Dynamics of Dyads in Social Networks: Assortative, Relational, … · 2004. 5. 7. · SO36CH05-Uzzi...

SO36CH05-Uzzi ARI 2 June 2010 23:8

Dynamics of Dyads inSocial Networks: Assortative,Relational, and ProximityMechanismsMark T. Rivera,1 Sara B. Soderstrom,1

and Brian Uzzi1,21Department of Management and Organizations, Kellogg School of Management,Northwestern University, Evanston, Illinois 60208;email: [email protected], [email protected],[email protected] University Institute on Complex Systems and Network Science (NICO),Evanston, Illinois 60208-4057

Annu. Rev. Sociol. 2010. 36:91–115

First published online as a Review in Advance onApril 12, 2010

The Annual Review of Sociology is online atsoc.annualreviews.org

This article’s doi:10.1146/annurev.soc.34.040507.134743

Copyright c© 2010 by Annual Reviews.All rights reserved

0360-0572/10/0811-0091$20.00

Key Words

embeddedness, homophily, complexity, organizations, networkevolution

Abstract

Embeddedness in social networks is increasingly seen as a root cause ofhuman achievement, social stratification, and actor behavior. In this arti-cle, we review sociological research that examines the processes throughwhich dyadic ties form, persist, and dissolve. Three sociological mech-anisms are overviewed: assortative mechanisms that draw attention tothe role of actors’ attributes, relational mechanisms that emphasize theinfluence of existing relationships and network positions, and proximitymechanisms that focus on the social organization of interaction.

91

Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


INTRODUCTION

If research activity and impact are indicatorsof scientific value, the capital of social networkresearch has risen rapidly. Taking a census ofall papers published in the top two sociologyjournals as defined by ISI’s journal impact fac-tor, the American Journal of Sociology (AJS) andthe American Sociological Review (ASR), therehas been a 20-year increase in the share of pa-pers that list “networks” as a keyword. In 1980,1990, 2000, and 2005, the share of papers onnetworks was 1.2%, 2.2%, 7.8%, and 11.6%.Further, Figure 1 shows the increasing citationimpact of network papers. The ranking of the15 most cited papers in AJS and ASR by cita-tions per year indicates that 10 out of 30 are net-work papers. Network papers account for 5 ofthe top 20 papers published in the Administra-tive Science Quarterly (ASQ) (Figure 1 inset). Inphysics and communications, network researchhas achieved similar prominence (Monge& Contractor 2003; Barabási 2008, 2009;Newman 2008).

One impetus behind this explosion of workis the increasing number of stratification vari-ables across persons, firms, and institutions thatare now understood to be influenced by net-works. Disciplines that once focused on indi-vidual attributes to explain behavior now seeembeddedness in networks as a determinantof job attainment and advancement, creativity,obesity, mortality, neighborhood cohesion, po-litical mobilization, state formation, markets,prices, digital ties, and the competitiveness offirms and states (Granovetter 2005). As thisemergent body of research establishes the im-portance of networks as determinants of ac-tors’ behavior and outcomes, understanding theantecedents of social networks has become anincreasingly important area of inquiry. Concep-tualizations of networks as static substrata ofsocial interaction are giving way to an image ofnetworks as continuously evolving over time. Increating, maintaining, and dissolving social ties,people, groups, and organizations are continu-ously altering their networks and the networksof those around them.

This review examines journal articles thatexplain how social networks evolve over time.As an orienting principle, we cover researchthat examines network change at the level ofthe network dyad. We focus on research thatexplicates the existence, creation, persistence,and dissolution of social relationships amongsocial actors and the implications that theseprocesses have on the evolution of networksover time. For readers less acquainted with net-work research, we begin with a brief clarifi-cation of terminology (see also Wasserman &Faust 1992). A social network can be said toexist wherever distinct social actors (also callednodes) are connected by more or less persis-tent ties (or relationships). Both actors and tiescan be represented by a variety of different con-structs and can consequently span levels of anal-ysis ranging from friendships among individu-als to interorganizational ties between firms tocommercial trade between states. Ties connect-ing actors can be valued, indicating strongeror weaker connections (such as friendships andbest friendships), or binary, simply indicatingthe presence or absence of a tie. Ties can alsobe directed (one person chooses another as afriend, who may or may not choose him back)or undirected. Moreover, actors can simulta-neously be connected by several different rela-tions. Schoolchildren can be connected by re-lations of friendship, trust, or animosity, eachof which may inform different aspects of a stu-dent’s social world. In practice, most researchwithin the social sciences—and by extensionmost of the research in this review—has focusedon single-relation, binary networks.

A TYPOLOGY OF MECHANISMSOF CHANGE IN SOCIALNETWORKS

We identify three perspectives on networkchange—each deriving its origins from a uniquesociological lineage: (a) assortative perspectivesthat emphasize compatibilities and comple-mentarities between actors’ attributes; (b) rela-tional perspectives that draw on the structural-istic assumption that trust, information, and

92 Rivera · Soderstrom · Uzzi

Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


Figure 1The increasing role of network papers in sociology journals, 1970–present. Note: The plot shows thefifteenth most highly cited papers according to the number of citations gained per year in the AmericanJournal of Sociology and American Sociological Review (Administrative Science Quarterly, inset) from 1970 to thepresent (source: http://www.harzing.com). The papers in ranked order are Coleman (1988, 522/yr),Granovetter (1985, 464/yr), DiMaggio & Powell (1983, 356/yr), Granovetter (1973, 314/yr), Meyer &Rowan (1977, 225/yr), Massey (1990, 204/yr), Uzzi (1996, 143/yr), Hannan & Freeman (1977, 110/yr),Inglehart & Baker (2000, 104/yr), Hannan & Freeman (1984, 102/yr), Swidler (1986, 102/yr), Burt (2004,97/yr), Meyer et al. (1997, 96/yr), Powell et al. (2005, 94/yr), Gulati & Gargiulo (1999, 88/yr), Sewell(1992, 80/yr), Portes & Sensenbrenner (1993, 79/yr), Snow et al. (1986, 79/yr), Williamson (1981, 66/yr),Pickering (1993, 66/yr), McPherson et al. (2006, 63/yr), Sampson et al. (1999, 62/yr), Cohen & Felson(1979, 60/yr), McCarthy & Zald (1977, 58/yr), Orloff (1993, 56/yr), Barro & McCleary (2003, 59/yr),Korpi & Palme (1998, 52/yr), Budig & England (2001, 51/yr), Duncun et al. (1998, 49/yr), and Uzzi (1999,47/yr). The papers in ranked order for ASQ are Cohen & Levinthal (1990, 476/yr), Uzzi (1997, 221/yr),Powell et al. (1996, 209/yr), Hansen (1999, 178/yr), Henderson & Clark (1990, 166/yr), Williamson (1991,166/yr), Abrahamson & Fairchild (1999, 127/yr), Brown & Eisenhardt (1997, 119/yr), Tushman & Anderson(1986, 113/yr), Ahuja (2000, 106/yr), Burt (1997, 104/yr), Weick & Roberts (1993, 100/yr), Karasek (1979,99/yr), Edmondson (1999, 95/yr), and Weick (1976, 91/yr). Papers coded as network papers are in bold.

introductions are conferred through actors’positions in existing social networks; and(c) proximity perspectives that focus on theorganization of social interaction in time andspace. This rubric places mechanisms intobuckets that are categorically distinct yet in-timately interwoven in the evolution of realsocial networks. For example, people may pref-erentially create social networks that corre-spond to their political beliefs but also posi-tion themselves into associations and contextsthat are populated with people who have similar

political preferences (Huckfeldt & Sprague1987). Independently, each perspective has re-vealed empirical regularities that often general-ize across contexts, actors, and ties, giving rise tointeresting self-similar patterns across interper-sonal, interfirm, and even interstate research.However, owing to different theoretical foun-dations and explanatory variables, research ineach stream has tended to occur in isolation.We articulate and review progress in each area,but we also point future research toward a morerobust integration.

www.annualreviews.org • Dynamics of Dyads in Social Networks 93

Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


ASSORTATIVE MECHANISMS

Assortative mechanisms speculate that thecreation, persistence, and dissolution of socialrelationships are all outcomes that rely onthe compatibility and complementarity ofactors’ attributes. Thus, assortative hypothesesare generally pair-level associations betweenactors’ similarities or dissimilarities and theirconsequent propensity to form connections.Lazarsfeld & Merton’s (1954) analysis of friend-ship selection among adults coined the term ho-mophily to refer to “a tendency for friendshipsto form between those who are alike in somedesignated respect” (p. 23) and gave birth to arich tradition of research that has investigatedthe tendency toward assortativity in a varietyof social contexts and relations. A more recenttrend has begun to note that outright similarityin actors’ attributes may not characterize socialconnections and that in some instances actorsmay seek a balance between similarity on somedimensions and differentiation, or heterophily,on others. Recent studies have also extendedthe assortative perspective to explain the evo-lution of ties between firms as well as betweenindividuals. In each instance, the driver of net-work change is thought to be an interaction ofactors’ attributes that collectively compels themto create, maintain, or dissolve relationships.

