Ties That Last Tie Formation and Persistence in Research Collaborations over Time

Abstract

Using a longitudinal dataset of research collaborations over 15 years at Stanford University, we build a theory of intraorganizational task relationships that distinguishes the different factors associated with the formation and persistence of network ties. We highlight six factors: shared organizational foci, shared traits and interests, tie advantages from popularity, tie reinforcement from third parties, tie strength and multiplexity, and the instrumental returns from the products of ties. Findings suggest that ties form when unfamiliar people identify desirable and matching traits in potential partners. By contrast, ties persist when familiar people reflect on the quality of their relationship and shared experiences. The former calls for shallow, short-term strategies for assessing a broad array of potential ties; the latter calls for long-term strategies and substantive assessments of a relationship’s worth so as to draw extended rewards from the association. This suggests that organizational activities geared toward sustaining persistent intraorganizational task relationships need to be different from activities aimed at forging new ones.

Full text

http://asq.sagepub.com/
Quarterly
Administrative Science
http://asq.sagepub.com/content/58/1/69
The online version of this article can be found at:
DOI: 10.1177/0001839212474272
2013 58: 69 originally published online 16 January 2013Administrative Science Quarterly
Linus Dahlander and Daniel A. McFarland
Ties That Last : Tie Formation and Persistence in Research Collaborations over Time
Published by:
http://www.sagepublications.com
On behalf of:
Johnson at Cornell University
can be found at:Administrative Science QuarterlyAdditional services and information for
http://asq.sagepub.com/cgi/alertsEmail Alerts:
http://asq.sagepub.com/subscriptionsSubscriptions:
http://www.sagepub.com/journalsReprints.navReprints:
http://www.sagepub.com/journalsPermissions.navPermissions:
What is This?
- Jan 16, 2013OnlineFirst Version of Record
- Feb 8, 2013Version of Record >>
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Administrative Science Quarterly
58 (1)69–110
ÓThe Author(s) 2013
Reprints and permissions:
sagepub.com/
journalsPermissions.nav
DOI: 10.1177/0001839212474272
asq.sagepub.com
Ties That Last:
Tie Formation
and Persistence
in Research
Collaborations
over Time
Linus Dahlander
1
and Daniel A. McFarland
2
Abstract
Using a longitudinal dataset of research collaborations over 15 years at
Stanford University, we build a theory of intraorganizational task relationships
that distinguishes the different factors associated with the formation and per-
sistence of network ties. We highlight six factors: shared organizational foci,
shared traits and interests, tie advantages from popularity, tie reinforcement
from third parties, tie strength and multiplexity, and the instrumental returns
from the products of ties. Findings suggest that ties form when unfamiliar
people identify desirable and matching traits in potential partners. By con-
trast, ties persist when familiar people reflect on the quality of their relation-
ship and shared experiences. The former calls for shallow, short-term
strategies for assessing a broad array of potential ties; the latter calls for
long-term strategies and substantive assessments of a relationship’s worth
so as to draw extended rewards from the association. This suggests that
organizational activities geared toward sustaining persistent intraorganiza-
tional task relationships need to be different from activities aimed at forging
new ones.
Keywords: research collaborations, network ties, tie formation, tie persis-
tence, long-term ties, task relationships
Organizations frequently require and benefit more from interpersonal ties that
persist over time than they do from new ties. Repeated collaborations have
fewer startup costs than new ones, they entail greater certainty and trust, and
the individuals engaged in long-standing ties frequently communicate better
(Marsden and Campbell, 1984; Uzzi, 1997). From this follow all manner of ben-
efits, from easier and more effective communication to a more complete
1
ESMT European School of Management and Technology
2
Stanford University
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

transfer of information and reciprocal forms of exchange (Katz, 1982). In short,
organizations are often more interested in factors that nurture the continuation
of extant collaborations than they are in creating an expanded portfolio of new
contacts.
Unfortunately, research on social network dynamics and organizational net-
works seldom focuses on the processes that sustain and nurture intraorganiza-
tional collaborations. Social network research focuses almost exclusively on
factors of tie formation (Coleman, 1974; Snijders, 2001; Levi Martin and Yeung,
2006; Robins et al., 2007; Baldassari and Diani, 2007). As Burt (2002: 343) said,
all the focus is on formation and there is ‘‘almost no research on the stability of
interpersonal relationships.’’ In addition, social network research typically con-
cerns personal ties such as friendship, which frequently occur beyond the orga-
nizational setting and are distinct from the more instrumental forms of work
collaboration central to a firm’s functioning (e.g., see Lazarsfeld and Merton,
1954). The goal of this paper is to fill this gap in the literature: to direct attention
towards intraorganizational task relationships and the factors leading to their
formation and persistence.
One must turn to interorganizational research to find a discussion of how
collaborations form and persist (Palmer, 1983). Some of this work focuses on
the persistence of exclusive firm-client relations (Seabright, Levinthal, and
Fichman, 1992; Baker, Faulkner, and Fisher, 1998; Broschak, 2004).
Researchers have found that such ties dissolve when the clients’ resource
needs shift and they find alternatives; and they persist when pre-established
relations are strong enough to focus the client’s resource needs on what the
firm can afford (Seabright, Levinthal, and Fichman, 1992). In other studies,
researchers have found that interorganizational collaborations form when com-
mon third-party ties bring them together, or when firms enter alliances to
access critical resources and information about each other (Palmer, 1983).
Conversely, these alliances dissolve when such factors are uncertain or partial
(Gulati and Gargiulo, 1999; Ingram and Torfason, 2010). This literature nicely
highlights interorganizational factors we might draw upon, but they do not expli-
citly concern intraorganizational relationships.
Missing from the literature is a focus on intraorganizational task relationships
and their dynamics, arguably a core feature of many organizations. There has
also been little discussion of how the process of forming work collaborations
may differ from the process of sustaining them. On purely intuitive grounds, it
makes sense that the factors associated with the formation of task relations
will differ from those that lead them to persist. After all, the tie formation pro-
cess is one in which factors bring strangers into a relation, while the tie persis-
tence process is one in which factors guide people who are familiar with each
other to repeat and extend their association. As such, the factors initially bring-
ing us together may not be the same as those that keep us together. But
research on personal networks assumes that the factors guiding tie formation
also guide tie persistence. For example, after a long discussion of tie formation,
McPherson, Smith-Lovin, and Cook (2001: 436–438) called for future research
on tie persistence and speculated that the processes driving tie persistence
likely mimic those of tie formation, but with weaker effects. From extant
research on personal tie formation in organizations, we know that ties form
due to people having organizational foci in common, homophily, and attraction
for status reasons (Ruef, Aldrich, and Carter, 2003). Hence, if we extend the
70 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

