter-net companies up to the year 1996 and to organiza-tions that appeared for the first time in other years. Thefinal sample consisted of an unbalanced panel includ-ing 12,774 company-year observations for 2,314 com-panies over a period of 10 years. I compiled data onboard interlocks using a list of directors of S&P 1500firms from the Investor Responsibility Research Center(IRRC) database and a list of top executives from theCompustat ExecuComp database. For each firm, I identi-fied a list of unduplicated names among the directors andtop executives.2 I then defined interlocked companies asthose sharing at least one common name on their listsof directors and executives.3 The data on firm size andprofitability were collected from the Compustat Indus-trial Annual database. Data on the number of articlesabout the Internet sector were collected from the Factivadatabase. Data on the input and output trade relation-ships between industries were collected from the Bureauof Economic Analysis (BEA) of the U.S. Department ofCommerce.

I used the 2002 North American Industry Classifica-tion System (NAICS 2002) to identify companies withinthe Internet sector. According to the revision report pub-lished by the Office of Management and Budget (2000),NAICS 2002 was the first industry classification sys-tem that distinguished businesses that operate mainlyvia the Internet from those operating in conventionalways. For observations before 2002, I used a correspon-dence table comparing NAICS 1997 and NAICS 2002 tomatch data with the NAICS 2002 classifications. In this

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


paper, I identify the Internet sector as consisting of firmsoperating via the Internet (NAICS classifications 454111and 425110), publishing and broadcasting content on theInternet (516), providing Internet access services (5181and 5182) or Web design (5415), or providing Internet-related training and services (611420). I list all theindustries classified as belonging to the Internet sector inOnline Appendix 1 (http://orgsci.journal.informs.org/).

Dependent Variables and EstimationThe first dependent variable is the formation of Internetinterlocks, which is specified as the number of a conven-tional firm’s interlocking companies, newly added withina year, that lie within the Internet sector. The unit ofanalysis is a conventional firm in a year. This count mea-sure can better capture the influence of fashions than theproportion of interlock partners because the proportionof interlocks that are Internet companies measures notonly new additions but also historical stock. Moreover,this count measure also has the advantage over othermeasures, such as the proportion of a firm’s interlockpartners initiated in each year that were Internet compa-nies, in that it avoids the problem of having to drop allfirms that did not add one interlock in a year (55% ofthe sample). However, using these proportion measuresproduces similar patterns of results as reported here.

I restricted my analysis to nonredundant ties—a tieto an Internet company was counted as a new initiationonly if one focal firm was not connected to that Internetcompany during the previous year. This definition of aninterlock tie helped to rule out potential bias caused bypersonnel replacement, because firms may occasionallyadd or drop ties when replacing a previous director witha new one. In addition, I manually checked the proxystatements of all the S&P 1500 firms excluded from thesample for two consecutive years to ensure that I hadnot mistakenly classified an unobserved tie in year t asa newly added tie in year t + 1. From 1996 to 2006,I identified 781 unique interlock ties between conven-tional firms and Internet companies, 601 of which wereinitiated during this period, and 507 of which were iden-tified as newly founded ties.4

Because the tie formation data constitute an unbal-anced panel, I used generalized estimating equations(GEEs) (Liang and Zeger 1986), which estimate quasi-likelihood models and accommodate nonindependentobservations.5 Because the dependent variable consistsof nonnegative integers with overdispersion, I estimatedGEE negative binomial models with a log link func-tion. I specified an unstructured correlation matrix,which allows any form of correlations between observa-tions, a specification that is also favored by the quasi-likelihood information criterion test (Pan 2001, Cui2007). I also report robust variance estimators.

One methodological concern arose in applying themodel: attrition in the unbalanced panel. Attrition is

problematic when the entry or drop-off patterns ofobservations are not random (for a review, see Baltagiand Song 2006). Verbeek and Nijman (1996) distin-guished between “ignorable” and “nonignorable” pat-terns of missing values in an unbalanced panel. If sampleattrition is ignorable for the parameters of interest, thestandard panel data method can be applied to generateconsistent estimation (Baltagi and Song 2006, p. 510).To test whether the attrition was ignorable, I used thevariable addition test proposed by Verbeek and Nijman(1992, p. 688). The intuition is that if the missingobservations are random, indicators of a firm’s exis-tence pattern in the sample should not be associated withthe outcome of interests after controlling for observedcovariates. Following Contoyannis et al. (2004), the testvariables included (1) an indicator of whether a firmwas observed in the subsequent year, (2) an indicatorof whether a firm was observed in all the years, and(3) a count of the number of years observed for eachcompany. The results show that none of these variablesis related to the dependent variable at the 0.05 signifi-cance level, and thus the attrition issue in the sample isignorable.

The second dependent variable is the dissolution ofInternet interlocks, for which the unit of analysis is aninterlock tie between a conventional firm and an Internetcompany. An interlock tie was coded as dissolved in ayear if two firms shared at least one common directoror executive in the previous year but not in the year inquestion. Ties that survived beyond 2006 were coded asright censored. For ties that disappeared from the samplebecause either the conventional partner or the Internetpartner (or both) dropped out of the sample, I manu-ally checked the proxy statements of the two companies.If an interlock tie failed in a year when either or bothpartners exited from the sample, the tie was coded as dis-solved. Otherwise, the tie was coded as right censored.Moreover, I also considered the chance of tie reconstitu-tion and restricted the definition of dissolution to thoseties that failed in one year and were not reconstitutedwithin the next three years. I treated the 12 ties thatwere reconstituted within the next three years as unbro-ken ties. From 1996 to 2006, 503 tie dissolutions wereobserved.

I modeled tie dissolution using r4t5, the instantaneousrisk of failing. This hazard rate of failure is defined asthe limiting probability of a failure between t and t+ãt1given that an interlock tie existed at t, calculated over

ãt2 r4t5= limãt→o

Pr4fail1 t +ãt � exist_t5

ãt0

Parametric estimates of the hazard rate require assump-tions about the effect of time on failure. I used the piece-wise exponential model, which allows the rate of failureto vary in an unconstrained way over preselected ageranges. Constants (baseline failure rates) were estimated

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


for each age period (0–1, 2–3, 4–6, and 6+).6 I esti-mated a piecewise exponential model of the form r4t5=

e�Xe�l , if t ∈ ll, where X is the vector of covariates,� is the associated vector of coefficients, and �l is aconstant coefficient associated with the lth age period.The life history of each tie was broken into one-yearspells to incorporate time-varying covariates, yielding2,259 spells; only 2,120 spells were actually used in theestimation because of missing values.

One issue in modeling the failure hazard of interlockties is the problem of unobserved heterogeneity, whichhappens when there are unobserved factors that makesome ties more susceptible to failure than others. Forexample, firms that had adopted Internet ties in ear-lier years might have (unobserved) structural reasons tomaintain these ties longer. To control for the unobservedheterogeneities, I added a conventional firm-specificfrailty variable into the piecewise exponential hazardestimation. The frailty-augmented estimation multipliesthe standard hazard function with a firm-specific �i,which accounts for the effect of one or more omittedvariables (Gutierrez 2002).7 Ties with �i larger than itsmean have a higher hazard of failure, whereas those with�i smaller than its mean have a lower hazard of failure.In addition, a very nice feature of the frailty model isthat it estimates the frailty variance �, an insignificantestimation of which indicates little unobserved hetero-geneity. More about the unobserved heterogeneity willbe discussed in the Alternative Explanations section.

Independent VariablesThe two most common variables for measuring fadsand fashions are media discourse and peer action(Abrahamson and Fairchild 1999, Strang and Macy2001). It is worth emphasizing that the focus of thispaper is on examining the attenuation effect of pri-vate information on contextual influences, and thusI do not attempt to distinguish whether media reportsand peer actions are indicators of a fashion or inter-mediaries through which a fashion affects organiza-tions. Instead, I treat media reports and peer actionsas contextual forces that exert conforming pressureon an organization’s networking decision, and on thispoint both the proxy and the intermediary accountsagree. Media discourse is an important source of socialinformation (Ruef 2000, Fiss and Hirsch 2005, Sineet al. 2005, Greve et al. 2006). A high volume ofmedia discourse about organizations of a certain typeincreases the public’s exposure to these organizations,and valenced discourse about their merits increases theirlegitimacy. Media reports are likely to serve as a sourceof social proof because they contain information gen-erated by opinion leaders such as journalists, profes-sionals, and experts (Pollock and Rindova 2003). Theactions of peers constitute another important source ofsocial influence. If a large number of organizations have

adopted ties with organizations of a certain type, thisenhances the cognitive legitimacy of these organizations.Moreover, firms within the same industry have simi-lar resources and face similar constraints, and there-fore their actions are particularly suitable to serve asa reference for social proof. In addition, the adoptionbehaviors of geographically proximate organizations areespecially contagious because geography represents acrucial parameter of information flow between organi-zations (Kono et al. 1998, Almeida and Kogut 1999,Sorenson and Audia 2000).