Dynamics of Homophily

In a landmark review of more than 100 socio-logical studies, McPherson et al. (2001) foundthat people display a strong tendency towardhomophily—the greater the similarity betweentwo individuals the more likely they are to es-tablish a connection. Homophily’s potency asa social mechanism lies in its apparent univer-sality. Similarities in the human attributes ofpotential alters, such as age, gender, religion,ethnicity, values, intelligence, and education,appear to characterize the formation and disso-lution of connections as varied as online chats,best friendships, and marriage.

Homophily appears to strongly affect at-tachment because people expect a priori that

self-similar alters are more likely to acceptthem (Ely 1995, Ibarra 1995), to be trustwor-thy (Allen & Wilder 1979, Lincoln & Miller1979), and to hold beliefs that affirm theirown (Byrne & Wong 1962, Stein et al. 1965,Rokeach & Mezei 1966, Stein 1966, Allen &Wilder 1979), thereby mitigating the potentialconflicts, misunderstandings, and monitoringcosts that come with making connections. As aresult, these benefits appear to increase with thestrength of the relationship. Thus, homophilymore strongly affects attachment in intimate re-lations (Burgess & Wallin 1943, Blackwell &Lichter 2004, Qian & Lichter 2007) and closefriendships than in casual ties. For example, VanDuijn et al. (2003) found that gender homophilywas positively associated with the formation of“real friends” but not “friendly” or “neutral” re-lations. Similarly, Leenders (1997, p. 149) con-cluded that gender homophily was “profoundlypredominant in ‘best friend’ networks, but losesmuch of [its] importance when ‘friends’ are alsoconsidered.”

Although innumerable attributes may im-part a homophily bias, it is a rare attributethat has the strongest influence on attachment.Mehra et al. (1998, p. 443) asked 159 MBAstudents to identify other students they “consid-ered especially similar to themselves” or “con-sidered to be personal friends.” Controlling forthe baseline availability of similar others, mi-norities (t = −2.03, df = 19.5, p < 0.05) andwomen (t = 3.82, df = 157, p < 0.001) iden-tified more strongly with homophilous othersthan did members of the majority categories(whites and men) and were also more likelyto name self-similar others as personal friends.Mehra et al. reason that rare attributes are morelikely to be salient to individuals and conse-quently central to their self-identity: “[T]woAfrican Americans in a crowd of whites willtend to notice and identify with each other be-cause of their common race; however, when ina group of other African Americans, the sametwo people are unlikely to notice or identifywith each other” (Mehra et al. 1998, p. 442).In this way, rare attributes strongly impactattachment.


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


If people are more likely to create so-cial ties with self-similar others, there is alsoevidence that they are more likely to main-tain homophilous relationships than they areto maintain heterophilous ones. Lazarsfeld &Merton’s (1954) work on homophily positedthat heterophilous relationships would be morelikely to break apart as differences in opinionsand beliefs led to interpersonal conflict. Onearea where this hypothesis has been tested isin research on the dissolution of romantic re-lationships. Felmlee et al. (1990) found thatthe hazard rate for detachment of interracialromantic relationships was more than threetimes higher than that exhibited by raciallyhomophilous couples (antilog coef = 3.06,p < 0.10), and several studies within fam-ily relations have noted a higher divorcerate among racially heterophilous individuals(Heaton 2002, Bratter & King 2008). In thecontext of friendships, tendencies toward ho-mophily may similarly be reflected in the dis-solution of heterophilous ties—Mollica et al.(2003) examined the friendship network ofgraduate students and found that homophilousfriendships were significantly more likely topersist more than three and a half months thanwere heterophilous ones (see also Hallinan &Williams 1987). Here again, the relative rar-ity of the attribute moderated the effect of ho-mophily: more than 95% of Hispanic-Hispanicties persisted, compared with 65% for white-white (the majority race in that school) ties.

Importantly, homophily shapes attachmenteven in contexts where diversity is explicitlyvalued and encouraged. In the friendship net-work of 114 newly admitted MBA students in alarge U.S. university, Mollica et al. (2003) foundthat six weeks into a two-year program studentshad a high degree of racial homophily in theirpersonal friendship networks—a bias that per-sisted more than 14 weeks despite the promo-tion of diversity by school administrators andthe assignment of students into heterogeneousclasses and teams. Ingram & Morris (2007) usedelectronic global positioning devices to monitorthe relationships during an MBA mixer, whichwas aimed at helping MBA students form new

ties with students with different characteristicsand backgrounds. In spite of attendees’ declara-tions of wanting to meet new and different peo-ple, they found that students were more likelyto engage with a group of people in which atleast one other person in the group was of thesame race or of the same approximate physi-cal attractiveness. Similarly, Ruef et al. (2003)found that the composition of new businessstart-ups is driven by similarity in gender, eth-nicity, and occupation rather than by functionaldiversification.

The role of digitally mediated relationshipsis an important new area of research on assorta-tivity, homophily, and network building. Theserelationships form or are maintained throughinstant messaging, blogs, online-communitymembership, or email, rather than face-to-face contact, which can potentially mask visi-ble characteristics of potential contacts (Adamic& Adar 2003, Eagle et al. 2007). One recentstudy, impressive for its ability to record theempirical patterns of vast numbers of people ona planetary scale, was conducted by Leskovec& Horvitz (2007). They examined the globalcommunication network of Microsoft Mes-senger instant-message communications andfound that, among 30 billion online conversa-tions between approximately 240 million indi-viduals, people tended to communicate morewith others with similar ages, languages, andlocations. A study on the determinants of mes-saging on an online dating Web site foundthat similarity in daters’ physical attractiveness,smoking habits, education level, religion, andrace all predicted the frequency of attemptsto initiate a romantic relationship (Fiore &Donath 2005).

Digitally mediated contexts of social interac-tion such as Web sites and online communitiesmay also impart a homophily bias in people’ssocial choices by selectively exposing people toself-similar others. Participation in online com-munities requires regular access to a personalcomputer—a prerequisite that biases commu-nities toward higher-income strata. Differencesin the demographic makeup of online com-munities may similarly affect the availability of


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


contacts in different groups. Hargittai (2007)found that Hispanic students were signifi-cantly less likely to use the online networkingsite Facebook and much more likely to useMySpace, whereas the opposite was truefor white and Asian students. This suggeststhat online networking sites might furthercontribute to homophily by organizing per-sons into groups of self-similar individuals.Surprisingly, in cyberspace, where individualshave a greater sense of anonymity, homophilyappears to shape social connections.

Dynamics of Heterophily

Although persons appear to form ties accord-ing to the homophily principle, it is also self-evident that diverse network relationships exist:“[t]he simple facts of the matter are that whensociety confronts difficult problems—puttingpeople on the moon, curing diseases, design-ing new products, crafting changes in the taxcode—we create teams of diverse people” (Page2007, p. 322). In some cases, diversity might bemandated—boards of directors of large compa-nies must represent dissimilar functional spe-cializations (e.g., law, science, or nonprofit) orconnections to diverse financial or political re-sources (Westphal & Milton 2000, Mizruchi2004). However, heterophilous ties are alsoformed voluntarily. Research on collaborationnetworks is one area that explores the increas-ing role of heterophily. Collaboration networksare ubiquitous and occur whenever people worktogether in teams—directors on a board, actorson a cast, writers on a movie, and so forth. Inthis work, the diversity of social ties can be in-ferred from coauthorship on the same articles(Newman 2001d, Moody 2004, Guimera et al.2005, Wuchty et al. 2007, Jones et al. 2008),codirectorships on the same board (Davis et al.2003, Westphal & Stern 2007), coperformancein the same artistic production (Watts 1999,Uzzi & Spiro 2005, Uzzi 2008), and so on.