argument of McPherson, Smith-Lovin, and Cook, these same factors should
extend to the persistence of task relations. That is, same-gender individuals
should be more likely to begin collaborating, and such collaborators should per-
sist because they share a greater sense of interpersonal understanding and
security. In this reasoning, traits remain salient to tie formation and
persistence—the persons associating do not grow accustomed to visible traits
as they shift from being newly relating strangers to repeatedly collaborating
individuals who are familiar with each other.
Other literature suggests that the processes of tie formation and tie persis-
tence may differ. According to a consistent strand of research on interorganiza-
tional networks, ties persist because of path dependence and inertia (March
and Simon, 1958; Cohen, March, and Olsen, 1972). The general argument is
that once a relationship forms, it takes on a life of its own and sustains itself
via its history or a logic of attachment (Stinchcombe, 1965; Seabright,
Levinthal, and Fichman, 1992). That is, once a tie is formed, people tend to
satisfice and stay in their current collaboration despite the availability of poten-
tially better matches. This argument highlights the features of a tie itself as fac-
tors for sustaining the tie into the future, and these features are not logically
relevant to tie formation.
Both arguments may be partially correct when it comes to explaining the for-
mation and persistence of intraorganizational task relationships. The processes
of tie formation and tie persistence occur in different contexts, and these pro-
cesses draw on different informational cues and resources. Individuals who
have already collaborated have different information than those who have
never met and who lack firsthand information. Some cues are only available
after a tie has formed, and other cues are available at both stages of formation
and persistence. A theory of tie formation and persistence thus needs to deter-
mine when a factor is salient and whether it influences the formation and per-
sistence of ties in distinct ways. Tie persistence occurs in the context of an
existing relationship between people. When deciding whether to sustain that
tie, its participants reflect on a rich assortment of information concerning the
quality of shared experiences and returns from their tie. By contrast, tie forma-
tion occurs in the context of two unfamiliar individuals meeting and seeking
points of similarity, mostly in appearances and credentials. As such, tie forma-
tion occurs in an uncertain and broad context of strangers applying short-term
strategies to quickly assess and forge ties on the basis of shared traits. Tie per-
sistence occurs in a more certain context of familiar others in which people
apply long-term strategies to make substantive assessments of a relationship’s
worth so as to draw extended rewards from the association. Tie formation and
tie persistence therefore represent qualitatively different stages of relational
decision making and a shift in context that is quite profound for studies of
social networks and organization theory. To explain how they differ, we build a
systematic, empirical study of the factors promoting the formation and persis-
tence of work collaborations in organizations.
Distinguishing tie persistence and tie formation as distinct network dynamics
requires a large and rich longitudinal dataset on intraorganizational networks.
Needless to say, such information is hard to come by. Given the prevalence of
the current knowledge economy, and the centrality of knowledge creation
activities within it, we turned to universities and academics’ collaborations in
written publications and grant applications (Weingart, 2000; Peterson, 2007).
Dahlander and McFarland 71
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Publications and grants are central to a research university’s prestige and the
faculty member’s prospects for tenure. Collaborations that increase the quality
and quantity of publications and grants are a premium resource in these
organizations. In fact, the prevalence of collaboration among scientists has
increased steadily over the last three decades (Wuchty, Jones, and Uzzi, 2007).
Collaborations are particularly common in the natural sciences because of the
need to access new instrumentation (De Solla Price, 1963) and address com-
plex problems too difficult for any single individual to solve (Basalla, 1988).
While some information on publications is readily available, it must be linked to
a rich assortment of information about the faculty so as to control for alterna-
tive explanations, evaluate the relative importance of different structural fac-
tors, and explain the variation in tie formation and tie persistence. We turned to
uniquely detailed information about the Stanford University faculty in 1993–
2007 that allowed us to identify when new, untenured faculty arrived at the uni-
versity and began to form their first ties and explain why some of these ties
persisted over time to become repeat collaborations.
TIE FORMATION AND PERSISTENCE IN RESEARCH COLLABORATIONS
In analyzing the collaborations of academics, we considered six different fac-
tors of tie formation and persistence to develop a set of hypotheses: (1) shared
organizational foci, (2) homophily in attributes and interest, (3) tie advantages
from popularity (cumulative advantage), (4) tie reinforcement from third parties
(triadic closure), (5) strength and multiplexity of ties (tie inertia), and (6) the
instrumental returns of a tie’s products (means-ends rationalization). Although
we examined the hypotheses in the context of research collaborations, we
believe they apply more broadly to instrumental relationships than to socio-
emotional relationships (Morris, Podolny, and Sullivan, 2008). The instrumental
ties we study concern task relationships formed with the intention to derive
some kind of reward or create a joint product. Our analyses focus on the struc-
tural factors of tie formation and persistence, acknowledging that some scho-
lars offer a more cultural perspective of networks (Emirbayer and Goodwin,
1994; Ruef, 2002).
Organizational Foci
It is common for network scholars to identify a network process without paying
close attention to organizational structure, yet as Brass and colleagues (2004:
796) suggested, people’s and groups’ positions locate formal actions in physi-
cal space and in workflows and hierarchies, ‘‘restricting their opportunity to
interact with some others and facilitating interaction with still others.’’ While
most interaction is voluntary, organizational structures bring together people
who might not otherwise associate. Feld (1981) characterized these shared
positions and activities as ‘‘foci,’’ or social, psychological, legal, or physical enti-
ties around which joint activities are organized. When individuals share organi-
zational foci, they are more likely to form social bonds. Festinger, Schachter,
and Back (1950) even suggested that foci trump other important factors of col-
laborations such as homophily. Recent work has also proposed that shared foci
deepen existing relationships (Reagans, 2011). It is less clear, however, how
this differs in the formation and persistence of ties.
72 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

At least two organizational foci are relevant in the context of universities:
departments and research centers. Departments are the most obvious foci
around which collaborations form. Departmental faculty are often located near
one another, share similar research training, and engage in many joint activities
such as student advising, teaching the same sets of students, performing simi-
lar research, and engaging in joint decisions about hiring, promotion, and so on.
As such, departments are constrained foci (Coser, 1974; Merton, 1976: 25),
and they come to represent formal, jurisdictional boundaries within the modern
university (Blau, 1973; Abbott, 2001). Intradepartmental ties should therefore
be more likely to form and persist over time.
That said, interdepartmental collaborations also occur, though at a lesser rate
(Friedkin, 1978). Cross-departmental ties are partly driven by scientists seeking
to solve particular kinds of problems by using expertise from several different
disciplines (Jacobs and Frickel, 2009). But ties spanning departments are hard
to form and sustain because they expose scholars to dissonant ideas that are
not necessarily compatible with their frames of reference. Burt (2002) argued
that bridges across disconnected groups decay faster than other ties because
they require more negotiations between disparate interests. Although such ties
may be beneficial, they can also be time-consuming to maintain, as there are
‘‘fewer individuals involved to carry the cost of the collaboration’’ (Burt, 2002:
344).
The university’s organizational structure has been augmented with a second
class of organizational foci—interdisciplinary research centers—to support and
sustain some of these interdepartmental collaborations. In general, research
centers are secondary foci to the primary foci of departments. Departments
are typically the locus of pay, career tracks, and disciplinary affiliations and
therefore hold more sway over faculty members’ behavior. In contrast,
research centers bring faculty together on visible research projects and topics
that often span departments, and they frequently sustain interaction with
research funding. Some faculty members even find interdisciplinary discussion
more enlivening because there are discussions of different methods, writing
styles, and epistemic cultures (Lamont, 2009). In fact, once faculty members
reach tenure, the constraints on interdisciplinary collaboration diminish, and
they may find extracurricular associations to be more novel and interesting.
In general, we expect organizational foci like departments and research cen-
ters to be positively associated with tie formation and tie persistence but not to
the same extent. The underlying mechanism is one of exposure: faculty who
become aware of one another are more likely to affiliate. We thus expect the
effect of foci to be salient for tie formation but not for tie persistence. Most
connected persons will already be proximate to one another, and they will
incrementally seek some degree of novel experience.
Hypothesis 1 (H1): Organizational foci have positive effects on tie formation but not
on tie persistence.
Homophily
A salient factor for research collaboration is homophily, or the tendency of peo-
ple to select collaborations with others similar to themselves. Lazarsfeld and
Merton (1954: 24) drew ‘‘a distinction between status-homophily (observed
tendencies for similarities between the group-affiliation of friends or between
Dahlander and McFarland 73
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