To create the measure of valenced media reports,I first collected the number of articles from the Factivadatabase appearing in any one year from 1997 to 2006that contained the keywords “Internet company,” “dot-com,” “e-business,” “e-commerce,” or “new economy.”Figure 1 shows the number of articles containing theseInternet-related keywords over time and confirms thatthe number of media reports rose and fell along withthe boom and bust, respectively, of the Internet bub-ble. Moreover, to further evaluate the valence of mediadiscourse, I randomly selected 1,000 articles from fourleading business periodicals, the Wall Street Journal,Financial Times, Business Week, and Forbes, during theperiod from 1997 to 2006 (i.e., 100 articles per year).Two trained research assistants each coded 600 articlesto determine whether an article manifested a positivetone toward Internet companies. In addition, the tworesearch assistants both evaluated 20 randomly selectedarticles for each year. From these 200 overlapped evalu-ations, the Cohen’s kappa of 0.869 indicated high inter-rater agreement.

The measure of valenced media attention is the prod-uct of the percentage of articles with positive tones and

Figure 1 Media Reports on Internet-Based Business Models

0

50,000

100,000

150,000

200,000

250,000

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

No.

of a

rtic

les

Internet companyDot-comE-businessE-commerceNew economy

Source. Factiva Database.

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


the total of number of articles in a year. This variablemeasures the total number of articles that manifest posi-tive attitudes toward Internet companies in a year. Thereare at least three advantages in adopting this measure.First, leading business periodicals set the tone of mediadiscourse, and thus coding their articles can accuratelycapture the prevailing media attitude toward Internetcompanies. Second, using the randomly selected arti-cles reduces the heavy burden of coding the more than1.2 million articles into a manageable task. Third, usingthe product term avoids the collinearity problem whenentering the volume and the valence of media reportsinto one regression simultaneously. In addition, I alsotried other measures of media attention by using the sim-ple amount of articles or by using the product of thepercentage of articles with positive tones and the totalnumber of articles in the four leading periodicals in ayear. These measures generate results similar to thosereported in this paper. Hypothesis 1 predicts that mediaattention will be positively related to the formation ofinterlock ties with the Internet sector. Hypothesis 2 pre-dicts that media attention will not be significantly relatedto the hazard of dissolution of an interlock tie with theInternet sector.

I measured the percentage of adopting firms withinthe same industry by using the percentage of otherfirms with the same three-digit NAICS code that inany one year were interlocked with Internet companiesfrom among the S&P 1500 firms. I also tried usingthe two-digit industry code and found the same patternof results. I measured the percentage of adopting firmswithin the same geographical region by using the per-centage of other conventional firms headquartered withinthe same metropolitan statistical area (MSA) that in anyone year were interlocked with Internet companies fromamong the S&P 1500 firms. MSA is a relatively smallgeographical unit that scholars have used to define localbusiness regions (e.g., Stuart and Sorenson 2003). Thelist of MSAs and their defined scope are collected fromthe U.S. Census Bureau. Hypothesis 1 predicts that thepercentage of adoptions within an industry and within anMSA will be positively related to the formation of inter-lock ties with the Internet sector. Hypothesis 2 predictsthat adoptions within an industry and within an MSAwill not be significantly related to the dissolution of aninterlock tie with the Internet sector.

A conventional firm’s contact with the Internet sectorin the previous year is a dummy variable that is codedas 1 if the conventional firm had Internet interlocks inthe previous year. To test the attenuation effect of pre-vious contact on social influence, I created interactionsbetween this variable and the three independent variablesthat measure fashions: media attention, adoption withinan industry, and adoption within an MSA. Hypothesis 3predicts that these three interaction effects will be neg-atively related to the formation of interlocks. Similarly,

when analyzing the dissolution of interlocks, I createdinteractions between a conventional firm’s previous con-tacts in the founding year of an interlock tie and the threevariables that measure fashions in the founding year ofan interlock tie. Hypothesis 4 predicts that interactionsbetween previous contact and media attention, adoptionwithin an industry, and adoption within an MSA in theyear of tie formation should all have negative effects onthe hazard of tie dissolution.

Control VariablesA few common control variables were used for boththe formation and dissolution analyses of interlocks. Thefirst one is a conventional firm’s size, measured as itstotal assets in the previous year. The second is a conven-tional firm’s profitability, measured as a firm’s ROA (itsnet income divided by its assets) in the previous year.The third is the trade relationship between the industriesto which the conventional firms belong and the Internetsector in the previous year. Resource dependence theorypredicts that board interlocks are an important mecha-nism for firms to manage their external resource depen-dence on suppliers and customers (Pfeffer and Salancik1978, Burt 1983). I use the sum of input dependence (thepercentage of inputs provided by Internet-related indus-tries out of the total inputs that are used by each industryto produce its output) and output dependence (the per-centage of commodities produced by each industry thatare used by Internet-related industries) to measure traderelationships. The data were collected from the bench-mark industry trade data reported by the BEA in 1997and 2002. The benchmark data are the most detailedindustry-level trade data available, reporting trade rela-tionships among more than 483 industries at the six-digitNAICS code level. For the years in which benchmarkdata are not available, linear interpolation values havebeen inserted.

For the formation analysis of interlocks, I controlledfor four other variables that may be related to the con-struction of networks. First, I controlled for the annualpercentage of firms that are Internet companies withinthe sample to rule out the possibility that the prevalenceof interlock ties with Internet companies was a spuriousresult from the more available Internet companies withinthe sample during the Internet boom period. Second, Icontrolled for the number of interlocking partners that afirm had in the previous year to control for the endoge-nous formation of networks (Gulati and Gargiulo 1999).Third, I controlled for the spatial availability of Internetcompanies by measuring the number of Internet compa-nies among the S&P 1500 firms that were located withinthe MSA where the focal firm was headquartered in ayear. Kono et al. (1998) found that corporate interlocksare a geographical phenomenon and are constrained byspatial boundaries. Finally, I controlled for the calen-dar year of observation. In additional analyses, I also

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


included the square term of the year to control for theboom and bust effects of Internet companies around themidpoint of the data; however, year2 is highly corre-lated with the year and does not affect the hypothesizedeffects, and it has thus been omitted.

For the dissolution analysis of interlocks, I controlledfor four other variables that may be related to the fail-ure of networks. First of all, I controlled for the size ofan Internet company, using its total assets in the previ-ous year. Second, I controlled for the profitability of anInternet company using its ROA from the previous year.8

Third, I controlled for the direction of interlock ties. Adirectional tie (i.e., a sent or received tie) is connectedby a director or an executive whose primary affiliationis one network partner. In contrast, a neutral tie is con-nected by a director whose primary affiliation is neitherof the two network partners. Network partners shouldhave more control over the dissolution of directional tiesthan over neutral ties. Fourth, I controlled for the leftcensoring of interlock ties. If the year when both part-ners first appeared in the S&P 1500 database is later thanthe founding year of their interlock tie, the left censor-ing dummy is coded as 1. Left censoring is not a severeproblem in the current research context because Inter-net companies are a new organizational population, andmost interlock ties were formed after 1996. Neverthe-less, I further checked the robustness of the results withregard to the left censoring issue by restricting analysesto those ties founded after 1996, and I found a simi-lar pattern of results. Finally, in additional analyses, Ialso controlled for the continuous age of an interlock tie,defined as the number of years between a tie’s foundingyear and the current year. I omitted this variable fromthe reported results, however, because it is highly corre-lated with tie age dummies, is not significant by itself,and does not affect hypothesized effects.

ResultsTable 1 reports the descriptive statistics and correla-tions of all variables used in the interlock formation

Table 1 Descriptive Statistics for Tie Formation Estimation (N = 121774)

Variable Mean S.D. 1 2 3 4 5 6 7 8 9 10 11

1 Network formation 0003 0.202 Internet company (%) 0003 0.00 00003 Internet company in MSA 1030 1.55 0005 00094 Interlock number (t − 1) 6010 6.28 0012 −0007 00085 Trade relationship 0007 0.36 0001 0003 0005 −00096 Year 21001044 2.82 −0001 0065 0001 −0012 00027 Firm asset (in billions) 0001 0.06 0005 0003 0011 0029 −0003 00068 Firm profitability 0004 0.14 −0001 −0003 −0001 0004 −0002 0000 −00029 Media report (in 100,000s) 0082 0.94 0006 0029 0009 0004 0000 −0024 −0002 0000

10 Adoption in industry 0012 0.08 0012 0004 0008 0010 0014 0002 0004 −0001 0.1511 Adoption in MSA 0013 0.11 0006 0004 0041 0008 0005 0002 0005 −0001 0.06 0.0912 Internet tie (t − 1) 0011 0.32 0010 0002 0011 0033 0004 0001 0015 0001 0.01 0.19 0.10

Note. The unit of analysis is a conventional firm in a year.

analysis. Table 2 reports the GEE negative binomialanalysis results on tie formation. Model 1 tests themain effects of the three variables that measure socialinfluence. The results show that a conventional firm ismore likely to form interlocks with an Internet com-pany if there is a high volume of positive media dis-course (b = 00127, p < 0005), a high percentage of otherfirms within the industry that have adopted Internet inter-locks (b = 50055, p < 0001), or a high percentage ofother conventional firms within the same MSA that haveadopted Internet interlocks (b = 10512, p < 0001) in ayear. I report the three main effects together in Model 1to save space, but they remain robust if tested separately.These findings not only confirm Sorenson and Stuart’s(2008) conclusion that fads and fashions are drivingforces in the formation of distant ties, but also extendthe literature from an examination of the popularity ofa common setting to consideration of the popularity ofpartners themselves. Fads and fashions strongly moti-vate organizations to overcome structural barriers and toform networks with fashionable partners. Hypotheses 1is supported.