Research on teams in which dyads are nestedwithin larger group structures suggests thatpeople are likely to collaborate with otherswho posses qualities, skills, and know-how that

are complementary to their own and relevantto solving a particular problem or objective.With regard to teams in science, Moody (2004,p. 217) remarked, “in high-growth, fast chang-ing specialties, we would expect to see morecoauthorship because it is easier to bring ina new author than it is to learn new materialoneself.” Indeed, Moody (2004) found that re-searchers who publish quantitative work, whichrequires specialization in theory and methodol-ogy, are more than five times more likely tohave coauthored a paper than someone whohas not published quantitative work. Examin-ing the formation of task-related ties—askingfor assistance or support from a colleague—inthree different organizations, Casciaro & Lobo(2008) found that a key fact of organizationallife is that people seek out others whom theybelieve to have valuable and complementarytask-related skills. Page (2007) found that inscience the trend toward heterophilous ties hasbeen increasing over time. He reported that thefirst ten Nobel Prizes in Physics were shared by14 individuals, whereas the last ten have beenshared by 27, a twofold increase in collabora-tion. This pattern has also been found in chem-istry and sociology (Moody 2004) and in pre-publication teams that form to submit grants(Cummings & Kiesler 2005). In collaborationsof creative and economic production, connec-tions can be valuable or sought after preciselybecause they connect people with differentand complementarity attributes, qualities, andcapabilities (Lin et al. 1981, Lin 1999).

Alliance research—in which actors areorganizations and ties are formal or informalcollaborative arrangements between firms—has perhaps examined the role of heterophilyin network formation most fully. It has becomeconventional wisdom that network ties areformed with dissimilar rather than similarcollaborators. Powell et al. (2005) examinedthe determinants of attachment among uni-versities, venture capital firms, public researchorganizations, and large pharmaceutical firmsduring the first ten years of the biotechnologyfield. They note that in this field “[n]o singleorganization [had] been able to internally


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


master and control all the competencies re-quired to develop a new medicine,” which mayhave led organizations to form collaborativerelationships with firms that were distinct fromthem in activity specialization. Research on in-vestment banks’ tie formation (Eccles & Crane1988; Podolny 1993, 1994) has found that di-versity is actively pursued. Chung et al. (2000)concluded that banks preferentially attached toother banks that (a) specialized in selling stockto different types of investors than themselves,(b) were located in different geographic areas,and (c) specialized in stock offerings of compa-nies in different industries. Similar preferencesfor heterophilous partners in interorganiza-tional relations have been found for industryniche (Gulati 1995, Chung et al. 2000), strate-gic grouping (Nohria & Garcia-pont 1991),venture syndicates (Sorenson & Stuart 2008),and organizational size (Shipilov et al. 2006).Compared with interpersonal ties, interorgani-zational links may be more prone to heterophilybecause the trust orientation of homophilybetween individuals may be functionally sub-stituted by contracts between organizations,and the desire for expressive benefits found ininterpersonal networks may be absent in firm-to-firm relationships (Granovetter 1985). Onestudy of task-related ties found that organiza-tional members seek out contacts they believeto have complementary and relevant skills, butonly if they perceive those people as being en-joyable to work with (Casciaro & Lobo 2008).

One indication of the instrumental natureof heterogeneous relationships is that collabo-rative ties between actors with complementaryattributes are often short term and oriented to-ward the completion of a project or goal. “Con-nections are often forged with a specific goal inmind, such as taking a company public or sellingand distributing a new medicine. Once the taskis completed, the relationship is ended and suc-cessful collaborators depart gracefully” (Powellet al. 2005, p. 1138).

The increasing dominance of diverse teamsand repeated heterophilous ties suggest thatmore research may be needed on whatBlau (1974) called multiform heterogeneity.

Multiform heterogeneity means that networkties may be between individuals who are simul-taneously similar and different, sharing simi-larities on some dimensions and differences onothers. It may ultimately be an oversimplifica-tion to refer to a relationship as homophilousor heterophilous, as few individuals do not dif-fer in at least some dimensions and match in atleast a few others (Blau 1974, p. 622). An impor-tant question is whether attachment is encour-aged by homogeneity across numerous dimen-sions (McPherson & Ranger-Moore 1991) orby a balance between similarity on some, such asrace and gender, but differences on others, suchas skill, knowledge, or connections. At leastsome early research supports the latter position.Casciaro & Lobo’s (2008) field research sug-gests that people preferentially collaborate withothers who have complementary specializationsbut similar demographic traits that facilitatecommunication and trust. Organizations, too,appear to seek a balance between similarityon some dimensions, such as status (Podolny1993), and dissimilarity on others, such as in-dustry niche (Gulati 1995, Chung et al. 2000),strategic grouping (Nohria & Garcia-Pont1991), geographic location (Gulati & Gargiulo1999, Chung et al. 2000), size (Shipilov et al.2006), or embedded versus arm’s-length ties(Lazerson 1995; Uzzi 1996, 1997). Future re-search that evaluates the choices and balance ofhomophily and heterophily is needed to under-stand the conditions under which one processor the other dominates and the balance that isstruck in different social settings.

RELATIONAL MECHANISMS

Drawing on a different theoretical heritage, asecond perspective places importance on directand indirect connections linking individuals.This perspective is sometimes referred to asa “within-the-network” approach, as the focalpredictor of network change is hypothesizedto be the shape and structure of the network ina prior time period (Stuart & Sorenson 2007,p. 220). Arising out of Simmel’s notion of thetriad, the classic relational hypothesis is of


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


network closure—the proposition that actorsseparated by one intermediary are the mostlikely to become connected in subsequenttime periods (Simmel 1908 [1950], Davis et al.1971). Recent work has expanded relationalperspectives to emphasize the role of priordirect ties (Baker 1990, Gulati 1995, Guimeraet al. 2005), indirect connections throughnumerous third parties (Newman 2001a,Kossinets & Watts 2006), and prior networkdegree centrality (Newman 2002, Uzzi 2008).Each mechanism originates from the struc-turalistic assumption that the selection ofrelationships, the maintenance of existing ones,and the dissolution of old ones are conditionedby trust, information, and opportunities forinteraction that are structured by the network.

Dynamics of Reciprocity

In cases in which ties are meaningfullydirectional—one actor chooses another whomay or may not also choose them—an impor-tant relational mechanism is the tendency to-ward reciprocity. That is, the bestowing of afriendship from i to j tends to be quickly fol-lowed by a reciprocal offering of a friendshipfrom j to i. Consequently, an important predic-tor of tie formation from one individual, i, toanother, j, is whether or not j previously had adirected tie with i. Reciprocity appears to occurfor several reasons. In interpersonal relation-ships, one-way ties are likely to become recip-rocal because people tend to like others wholike them (Newcomb 1956, Backman & Secord1959, Sprecher 1998, Montoya & Insko 2008).Similarly, reciprocation relative to a first ad-vance of friendship decreases the chance of be-ing rebuffed (Goffman 1963).

Early evidence for the temporal develop-ment of reciprocity in friendship networkscomes from grade-school data (Sørensen& Hallinan 1977, Eder & Hallinan 1978,Hallinan 1978, Tuma & Hallinan 1979, Runger& Wasserman 1980, Hallinan & Williams1989, Doreian et al. 1996). Hallinan (1978)measured the friendship networks of fourth andsixth grade students at five elementary schools

in the Midwestern United States at seven pointsin time. She found that the most likely newfriendship to be created was the reciprocationof an extended friendship from another student.Using the same data, Runger & Wasserman(1980) showed that students were betweenthree and nine times more likely to create anew friendship with another student in thefuture if that student had already selected them.Analyzing longitudinal networks of friendshipformation among 17 formerly unacquaintedcollege students, Doreian et al. (1996) foundthat one of fastest structural tendencies in adeveloping network was a bias toward recipro-cated relationships. In another analysis on new-comer networks, Mollica et al. (2003) measuredthe friendship network of 114 MBA studentssix weeks after matriculation into a graduateprogram and three and a half months later.They found that just six weeks after studentshad met, the level of reciprocity in their friend-ship network was much higher than chancelevels; approximately 37% of friendships werereciprocated. The reciprocation of extendedfriendships continued over the following eightweeks such that 52% of ties were reciprocatedby the end of the observation period.