their positions within a group) and value-homophily (observed tendencies
toward correspondence in the values of friends).’’ McPherson, Smith-Lovin,
and Cook (2001: 419) extended Lazarsfeld and Merton’s distinction:
Status homophily includes major sociodemographic dimensions that stratify
society—ascribed characteristics like race, ethnicity, sex, or age, and acquired charac-
teristics like religion, education, occupation, or behavior patterns. Value homophily
includes the wide variety of internal states presumed to shape our orientation toward
future behavior.
Collaborations frequently form in broad contexts in which many unfamiliar
others are present and actors use homophily to quickly winnow the field.
Hence, we expect tie formation will be most associated with status homophily
because it is based on characteristics people can often glean from visual cues.
By contrast, tie persistence should occur in contexts in which more firsthand
information is present. Therefore, we expect tie persistence will be associated
with value homophily.
Status homophily. Many scholars submit that, among a sea of possible
ties, we are likely to forge collaborations with similar others. An expansive liter-
ature shows that people are more likely to associate when they share attri-
butes of age, gender, education, and ethnicity (Lazarfeld and Merton, 1954;
McPherson and Smith-Lovin, 1987; Ibarra, 1993; McPherson, Smith-Lovin, and
Cook, 2001). Homophily is often seen as an explanation for initial tie formation,
and some also believe it explains why some relationships persist longer than
others. McPherson, Smith-Lovin, and Cook (2001) encouraged research in this
area, as there is very limited empirical evidence and few theoretical arguments
relating status homophily to tie persistence. The few studies available note that
factors generating tie persistence appear to mimic tie formation, but with
smaller effect magnitudes. For example, Hallinan and Williams (1989) found
that friendship ties in secondary schools persisted longer for same-gender and
same-race relationships than for other ties. Given the uncertainty in tie forma-
tion contexts, it seems reasonable to believe that attribute homophily (based
on phenotypic comparisons) will be an efficient search strategy and means of
reducing uncertainty, but it will be less salient in a familiar context of preexist-
ing ties that transcend these initial scope conditions.
Hypothesis 2a (H2a): Status homophily as manifested in similar attributes has a posi-
tive effect on tie formation but not tie persistence.
Value homophily. Lazarsfeld and Merton (1954: 25) noted how ‘‘the
dynamic role of similarities and differences of these values in forming, main-
taining, or disrupting friendships . . . requires notice in its own right.’’ But mea-
suring value homophily has been inherently difficult (Ingram and Morris, 2007).
A great deal of research has measured it via attitudinal similarities, but others
have highlighted similarities in knowledge and shared experiences (McPherson,
Smith-Lovin, and Cook, 2001). We consider how tie formation and tie persis-
tence correspond with individuals reading the same research. When two indi-
viduals read similar work and engage in similar research topics, the challenges
of communicating with and comprehending one another decrease, and this
74 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

likely increases the chances of tie formation and tie persistence. At very high
levels, however, there may be a cost to too much knowledge overlap. The mar-
ginal benefit of shared knowledge decreases because there is a risk of redun-
dant ideas and few opportunities for complementarities between individuals. At
the extreme, one could even expect negative returns to value homophily—
individuals who are too similar have no new information to exchange and
explore, and they begin to encroach on each other’s turf. In the world of acade-
mia, scholars often seek a niche for their own work by differentiating it from
their colleagues’. To collaborate with someone who is too similar may be
undesirable.
We thus expect that shared knowledge will have an inverted U-shaped rela-
tionship with tie formation and tie persistence. The benefits are likely to have a
greater effect at low levels because shared knowledge establishes a degree of
common ground and mutual interest while preserving room to explore unique
new experiences and perspectives. When two individuals read the exact same
texts and are substitutable, they have nothing new to exchange and are likely
to assume a competitive relationship. Given these arguments, a balance
between similarity and difference of work should be most conducive to produc-
ing new and persistent ties.
Hypothesis 2b (H2b): Value homophily, as manifested in intellectual similarity, has
an inverted U-shaped relationship with tie formation and persistence.
Cumulative Advantage
Research on the Matthew effect maintains that popular individuals attract more
overtures from others for ties (Merton, 1973; Bothner, Podolny, and Smith,
2011) and thereby have greater returns to research productivity (Azoulay, Graff
Zivin, and Wang., 2010; Oettl, 2012; Waldinger, 2012). In the context of acade-
mia, there are two especially salient sources of cumulative advantage: the
number of ties an individual maintains and the amount of financial resources
available to that person. Grant funding is an important source of power in uni-
versities (Pfeffer and Moore, 1980).
The halo effect of cumulative advantage through approaching sought-after
individuals is especially attractive for tie formation, but it may be less pro-
nounced for tie persistence. By virtue of collaborating with many other individu-
als, a popular individual has greater social capital to draw on, but his or her time
is also more in demand and scarcer. There is a tension between the access to
fungible resources from working with people of great social standing and the
danger of not getting sufficient attention from them (Burt, 1992). This creates a
situation in which individuals with some ties are likely to be attractive collabora-
tors, while those with too many ties are less attractive.
This speaks to the general assertion that there is a finite number of ties an
individual can credibly maintain (Jackson, 2008). Each tie requires attention,
and individuals have limited attention to distribute (March and Simon, 1958).
Simon (1971: 41) noted that ‘‘a wealth of information creates a poverty of
attention and a need to allocate that attention efficiently among the overabun-
dance of information sources that might consume it.’’ This points to an upper
limit of ties an individual can credibly maintain, because each tie requires com-
munication and effort (Ahuja, 2000). This is particularly evident for collabora-
tions that involve substantial dialogue and regular meetings. To the extent that
Dahlander and McFarland 75
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

one can only maintain a finite number of ties, individuals have to select among
different credible options. The relative importance of any given tie decreases
with the number of ties one must attend to. At some point, too many partners
can decrease the advantage derived from any particular tie (see Koput, 1997,
for a simulation model).
For forming ties, it is attractive to go after the most sought-after individuals,
but persistent returns from such a relation may be untenable in the long run.
Popular individuals seldom reciprocate all tie overtures equally, and they are
likely to contribute less time to most collaborations. Therefore, popular individu-
als will be less able to sustain their ties, and their partners may see little value
in doing so. People with a few ties are likely to sustain their ties longer because
they see learning benefits in multiple collaborations and have time to commit
to them. But people with too many ties may find that coordination costs
increase with each additional tie. This leads us to hypothesize an inverted
U-shaped relationship between the number of ties an individual maintains and
the likelihood of tie persistence. The tendency for well-connected individuals to
affiliate should be stronger for new tie formation than for persistence, as new
ties are cheap to produce, have few obligations, and additional rewards have
yet to be sought.
Hypothesis 3 (H3): Individuals who are more connected and have more resources
will facilitate tie formation and have an inverted U-shaped relation with tie
persistence.
Triadic Closure
Sociologists have long described how the commitment to a relationship is
contingent upon the broader network of ties in which it is embedded
(Simmel, 1950). One of Coleman’s (1988) observations was that dense,
cohesive communities establish cooperative norms through an increased
capacity for monitoring and sanctioning behaviors. From this, we infer that
ties are more likely to form between individuals with shared friends but will
also persist when they exist in a closed triad or group. This argument is
rooted in Simmel’s (1950: 136) observation that ‘‘the sociological structure of
the dyad is characterized by . . . the intensification of [the] relation by a third
element, or by a social framework that transcends both members of a dyad.’’
A completely connected triad transforms dyadic ties by mitigating the pursuit
of an individual’s self-interests, reducing the bargaining power of single indi-
viduals, and facilitating cooperation and conflict resolution (Krackhardt, 1999).
This is supported in Krackhardt’s (1998) study of a college dorm, in which he
found that ties are more ‘‘sticky’’ if they have mutual friends. Reagans and
McEvily (2003) proposed that social cohesion around a relationship affects
the willingness and motivation of individuals to invest time, energy, and
effort in sharing knowledge with others. Even in the extreme case, in which
two individuals face obstacles to collaborate because of conflicts of inter-
ests, it can be difficult to disengage because the relationship is embedded in
a larger group, such as a lab.
Hypothesis 4 (H4): Individuals who are connected by indirect ties are more likely to
form and maintain ties.
76 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Tie Inertia
Some view tie dissolution as the inverse of tie formation, and this approach
suggests that two individuals would end their relationship when the tie could
no longer stimulate the necessary interest to produce future outcomes, but this
view of tie dissolution ignores the concept of attachments (Seabright,
Levinthal, and Fichman, 1992). Attachment is a commitment that emerges
from shared experiences and investments in a relationship. This may lead to
people maintaining relationships that are no longer fruitful, despite their having
more attractive alternatives. When a relationship is in place, the two partners
have more information available about one another than about unaffiliated indi-
viduals, and they have accumulated a variety of shared experiences. The fac-
tors of tie persistence are logically independent of those from tie formation
because they require a tie to already exist. In particular, inertia and attachment
correspond with two qualities of ties: tie strength and tie multiplexity.
Characteristics of ties afford participants motives to sustain a relationship.
Prior work has found that strong ties are proximate, entail frequent interaction
and shared experiences, and correspond with a sense of attachment or feel-
ings of closeness (Marsden and Campbell, 1984). A great deal of research has
elaborated on the qualities of strong ties, making its further elaboration unne-
cessary here (Granovetter, 1973, 1983).
Another tie quality—multiplexity—suggests that greater dimensionality in a
tie corresponds with greater value. Multiplexity refers to the number of differ-
ent types of relationships two persons share (Burt, 1980; Wasserman and
Faust, 1994). Although network scholars typically focus on a single activity
(Brass et al., 2004), there has been recent interest in multiple networks and
multiple ties, which can serve more than one purpose (Gould, 1991; Lomi and
Pattison, 2006). For example, in some of these studies ties from marriage are
often used for the purposes of business (Padgett and McLean, 2006). As such,
a unidimensional tie can be expanded and layered further for other purposes,
extending its value and persistence. Elsewhere, studies find that multiplex ties
act much like strong ties and have greater influence on productivity (Marsden
and Campbell, 1984; Rawlings and McFarland, 2011).
Multiplex ties are rare because they require greater investment to establish
interactions in multiple domains (Burt, 1980), but they are particularly valuable
for career outcomes (Ibarra, 1992). For example, Coleman (1988) argued that
multiplexity increases the overall resources available because resources in one
dimension of the relationship can be appropriated for use in another. Uzzi
(1997) built on this argument and described how multiplexity not only increased
the pool of resources but afforded flexibility and adaptation in the face of uncer-
tainty. Multiplex ties therefore enable people to transpose one type of associa-
tion to another and to overcome challenges that arise in any one type of
association. In addition, the breadth of association brings experiential ‘‘variety’’
to the dyad, making it less stale and more interesting over the long term. Last,
multiplex ties afford greater certainty about and understanding of interaction
partners (Padgett and Ansell, 1993). When we observe one another’s behaviors
in multiple settings, we forge a deeper understanding of a person and the sty-
lized habits shaping his or her behavior across contexts (Blau, 1964). Therefore,
even though multiplex ties are less common, they are a relevant factor in tie
persistence because they afford adaptability in the face of uncertainty, bring
Dahlander and McFarland 77
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