Model 2 includes the main effect of contact with Inter-net companies in the previous year. The coefficient ispositive but not statistically significant (b = 00121, n.s.).Models 3–6 test the interaction effects between previouscontact and the three social influence variables. In thefull model (Model 6), the attenuation effects of previouscontact achieve statistical significance for all three vari-ables: media attention (b = −00359, p < 0005), adoptionwithin the industry (b = −40484, p < 0001), and adoptionwithin the MSA (b = −20986, p < 0005). The goodnessof fit for the full model is also significantly improved overthe previous nested models. Combined with the maineffects of social influence variables, Model 6 shows thatalthough conventional firms without direct contact weremore likely to form interlocks with Internet companieswhen media attention was high, those with direct contactdid not increase, or even slightly decreased, their ten-dency to form Internet interlocks when media attention

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


Table 2 GEE Negative Binomial Estimation of Tie Formation

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

Internet company (%) −270772 −270197 −290078 −280970 −270040 −32016642400105 42309975 42400685 42309455 42309985 42400235

Internet company in MSA 00075∗∗ 00073∗∗ 00075∗∗ 00075∗∗ 00075∗∗ 00079∗∗

4000335 4000335 4000325 4000335 4000335 4000335Interlock number (t − 1) 00067∗∗∗ 00065∗∗∗ 00066∗∗∗ 00063∗∗∗ 00065∗∗∗ 00064∗∗∗

4000065 4000075 4000075 4000075 4000075 4000075Trade relationship 00183 00181 00177 00159 00184 00153

4001405 4001395 4001385 4001395 4001405 4001415Year 00026 00025 00027 00029 00023 00030

4000295 4000295 4000295 4000295 4000305 4000295Firm asset −00245 −00267 −00277 −00326 −00203 −00277

4006975 4007045 4007085 4006475 4007335 4007025Firm profitability −00384∗∗ −00388∗∗ −00381∗∗ −00403∗∗ −00384∗∗ −00379∗∗

4001825 4001815 4001815 4001755 4001815 4001775Media report 00127∗∗ 00122∗∗ 00188∗∗∗ 00115∗∗ 00124∗∗ 00232∗∗∗

4000685 4000695 4000645 4000695 4000685 4000615Adoption in industry 50055∗∗∗ 40981∗∗∗ 40977∗∗∗ 60298∗∗∗ 50015∗∗∗ 60328∗∗∗

4003905 4003995 4003985 4004425 4003925 4004425Adoption in MSA 10512∗∗∗ 10517∗∗∗ 10474∗∗∗ 10511∗∗∗ 20004∗∗∗ 10966∗∗∗

4004295 4004265 4004335 4004025 4003805 4003795Internet tie (t − 1) 00121 00266 00977∗∗∗ 00568∗∗ 10640∗∗∗

4001665 4001915 4002375 4002585 4003045Internet tie 4t − 15 −00213 −00359∗∗

∗ Media report 4001835 4002065Internet tie 4t − 15 −40636∗∗∗ −40484∗∗∗

∗ Adoption in industry 4100805 4009885Internet tie 4t − 15 −20870∗∗ −20986∗∗

∗ Adoption in MSA 4104255 4105225Constant −560879 −540568 −580720 −620287 −500597 −640866

45805165 45804225 45804735 45803205 45805955 45803825

N 12,774 12,774 12,774 12,774 12,774 12,774�2 4060273 4120278 4300708 4660776 4620276 4700341

Notes. The unit of analysis is a conventional firm in a year. Standard errors are in parentheses.∗∗p < 0005; ∗∗∗p < 0001 (one-sided test for hypothesized variables and two-sided test for other variables).

was high. Similarly, although conventional firms tendedto increase their tendency to build Internet interlockswhen a high percentage of peers had done so, the magni-tude of these increases for those with direct contact wassubstantially smaller.

An alternative account related to direct contact is thatthe attenuation effect of direct contact comes not fromthe flow of knowledge but from a substitution effect. Inother words, organizations that had had direct connec-tions to the Internet sector in the previous year did nothave as strong an incentive to establish new ties as thosewithout direct connections, because their need for infor-mation had already been satisfied by their previous con-nections. If this account is true, we should expect a neg-ative main effect of prior contact by a conventional firmon its tendency to build additional ties. However, a care-ful examination of Table 2 not only shows that the vari-able of previous contact has an insignificant main effectwhen tested alone but also shows significantly posi-tive coefficients when controlling for interaction effects.

Without strong conforming pressure (i.e., three fashionvariables were set at their means in Model 6), con-ventional firms with prior contacts built 66.3% morenew ties with the Internet sector than those withoutprior contacts. Instead, these results lend some sup-port to the endogenous account of network formation,which suggests that existing network structures facil-itate the construction of future networks (Gulati andGargiulo 1999, Rosenkopf and Padula 2008, Zaheer andSoda 2009). Moreover, the substitution account alonehas difficulty explaining why the attenuation effect isstronger when fads and fashions are high. Thus, infor-mation substitution is unlikely to have driven the attenu-ation effect of direct contact on tie formation. Together,these results suggest that a conventional firm that hashad prior contact with the Internet sector is significantlyless responsive to social influence when forming newinterlocks with Internet companies, lending support toHypothesis 3.

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


Table 3 Descriptive Statistics for Tie Dissolution (N = 21120)

Variable Mean S.D. 1 2 3 4 5 6 7 8 9

1 Directional tie 0.57 0.652 Trade relationship 1.76 0.48 −00143 Left censored 0.38 0.49 −0002 00034 Internet company asset (in billions) 0.01 0.02 −0005 0010 −00045 Internet company profitability 0.03 0.15 −0008 0004 0000 00056 Conventional firm asset (in billions) 0.04 0.13 −0005 0007 −0007 0013 −00037 Conventional firm profitability 0.04 0.14 0006 −0002 0001 0009 0002 −00048 Media report (in 100,000s) 0.84 0.95 −0003 0001 0007 −0002 0005 −0002 −00029 Adoption in industry 0.17 0.08 −0004 −0002 0003 0005 −0001 −0001 −0001 0.06

10 Adoption in MSA 0.17 0.11 −0002 −0001 0002 0005 −0003 0011 0002 0.02 0.04

Note. The unit of analysis is an existing interlock tie between a conventional firm and an Internet company in a year.

Table 3 reports the descriptive statistics and correla-tions of variables used in the interlock dissolution anal-ysis. Table 4 reports the piecewise exponential frailtyhazard model of tie dissolution. Model 7 reports a pri-mary model with all control variables. Models 8–11 testthe main effects of social influence on the dissolutionof networks. None of the three variables has a robustimpact on the hazard of tie dissolution. Moreover, noneof these models shows significant improvement in thegoodness of fit over the primary model, confirming theinsignificant effect of fashions on the dissolution of ties.Together, these results suggest that the dissolution of

Table 4 Piecewise Exponential Frailty Hazard Model on Tie Dissolution

Model 7 Model 8 Model 9 Model 10 Model 11

Directional tie 00195∗∗∗ 00195∗∗∗ 00199∗∗∗ 00194∗∗∗ 00198∗∗∗

4000745 4000745 4000745 4000745 4000745Trade relationship −00150 −00151 −00153 −00156 −00159

4000875 4000875 4000875 4000875 4000875Left censored 00167 00169 00176 00171 00179

4001185 4001185 4001195 4001185 4001195Internet company asset −10758 −10758 −10624 −10622 −10512

4201525 4201525 4201625 4201525 4201625Internet company profitability −00248 −00247 −00252 −00240 −00243

4002875 4002875 4002875 4002855 4002855Conventional firm asset 00325 00324 00325 00395 00392

4003275 4003275 4003265 4003305 4003295Conventional firm profitability −00413 −00414 −00421 −00406 −00414

4002715 4002715 4002715 4002735 4002735Media report −00007 −00005

4000495 4000505Adoption in industry −00425 −00372

4005345 4005375Adoption in MSA −00706 −00681

4005135 4005145

N 21120 21120 21120 21120 21120Log likelihood 214140277 214140287 214140595 214150223 214150477�2 110530750 110530739 110530458 110520878 110520643

Notes. The unit of analysis is an existing interlock tie between a conventional firm and an Internetcompany in a year. Standard errors are in parentheses. Dummies of tie age were included in theestimation but omitted from reporting.