However, reciprocity in social networksdoes not occur solely because people recipro-cate offered relationships. It also occurs as ac-tors withdraw unreciprocated ones. Hallinan(1978) reported an interesting pattern in theevolution of friendship networks. She foundthat while one-way friendships were signifi-cantly more likely than null relationships toblossom into reciprocated ties, they also had ahigh base rate of dissolution. One-way friend-ships were approximately two times more likelyto entirely dissolve than they were to be re-ciprocated. This suggests that although a one-way tie can make a reciprocal tie more likelyto arise relative to a nonexistent relationship,one-way friendships are fundamentally short-lived if they fail to become reciprocal. Severalfactors have been identified as potential causesof the fragility of one-way ties. If a one-waytie is not reciprocated relatively quickly, in-dependent of its potential value, it can evoke


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


embarrassment (Goffman 1963), distress, ormutual discomfiture (Heider 1958). Moreover,in an influential paper on the development ofstatus hierarchies, Gould (2002, p. 1151) com-ments “[s]omeone who pays less attention toyou than you pay to her implicitly asserts thatshe is superior to you in status. If you do notrespond by withdrawing your attention, youhave implicitly agreed.” In this way, extendinga tie to someone who does not reciprocate maysignify deference to another person’s higherstatus. Friendships in Hallinan’s schools wereconsistent with this hypothesis. Students wereroughly half as likely to withdraw a reciprocatedfriendship as they were to withdraw an unre-ciprocated one (Runger & Wasserman 1980).Mollica et al.’s (2003) MBA study also showedthat students were significantly more likely tomaintain reciprocated friendships more thanthree and a half months than they were to main-tain unreciprocated ones.

Thus, reciprocity is associated with stabil-ity. Over time, unreciprocated ties are inher-ently unstable, either becoming reciprocated orwithdrawn. A relatively open area of inquiry isto understand why some ties are reciprocatedand others are not. For example, are high-statusindividuals the recipients of many one-way ties,only a few of which they reciprocate? Are ho-mophilous ties less or more likely to be recip-rocated given their frequency? Do certain con-texts promote reciprocation while others arecharacterized by many failed one-way ties?

Dynamics of Repetition

In relationships that can be ongoing in time, animportant relational property is the frequencyof repetition. Repeated ties figure prominentlyin the networks literature in many types of ties,both social and economic. In addition to be-ing a measure of the strength of a relationship,repetition is also taken as an indicator of trust(Gulati & Gargiulo 1999, Uzzi & Lancaster2004) and social embeddedness in economicexchange (Uzzi 1996). The frequency of a re-peated tie has also been correlated with al-truism, joint problem solving, and information

exchange (Uzzi 1996, 1997); deals in exchanges(DiMaggio & Louch 1998, Uzzi 1999); prices(Uzzi & Lancaster 2004); and collusion (Ingram& Roberts 2000).

Uzzi & Spiro’s (2005) work on the ca-reer histories of about 5000 artists who madeBroadway musicals between 1900 and 1995and the evolution of their network providesevidence of the tendency toward repeat re-lationships over time. Uzzi & Spiro (2005)showed that over time collaborators show asignificant propensity to work with personswith whom they have worked in the past.Guimera et al. (2005) used agent-based simu-lations to show that the network structure offour scientific disciplines—astronomy, ecology,economic, and psychology—exhibited similarlyhigh rates of repeated ties. Disaggregating thecoauthorship networks in these fields over thetop seven journals in each field, the authors wereable to compare the rates of repeated ties acrossthese diverse networks. They showed that therate of repeated ties varied from about 50%to approximately 99%, which suggests that re-peated ties are common and, for some networks,are the norm.

Even in markets where relationships arefrequently thought of as being arm’s-length,one-shot engagements, we find relatively highrates of repeated, reciprocal ties. Baker (1990)looked at the relationships between large cor-porations and their investment banks with aneye to how many banks they maintain relation-ships with and the repetition of dyadic financialrelationships. He showed that corporationstend to have ongoing relationships with a coreset of financial partners, adding others on anad-hoc basis for special deals. This condition isalso found between investment banks (Podolny1994) and large law firms and their clientcorporations (Uzzi & Lancaster 2004). Uzzi(1996) showed that in the competitive marketsof the apparel industry, many contractorsconcentrate their relationships with certainmanufacturers—establishing ongoing, repeatrelationships—and those contractors who failto establish these embedded ties tend to fail atgreater rates, suggesting that the orientation


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


toward repetition is important to the function-ing of the network and the actors embeddedwithin it. Thus, in networks of economicexchange and friendship, the ties that exist atone point in time are likely to be repeated inthe future.

Dynamics of Clustering and Closure

In comparing networks as diverse as biolog-ical food webs to the World Wide Web tointerlocking corporate directorates, Newman& Park (2003) found that a characteristic thatdistinguishes social networks from biologicaland technological networks is clustering, a con-dition originally proposed by Simmel (1908[1950]). Clustering means that social networkstend to have a high density of closed triads,or, colloquially, people tend to become friendswith the friends of their friends. In their respec-tive networks, corporate board members (Daviset al. 2003), Hollywood movie actors (Watts1999), Broadway musical artists (Uzzi & Spiro2005), inventors (Fleming et al. 2007), scientists(Newman 2001b), lawyers (Lazega & Pattison1999), and organizations, tied together throughalliances (Kogut & Walker 2001, Baum et al.2003), all tend to be connected to other actorswho are themselves connected.

Several rationales have been proposed toexplain the almost universal observation thatsocial networks exhibit nontrivial clustering.Granovetter (1973, p. 1362) suggested that peo-ple are inclined to create ties with the friends oftheir friends (or the business associates of theirbusiness associates, etc.) because people whospend time with a common third are likely toincidentally encounter each other, even if theyare not explicitly introduced (cf. Feld 1997).Others have noted that sharing a third partyprovides information about potential connec-tions through referrals and gossip and that thisinformation eases the process of developing anew tie by decreasing the uncertainty and riskof a new connection (Burt & Knez 1995). Inthis respect, proximity in a social network iscorrelated with increased information about thetrustworthiness of a potential contact. “Better

than the statement that someone is known to bereliable is information from a trusted informantthat he has dealt with that individual and foundhim so,” remarked Granovetter (1985, p. 490).Finally, research on social capital has notedthat embeddedness among third parties can alsopromote collectively oriented norms, therebyminimizing opportunism (Coleman 1990;Granovetter 1985; Uzzi 1997, p. 48) and mit-igating conflict (Simmel 1908 [1950], p. 135;Portes & Sensenbrenner 1993; Krackhardt &Handcock 2008).

A measurable implication of these processeson network dynamics is that open triads tendto close over time. In one of the first empir-ical studies on the closure process, Hammer(1980) examined longitudinal data of interper-sonal interaction in three settings: a neighbor-hood church, a coffeehouse, and a textile fac-tory. In each network, ties between individualswere measured by observing interactions overa period of time (ranging from one month forthe coffeehouse and church to six months forthe textile plant). A tie was said to exist if peoplewere observed interacting (usually conversing)with one another. Providing early support forthe closure process, Hammer found that in allthree settings pairs of unacquainted individu-als were significantly more likely to begin in-teracting if they had a common acquaintance.This early study alluded to the generality of theclosure mechanism; Hammer’s church and cof-feehouse were located in Manhattan, the textilefactory in Africa.

The closure hypothesis of social attachmenthas been further corroborated by recent studiesthat have vastly increased the size and scope ofresearch through the use of digital communi-cation data. Kossinets & Watts (2006) trackedthe emails of 45,553 students, faculty, and staffat a large research university over an academicyear with the goal of identifying factors thatpredicted communication between individuals.One of the predominant influences on social at-tachment that they uncovered was Simmel’s tri-adic closure. Having a mutual contact dramat-ically increased the tendency for two formerlyunconnected students to begin communicating


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


via email: Among students who did not share acommon class—and were therefore less likelyto incidentally meet—having a mutual contactincreased the probability of communication bya staggering 140 times. Even students who sat ina class together were three times more likely tobegin communicating with each other via emailif they both emailed a common third individ-ual. Sharing common third parties has also beenfound to influence the formation of new col-laborative ties between researchers. Newman(2001d) examined the structure of scientific col-laboration networks of more than 1.6 millionresearchers in biology, physics, medicine, andcomputer science between 1995 and 1999. Twoscientists were considered to have a tie if theyhad coauthored a paper together. He found asimilar result—across each discipline, two sci-entists had a 30% or greater probability of col-laborating if they had both collaborated with acommon third researcher.