the thrill of variety, and afford greater certainty. This claim is supported at the
organizational level, where Palmer (1983) argued that overlapping ties will be
disbanded at a slower pace.
Hypothesis 5 (H5): Characteristics of ties, such as multiplexity and tie strength, influ-
ence the probability that ties will persist. This effect is only available after the tie
has formed.
Means-ends Rationalization
It is reasonable to assume that the products of a tie would have feedback
effects on the tie’s persistence, but this effect has been unexplored. Burt
(2002: 343) acknowledged that ‘‘people in . . . relationships often discover that
they do not enjoy one another, or cannot work well together, so they disen-
gage in favor of more compatible contacts.’’ Missing from his remark, how-
ever, is a reflection on the outcome of a collaboration and whether that is the
reason for the tie’s continuation. Independent of whether two individual scho-
lars enjoyed collaborating, they would be more positively inclined to continue
collaborating if their paper submissions successfully reached publication and
were widely cited.
The range of products that can arise from a social tie is quite broad—from
babies and ideas to meanings and problems—and each product has socially
recognized qualities and value. Most judgments of tie outcomes are only avail-
able in hindsight. Hence, individuals should repeat ties whose products have
resulted in success. They should drop ties when the product was a failure
(Levitt and March, 1988). For the collaborations of interest here, the tie prod-
ucts are articles, and they vary in how much recognition they receive. This
reflects how the wider social reinforcement for the activity being performed
affects the persistence of the tie. These factors feed back into decisions of tie
persistence such that more recognized collaborations will associate a sense of
reward with the shared activity. This links into March’s (1999: 141) observation
that ‘‘each node in a network learns from local experience, thereby adjusting
the local linkages....Locallearning depends on local judgments about the
‘success’ or ‘failure’ of experience with particular local links.’’
Hypothesis 6 (H6): Successful tie outcomes will facilitate tie persistence. This effect
is only available after the tie has formed.
Table 1 summarizes the factors involved in tie formation and persistence
and the hypotheses we developed about their effects.
METHODS
Data
We tested our hypotheses using data from a variety of different archival
resources covering fifteen years of data on Stanford University’s faculty mem-
bers, from 1993 to 2007. Collaborations in academia have become more popu-
lar over the last few decades for a variety of reasons (see, e.g., Wuchty, Jones,
and Uzzi, 2007). In academia, individuals can self-select whom they work with,
it is time consuming to develop the research, and there are significant rewards
to producing influential publications. By extension, choosing to collaborate with
78 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Table 1. Six General Factors for Tie Formation and Persistence
Factor General claims Tie formation Tie persistence Hypotheses
Organizational
foci
Individuals who
share organizational
foci are more likely
to be exposed to
one another and
thus more likely to
collaborate.
Strong positive
effect. The initial
exposure causes
individuals to form
ties.
No effect. The exposure
has already occurred,
so shared foci means
less once the tie is
established.
H1: Organizational foci have
positive effects on tie
formation but not on tie
persistence.
Status and
value
homophily
Individuals who
share status and
interest homophily
have more in
common and are
more likely to
collaborate.
Strong positive effect
for status
homophily. Shared
social attributes
cause individuals to
affiliate.
No effect. Ties become
less stereotypical once
the tie is established.
H2a: Status homophily as
manifested in similar
attributes has a positive
effect on tie formation but
not tie persistence.
Inverted U-shape
relationship
between interest
homophily and tie
formation. Shared
interests and similar
styles of work are
more compatible.
Inverted U-shape
relationship between
interest homophily and
tie persistence. A
sense of comple-
mentarity grows
stronger once the tie is
established.
H2b: Value homophily, as
manifested in intellectual
similarity, has an inverted
U-shaped relationship with
tie formation and
persistence.
Cumulative
advantage
Individuals who are
influential in terms
of past ties and
resources are
attractive
collaborators and
more likely to
collaborate.
Strong positive
effect. Cumulative
advantage suggests
that individuals with
status will be
attractive
collaborators.
Strong curvilinear effect.
The status effect of
collaborating with a
‘‘star’’ diminishes once
the tie is established.
H3: Individuals who are
more connected and have
more resources will
facilitate tie formation and
have an inverted U-shaped
relation with tie
persistence.
Triadic closure Individuals with
shared colleagues
have more detailed
information and are
more likely to
collaborate.
Strong positive
effect. Individuals
with shared friends
are more likely to
be aware of one
another and close
triads.
Strong positive effect.
Individuals that have
shared collaborators are
more likely to stick to
those relationships.
H4: Individuals who are
connected by indirect ties
are more likely to form and
maintain ties.
Tie inertia Individuals who are
connected by ties
that are expandable
for different
activities and
stronger are more
likely to continue
collaborating.
Not logically feasible.
The tie needs to be
in place.
Strong positive effect.
Experiences about the
collaboration itself
affect its future
continuation.
H5: Characteristics of ties,
such as multiplexity and tie
strength, influence the
probability that ties persist.
This effect is only available
after the tie has formed.
Means-ends
rationalization
The outcome quality
of collaborations
influences the
probability a
collaboration will
continue.
Not logically feasible.
The tie needs to be
in place.
Strong positive effect.
Collaborations that are
deemed successful by
relevant observers
affect the continuation
of the collaboration.
H6: Successful tie
outcomes will facilitate tie
persistence. This effect is
only available after the tie
has formed.
Dahlander and McFarland 79
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