∗∗p < 0005; ∗∗∗p < 0001 (one-sided test for hypothesized variables and two-sided test for othervariables).

Internet interlocks is not sensitive to the rise and fallof fashions in the Internet sector. These results confirmmy expectation that once a conventional firm has directcontact with the Internet sector, it is capable of makingindependent decisions and relies less on social informa-tion. Hypothesis 2 is supported.

Table 5 reports the descriptive statistics and correla-tions of variables used in the moderation analysis ofinterlock dissolution. Because the IRRC database doesnot provide data on directors prior to 1996, I could onlyobserve the percentages of Internet interlock adoptionwithin an industry and within an MSA beginning in

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


Table 5 Descriptive Statistics for the Dissolution of Ties Founded After 1996 (N = 11055)

Variable Mean S.D. 1 2 3 4 5 6 7 8 9 10 11 12 13

1 Directional tie 0.60 0.642 Trade relationship 1.77 0.46 −00133 Left censored 0.14 0.35 0006 00054 Internet company 0.01 0.02 −0016 0010 0000

asset (in billions)5 Internet company profitability 0.02 0.15 −0007 −0005 −0001 00076 Conventional firm 0.04 0.14 −0005 0007 0001 0016 0003

asset (in billions)7 Conventional firm profitability 0.04 0.13 0006 −0002 0001 0009 0003 −00048 Media report (in 100,000s) 0.80 0.90 0000 −0004 −0003 −0002 0003 −0002 00029 Adoption in industry 0.17 0.09 0000 −0001 0014 0008 0002 −0002 0002 0007

10 Adoption in MSA 0.18 0.11 −0002 0002 −0003 0011 0005 0008 0005 0005 000611 Media report at tie 0.87 0.98 0006 0008 −0008 −0008 −0011 −0010 −0003 0025 −0003 0.01

founding (in 100,000s)12 Adoption in industry at 0.16 0.08 0005 0002 −0002 0010 0000 −0002 0003 0005 0047 0.05 0016

tie founding13 Adoption in MSA at tie founding 0.16 0.11 0003 0000 −0013 0009 0003 0007 0002 0006 0002 0.67 0003 0.0814 Internet tie (t − 1) at tie founding 0.26 0.44 −0008 0002 0027 0019 −0002 0022 0001 −0000 0004 0.03 −0004 0.02 0.06

Note. The unit of analysis is an existing interlock tie between a conventional firm and an Internet company in a year.

1996. I could only observe a conventional firm’s pre-vious contact with the Internet sector (previous to theyear in which a tie was founded) for ties founded begin-ning in 1997. Thus, the attenuation analysis of previouscontact on social influence at the founding of an inter-lock tie is restricted to a smaller sample that includesties that were founded after 1996 and whose conven-tional partners did not appear in the sample for the firsttime. However, despite the reduction in sample size, thedescriptive statistics for the variables shown in Table 5are largely similar to those for the full sample shownin Table 3, except that the left-censored rate is lowerin the subsample. The similarity in descriptive statisticsprovides confidence about the representativeness of thesubsample.

Table 6 reports the piecewise exponential frailty haz-ard models that test the moderation effect on tie disso-lution of contact prior to the founding of a tie. Model12 presents a primary model with all control variablesand the three social influence variables for the yearwhen an interlock tie was founded. The results showthat there are no general predictions that can be maderegarding the effects of social influence variables presentat the founding of a tie on the hazard of tie dissolu-tion. However, the profitability of Internet companieshas a negative impact on the dissolution hazard of a tie(b = −00625, p < 0010; two-sided test), which suggeststhat the quality of a network partner is a factor in affect-ing the stability of an interlock tie. Model 13 includesthe main effect of a conventional firm’s previous con-nection to the Internet sector at the time of tie formation.The effect is not significant either (b = −00088, n.s.).Models 14–17 test the moderation effect of a conven-tional firm’s contact prior to initiation on the relationshipbetween social influence at the time of a tie’s founding

and the dissolution hazard of the tie. The results showthat for a conventional firm, previous contact at the timeof tie founding has negative moderating effects on theimpact of all three social influence variables, althoughonly the moderation effect on media attention is statis-tically significant (b = −00562, p < 0001). This negativemoderation effect remains robust when all three interac-tion effects are tested simultaneously in the full model,Model 17 (b = −00561, p < 0001). Setting other vari-ables at their means, the life length of Internet interlocksformed when media attention was high (i.e., one stan-dard deviation above the mean) by conventional firmswith previous contact was about 6 years, whereas thatby firms without prior interactions was only 3.3 years.These findings confirm the expectation that conventionalfirms with direct knowledge make wiser decisions inpicking up partners from a new sector, whereas thosewithout direct knowledge are more affected by bandwag-ons and pick up popular but not necessarily high-qualitypartners. Although similar patterns of results emergedfor all three social influence variables, the result holdsstatistical confidence only for media reports. The dif-ference may be because media reports are more likelyto be laudatory and contain biases, whereas peers havedirect knowledge to various degrees and engage in diver-sified actions, helping to reduce social conformity. Thus,Hypothesis 4 receives partial support.

Joint Estimation of the Formation andDissolution of Internet InterlocksThe above analyses of tie formation and tie dissolutionare conducted in two separate equations. The fact thatthe strong and significant impacts of social influencevariables on the formation of interlocks stands in sharpcontrast to their insignificant impacts on the dissolution

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


Table 6 Piecewise Exponential Frailty Hazard Model on the Dissolution of Ties Founded After 1996

(12) (13) (14) (15) (16) (17)

Directional tie 00058 00054 00060 00058 00053 000604001055 4001055 4001065 4001055 4001055 4001065

Trade relationship −00008 −00011 00001 −00017 −00015 −000054001335 4001345 4001345 4001345 4001345 4001355

Left censored −00160 −00126 −00121 −00128 −00144 −001384002105 4002195 4002205 4002195 4002235 4002235

Internet company asset −30905 −30627 −40135 −30554 −30661 −401994303425 4303725 4304095 4303815 4303765 4304285

Internet company profitability −00625 −00645 −00725∗∗ −00637 −00623 −007094003445 4003475 4003695 4003485 4003505 4003715

Conventional firm asset 00143 00192 00088 00186 00224 001224004305 4004385 4004345 4004355 4004435 4004405

Conventional firm profitability −00493 −00488 −00487 −00493 −00487 −004864003005 4002995 4003095 4003005 4002995 4003095

Media report 00055 00055 00040 00054 00056 000414000735 4000735 4000735 4000735 4000735 4000735

Adoption in industry −00797 −00797 −00567 −00766 −00808 −005714008535 4008545 4008425 4008585 4008555 4008445

Adoption in MSA −00292 −00310 −00198 −00299 −00271 −001584008115 4008065 4008215 4008085 4008065 4008205

Media report at tie founding −00001 −00001 00000 −00001 −00001 000004000015 4000015 4000015 4000015 4000015 4000015

Adoption in industry at tie founding 00439 00434 00443 00725 00457 004904009755 4009765 4009635 4100695 4009785 4100795

Adoption in MSA at tie founding −00548 −00516 −00627 −00521 −00430 −005404008485 4008435 4008555 4008445 4008605 4008735

Internet tie 4t − 15 at tie founding −00088 00293 00122 00034 004264001625 4002045 4003595 4003225 4004305

Internet tie 4t − 15 at tie founding −00562∗∗∗ −00561∗∗∗

∗ Media report at tie founding 4002135 4002165Internet tie 4t − 15 at tie founding −10305 −00107

∗ Adoption in industry at tie founding 4200115 4109475Internet tie (t − 1) at tie founding −00727 −00704

∗ Adoption in MSA at tie founding 4106835 4106675

N 11055 11055 11055 11055 11055 11055Log likelihood 113260694 113260843 113310211 113270056 113260936 113310305�2 5020274 5020090 4950533 5010812 5020006 4950431

Notes. The unit of analysis is an existing interlock tie between a conventional firm and an Internet company in a year. Standard errors arein parentheses. Dummies of tie age were included in the estimation but omitted from reporting.

∗∗p < 0005; ∗∗∗p < 0001 (one-sided test for hypothesized variables and two-sided test for other variables).

of interlocks provides a straightforward confirmation ofthe asymmetric effects specified in H2. However, a jointestimation of the formation and dissolution of interlockswould allow for a direct comparison of the magnitude ofthese coefficients. I therefore adopted the joint estima-tion of the two dependent variables using a seeminglyunrelated analysis.9 The seemingly unrelated analysis isdeveloped by Zellner (1962), who shows that the seem-ingly unrelated regression can be conceived as a general-ized least square estimation in the case of linear regres-sions. The seemingly unrelated analysis for nonlineardependent variables, however, has not been widely incor-porated into prepackaged software.10 An added chal-lenge is that the dependent variable for tie formation is acount of the number of new initiations, but the dependent

variable for tie dissolution is a hazard rate of tie disso-lution. The current econometric technique is not read-ily available to account for the interdependence betweentwo different forms of distribution (Kim 2006, Simonsand Roberts 2008). Following Kim (2006) and Simonsand Roberts (2008), I adopted the seemingly unrelatedestimation of two equations with the same type of dis-tributional function. In particular, I conducted two setsof seemingly unrelated analyses, one with the bivariatenegative binomial and the other with the bivariate probitas the underlying joint distribution. Triangulation basedon two different forms of distributions will provide anespecially rigorous check of the robustness of my results.