The effect of sharing mutual acquaintanceson attachment appears to be additive—eachadditional mutual acquaintance shared by anunconnected dyad additionally increases thelikelihood that they will become connected.Hammer (1980) found that in each of heranalyzed settings having two shared acquain-tances increased the probability that a pair ofindividuals would create a tie above and be-yond the probability observed for people whoshared only one. Newman (2001a) providescompelling evidence for additivity in closure.He examined the scientific collaborations cat-alogued in the National Institute of Health’sMedline R© database, a network containing morethan 1.6 million researchers in biology andmedicine, and found that two researchers whoshared one prior collaborator were 40 timesmore likely to collaborate than were researcherswho did not share any collaborators. Sharingtwo, three, four, and five prior collaborators in-creased the probability of collaboration by 100,140, 170, and 200 times, respectively, than theprobability experienced by researchers with noprior common collaborators. Thus, scientists’current relationships powerfully affected theirselection of future collaborators.

Clustering occurs even between individualswho are separated by more than a single inter-mediary. In a social network, the distance fromone actor, i, to another, j, is denoted dij and rep-resents the length of the shortest path betweeni and j.1 Individuals who are not connectedthemselves, but who are connected to a com-mon third actor, are separated by a distance of2, people separated by two acquaintances havea distance of 3, and so on. Research has foundthat shorter network distances are correlatedwith increased tendencies toward connections.Early evidence can again be found in Hammer’s(1980) analysis of interaction networks. Shefound that the network distance between twopeople at one point in time was negatively re-lated with the probability of a tie being createdbetween them in a subsequent time period.Patrons of the coffeehouse, employees in theAfrican textile factory, and churchgoers whoshared a contact (dij = 2) were significantlymore likely to begin conversing during theobservation period than were individuals sepa-rated by two intermediaries (dij = 3). Kossinets& Watts’s (2006) study on the creation of emailcommunication ties similarly found that net-work distance at one point in time is negativelyrelated to the probability of attachment: Amongstudents who were not currently in a class to-gether, dyads separated by two intermediaries(dij = 3) were approximately 30 times lesslikely to initiate a new tie than were individualsseparated by only one (dij = 2). Given evidencethat on average people are regularly linked by ashort “6 degrees of separation” (Milgram 1967,Leskovec & Horvitz 2007), it is likely thatthe effect of network distance on attachmentdecays rapidly as people become separated bymore and more intermediaries. Thus, mutualacquaintances and proximity within a socialnetwork have notable effects on social attach-ment: In many networks, new ties are morelikely to be formed between people who are not

1This definition of distance is made more nuanced by consid-erations of valued ties, or of the directionality of ties, whichwe do not here include (see Wasserman & Faust 1992, chap-ter 4.3–4.5, for more on these considerations).


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


only close in a network, but are close throughnumerous independent intermediaries.

Ties that connect people who share severalcommon alters are said to be structurally em-bedded and are not only more likely to arisein a social network, but are also more likely towithstand the test of time. In one study, Martin& Yeung (2006) looked at the relationships be-tween members of 60 voluntary communes tosee what factors predicted the survival of inter-personal relationships between 1974 and 1986.Even though all ties examined were betweenpeople who had once lived together and conse-quently had relatively strong social ties, peoplewith mutual acquaintances were more likely tostay in touch for more than 12 years and ex-hibited more frequent interaction. Burt (2000)used survival analysis to examine the hazardrate of business relationship decay among 345bankers during a four-year period. At the end ofeach year, bankers were asked to identify indi-viduals with whom they had “frequent and sub-stantial business contact in the preceding year.”Bankers experienced a high degree of turnoverin their year-to-year business networks. On av-erage, less than 35% of the ties they had in agiven year were still extant the following year.Even in this rapidly evolving context of strongand weak ties, Burt found that ties that were em-bedded in triads were slower to decay than werenonembedded ones (p < 0.001). An apparentparadox is that ties bridging distant parts of so-cial networks—connections that are dispropor-tionately important for global connectivity, in-formation flow, and human achievement—arenot only the least likely to be created in the firstplace, but are also the most likely to decay overtime.

It might therefore be expected that socialnetworks close over time, continuously becom-ing more clustered. Yet networks can both openand close. Using a complete population data setof about 5000 creative artists who cocreated thehundreds of Broadway musicals that appearedbetween 1877 and 1995, Uzzi & Spiro (2005)and Uzzi (2008) studied the changing proper-ties of small world networks with a focus on thetemporal formation and distribution of closed

triads, or clustering, in the network. Using theclustering coefficient ratio, a variable that mea-sures the ratio of actual clustering to that ex-pected by the random creation of links in anetwork of the same size, they found that thisnetwork increased and decreased in clusteringover time in a punctuated way. Clustering de-creased after a shock perturbed the network.For example, the clustering coefficient of thenetwork sometimes dropped as much as 30%after large shocks such as the Great Depression,the rise of movies and television, and the AIDSepidemic. This suggests that large upheavals ina network or shocks that change the environ-mental or market conditions within which anetwork is embedded may lead to the break-down of clusters and a consequent decrease inthe clustering coefficient ratio. Perhaps shocksweaken common ties to the same third partyor make interaction with a common third partyless predictable (Uzzi & Spiro 2005).

Not all social networks exhibit clustering.In fact some social networks exhibit the oppo-site tendency. Networks that have few, if any,closed cycles may exist where closure is nor-matively prohibited or logistically impossible.Kinship networks where ties represent ge-nealogical descent are an obvious example oftaboo norms against closure. More sociologi-cally interesting examples can be found in ro-mantic dating networks, which have implica-tions for the spread of ideas, infectious diseases,and new practices. Bearman et al. (2004) ex-amined the romantic and sexual networks ofmore than 800 high school students in a midsizeMidwestern town. Conspicuously absent fromthe network was the existence of many closedtriads or cycles of length 4. Bearman et al.speculated that such short cycles do not fre-quently occur because of a social prohibitionagainst dating the former lover of your cur-rent lover’s ex-lover, and because the closureof triads would require a high rate of same-sex unions. Consequently, even in Hollywood,where revolving door romances among thestars could increase the likelihood of closure,closed triads, or cycles of length 4, are unlikely.Independent of their personal attractiveness


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


(or not), Billy Bob Thornton is unlikely to dateJennifer Aniston because their exes, AngelinaJolie and Brad Pitt, became a couple first.

Dynamics of Degree

In the 1920s and 1930s, Moreno (1934)sketched the social networks of prisoners atSing Sing prison and schoolchildren at theHudson School for girls in upstate New York,thus giving birth to the sociogram (Moreno1934). Reflecting on these first sociograms,Moreno and a collaborator, Jennings (1943),discovered that some individuals tended to beat the center of their networks (stars) while oth-ers remained on the periphery (isolates). Thisrealization gave rise to the concept of networkcentrality. It has since developed into a myr-iad of related measures that capture differentaspects of an actor’s connectedness and status,including betweenness, closeness, degree, pres-tige, fitness, attractiveness, rank, and broker-age (Borgatti & Everett 2006). Most work findsa positive association between an actor beingin a central position in the network and thatactor’s goal achievement, including creativity(Burt 2004, Uzzi & Spiro 2005), job attainment(Granovetter 1973), professional advancement(Brass 1984; Burt 1992, 2004), political in-fluence (Fernandez & Gould 1994), prestige(Burris 2004), rate of alliance formation(Powell et al. 2005), and dissolution (Saavedraet al. 2008).

In addition to shaping outcomes, centralityhas important effects on the evolution of actors’social networks. Degree centrality—the num-ber of ties that an actor possesses—has receivedparticular attention (Zeleny 1940, Katz 1953,Price 1976, Freeman 1979, Friedkin 1991,Marsden 2002). In many social networks, re-searchers have found that most actors have onlya few ties, while a small number have extraordi-narily many (Borgatti & Everett 1999, Albert &Barabási 2002). An analysis of the online friend-ships of 4.2 million people on Facebook foundthat a few individuals had more than 10,000friends, more than 55 times as many as the av-erage user’s 180 (Golder et al. 2006). Liljeros

et al. (2001) found that degree centrality is sim-ilarly skewed in sexual contact networks, wheresome superconnecter actors acquire as manyas 1000 partners. Davis et al. (2003) examinedthe network of corporate board memberships inthe United States between 1982 and 2001 andfound that the average director had ties to 16other directors, but a few were connected to asmany as 100. Similar patterns exist in the rapmusic industry cosinger network (Smith 2006),the Hollywood movie coappearance network(Barabási & Albert 1999), and numerous coau-thorship networks in academia (Price 1965;Newman 2001c,d; Jeong et al. 2003; cf. Moody2004; Guimera et al. 2005; Wuchty et al. 2007).