another individual is a big commitment that can have serious implications for
careers. Individuals learn from their experiences working together, and if they
decide to collaborate repeatedly, they will likely produce a stream of successful
publications. These in turn help individuals achieve indefinite tenure and
advance their research area. For this reason, we study the first instance of col-
laboration and the factors leading to collaborations persisting over time.
The data we used for this paper were retrieved from several sources at
Stanford’s central office and include complete information about appointment
dates, department, courtesy and joint appointments, tenure status, ethnicity,
gender, and age. We also collected information from CVs and websites about
what year and from which university faculty members received their highest
degrees. This resulted in detailed longitudinal information for all 3,052 faculty
members who were in the faculty for the whole or some part of the study
period. These faculty are academic council members, meaning they have vot-
ing privileges in their departments. This includes teaching professors, research
professors, and clinical line faculty at the Medical School, all of whom can vote
in their departments. We disregarded lecturers, acting assistant professors, vis-
iting professors, consulting professors, and outside members of dissertation
committees. In this way, we bounded the network of faculty at those who are
more permanent members of the faculty (not as temporary as yearly appoint-
ments) and who have a say in departmental decisions. This reduces the list of
faculty to some extent but mostly removes peripheral individuals who occa-
sionally show up and have little effect on all of the other network statistics.
Wang et al. (2012) showed that such people have little effect on measurement
error. Figure 1 shows the proportion of faculty included in our sample and how
Figure 1. Distribution of faculty in different schools over time between 1993 and 2007.
Graduate School of Business
School of Earth Sciences
School of Education
School of Engineering
School of Law
School of Medicine
School of Humanities and Sciences
(Sciences)
School of Humanities and Sciences
(Social science)
School of Humanities and
Sciences (Humanities)
0
.2
.4
.6
.8
1
Proportion
1993 1998 2003
80 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

they are associated with each of the schools at Stanford. We divided the
School of Humanities and Sciences into three different areas of humanities,
social sciences, and natural sciences because they conduct distinctive forms of
research (Kagan, 2009). Appendix A describes all of the schools and their asso-
ciated departments.
We used departments to calculate the shared foci variable, but we report
schools in figure 1 to show trends visually (there are more than sixty depart-
ments at Stanford). We separated the School of Humanities, Social Science,
and Sciences as shown in the figure. As the graph shows, the School of
Medicine is the largest school and has also grown over our study period. In our
analyses, we accounted for departmental differences using fixed individual-
level effects and also included time dummies to capture time trends in colla-
boration patterns.
We obtained records on several types of faculty work activities to construct
different types of collaboration networks. First, we obtained records on all dis-
sertations defended at the university. These records include information
about each dissertation, including the year filed, name of the student, depart-
ment affiliation, and the names of dissertation committee members. We
used these data to construct affiliation networks showing how faculty mem-
bers were connected through co-advising of doctoral students. Second, we
collected yearly data from the university’s sponsored project office that
tracks all successful and unsuccessful grant applications submitted by the
faculty. We used this information to investigate faculty members’ collabora-
tion in writing grants. Third, we obtained a license and downloaded the full
content of all publications listed in Thomson’s ISI Web of Knowledge data-
base with Stanford in the address field. ISI is generally considered the most
comprehensive database for scholarly work and includes thousands of scho-
larly journals and reports. One obvious problem noted by earlier researchers
(Newman, 2008; Azoulay, Ding, and Stuart, 2009) is that some individuals
use different spellings in their publication record, and there are several differ-
ent individuals with the same name. Through a series of matching rounds,
we matched each faculty member with his or her publications. We evaluated
false positives and false negatives for each round of matching and reesti-
mated our regressions using networks with different precision for matching
faculty members and their publications (see Appendix B for details). We used
these data to create networks of faculty members collaborating on publica-
tions over time.
The records for each type of work activity are initially represented as an
affiliation network in which multiple individuals share events in the form of pub-
lications, grants, and dissertations. We multiplied the affiliation matrix
M
with
its transpose
M’
to create a collaboration network of how many times each
faculty member collaborated with another in a given year.
Dependent Variables
Tie formation. For our first analysis, we developed a yearly dataset of each tie
created between two individuals through coauthoring publications or applying
for a grant together. We focused on these types of collaborations because they
reflect the faculty member’s instrumental efforts at knowledge creation,
Dahlander and McFarland 81
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

whereas other forms of collaboration (like student co-advising) may be driven
by exogenous factors (e.g., the students select advisors). To avoid problems of
left-censoring, we analyzed how newly hired, untenured faculty members with
a Ph.D. they defended in the three years prior to their arrival at Stanford
forged ties with other Stanford faculty when they arrived. When manually
checking CVs, we found that these individuals had no or very little prior his-
tory of collaboration with Stanford faculty members. This variable was coded
as a dummy taking the value of 1 if a tie between individuals iand jis formed
in year t. Most academics know all too well that some paper projects fail to
get published, just as some alliance efforts between organizations do not
materialize. Because it is virtually impossible to get data on papers that were
rejected from journals and never made it to print, we undertook a comple-
mentary analysis in which failed attempts are observable. The second depen-
dent variable takes the same approach as above, but it distinguishes
between successful grants that won a financial award and unsuccessful
grant applications that never received funding. This allowed us to examine if
there were any differences across types of ties and assess whether counting
publication ties as manifested in publications is problematic and subject to a
potential success bias.
Tie persistence. Our second analysis focused on the survival of the first tie
initiated between iand jduring the course of our study. We can directly
observe when different ties come into being, but not when a tie ceases to
exist. A tie between individuals iand jcreated in year ttakes the value of 1 in
t+n (Singer and Willett, 1993). Given our analytical approach, a tie that is never
repeated is treated as right-censored. This approach allowed us to estimate the
likelihood of a tie being repeated between two academics, conditional on the
tie being at risk. We coded this dummy variable to take the value of 1 if the tie
between individuals iand jwas repeated in year t. We did this for both publica-
tion and grant ties.
Independent Variables
Organizational foci. To assess hypothesis 1, we assessed how shared orga-
nizational foci differ for tie formation and persistence. An often-cited barrier
to tie formation in academia is departmental boundaries that represent for-
mal, jurisdictional boundaries within the modern university (Abbott, 2001).
University departments constitute different units in which members have
similar cultures and intellectual heritages. As Blau (1973) said, one conse-
quence is that ties across departmental boundaries are rare and more diffi-
cult to form because there are different underlying logics of doing work and
assessing its importance. Other scholars have a more optimistic view of the
rate at which we will observe interdisciplinary ties. For instance, Friedkin’s
(1978) analysis of the physical sciences showed that university networks
sometimes span departments through interdisciplinary collaborations. To
control for these explanations, we controlled for two forms of propinquity in
departments and centers.
We developed a dummy that measured whether individuals iand jare from
the same or different departments in year t. A few departments merged or
changed names during our study period, which we accounted for in our
82 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

analysis. We have divided the School of Humanities and Sciences into three
sub-schools (Humanities, Sciences, and Social Science), as they deal with dif-
ferent types of research. Because there are no departments in three of the pro-
fessional schools (Law, Education, and the Graduate School of Business), we
treated those schools as departments.
1
Universities have also undergone major
transformations to incorporate interdisciplinary centers that bring researchers
together to solve pressing societal problems. We thus constructed a dummy
variable that captured whether individuals iand jbelonged to the same center
at Stanford in year t.
Homophily. We tested the homophily factor by measuring both status
homophily and value homophily (Lazarsfeld and Merton, 1954). We tested the
status attribute homophily argument in H2a using several variables that are
especially salient in the context of collaboration among academics (Burt, 2000;
McPherson, Smith-Lovin, and Cook, 2001). The first three—gender, age, and
ethnicity—are ascribed status characteristics, and the last two—education and
tenure status—are achieved status characteristics.
Ascribed status. Individuals with the same gender are more likely to affiliate
(Hallinan and Williams, 1989), and same-gender ties are more likely to persist
(McPherson, Smith-Lovin, and Cook, 2001). We therefore used a dummy vari-
able, same gender, that captures whether iand jare the same gender. Another
homophily variable that causes ties to persist is age difference. Here we mea-
sured whether similarly aged individuals—by their absolute difference in age—
are more likely to have persistent ties. Individuals who share the same ethnicity
are also more likely to associate. We measured ethnicity similarity with dum-
mies indicating whether iand jhad the same ethnicity or different ethnicities.
The following classes of ethnicity are in the data: (1) African American,
(2) Asian, (3) Caucasian, (4) Hispanic, and (5) Native American. We tested the
effect of ethnic homophily with a dummy variable that takes a value of 1 for
the same ethnicity and 0 otherwise.
2
Achieved status. We used two measures of achieved status: educational
background similarity and tenure status similarity. Although the job market
induces disciplinary boundaries (Abbott, 1988), scientists can apply for jobs in
different departments. For instance, a trained sociologist could potentially apply
for jobs in sociology departments, business schools, and schools of education.
To measure educational background similarity for each faculty member, we col-
lected information about the subject area of the individual’s highest degree. For
each of these degrees, we used the taxonomy of subjects developed by the
U.S. Department of Education’s National Center for Education Statistics
(NCES). This classification was developed to facilitate assessment and report-
ing of educational programs. We coded a dummy variable as 1 if iand jhad the
1
In supplementary analysis not reported here we reran the analysis with schools as the relevant
organizational foci. This did not affect the substantive results, although the effect of same school
was slightly weaker than same department.
2
To investigate whether there are differences across ethnicities, we also developed separate dum-
mies for the different possible combinations of ethnicity. In the regressions reported, we report
whether the two individuals have the same ethnicity, but alternative estimations with different com-
binations of ethnicity are available upon request. They do not change the results relevant to our
hypotheses.
Dahlander and McFarland 83
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