First, I adopted the seemingly unrelated negative bino-mial model to analyze the formation and dissolution

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


of ties. The unit of analysis is a conventional firm ina year. For the equation to estimate tie formation, thedependent variable is the number of ties with Internetcompanies that a focal conventional firm newly formedin a year, the same as that used in the GEE negativebinomial model. The sample includes all conventionalfirms in a year. For the equation to estimate tie dissolu-tion, the dependent variable is the number of ties withInternet companies that a conventional firm dropped ina year. The sample includes all conventional firms thathad at least one tie to the Internet sector in the previ-ous year. I adopted the basic seemingly unrelated nega-tive binomial (SUNB) model developed by Winkelmann(2003). However, my model is more complex in that thedissolution model estimates a conditional likelihood—tie dissolution can only happen for firms that had atleast one tie at the beginning of a year, and the numberof dissolutions cannot exceed the number of ties heldby a firm at the beginning of a year. I used the max-imum likelihood method to estimate the parameters ofthe conditional SUNB model and MATLAB to developthe estimation code.11

Second, I adopted the seemingly unrelated bivariateprobit model to analyze the formation and dissolution ofties. The unit of analysis is an interlock tie between aconventional firm and an Internet company in a year. Forthe tie formation analysis, the dependent variable is adummy variable that equals 1 for a realized tie between aconventional firm and an Internet company. The sampleincludes all realized ties and 10 randomly selected con-ventional firm–Internet company ties that did not happenbut were possible for each realized tie in a year.12 Outof the 2,259 realized tie-years, 1,944 were actually usedto include the lagged control variables (i.e., the numberof interlock partners and the number of Internet partnersin the previous year), and 19,727 out of 21,384 matchedobservations (1944 × 11) were actually included in theregression because of missing values.

For the tie dissolution analysis, the dependent variableis a dummy variable that equals 1 if a realized tie dis-solved in a year. The sample includes all the realized tiesbetween a conventional firm and an Internet company atthe beginning of a year. Because the event of dissolutioncan only be observed for those realized ties, the seem-ingly unrelated bivariate probit model is computation-ally equivalent to a Heckman probit model, in which thetie formation analysis serves as the first-stage selectionmodel. Moreover, I reported the Huber–White estimatorof standard errors clustered by conventional partners toaccount for interdependence between observations.

The joint estimation results were reported in OnlineAppendices 2(a) and 2(b). The conditional SUNB modelshows that media reports, the percentage of adopterswithin the same MSA, and the percentage of adopterswithin the same industry all have strong positive effectson the formation of ties but no significant effect on

the dissolution of ties, consistent with the previouslyreported results of independent estimations. The like-lihood ratio (LR) tests further confirm that the fash-ion variables have a stronger impact on tie formationthan on tie dissolution (�2415 = 4004, p < 0005 formedia reports; �2415 = 6028, p < 0005 for the industrypeer effect; and �2415 = 3094, p < 0005 for the MSApeer effect). Similarly, the Heckman probit model showsthat the percentages of firms within the same MSAand within the same industry both have strong positiveeffects on the formation of ties but no significant effecton the dissolution of ties. The LR tests further confirmthat the two fashion variables have a stronger impacton tie formation than on tie dissolution (�2415= 18093,p < 0001 for the industry peer effect; and �2415= 5047,p < 0005 for the MSA peer effect). The fashion vari-able measured by media reports is omitted from theestimation because the proportional matching methodprecludes the variance at the year level. Together, therobustness of my results across different model speci-fications (negative binomial versus hazard rate), sampleselections (full versus matched sample), and levels ofanalysis (firm versus dyad) provides especially rigoroussupport for the asymmetric effects of fashion variableson the formation and dissolution of ties.

Alternative ExplanationsDespite the strong evidence supporting the attenuationeffect of private information on social influence, thereare a few alternative explanations. The first alternativeexplanation argues that firms that adopted Internet inter-locks before the fad period might have had special struc-tural reasons (other than learning) that prevented themfrom dropping Internet ties compared with firms that hadnot adopted such ties in the pre-fad period. First, thefrailty-augmented piecewise exponential hazard modelhas already taken into account the unobserved hetero-geneities at the firm level. In addition, the estimationof the variance of firm-specific heterogeneities rejectsthe hypothesis that some firms are more susceptible totie failures than others after controlling for observables,further confirming that there is little unobserved hetero-geneity in the conventional firm-specific hazards of tiefailure. Besides the statistical correction, I also directlytested the argument that firms that established Inter-net interlocks before the fad were likely to retain theirties longer by creating dummy variables that indicatewhether a tie’s conventional partner had Internet inter-locks before 1999 or 2000. If the special structure argu-ment was true, then we should expect that Internet tiescreated by these firms in general survived longer, regard-less of whether or not they were created during the fadperiod. However, the results show that neither of thesevariables is significant, again lending no support for thisargument on unobserved structure.

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


The second alternative account explaining the lackof responsiveness of interlock dissolution to fashionsinvolves a time-lag effect attributable to the appointmentterms of directors. The typical term for an initial direc-tor appointment is three years; consequently, firms maynot be able to immediately abandon the interlocks thatwere constructed in the previous one or two years. Totest the potential time-lag effect, I lagged the three fash-ion variables by one, two, and three years, testing theirrespective impact on the dissolution of interlock ties.None of these variables generates a significant impacton the dissolution of networks, which helps rule out thisalternative explanation attributed to director terms.

Finally, I conducted two sets of robustness checksregarding other forms of indirect learning. One has todo with the fact that fashion cycles per se may conveyknowledge. Diverse information is conveyed when pos-itive views during the boom period stand in contrast topessimistic opinions during the bust era. In unreportedanalyses, I compared the sensitivity of network forma-tion during the fashion cycle (1997–2003) with that ofthe post-fashion era (2004–2006) and found support forthe fashion-cycle learning argument. The formation ofInternet interlocks remains highly sensitive to all threefashion variables during the fashion cycle, but it is onlysensitive to adoption within the same industry in thepost-fashion era (2004–2006). The robust effect of peeractions within the same industry may be attributed tothe fact that industry peers may be more likely to serveas a social comparison group because of the perceivedrelevance of their actions.

The second form of indirect learning may happenthrough relational connections such as trade. Firms thatare in a close trade relationship with the Internet sectorare likely to have more opportunities to obtain informa-tion about the Internet. In unreported analyses, I testedthe moderation effects of trade on the formation and dis-solution of networks. The results showed that, despitethe absence of a significant attenuation effect on theformation of networks, a close trade relationship sig-nificantly reduces the failure hazard of ties that arefounded in a hurry with peers within the same MSAarea. Together, although evidence derived from theseindirect forms of learning is weaker and noisier, it gener-ally supports the argument that knowledge and learningfacilitate independent decision making and reduce theinfluence of social conformity.

Discussion and ConclusionIn this study of the evolution of conventional firms’interlocks with Internet companies during the Internetbubble period, I have found that fashions had asym-metric effects on the formation and dissolution of net-works. The popularity of network partners significantlyincreases the formation of networks with distant actors

segregated by structural barriers but does not predict thedissolution of these ties. Building on the contact theory,I proposed that as direct contact affords opportunitiesfor learning, it reduces a conventional firm’s relianceon social information to judge the quality of networkpartners. As a result, the formation of distant networksis more sensitive to fashions than is their dissolution.Consistent with this argument, I further found that previ-ous contact can prevent conventional firms from engag-ing in social adherence when forming networks. Thosewith direct contact were less likely to be induced byfashions to construct ties with popular network partners.Moreover, the ties they did form during the heyday of afashion survived longer than those formed by other con-ventional firms without direct contact during the heydayof a fashion. These findings contribute to the literaturein at least four respects.

First, this study expands the contextual perspectiveof network evolution. The contextual perspective pro-vides a promising avenue for explaining how bridgingties across local sectors are formed. The promise ofthe contextual perspective rests on a change of envi-ronments that gives heterogeneous actors a chance tomingle (Sorenson and Stuart 2008). However, changealso makes networks instable. Although scholars haveshown that the change in social contexts can explain theformation of distant ties, they have not shown whetherthese ties will dissolve as soon as social contexts changeagain. It may not be surprising that networks graduallydissolve as the conditions that facilitate their formationand maintenance weaken or disappear. However, as longas the responsiveness of network formation to contex-tual factors is stronger than that of network dissolution,contextual factors should remain a powerful force thatdrives diverse actors together and shortens the distancebetween local clusters. One unique contribution of thispaper is to show that the dissolution of networks is amuch stickier process than their formation.