One reason social networks develop such ahigh variance in actors’ degree centrality is thatthe number of ties an actor possesses (degree)affects processes of attachment. One prominentmodel, termed preferential attachment, positsthat the rate at which actors acquire new ties isa function of the number of ties they currentlyhave (Price 1976, Barabási & Albert 1999).Implications of preferential attachment echoMerton’s (1968) “Matthew Effect.” Social con-nections tend to accrue to those who alreadyhave them, the consequence of which is thatsmall differences in actor centrality compoundover time into a distinct cumulative advantage.

Preferential attachment purportedly occursbecause actors looking for new connectionsuse an actor’s degree as a proxy for hisor her fitness. In this way, the deceptivelysimple process has significant ramificationsfor acquiring and reinforcing centralitybecause it suggests that central actors ben-efit from a rich-get-richer phenomenon ofcumulative advantage. In a paper that ex-amined this process across varied types ofsocial networks, Jeong et al. (2003) explored(a) the coauthorship network in the neuro-science field, (b) the citation network of paperspublished in Physical Review Letters, and (c) theHollywood cocast actor network from 1892–1999. They found that neuroscientists, Holly-wood actors, and physicists who had relativelymore links than their respective peers tendedto attach to other actors at a higher rate than


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


did their less connected counterparts. Newman(2001a) similarly found that researchers inphysics and biology who already had a largenumber of collaborators were more likely toaccumulate new collaborators in the future.In a first empirical test of the preferential at-tachment model that used a specially designedmethodology for testing preferential attach-ment (Clauset et al. 2009), Uzzi (2008) showedthat the creative artists who made Broadwaymusicals acquired new ties in proportion to thenumber of ties they already possessed.

The relationship between centrality andtie accumulation is likely to be diminished byfactors that moderate an actor’s resources fortie acquisition or maintenance such as time,money, or even natural aging. In many con-texts, highly connected actors may accumulatenew ties up to a carrying capacity, after whichthey cease to acquire more (Amaral et al. 2000,Guimera et al. 2005, Uzzi & Spiro 2005).As a result, networks of relationships thatrequire significant investments of time, suchas friendship networks (Amaral et al. 2000,Newman et al. 2001), interorganizationalcollaboration networks (Kogut et al. 2007),and corporate board membership networks(Newman et al. 2002, Davis et al. 2003,Conyon & Muldoon 2006), have less extremedifferences between the most and least centralactors than do networks of relationships whereincreasing one’s degree requires only a smallcost, or one-time cost, such as Facebook friendsor networks of sexual contacts (Liljeros et al.2001). Research on interfirm networks hasfound that the influence of degree centralityon attachment can also be minimized by otherfactors that are stronger proxies for fitness. Forexample, Powell et al.’s (2005) analysis of the in-terorganizational collaboration networks foundthat in many cases firms exhibited a preferencefor novelty over preferential attachment—withwell-established, highly connected firmscollaborating with younger, less-connectedorganizations. Kogut et al. (2007) examined theemergence of the U.S. venture capital syndica-tion network and found that new entrant firmssometimes entered the network with more

cutting-edge knowledge and consequentlysurpassed some incumbent firms in the numberof partners they accumulated. Whether apreference for novelty may similarly enablesome individuals—for example, rising stars—to surpass incumbents in the extent of theirconnectedness is a relatively unexplored area.

Not only are high-degree actors more likelyto form attachments than noncentral actors,but they are also more likely to form attach-ments with each other—popular actors attachto other popular actors, whereas lower-degreeactors tend to attach to other low-degree actors.In a study on the bases of community powerin a Midwestern U.S. town, Perrucci & Pilisuk(1970) identified the individuals who had themost ties to different organizations and con-stituencies. They encountered an interestingresult: Among eight of the elite individuals whohad ties to multiple organizations within thecommunity of 50,000 people, 53 of 56 possiblesocial ties existed. In other words, each of theeight elite individuals named nearly all of theother eight as people they saw socially. Thisfinding supported assortativity among elite ac-tors, but also strongly suggested that the mostwell-connected individuals in the communitypreferred to attach to other well-connectedindividuals. Recently, Conyon & Muldoon(2006) collected data on corporate director-ships at 1733 unique firms in the United States,Germany, and the United Kingdom. Despitethe different regulatory environments charac-terizing each locale, they found that in eachcase directors who sit on a lot of boards tendto do so with other high-degree directors. In aprovocative paper, Newman (2002) comparedfive different social networks—including coau-thorship networks and a network of corporatedirectors—to six biological and technologicalnetworks and found that only social networksexhibited a tendency for the most connectedactors to connect among themselves. In hisstudy of the Broadway musical industry, Uzzi(2008) found that this tendency occurs at the in-terteam level as well (where teams are the nodesand ties between teams are made by a sharedteammate)—production teams that are highly


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


central are more likely to connect to othercentral teams.

A literature is emerging on degree central-ity and detachment. Baker et al. (1998) studiedthe longevity and detachment of large compa-nies’ links with their advertising agencies circa1985–1987. They argued that conditions affect-ing breakup were due to competitive, market,and institutional forces that could be measuredas network variables. Their work indicated thata company’s degree, as measured by the num-ber of ad agencies it employed, was positivelyrelated to the probability of breakup of the firm-agency relationship. Their rationale for prefer-ential detachment is that firms that employed alarge number of advertising agencies had an in-creased variety and availability of alternatives toany one focal agency. Thus, echoing processesof attachment, detachment can be affected byactors’ degree. However, Saavedra et al. (2008)found that detachment is not the reverse pro-cess of attachment. Using 20 years of data onthe exchanges between manufacturers and con-tractors in the New York City garment indus-try, they found that even though firms formties through preferential attachment, the linksbetween the highest- and lowest-degree nodeshave the lowest probability of breakup. Theirexplanation is that in the preferential attach-ment model a node’s degree is a stand-in forits true fitness because direct performance dataare costly to gather before the relationship ismade. By contrast, during breakup there is al-ready a link in place, which enables the qualityof the relationship to be judged based on first-hand accounts and a history of past interactions.Saavedra et al. (2008, p. 16470) conclude thata preference to maintain connections betweenhigh- and low-degree actors results in “a re-markably robust topology,” as these connec-tions preserve the global connectivity of a net-work even when undergoing severe decline.

PROXIMITY MECHANISMS

Whereas assortative mechanisms are derivedfrom information on actors’ attributes, and rela-tional mechanisms originate from information

about actors’ relationships, proximity mecha-nisms put the source of network change at thelevel of actors’ social and cultural environments.The most elementary proximity hypothesisis that interaction increases with geographic/physical propinquity. Being proximate isthought to encourage chance encounters andopportunities for interaction, which can leadto the formation of new relationships and themaintenance of existing ones. More complexconjectures can be drawn from the heteroge-neous organization of social activities into focithat bring actors together and occasion the for-mation of positive sentiment, opportunities forinteraction, shared goals, and cultural norms ofsociability.

Dynamics of Proximity and Social Foci

“Cupid may have wings, but apparently theyare not adapted for long flights,” remarkedBossard (1932, p. 222) after examining the phys-ical distances between 10,000 applicants formarriage licenses in Philadelphia. More thanone in six couples filing for a marriage li-cense had previously lived within a single cityblock of each other, providing early evidencefor the important role of proximity on socialattachment. Using electronic footprint data,recent studies have found additional evidencefor the large impact of proximity on attach-ment. A study of email networks among Dart-mouth students found that sharing a dormi-tory increased the volume of emails sent by twotimes, living on the same floor by another twotimes, and sharing a room by a further threetimes (Marmaros & Sacerdote 2006). Withina large multidivisional company, Kleinbaumet al. (2008) show that individuals who are inthe same business unit, subfunction, and of-fice location communicate in emails, calendarmeetings, and teleconferences approximately1000 times more than do pairs of employeesthat do not share any of those categories. Atthe planetary scale, Leskovec & Horvitz (2007)showed that the frequency and duration of 1.8billion instant-message conversations between180 million people worldwide decreased as ge-ographic distance increased. Even when digital


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


communication can make the world flat and inone electrifying moment “end the tyranny ofdistance” (to paraphrase Samuel Morse), peo-ple still tend to connect to those comparativelyfew others who are spatially proximate.

Proximity similarly influences the persis-tence of social ties by moderating the effortrequired to maintain relationships. Martin &Yeung’s (2006) analysis on the determinants oftie persistence of former residents of voluntarycommunes found that geographic proximitywas a significant predictor of detachment; peo-ple were much more likely to remain friendsover a 12-year period if they were geograph-ically close. In an analysis of the evolution of66 high school graduates’ personal networksover a 6-year period, Bidart & Lavenu (2005)similarly found that changes in geographiclocation resulted in the loss of significantportions of people’s networks. They note thatsome individuals lost as much as one-third oftheir old personal networks as they transitionedto a new location (pp. 365–68). Proximity mayhave increased importance during later periodsof people’s lives when they are less mobile.For example, Ikkink & van Tilburg (1999)found that distance was negatively related tothe longevity of older adults’ social-supportties, and Fischer (1982) showed that age wasassociated with social networks that were moregeographically constrained.