same two-digit degree code.
3
Our second measure of achieved status is tenure
status similarity. Individuals who are at similar stages of their careers may be
more likely to collaborate. For instance, untenured professors are pressured to
publish in certain journals to be promoted. A related argument refers to the
career logics of faculty members. Untenured scholars need to build a reputa-
tion within their discipline. Junior and senior colleagues have symbiotic rela-
tions in which the junior person brings the senior colleague new
methodological skills and the senior person brings the junior colleague access
to resources.
Value homophily. To test our hypothesis 2b that intellectually similar individ-
uals form and sustain ties, we compared the citations used in their publications
at t-1. Citations have commonly been regarded as good indicators of intellectual
similarity (Zuckerman and Merton, 1972). The underlying logic is that shared
citations demonstrate that scholars are building on similar kinds of prior work
and define the audience they are trying to approach. A greater proportion of
shared cites is therefore an indicator that two individuals work in the same
intellectual space. To measure reference similarity for each individual iand jin
year t, we measured the extent to which two individuals cite the same refer-
ences. Consider faculty member i’s references as a set A. To compare with
another faculty member j’s set B, we calculated the intersection of A ∩B over
the union A ∪B. This allowed us to capture the similarity of sets iand jwhile
acknowledging the fact that faculty differ with respect to how much they have
referred to prior work in the past or more formally, that the sets are of different
sizes. We ignored publications that iand jhad done jointly (as that would imply
that iand jhave a perfect overlap). We hypothesized that there are diminishing
returns to similarity; there is a greater effect from an increase in similarity at
lower levels than at higher levels. To test this, we included a squared term for
reference similarity.
Cumulative advantage. We expected individuals to have a cumulative
advantage as a result of collaboration centrality and amount of grant funding.
We argued that individuals with high collaboration centrality are likely to
attract further associations. We would expect this effect to taper off when
people get overwhelmed by a too-large number of ties; even the most extro-
verted scholars will be unable to sustain huge numbers of collaborations. To
measure collaboration centrality, we calculated the respective degree cen-
trality in the network for individuals iand jin year t-1, using a decay time of
three years for ties. We included a squared term to test for a possible curvi-
linear association with tie formation and tie persistence. To assess whether
ties form between individuals who occupy similar positions in networks, we
also calculated the absolute differences between iand j’s collaboration cen-
trality in year t-1 (collaboration centrality difference). This allowed us to cap-
ture the possibility that highly central individuals may be suspicious of
reaching out to peripheral ones, so that ties form between individuals with a
3
All two-digit codes can be found at http://nces.ed.gov/pubs2002/cip2000/. We merged categories
51, Health Professions and Related Clinical Sciences, and 60, Residency Programs, into one cate-
gory because they are closely interrelated for researchers at the Medical School and separating
them would overestimate collaborations between researchers with different educational
backgrounds.
84 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

similar social standing. Grant funding is an important source of power within
universities (Pfeffer and Moore, 1980). We calculated the amount of grant
funding that faculty iand jraised, respectively, in t-1. People with grants are
often attractive collaborators as they can build groups around themselves.
We calculated the absolute differences between funding for iand jto capture
whether faculty form ties with people in a similar funding position (grant
resource difference).
Triadic closure. We posited that shared third parties constrain individuals
from breaking up (Simmel, 1950; Coleman, 1990: 318–319). To test H4, we
measured this effect by coding a dummy variable taking the value of 1 when an
indirect tie linked iand jin year t (indirect ties). A common problem in studies of
networks is the ‘‘boundary problem,’’ as one is forced to make an assumption
of where the network ends. To assume that ties end at Stanford is obviously
not realistic, and thus we searched extensively for information about external
collaborators. This allowed us to analyze how triadic closure may occur through
a shared collaborator inside and outside of Stanford.
Tie inertia. We proposed in H5 that ties, once in place, get a life of their
own. This creates inertia, a tendency to stick to ties that are already formed, an
effect that will carry depending on the strength of the ties and its multiplexity.
We measured tie strength by counting the number of times iand jcollaborated
on a publication in year t. To measure multiplexity, we investigated three net-
works: mentoring in dissertation committees, resource acquisitions in grant
teams, and knowledge creation in coauthored publications. We counted
whether iand jhad ties in the grant and dissertation network in year t. For
instance, a tie between two individuals in the publication network gets a score
of 1 if they had a tie in the dissertation or grant network and a score of 2 if they
had a tie in both networks.
Means-ends rationalization. After the first tie has formed, there are social
cues that individuals can use to decide whether to continue the relationship.
We argued in H6 that the success of an intellectual collaboration affects its per-
sistence. Hence, we formulated an experiential learning hypothesis that individ-
uals repeat activities that are deemed successful. We measured the outcome
success of a collaboration by counting the annual number of forward citations
per team member.
Control variables. We accounted for individual differences by controlling
for individual i’s and j’s tenure status, gender, and ethnicity, all of which may
influence the likelihood of collaborating. To account for individual unobserved
heterogeneity, we built the models stepwise and included individual fixed
effects. Also, the number of papers produced every year has increased over
time because there are more scholars pursuing research, and the competition
has increased productivity. We therefore included separate year dummies to
account for this trend.
Table 2 summarizes the variables used in the study, their definitions, and
the data sources used to construct them.
Dahlander and McFarland 85
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Table 2. Definition of Variables and Their Data Sources
Variable Description Data sources
Tie formation (publication) Dummy = 1 if the first publication tie
forms between individual iand jin
year t.
ISI Web of Knowledge
Tie persistence (publication) Dummy = 1 if the publication tie
between iand jis repeated in year t.
ISI Web of Knowledge
Tie formation (grant application) Dummy = 1 if the first grant application
tie forms between individual i and j in
year t.
University’s sponsored project
office
Tie persistence (grant application) Dummy = 1 if the tie grant application
between iand jis repeated in year t.
University’s sponsored project
office
Organizational foci
Same department Dummy = 1 if iand jare both in the
same department.
Official university records
Same center Dummy = 1 if iand jare both affiliated
with the same center.
Official university records and
manual coding
Status homophily
Same educational background Dummy = 1 if iand jreceived their
highest degree in the same two-digit
degree code.
Coded data on the subject area of
the highest degree
Gender similarity Dummy = 1 if iand jare the same
gender.
Official university records
Age difference Absolute difference in age between i
and j.
Official university records
Same ethnicity Dummy = 1 if iand jhave the same
ethnicity.
Official university records
Same tenure status Dummy = 1 if iand jhave the same
tenure status in year t-1.
Official university records
Value homophily
Reference similarity The intersection of iand j’s previous
references over the union in year t-1.
All references included in the ISI
publications
Cumulative advantage
Collaboration centrality difference Absolute difference in degree centrality
in and outside Stanford between iand
jin year t-1.
Official university records and ISI
Web of Knowledge
i/j’s collaboration centrality i/j’s degree centrality in year t-1.
Amount of grant resources difference Absolute difference in grant resources
between iand jin year t-1.
University’s sponsored project
office
i/j’s grant resources i/j’s grant funding in year t-1. University’s sponsored project
office
Triadic closure
Indirect ties Dummy = 1 if there is at least one
indirect tie between iand jin year t-1
through having a shared collaborator
at Stanford or at another university.
ISI Web of Knowledge
Tie inertia
Tie strength Number of published papers or applied
grants that iand jdid together in year
t-1.
Tie multiplexity Number of different networks in which
iand jhave a tie (dissertation, grant,
and publication network) in year t-1.
Official university records on
dissertation committees and grant
data from university’s sponsored
project office
(continued)
86 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Estimation Technique
To understand why some ties form and persist, we used a longitudinal dataset
in which we could observe the formation of new ties and their repetition over
time. We considered all ties created between individuals iand jin year t.To
avoid problems of left-censoring, we investigated newly hired, untenured
faculty members who arrived at Stanford fresh out of their Ph.D. programs.
We thus constructed all possible ties these faculty members could form with
other faculty members during our observation period. This resulted in a little
more than 5 million possible ties, though only a small fraction of these possible
ties actually formed.
We used a discrete-time survival analysis to estimate time to the first colla-
boration and, among those collaborations that did form, time to the second col-
laboration. In this approach, a possible dyad enters our analysis when both
faculty members were present at the university. They are considered ‘‘at risk’’
to collaborate in subsequent time periods. We included dummies for each year
following the initial exposure. For the tie persistence analysis, we used a similar
approach, but here we included dummies for each year after the tie was
formed. The time period in which a potential tie first exists in the data is set at
0, independent of the year it was formed. Here a tie enters the analysis condi-
tional upon its being formed. To analyze tie formation and tie persistence, we
used discrete-time survival models using logistic regression with indicator vari-
ables for each of the time periods. The discrete time hazard rate is the unob-
served rate at which events occur in the data (Singer and Willett, 1991, 2003).
Thus discrete-time hazard h
j
is the conditional probability that a tie will be
formed or repeated in time period j, given that the respective event has not
occurred prior to j:
hj=P½T=jjT≥j
where Tis the discrete random variable that indicates the time period jwhen
the event occurs.
Table 2. (continued)
Variable Description Data sources
Means-ends rationalization
Outcome success Number of forward cites of the articles
produced by iand jin year t.
ISI Web of Knowledge
Controls
Individual characteristics
i/j’s gender Dummy = 1 if i/jis male. Official university records
i/j’s tenure status Dummies = 1 if i/jfor each tenure
status in year t-1.
Official university records
i/j’s ethnicity Dummies = 1 for each ethnicity of i
and j.
Official university records
Exclusion restriction
Appointment year difference Absolute difference in appointment
year at Stanford between iand j.
Official university records
Dahlander and McFarland 87
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