The contextual perspective also introduces a longitu-dinal and dynamic view of network evolution. Throughtaking into account macrocontextual factors, this paperanswers the call of economic sociologists to embed orga-nizational decisions into social contexts. Moreover, thispaper brings a novel angle to the study of interorgani-zational networks and social contexts by treating net-work evolution as the subject of interests, thus departingfrom the tradition deriving from Granovetter (1985) thattreats networks as antecedents of economic outcomes.Meanwhile, examining the attenuation effect of firm-level characteristics also answers the call of institutionaltheorists to attend not just to consensus and conformitybut also to conflict and change in social structure (Scott2004). Sorenson and Stuart (2008) focused on the maineffects of fads and fashions in attracting distant orga-nizations. This study extends Sorenson and Stuart byconsidering how individual organizations may respond

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


to fads and fashions differently. Firms’ direct knowledgeplays a critical role in attenuating the effectiveness offashions as a driving force of network evolution. Incor-porating organization-level heterogeneities helps to setup the boundary conditions of the theory of contextualfactors as antecedents of network evolution.

Second, this study contributes to the network evo-lution literature by simultaneously examining networkformation and dissolution. This study extends the lit-erature to investigate the dissolution of networks andthus presents a more complete picture of network evo-lution. Moreover, the dissolution of networks itself is anunderstudied topic in the literature on network evolution.By showing that network formation and dissolution aredriven by asymmetric forces, this study not only con-tributes to the need for investigations on the dissolutionof networks but also suggests that the dissolution of net-works is a more complicated process.

By highlighting the complicated nature of networkdissolution, this paper also opens up promising avenuesfor future research. The core argument of this paper isthat the absence of knowledge that leads to social con-formity in the upswing of a fashion is overcome in thedownswing of a fashion because of direct contact estab-lished through networking. Thus, independently of otherforces, we should expect an asymmetric relationshipbetween fashions and the formation and dissolution ofnetworks. However, direct contact also opens up oppor-tunities for various forms of social interaction. It is plau-sible to argue that certain effects of social interactionsmay prevent the dissolution of networks with low-qualitypartners and result in a stickier process of network disso-lution (e.g., Sorenson and Waguespack 2006). However,this account is inadequate for explaining the observedasymmetry for at least three reasons. One is that theaverage life span of an interlock tie between a conven-tional firm and an Internet company during the periodof my observation is 2.76 years, slightly shorter thanthe common 3-year term of appointment, suggesting thatnetwork dissolution is fairly frequent. Second, there isevidence suggesting that a network partner’s quality (asis indicated for example by profitability) affects the haz-ard of tie failure. Third, the effects of social interactionsalone cannot account for the variance in the process oftie dissolution, such as the moderation effect of directcontact in reducing the failure hazard of ties foundedduring fashion heydays. Future scholars should use inter-views or surveys to examine the social and emotionalprocesses associated with direct contact and to investi-gate how interlock networks may evolve with the inter-actions between directors and corporate management.

My finding that ties created by conventional organiza-tions that have prior connection survived longer is con-sistent with the prediction of network matching theorythat well-matched ties will be more stable (Vissa 2011).

Using the dyad-level similarity and dissimilarity to pre-dict the chance that two actors will be connected, thematching theory particularly emphasizes the criteria thatorganizations adopt to pick up network partners (Mit-suhashi and Greve 2009, Vissa 2011). The findings ofthis paper are consistent with the matching theory inthree respects. First, the quality of matching is directlyrelated to the quality of organizations’ decisions in pick-ing up network partners. Second, the finding that organi-zations abandoned fashion-induced ties confirms Vissa’s(2011) point that organizational decision makers are onlypartially accurate in picking up networks partners. Third,the finding that direct connection transfers knowledge isalso consistent with Mitsuhashi and Greve’s (2009) find-ing that existing ties improve matching quality becausethe assessment of prospective partners requires privateinformation that circulates only through direct connec-tions. This paper also has two implications for the net-work matching theory: (1) the criterion of matching maynot be static but changing and socially constructed, and(2) the criterion of matching may be extended fromthe level of dyads to micro–macro interaction. Futureresearch that adopts the matching perspective can inves-tigate how matching criteria evolve over time and howcontextual factors may interact with dyad-level matchingcriteria.

Third, this study contributes to the organizationallearning and interorganizational network literature. Thenumber of papers that span the boundaries between learn-ing theories and the interorganizational network litera-ture has been growing. A key linkage between the twotheoretical domains is the role of networks as a chan-nel of learning (Powell 1990). Research has found thatnetwork-enabled learning ranges from straightforwardimitation (e.g., Davis 1991, Haunschild 1993) to sophisti-cated inference (Beckman and Haunschild 2002). Recentresearch has reversed the order of reasoning, suggest-ing that organizations purposely build certain types ofnetworks to discover new information and opportunities.Along this line of reasoning, Beckman et al. (2004) usedthe concept of exploitation and exploration in organi-zational learning theories to predict firms’ networkingbehaviors in building new ties or reinforcing existingones. Similarly, Baum et al. (2005) argued that orga-nizations’ learning from their past actions can explaintheir performance-driven network construction. My paperextends this literature by simultaneously treating directnetworks as a source for acquiring valuable informa-tion and the construction and termination of networks asresults of organizations’ attempts to cope with uncertain-ties in their environment. Moreover, this paper suggeststhat the function of learning is not limited to refiningorganizations’ skills or practices but also includes shield-ing them from external irrational exuberance.

In particular, this paper extends the contact theory byexpanding the scope of application from individuals to

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


organizations. Decades of research on the contact the-ory have focused on how an individual’s contact without-group members reduces his or her intergroup preju-dice. Early research on the contact theory mainly investi-gated how racial segregation and integration may changeindividuals’ attitudes (e.g., Bradburn et al. 1971, Brooks1975). Although recent studies have expanded the scopeof investigation from racial groups to the homeless(Lee et al. 2004), homosexuals (Herek and Capitanio1996), the mentally and physically disabled (Pettigrewand Tropp 2006), and even computer programmers(McGinnis 1990), most of these studies addressed theindividual level. My findings that the behavior of orga-nizations with direct contact is less affected by sociallypopular beliefs extend the contact theory to explainorganization-level phenomena. In addition, future orga-nizational scholars can profit from applying the contacttheory to study the organizational-level phenomena bytesting the conditions that facilitate integration and bust-ing stereotypes.

Nevertheless, readers should be reminded that thisstudy only considers one type of direct contact. Inthe current research setting, the limited scope of directcontact is less problematic because Internet companiesbelong to a new organizational population and thus otherchannels of connection may not have been highly devel-oped. However, future researchers who apply the contacttheory to account for other organizational phenomenashould be aware of a full set of connection channels andconsider the stage of development in interorganizationalrelationships.

Last, but not least, this study contributes to the inter-lock literature. Early scholars adopted resource depen-dence and class perspectives to study the formation ofboard interlocks (Borgatti and Foster 2003). They per-ceived interlocks as an instrument for managing organi-zations’ resource dependence (Pfeffer 1972, Burt 1983,Mintz and Schwartz 1985, Mizruchi and Stearns 1994)and for maintaining the power and cohesion of capital-ist elites (Domhoff 1967, Palmer 1983, Pennings 1980,Useem 1979). Since the 1990s, board interlock stud-ies have adopted an information perspective and vieweddirectorate ties as an important channel for transferringinformation and sharing knowledge about effective orga-nizational practices (e.g., Galaskiewicz and Wasserman1989, Davis 1991, Haunschild 1993, Palmer et al. 1993,Beckman and Haunschild 2002). More recent inter-lock studies have also emphasized the constraints ofgeography on the formation of interlocks (Kono et al.1998) and the imprinting effects of community struc-tures (Marquis 2003). One implication of these inter-lock studies is that interlock networks tend to be stable(Beckman and Haunschild 2002, Sullivan et al. 2007)when interlock networks play an important role in main-taining elite cohesion (Useem 1984) or diffusing orga-nizational practices (Davis 1991, Haunschild 1993) and

when the development of networks is constrained bygeographical boundary (Kono et al. 1998) or historicalorigins (Marquis 2003). However, this paper suggeststhat significant changes in interlock networks are possi-ble. This paper adds to this vibrant body of literature bydemonstrating that fads and fashions are yet another setof antecedents in the change of board interlocks.

Electronic CompanionAn electronic companion to this paper is available as part ofthe online version that can be found at http://orgsci.journal.informs.org/.

AcknowledgmentsThe author is thankful to Eric Abrahamson, Peer Fiss, JosephGalaskiewicz, Paul Ingram, James Kitts, Ozgecan Kocak, KoKuwabara, Olav Sorenson, Bilian Sullivan, Botao Yang, andseminar participants at Columbia University and the Univer-sity of Southern California for suggestions. The author is alsograteful to the anonymous Organization Science reviewers fortheir constructive comments.