The idea of proximity may involve but is notdefined by geographic closeness (Gieryn 2000).Rather, social interaction is organized into focisuch as “social, psychological, legal, or physicalentit[ies] around which joint activities are or-ganized” (Feld 1981, p. 1016). The notion ofsocial foci draws attention to collectivities suchas workplaces, clubs, groups, and associationsto which individuals can belong and that can-not be defined in terms of physical geographyalone (Feld 1981, Grossetti 2005). Sharing fociof activity affects attachment above and beyondphysical proximity, as common interest in anactivity may emphasize members’ shared inter-ests instead of their discordant ones (Uzzi &Dunlap 2005), downplay status or rolediscrepancies (Moody 2001), and bring people

together in mutually rewarding situations thatpromote positive sentiment (Homans 1961;Feld 1981, p. 1017). Moreover, foci such asneighborhoods, communities, and organiza-tions may develop cultures that further inducethe creation of social relationships (Entwisleet al. 2007). If networks are the fabric of inter-personal interaction, social foci are the loomsin which they are woven.

Kono et al. (1998) examined the influenceof social foci and geographic proximity onfirms’ choices of corporate board members.They found that firms were more likely toshare corporate board members (referred toas a corporate interlock; see Mizruchi 1996)if those firms had headquarters in proximatelocations. However, upon closer inspection,Kono et al. (1998) found that membership inexclusive upper-class clubs in the same city asthe headquarters was the actual driver of boardinterlocks. These clubs were important socialfoci that enabled interaction between elites whothen shaped interfirm ties. More broadly, sev-eral studies have documented the influence thatthe foci of shared workplaces play in fosteringsocial ties. Kossinets & Watts’s (2006) analysisof the formation of email communication tiesfound that a significant predictor of tie forma-tion was whether or not students were in a classtogether. Golder et al. (2006) recently lookedat 284 million Facebook messages sent by4.2 million students at 496 different uni-versities and found that nearly 60% of allmessages were exchanged by students at thesame university.

Given advances in technologies that facil-itate communication, geographically dispersedsocial foci may be playing a more important rolein organizing attachments while geography isplaying a less important role. Jones et al. (2008)examined all the collaborations among U.S. sci-entists worldwide from 1975 to 2005 for 650U.S. research universities. They calculated thefraction of work published by scientists as soleauthors, coauthors in teams with other scien-tists from their home or nearby universities, andcoauthors in teams with other scientists at otheruniversities at a distance. Their work showed


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


that in all areas of science the greatest growthrate in collaborations is in university collabo-rations at a distance. In social science, for ex-ample, multiuniversity collaborations made upabout 40% of all team collaborations in 2005.The fraction of collaborations of coauthoredwork done with researchers from their home ornearby universities was also 40%. These find-ings allude to the possibility that shared focisuch as academic associations, conferences, andconsortia oriented around academic disciplinesand research areas may increasingly bring to-gether researchers from geographically distinctlocales. As technological advances in commu-nication proliferate, the effect of distance onattachment may increasingly be supplanted bythe influence of shared foci that supersede theconfines of geographic propinquity (Wellman1979, 1996; Hampton & Wellman 2001).

Similar to how changes in geographic dis-tance influenced detachment, the loss of sharedfoci induces the breaking apart of network ties.Bidart & Lavenu’s (2005, p. 361) study of net-work evolution found that when transitioningfrom school to work, young people “lost theirschoolmates in a fairly extreme way. . . . Theyhad relatively extensive networks, which under-went a major replacement of ties at the time oftheir entry into working life.” Nevertheless, “[a]relation may last longer than the circle [foci],which made its initial construction possible. Wemaintain relations with former school or uni-versity companions, former colleagues, and for-mer activists in political parties long since disap-peared” (Grossetti 2005, p. 292). Relationshipsthat carry over from prior foci play an addition-ally important role of connecting the differentsocial groups and communities to which peo-ple belong. Ties formed during university yearsmay have a large impact on the social capital ofgraduates throughout their careers, as Cohenet al. (2008) recently showed with regard to therole of MBA alma mater and access to privateinformation in the stock market.

What is much less understood in this liter-ature is an understanding of the aspects of fociof activity that lead to strong tendencies towardsocial connections—i.e., a theory of why some

foci encourage connections and others do not.In his study of New York City child care cen-ters, Small (2009) has begun to address this.He found that mothers often expanded theirnetwork through relationships developed at thecenters; however, the extent of network devel-opment was strongly influenced by institutionalpractices at the centers. He found that peoplewere more likely to form ties when “they haveopportunities to interact, when they do so fre-quently, when they are focused on some activ-ity, when they are not competitors, and whenthey have reason to cooperate” (p. 15). Thiscorroborates experimental findings that sug-gest that perceptions of group belongingness,positive affective sentiments, and emotional re-actions are greatest when activities are charac-terized by highly interdependent activities andfeelings of shared responsibility (Lawler et al.2008). There is still much opportunity for fu-ture research in this area. Do shared activitiesthat focus on small group interaction, such asfour-person bridge teams (to allude to the fa-mous shared activity that created a connection,valued at $45 billion, between William Gatesand Warren Buffet), or large group activities,such as soccer or basketball, have a greater im-pact on the formation of network attachments?Do activities that have some gradation of win-ning or losing versus noncompetitive interac-tion (e.g., a book club) impact the creation ofties differently? How do activities that highlighta core aspect of personal commonality, such as ashared passion for the environment, help createconnections between people who are otherwisenonhomophilous or distantly linked throughcommon third parties?

CONCLUSION

Homans (1961, p. 1) commented that social be-havior is a “familiar chaos,” in that “[n]othingis more familiar to men than their ordinary, ev-eryday social behavior,” yet much of social in-teraction still appears chaotic. Change in socialnetworks is a familiar chaos. Although there arefew social phenomena more elementary thanthe creation and dissolution of relationships,


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


these processes are driven by sometimes com-plex mechanisms and generate networks thathave profound implications for human achieve-ment, beliefs, and outcomes. Researchers havehistorically approached this familiar chaos inthree ways: one focusing on actors’ attributes,another on their relationships, and a thirdon proximity. Each approach has yielded ro-bust and often generalizable findings. How-ever, these theoretical streams have also tendedto progress in relative isolation—favoring theirown theoretical foundations and explanatoryvariables. Further integration stands to ben-efit network research in at least three ways:(a) by explicating causal mechanisms, (b) by ex-ploring interactions between mechanisms, and(c) by evaluating the changing role of differentmechanisms as networks emerge and evolve.

Disambiguating causal relationships in thedynamics of relationship formation and dissolu-tion remains an elusive goal. Perpetually threat-ening the validity of each perspective covered inthis review is the possibility that its findings areepiphenomenal to the operation of the others.For example, Feld (1982) argues that propo-nents of homophily significantly overestimateactors’ preferences to interact with self-similarothers by neglecting to account for people’sself-selection into contexts that bring togetherdisproportionately homophilous actors. In-deed, a recent analysis by Kossinets & Watts(2009) found precisely this effect—in theirstudy, college students with similar gender, age,majors, academic years, and foreign/domesticstatus tend to take classes together, wherethey find occasion to create social ties. Thus,observed biases toward interaction with self-similar others are largely the result of priorselection into contexts of homophilous actors.Similar alternative explanations for clusteringhave also been proposed (e.g., Feld 1981). Fu-ture research that simultaneously accounts formechanisms in each perspective may furtheruntangle these theoretically elusive questions.

A second promising avenue for futureresearch is to explore interactions betweenassortative, relational, and proximity mech-anisms. One extant stream of research that

brings together assortative and relationalmechanisms explores the different networkingtendencies of men and women. For example,Eder & Hallinan (1978) found that male andfemale schoolchildren demonstrate differentsolutions to having a friend that is not friendswith their friends—boys were more likelyto add relationships to embed the isolatedindividual, whereas girls were more likely todelete a relationship to entrench their existingfriendship clique. Conversely, Plickert et al.(2007) found that women are more likely thanmen to reciprocate some relationships and,further, that reciprocity is more likely betweenindividuals that are spatially proximate. Fu-ture studies may further explore interactionsbetween assortative, relational, and proximitymechanisms by (a) evaluating whether closureis more likely between individuals who sharesocial foci or share attributes, or by (b) testingwhether homophily is greater (or lesser) withinsocial foci or whether homophily still mattersbetween people who share friends.