The discrete-time model does not directly measure the duration to a termina-
tion event. The models are estimated by constructing datasets in which each
tie has a separate observation for each year that it was deemed part of the risk
set. Ties that were never repeated are considered right-censored. We dropped
ties after either faculty member ior jleft the university because after they left
we lacked information about their ties. We used a logistic regression with sepa-
rate time dummies to allow for disjunctures in the hazard rate. In our models,
we suppressed the constant in order to be able to include all time dummies
(Singer and Willett, 2003).
4
Burt (2000) used a similar approach in his analysis
of tie decay among managers within a firm, but with fewer time periods at his
disposal.
The analysis has two parts: one predicting tie formation, and another one
predicting tie persistence. The tie formation analysis obviously includes more
ties because it includes all potential ties that could form in any given year. The
tie persistence analysis was conditional upon the tie being formed and included
fewer ties.
5
In one of our models, we also accounted for the fact that the tie
persistence analysis is conditioned upon a tie being formed: ties have to form
to persist. We accounted for this by developing the inverse Mills ratio as a
selection parameter (see Polidoro, Ahuja, and Mitchell, 2011). To estimate this,
we used one exclusion restriction variable that affected tie formation, but not
persistence: the absolute difference in year of arrival at Stanford. When people
join organizations, networks are initially more open to collaborations, so people
who arrive at the same time are more likely to collaborate. This, however, had
no effect on the persistence of ties.
Our analysis used two different empirical strategies to overcome some
inherent challenges to studying networks. We first used an approach devel-
oped by Cameron, Gelbach, and Miller (2011) that allows for clustering of stan-
dard errors based on more than one variable. So rather than using Huber-White
standard errors clustered by each untenured faculty member i, this approach
allowed us to cluster standard errors on both faculty members iand j.
Kleinbaum, Stuart, and Tushman (2013) applied this approach when analyzing
e-mail exchange between people within an organization. This approach calcu-
lates standard errors in three separate covariance matrices: one clustering on
faculty member i, one on j, and one on the intersection. The reported standard
errors are clustered on both iand jand estimated on a matrix formed through
adding the first two covariance matrices and subtracting the third (Kleinbaum,
Stuart, and Tushman, 2013). Although this is a significant methodological
advancement, a potential concern about unobserved heterogeneity remains at
the individual level, including disciplinary differences in where the individuals
received their degrees and the possibility that some partners may be more
attractive. We coped with these challenges by using fixed individual effects for
all faculty members iand j, an estimation strategy that also accounts for non-
independence of the observations. The fixed-effects estimations drop observa-
tions when there is no underlying variation in the dependent variable (when ior
jis not collaborating).
4
Models without the constant but with all of the time dummies are equivalent to models with the
constant dropping one dummy; they are merely different parameterizations of the same model.
5
We estimated tie persistence in two ways. Ties that are repeated in t+1 were left out of the anal-
yses in subsequent time periods and when all repetitions of collaborations were considered.
88 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

RESULTS
Tables 3 and 4 show the descriptive statistics for the main variables concerning
tie formation and tie persistence, respectively. It is worth noting that there are
only modest correlations between our variables. Table 5 shows the results for
the tie formation analysis. We estimated tie formation with a discrete-time sur-
vival analysis of the time members of a given dyad had jointly worked at
Stanford (time hazard dummies are suppressed in the table).
Models 1 through 3 report the results predicting a publication tie being
formed between two faculty members. Model 1 is the baseline and model 2
includes all the independent variables, both using two-way clustered standard
errors for faculty member iand j. Model 3 includes the same variables as
model 2, but it estimates this with fixed individual effects to account for unob-
served heterogeneity. Model 3 thus drops observations consisting of individu-
als never forming ties and variables that are time invariant (such as gender and
ethnicity). Note that when we compare the coefficients in model 2 and model
3, the signs and magnitudes remain similar. To further analyze why ties form,
and whether publication ties are subject to a success bias because we only
observe published papers, we used the same factors to predict grant applica-
tion ties between faculty members. These grant ties include both attempts that
received funding and those that failed. We used the same strategy as
explained above and report the results in models 1B–3B.
Table 6 shows results from the tie persistence analyses. Again we used
discrete-time survival models in which the time hazard dummies (suppressed
in the table) of years since first collaboration estimate tie persistence. Because
this analysis is conditional upon a tie being formed, we first developed the
inverse Mills ratio in which we have an exclusion restriction variable. The first
three models used two-way clustered standard errors, and the last used indi-
vidual fixed effects. In models 4 and 5, we mimicked the approach for tie for-
mation. In model 6, we tested the idea that features of the tie itself affect its
Table 3. Descriptive Statistics for the Main Variables in Tie Formation Analysis of Publication
Ties*
Variable Mean S.D 1 2 3 4 5 6 7 8 9 10 11 12
1. Tie formation 0.00
2. Same department 0.02 .08
3. Same center 0.00 .01 .01
4. Same educational
background
0.09 .03 .20 .01
5. Same gender 0.62 .01 .01 −.01 .01
6. Same ethnicity 0.66 .00 −.01 .02 −.01 .05
7. Same tenure status 0.20 .00 .01 .03 .00 .03 .05
8. Reference similarity 0.00 0.00 .11 .10 .02 .07 .00 −.01 .01
9. i’s degree centrality 25.70 68.25 .02 .00 .02 .05 .04 −.05 .08 .05
10. j’s degree centrality 34.80 76.92 .02 .01 .01 .01 .02 −.03 −.03 .04 .00
11. i’s grant amount 1302519 2913314 .00 .02 .03 .07 .03 .08 .23 .02 .13 .02
12. j’s grant amount 2884113 6100788 .00 .00 .02 .00 .02 .02 .01 .02 .01 .12 .01
13. Indirect tie 0.01 .14 .11 .02 .11 .02 .00 .10 .17 .18 .15 .07 .02
*The descriptive statistics for grant application ties are not reported to conserve space.
Dahlander and McFarland 89
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