Endnotes1The terms “fad” and “fashion” are used interchangeably inthis paper.2IRRC and ExecuComp also assign a unique ID for each direc-tor and executive. I identified unduplicated directors and exec-utives using their IDs, names, ages, and gender.3This definition of interlocks enables testing the formation ofall three types of interlock: sent, received, and neutral inter-locks. Haunschild and Beckman (1998) suggest that interlocksof all three types are influential channels of information andresources.4There were 94 ties founded between 1996 and 2006 that werenot counted as new ties because they emerged before both thepartners first entered my sample.5Fixed-effect models are often used to control firm-level unob-served heterogeneities but are inappropriate in the current con-text. Because the Internet companies are a new population andinterlocks with conventional firms were relatively infrequentduring the period of observation, most conventional firms didnot initiate interlock ties with the Internet sector. These obser-vations with an invariant dependent variable cannot contributeto the fixed-effect estimation and result in an 80% loss of thesample size. Nevertheless, the fixed-effect negative binomialand Poisson estimation results support the hypothesis that fash-ions have strong positive effects on the formation of interlockswith Internet companies.6These periods were determined by considering both the typ-ical director terms and the structure of my data. The mostcommon director term is three years, and thus six years equalsapproximately two terms.7The term �i is assumed to follow a predetermined distribu-tion with a fixed mean and variance �. This paper adopts acommonly used gamma distribution but assuming the inverseGaussian distribution produces similar results. The likelihoodof the observed data can be derived by calculating the individ-ual conditional likelihood and integrating out the frailty. Usingthe maximum likelihood algorithm, the regression parametersand frailty variance � can be estimated.

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


8An alternate measure of the profit potential of an Internetcompany is the ratio of its market value over total assets. How-ever, controlling for this variable would substantially reducethe sample size because of missing values in the market valuedata. Nevertheless, the reported results are not sensitive to con-trolling for the ratio of market value over total assets.9I thank an anonymous reviewer for pointing this out to me.10One exception is the bivariate probit model.11The technical details of the estimation of the conditionalSUNB model are available from the author upon request. Inaddition, Kim (2006) reported a GAUSS version of the uncon-ditional SUNB estimation code.12This approach of comparing actual ties with unrealized onesis same as that used in prior work on dyad-level analysis oftie formation (e.g., Sorenson and Stuart 2008, Mitsuhashi andGreve 2009, Vissa 2011). The use of a matched sample hasadvantages over the use of the full sample of all potential tiesbecause the latter generates too many observations of a singlefirm in the data, which can lead to systematic underestima-tion of standard errors. Moreover, in this paper, the realizedties account for only 0.3% of all possible ties. The proce-dure of proportional matching does not reduce statistical powerbecause rare events preserve most information, whereas thevast majority of the void potential ties are likely to be irrele-vant controls. In addition to the 1:10 match, I also tried the 1:1,1:3, and 1:5 matches, all of which produced similar results.

ReferencesAbrahamson, E. 1991. Managerial fads and fashions: The diffusion

and rejection of innovations. Acad. Management Rev. 16(3)586–612.

Abrahamson, E., G. Fairchild. 1999. Management fashion: Lifecycles,triggers, and collective learning processes. Admin. Sci. Quart.44(4) 708–740.

Abrahamson, E., L. Rosenkopf. 1993. Institutional and competitivebandwagons: Using mathematical modeling as a tool to exploreinnovation diffusion. Acad. Management Rev. 18(3) 487–517.

Allport, G. W. 1954. The Nature of Prejudice. Addison-Wesley, Read-ing, MA.

Almeida, P., B. Kogut. 1999. Localization of knowledge and themobility of engineers in regional networks. Management Sci.45(7) 905–917.

Baltagi, B. H., S. H. Song. 2006. Unbalanced panel data: A survey.Statistical Papers 47(4) 493–523.

Banerjee, A. V. 1992. A simple model of herd behavior. Quart. J.Econom. 107(3) 797–817.

Baum, J. A. C., T. J. Rowley, A. V. Shipilov, Y.-T. Chuang. 2005.Dancing with strangers: Aspiration performance and the searchfor underwriting syndicate partners. Admin. Sci. Quart. 50(4)536–575.

Beckman, C. M., P. R. Haunschild. 2002. Network learning: Theeffects of partners’ heterogeneity of experience on corporateacquisitions. Admin. Sci. Quart. 47(1) 92–124.

Beckman, C. M., P. R. Haunschild, D. J. Phillips. 2004. Friends orstrangers? Firm-specific uncertainty, market uncertainty, and net-work partner selection. Organ. Sci. 15(3) 259–275.

Bikhchandani, S., D. Hirshleifer, I. Welch. 1992. A theory of fads,fashion, custom, and cultural change as informational cascades.J. Political Econom. 100(5) 992–1026.

Borgatti, S. P., P. C. Foster. 2003. The network paradigm in organi-zational research: A review and typology. J. Management 29(6)991–1013.

Bradburn, N. M., S. Sudman, G. L. Gockel. 1971. Side by Side: Inte-grated Neighborhoods in America. Quadrangle Books, Chicago.

Brass, D. J., J. Galaskiewicz, H. R. Greve, W. Tsai. 2004. Takingstock of networks and organizations: A multilevel perspective.Acad. Management J. 47(6) 795–817.

Brooks, D. 1975. Race and Labor in London Transport. Oxford Uni-versity Press, London.

Broschak, J. P. 2004. Managers’ mobility and market interface: Theeffect of managers’ career mobility on the dissolution of marketties. Admin. Sci. Quart. 49(4) 608–640.

Burt, R. S. 1983. Corporate Profits and Cooptation: Networks of Mar-ket Constraints and Directorate Ties in the American Economy.Academic Press, New York.

Byrnes, N., P. C. Judge. 1999. Internet anxiety; part in envy, part infear, corporate America is embracing a radically new businessmodel. Business Week 3635 78–88.

Cialdini, R. B. 1993. Influence: The Psychology of Persuasion.William Morrow and Company, New York.

Contoyannis, P., A. M. Jones, N. Rice. 2004. The dynamics of healthin the British Household Panel Survey. J. Appl. Econometrics19(4) 473–503.

Cui, J. 2007. QIC program and model selection in GEE analyses.Stata J. 7(2) 209–220.

Davis, G. F. 1991. Agents without principles? The spread of the poi-son pill through the intercorporate network. Admin. Sci. Quart.36(4) 583–613.

Davis, G. F., H. R. Greve. 1997. Corporate elite networks and gover-nance changes in the 1980s. Amer. J. Sociol. 103(1) 1–37.

Domhoff, G. W. 1967. Who Rules America? Power, Politics, andSocial Change. Prentice-Hall, Englewood Cliffs, NJ.

Feld, S. L. 1981. The focused organization of social ties. Amer. J.Sociol. 86(5) 1015–1035.

Fiss, P. C., P. M. Hirsch. 2005. The discourse of globalization: Fram-ing and sense-making of an emerging concept. Amer. Sociol.Rev. 70(1) 29–52.

Galaskiewicz, J., S. Wasserman. 1989. Mimetic processes within aninterorganizational field: An empirical test. Admin. Sci. Quart.34(3) 454–479.

Gilovich, T. 1987. Secondhand information and social judgment.J. Experiment. Soc. Psych. 23(1) 59–74.

Granovetter, M. 1985. Economic action and social structure: Theproblem of embeddedness. Amer. J. Sociol. 91(3) 481–510.

Greve, H. R. 1995. Jumping ship: The diffusion of strategy abandon-ment. Admin. Sci. Quart. 40(3) 444–473.

Greve, H. R., J.-E. Pozner, H. Rao. 2006. Vox populi: Resource par-titioning, organizational proliferation, and the cultural impactof the insurgent microradio movement. Amer. J. Sociol. 112(3)802–837.

Gulati, R., M. Gargiulo. 1999. Where do interorganizational networkscome from? Amer. J. Sociol. 104(5) 1439–1493.

Gutierrez, R. G. 2002. Parametric frailty and shared frailty survivalmodels. Stata J. 2(1) 22–44.

Haunschild, P. R. 1993. Interorganizational imitation: The impact ofinterlocks on corporate acquisition activity. Admin. Sci. Quart.38(4) 564–592.

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


Haunschild, P. R. 1994. How much is that company worth?: Interorga-nizational relationships, uncertainty, and acquisition premiums.Admin. Sci. Quart. 39(3) 391–411.

Haunschild, P. R., C. M. Beckman. 1998. When do interlocks matter?Alternate sources of information and interlock influence. Admin.Sci. Quart. 43(4) 815–844.

Herek, G. M., J. P. Capitanio. 1996. “Some of my best friends” inter-group contact, concealable stigma, and heterosexuals’ attitudestoward gay men and lesbians. Personality Soc. Psych. Bull. 22(4)412–424.

Kim, D.-G. 2006. Multivariate analysis of accidental related outcomeswith respect to contemporaneous correlation and endogene-ity: Application of simultaneous estimation approaches. Unpub-lished doctoral dissertation, University of Arizona, Tempe.

Kono, C., D. Palmer, R. Friedland, M. Zafonte. 1998. Lost in space:The geography of corporate interlocking directorates. Amer. J.Sociol. 103(4) 863–911.