Finally, it is relatively unknown whetherdifferent mechanisms play greater or lesserroles as networks evolve. One area wherethis is especially apparent is in the genesisof social networks. Critiques of relationalmechanisms have noted that an emphasis onprior network structure unavoidably begsthe question: Where does the network comefrom in the first place (Podolny & Page 1998;Stuart & Sorenson 2007, p. 220)? Given theimportance of prior connections on informingactors about potential friends and colleagues,people in emerging networks may rely onother avenues of information gathering—suchas homophily and social proximity—to initiallyfind new friends and colleagues, and only laterrely on networks to facilitate the formation ofrelationships. This suggests that as networksevolve so too do the rules that govern theirevolution (Rivera & Uzzi 2009). In the faceof this complexity, what is clear is that under-standing these contingent effects of networkdynamics promises that future work will de-velop a greater understanding of the changingnetworks in which we are all embedded.


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


DISCLOSURE STATEMENT

The authors are not aware of any affiliations, memberships, funding, or financial holdings thatmight be perceived as affecting the objectivity of this review.

ACKNOWLEDGMENTS

The authors thank Helena Buhr, Brayden King, Sekou Bermiss, Kate Ashley, and Maxim Sytchfor their invaluable comments and revisions, as well as attendees of the 2008 Complex SystemsSummer School at the Santa Fe Institute. Research was sponsored by the Northwestern UniversityInstitute on Complex Systems and Network Science (NICO) and the Army Research Laboratory(Cooperative Agreement Number W911NF-09-2-0053). The views and conclusions containedin this review are those of the authors and should not be interpreted as representing the officialpolicies, either expressed or implied, of the Army Research Laboratory or the U.S. Government.

LITERATURE CITED

Abrahamson E, Fairchild G. 1999. Management fashion: lifecycles, triggers, and collective learning processes.Adm. Sci. Q. 44:708–40

Adamic LA, Adar E. 2003. Friends and neighbors on the Web. Soc. Netw. 25:211–30Ahuja G. 2000. Collaboration networks, structural holes, and innovation: a longitudinal study. Adm. Sci. Q.

45:425–55Albert R, Barabási A-L. 2002. Statistical mechanics of complex networks. Ref. Mod. Phys. 74:47–97Allen VL, Wilder DA. 1979. Group categorization and attribution of belief similarity. Small Group Behav.

10:73–80Amaral LAN, Scala A, Barthelemy M, Stanley HE. 2000. Classes of small-world networks. Proc. Natl. Acad.

Sci. USA 97:11149–52Backman CW, Secord PF. 1959. The effect of perceived liking on interpersonal attraction. Hum. Relat. 12:379–

84Baker WE. 1990. Market networks and corporate-behavior. Am. J. Sociol. 96:589–625Baker WE, Faulkner RR, Fisher GA. 1998. Hazards of the market: the continuity and dissolution of interor-

ganizational market relationships. Am. Sociol. Rev. 63:147–77Barabási A-L. 2008. Network Science Applications to Global Communications. Norwich Research Park, UK: NetSci.Barabási A-L. 2009. Network theory-the emergence of the creative enterprise. Science 308:639–41Barabási A-L, Albert R. 1999. Emergence of scaling in random networks. Science 286:509–12Barro RJ, McCleary RM. 2003. Religion and economic growth across countries. Am. Sociol. Rev. 68:760–81Baum JAC, Shipilov AV, Rowley TJ. 2003. Where do small worlds come from? Ind. Corp. Change 12:697–725Bearman PS, Moody J, Stovel K. 2004. Chains of affection: the structure of adolescent romantic and sexual

networks. Am. J. Sociol. 110:44–91Bidart C, Lavenu D. 2005. Evolutions of personal networks and life events. Soc. Netw. 27:359–76Blackwell DL, Lichter DT. 2004. Homogamy among dating, cohabiting, and married couples. Sociol. Q.

45:719–37Blau PM. 1974. Presidential address—parameters of social structure. Am. Sociol. Rev. 39:615–35Borgatti SP, Everett MG. 1999. Models of core/periphery structures. Soc. Netw. 21:375–95Borgatti SP, Everett MG. 2006. A graph-theoretic perspective on centrality. Soc. Netw. 28:466–84Bossard JS. 1932. Residential propinquity as a factor in marriage selection. Am. J. Sociol. 38:219–24Brass DJ. 1984. Being in the right place—a structural-analysis of individual influence in an organization.

Adm. Sci. Q. 29:518–39Bratter JL, King RB. 2008. “But will it last?”: marital instability among interracial and same-race couples.

Fam. Relat. 57:160–71Brown SL, Eisenhardt KM. 1997. The art of continuous change: linking complexity theory and time-paced

evolution in relentlessly shifting organizations. Adm. Sci. Q. 42:1–34


Ann

u. R

ev. S

ocio

l. 20

10.3

6:91

-115

. Dow

nloa

ded

from

arj

ourn

als.

annu

alre

view

s.or

gby

NO

RT

HW

EST

ER

N U

NIV

ER

SIT

Y -

Eva

nsto

n C

ampu

s on

07/

13/1

0. F

or p

erso

nal u

se o

nly.


Budig MJ, England P. 2001. The wage penalty for motherhood. Am. Sociol. Rev. 66:204–25Burgess EW, Wallin P. 1943. Homogamy in social characteristics. Am. J. Sociol. 49:109–24Burris V. 2004. The academic caste system: prestige hierarchies in PhD exchange networks. Am. Sociol. Rev.

69:239–64Burt RS. 1992. Structural Holes: The Social Structure of Competition. Cambridge, MA: Harvard Univ. Press.

313 pp.Burt RS. 1997. The contingent value of social capital. Adm. Sci. Q. 42:339–65Burt RS. 2000. Decay functions. Soc. Netw. 22:1–28Burt RS. 2004. Structural holes and good ideas. Am. J. Sociol. 110:349–99Burt RS, Knez M. 1995. Kinds of third-party effects on trust. Ration. Soc. 7:255–92Byrne D, Wong TJ. 1962. Racial prejudice, interpersonal attraction, and assumed dissimilarity of attitudes.

J. Abnorm. Soc. Psychol. 65:246–53Casciaro T, Lobo MS. 2008. When competence is irrelevant: the role of interpersonal affect in task-related

ties. Adm. Sci. Q. 53:655–84Chung S, Singh H, Lee K. 2000. Complementarity, status similarity and social capital as drivers of alliance

formation. Strateg. Manag. J. 21:1–22Clauset A, Shalizi CR, Newman MEJ. 2009. Power-law distributions in empirical data. SIAM Rev. 51:661–703Cohen LE, Felson M. 1979. Social-change and crime rate trends—a routine activity approach. Am. Sociol. Rev.

44:588–608Cohen L, Frazzini A, Malloy C. 2008. The small world of investing: board connections and mutual fund

returns. J. Polit. Econ. 116:951–79Cohen WM, Levinthal DA. 1990. Absorptive capacity: a new perspective on learning and innovation. Adm.

Sci. Q. 35:128–52Coleman JS. 1988. Social capital in the creation of human capital. Am. J. Sociol. 94:S95–120Coleman JS. 1990. Foundations of Social Theory. Cambridge, MA: Harvard Univ. PressConyon MJ, Muldoon MR. 2006. The small world of corporate boards. J. Bus. Financ. Account. 33:1321–43Cummings JN, Kiesler S. 2005. Collaborative research across disciplinary and organizational boundaries. Soc.

Stud. Sci. 35:703–22Davis G, Yoo M, Baker W. 2003. The small world of the American corporate elite. Strateg. Organ. 3:301–26Davis JA, Holland PW, Leinhardt S. 1971. Comments on Professor Mazur’s hypothesis about interpersonal

sentiments. Am. Sociol. Rev. 36:309–11DiMaggio P, Louch H. 1998. Socially embedded consumer transactions: For what kinds of purchases do

people most often use networks? Am. Sociol. Rev. 63:619–37DiMaggio PJ, Powell WW. 1983. The iron cage revisited: institutional isomorphism a

Dynamics of Dyads in Social Networks: Assortative, Relational, … · 2004. 5. 7. · SO36CH05-Uzzi...

Documents

Transcript of Dynamics of Dyads in Social Networks: Assortative, Relational, … · 2004. 5. 7. · SO36CH05-Uzzi...