Lee, B. A., C. R. Farrell, B. G. Link. 2004. Revisiting the contacthypothesis: The case of public exposure to homelessness. Amer.Sociol. Rev. 69(1) 40–63.

Liang, K.-Y., S. L. Zeger. 1986. Longitudinal data analysis using gen-eralized linear models. Biometrika 73(1) 13–22.

Lorsch, J. W., E. MacIver 1989. Pawns and Potentates: The Realityof America’s Corporate Boards. Harvard Business School Press,Boston.

Marquis, C. 2003. The pressure of the past: Network imprinting inintercorporate communities. Admin. Sci. Quart. 48(4) 655–689.

McGinnis, S. P. 1990. Descriptive and evaluative components ofstereotypes of computer programmers and their determinants.Unpublished doctoral dissertation, City University, New York.

Mintz, B., M. Schwartz. 1985. The Power Structure of American Busi-ness. University of Chicago Press, Chicago.

Mitsuhashi, H., H. R. Greve. 2009. A matching theory of alliance for-mation and organizational success: Complementarity and com-patibility. Acad. Management J. 52(5) 975–995.

Mizruchi, M. S. 1996. What do interlocks do? An analysis, critique,and assessment of research on interlocking directorates. AnnualRev. Sociol. 22 271–298.

Mizruchi, M. S., L. B. Stearns. 1994. A longitudinal study of borrow-ing by large American corporations. Admin. Sci. Quart. 39(1)118–140.

Mizruchi, M. S., L. B. Stearns, C. Marquis. 2006. The conditionalnature of embeddedness: A study of borrowing by large U.S.firms, 1973–1994. Amer. Sociol. Rev. 71(2) 310–333.

Office of Management and Budget. 2000. North American Indus-try Classification System—Update for 2002. Report, ExecutiveOffice of the President, Washington, DC.

Palmer, D. 1983. Broken ties: Interlocking directorates and intercor-porate coordination. Admin. Sci. Quart. 28(1) 40–55.

Palmer, D. A., B. M. Barber, X. Zhou. 1993. Late adoption of the mul-tidivisional form by large U.S. corporations: Institutional, polit-ical, and economic accounts. Admin. Sci. Quart. 38(1) 100–131.

Pan, W. 2001. On the robust variance estimator in generalized esti-mating equations. Biometrika 88(3) 901–906.

Pennings, J. M. 1980. Interlocking Directorates. Jossey-Bass, SanFrancisco.

Perkins, A. B., M. C. Perkins. 2001. The Internet Bubble: Insidethe Overvalued World of High-Tech Stocks—And What YouNeed to Know to Avoid the Coming Shakeout. Harper Business,New York.

Pettigrew, T. F. 1998. Intergroup contact theory. Annual Rev. Psych.49 65–85.

Pettigrew, T. F. 2008. Future directions for intergroup contact theoryand research. Internat. J. Intercultural Relations 32(3) 187–199.

Pettigrew, T. F., L. R. Tropp. 2006. A meta-analytic test of intergroupcontact theory. J. Personality Soc. Psych. 90(5) 751–783.

Pfeffer, J. 1972. Size and composition of corporate boards of direc-tors: The organization and its environment. Admin. Sci. Quart.17(2) 218–228.

Pfeffer, J., G. R. Salancik. 1978. The External Control of Organi-zations: A Resource Dependence Perspective. Harper & Row,New York.

Podolny, J. M., K. L. Page. 1998. Network forms of organization.Annual Rev. Sociol. 24 57–76.

Pollock, T. G., V. P. Rindova. 2003. Media legitimation effects in themarket for initial public offerings. Acad. Management J. 46(5)631–642.

Powell, W. W. 1990. Neither market nor hierarchy: Network formsof organization. B. M. Staw, L. L. Cummings, eds. Researchin Organizational Behavior, Vol. 12. JAI Press, Greenwich, CT,295–336.

Rao, H., H. R. Greve, G. F. Davis. 2001. Fool’s gold: Social proofin the initiation and abandonment of coverage by Wall Streetanalysts. Admin. Sci. Quart. 46(3) 502–526.

Rogers, E. M. 2003. Diffusion of Innovations, 5th ed. Simon &Schuster, New York.

Rosenkopf, L., G. Padula. 2008. Investigating the microstructure ofnetwork evolution: Alliance formation in the mobile communi-cations industry. Organ. Sci. 19(5) 669–687.

Ruef, M. 2000. The emergence of organizational forms: A communityecology approach. Amer. J. Sociol. 106(3) 658–714.

Schwartz E. I. 1999. Digital Darwinism. Broadway Books, New York.

Scott, W. R. 2004. Institutional theory: Contributing to a theoreticalresearch program. K. G. Smith, M. A. Hitt, eds. Great Mindsin Management: The Process of Theory Development. OxfordUniversity Press, Oxford, UK, 460–484.

Simons, T., P. W. Roberts. 2008. Local and non-local pre-foundingexperience and new organizational form penetration: The caseof the Israeli wine industry. Admin. Sci. Quart. 53(2) 235–265.

Sine, W. D., H. A. Haveman, P. S. Tolbert. 2005. Risky business?Entrepreneurship in the new independent-power sector. Admin.Sci. Quart. 50(2) 200–232.

Sorenson, O., P. G. Audia. 2000. The social structure of entrepre-neurial activity: Geographic concentration of footwear produc-tion in the United States, 1940–1989. Amer. J. Sociol. 106(2)424–462.

Sorenson, O., T. E. Stuart. 2008. Bringing the context back in: Set-tings and the search for syndicate partners in venture capitalinvestment networks. Admin. Sci. Quart. 53(2) 266–294.

Sorenson, O., D. M. Waguespack. 2006. Social structure andexchange: Self-confirming dynamics in Hollywood. Admin. Sci.Quart. 51(4) 560–689.

Strang, D., Y.-M. Kim. 2004. The diffusion and domestication ofmanagerial innovations: The spread of scientific management,quality circles, and TQM between the United States and Japan.P. T. S. Ackroyd, P. Tolbert, R. Batt, eds. The Oxford Handbookof Work and Organization. Oxford University Press, Oxford,UK, 177–199.

Strang, D., M. W. Macy. 2001. In search of excellence: Fads, successstories, and adaptive emulation. Amer. J. Sociol. 107(1) 147–182.

Stuart, T. E., O. Sorenson. 2003. Liquidity events and the geographicdistribution of entrepreneurial activity. Admin. Sci. Quart. 48(2)175–201.

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.


Sullivan, B. N., P. Haunschild, K. Page. 2007. Organizations nongratae? The impact of unethical corporate acts on interorganiza-tional networks. Organ. Sci. 18(1) 55–70.

Useem, M. 1979. Social organization of the American business eliteand participation of corporation directors in the governance ofAmerican institutions. Amer. Sociol. Rev. 44(4) 553–572.

Useem, M. 1984. The Inner Circle: Large Corporations and the Riseof Business Political Activity in the U.S. and U.K. Oxford Uni-versity Press, New York.

Verbeek, M., T. Nijman. 1992. Testing for selectivity bias in paneldata models. Internat. Econom. Rev. 33(3) 681–703.

Verbeek, M., T. Nijman. 1996. Incomplete panels and selectionbias. L. Mátyás, P. Sevestre, eds. The Econometrics of PanelData: A Handbook of the Theory with Applications. Kluwer,Dordrecht, The Netherlands, 449–490.

Vissa, B. 2011. A matching theory of entrepreneurs’ tie formationintentions and initiation of economic exchange. Acad. Manage-ment J. 54(1) 137–158.

Vissa, B. 2012. Agency in action: Entrepreneurs’ networking style andinitiation of economic exchange. Organ. Sci. 23(2) 492–510.

Winkelmann, R. 2003. Econometric Analysis of Count Data. Springer,New York.

Xiao, Z., A. S. Tsui. 2007. When brokers may not work: The culturalcontingency of social capital in Chinese high-tech firms. Admin.Sci. Quart. 52(1) 1–31.

Zaheer, A., G. Soda. 2009. The evolution of network structure:Where do structural holes come from? Admin. Sci. Quart. 54(1)1–31.

Zellner, A. 1962. An efficient method of estimating seemingly unre-lated regression equations and tests for aggregation bias. J. Amer.Statist. Assoc. 57(298) 348–368.

Zucker, L. G. 1988. Where do institutional patterns come from? Orga-nizations as actors in social systems. L. G. Zucker, ed. Insti-tutional Patterns and Organizations: Culture and Environment.Ballinger, Cambridge, MA, 23–49.

Lori Qingyuan Yue is an assistant professor at the USCMarshall School of Business. She received her Ph.D. in busi-ness administration from Columbia University. She studiesevolutions of market institutions and market structure.

INFORMS

holds

copyrightto

this

article

and

distrib

uted

this

copy

asa

courtesy

tothe

author(s).

Add

ition

alinform

ation,

includ

ingrig

htsan

dpe

rmission

policies,

isav

ailableat

http://journa

ls.in

form

s.org/.

asymmetric effects of fashions on the formation and

Documents