future continuation, namely, the strength of ties and their multiplexity and
whether successful collaborations are more likely to persist. Finally, model 7
includes fixed effects for individuals iand j. We also repeated this analysis for
grant application ties reported in models 4B–7B. In previous sections we theo-
rized about the differences between tie formation and tie persistence, so we
drew on the final models from tables 5 and 6 to interpret our findings.
Organization foci. We argued that people who are exposed to shared
organizational foci are more likely form a tie and persist in a relationship. Two
such foci are particularly salient in the context of universities: departments and
research centers. We theorized in H1 that this effect would be salient for tie
formation but not for tie persistence. Our results partly confirmed this
hypothesis. The coefficients for both the same department and research
center were significant for tie formation. But only same department had a
positive effect (although smaller) on tie persistence. The results are consistent
across both publication and grant application ties. This suggests that shared
organizational foci expose individuals to one another but do not necessarily
promote persistent ties. In many regards these results suggest that work
collaborations follow the golden rule of ‘‘out of sight, out of mind.’’ They are
also consistent with Reagans’ (2011) finding that foci create significantly
Table 4. Descriptive Statistics for the Main Variables in the Tie Persistence Analysis of
Publication Ties
Variable Mean S.D 1234567
1. Tie persistence 0.23
2. Same department 0.34 .10
3. Same center 0.03 −.01 −.02
4. Same educational background 0.44 .02 .22 −.02
5. Same gender 0.76 .01 −.02 .00 −.05
6. Same ethnicity 0.66 −.06 .01 .06 .05 .07
7. Same tenure status 0.38 −.09 .02 .06 −.02 .03 .08
8. Reference similarity 0.02 0.03 .08 .18 .04 .05 −.06 −.01 .04
9. i’s degree centrality 117.28 140.20 .17 −.08 −.05 −.08 .04 −.16 −.04
10. j’s degree centrality 125.99 152.51 .15 −.11 −.03 −.17 .03 −.10 −.02
11. i’s grant amount 392334 4846659 −.09 −.06 .02 .01 .02 .14 .27
12. j’s grant amount 7029847 14602201 −.03 −.08 .10 −.05 .03 .07 .14
13. Indirect tie 0.89 .07 −.02 −.05 .00 .05 −.04 .01
14. Cites per year by author 10.06 41.12 −.05 −.08 −.02 −.04 .06 .03 −.02
15. Multiplex tie 0.47 0.66 .11 .22 .14 .06 .02 .09 .22
16. Tie strength 2.9 9.3 .29 .06 −.02 −.04 .00 −.09 .02
Variable 8 9 10 11 12 13 14 15 16
9. i’s degree centrality .01
10. j’sdegree centrality .05 .55
11. i’s grant amount .01 −.08 −.09
12. j’s grant amount .01 −.10 .17 .10
13. Indirect tie .04 .17 .15 .09 .04
14. Cites per year by author .01 .07 .13 −.03 .03 .05
15. Multiplex tie .17 −.15 −.09 .24 .15 −.03 −.06
16. Tie strength −.03 .34 .28 −.08 −.05 .07 −.02 −.03
90 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

stronger ties. The magnitude of the effect on tie persistence is weaker than it
is for tie formation, however, suggesting that people are willing to walk farther
to sustain an existing collaboration than they are to form a new one.
Status homophily. In line with our reasoning, we found that status homo-
phily appears to be a stronger predictor for explaining why ties form initially
than for explaining subsequent interactions. The coefficients for same Ph.D.
training, same gender, and same ethnicity had positive effects on a publication
tie being formed. The exception was same-tenure status, which was insignifi-
cant. These findings are largely consistent with McPherson, Smith-Lovin, and
Cook’s (2001) argument that similar social attributes breed association. For tie
persistence, the coefficients for the status homophily variables were all insignif-
icant. The only significant variable was same educational background, which
has a negative sign, suggesting a pattern of heterophily rather than homophily.
For grant application ties, same educational background was significant, but
same gender and ethnicity had no effect. Lending support to hypothesis 2a,
these results show that while status homophily predicts a tie’s formation, it
has no correlation with whether that tie persists.
Value homophily. We argued in hypothesis 2b that intellectual similarity
between individuals has a curvilinear relationship to tie formation and persis-
tence. In support of this argument, the main effect of reference similarity was
positive and significant, and reference similarity squared was negative and sig-
nificant. Although the negative squared term suggests that too much intellec-
tual similarity can be detrimental to tie formation and persistence, there were
relatively few observations in this part of the distribution. We are therefore cau-
tious about the negative returns to similarity. Our results do support the
hypothesis of diminishing returns to intellectual similarity on tie formation and
persistence. The pattern is consistent across both publication and grant applica-
tion ties.
Cumulative advantage. We assessed hypothesis 3, the cumulative advan-
tage hypothesis, by looking at individuals’ collaboration centrality (in the publica-
tion and grant network, respectively) and amount of grant resources. We
argued that learning benefits could be outweighed by coordination costs cre-
ated by having too many ties. Each additional collaboration has a potential
downside: it increases coordination costs and the difficulty of credibly maintain-
ing each relationship. In support of this argument, we found that the coeffi-
cients for collaboration centrality for individuals iand jwere positive, and
collaboration centrality squared was negative and significant. This suggests that
the initial benefits of a collaboration experience can turn into a disadvantage.
We also investigated the absolute difference in collaboration centrality for indi-
viduals iand jand found that new ties were more likely to form between indi-
viduals of similar positions. In other words, a well-connected individual is more
likely to collaborate with other well-connected individuals than to reach out to
someone who is more peripheral. The signs of the coefficients for tie persis-
tence were similar to those for tie formation, but the magnitudes of the main
effects were smaller. This implies that, for tie persistence, junior faculty
Dahlander and McFarland 91
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Table 5. Tie Formation Results for Different Estimation Methods*
Publication Tie Grant Application Tie
Model 1: Model 2: Model 3: Model 1B: Model 2B: Model 3B:
Variable
Two-way
clustered
standard
errors
Two-way
clustered
standard
errors
Fixed
individual
effects
Two-way
clustered
standard
errors
Two-way
clustered
standard
errors
Fixed
individual
effects
Hypothesis 1
Same department 1.800
•••
2.260
•••
2.179
•••
2.474
•••
(0.123) (0.064) (0.122) (0.054)
Same center 1.334
•••
0.904
•••
1.072
•••
0.649
•••
(0.285) (0.184) (0.317) (0.156)
Hypothesis 2a
Same educational background 0.735
•••
0.611
•••
0.691
•••
0.829
•••
(0.114) (0.059) (0.087) (0.046)
Same gender 0.308
•••
0.168
••
0.117
•
−0.041
(0.062) (0.078) (0.064) (0.063)
Same ethnicity 0.466
•••
0.119 0.511
•••
0.109
(0.065) (0.074) (0.074) (0.070)
Same tenure status −0.460
•••
0.0820 −0.363
•••
0.126
•
(0.119) (0.088) (0.110) (0.071)
Hypothesis 2b
Reference similarity 0.245
•••
0.232
•••
0.299
•••
0.303
•••
(0.040) (0.010) (0.023) (0.012)
Reference similarity squared −0.004
•••
−0.004
•••
−0.006
•••
−0.006
•••
(0.001) (0.000) (0.001) (0.000)
Hypothesis 3
i’s collaboration centrality 1.060
•••
0.672
•••
1.431
•••
1.356
•••
(0.109) (0.078) (0.196) (0.039)
j’s collaboration centrality 1.253
•••
1.000
•••
1.381
•••
1.301
•••
(0.083) (0.084) (0.103) (0.037)
Collaboration centrality difference −0.627
•••
−0.582
•••
−0.820
•••
−0.866
•••
(0.067) (0.026) (0.115) (0.022)
i’s grant resources −0.011 0.041 0.130
•••
0.096
•••
(0.069) (0.043) (0.045) (0.031)
j’s grant resources 0.455
•••
0.437
•••
0.477
•••
−0.255
••
(0.132) (0.143) (0.175) (0.104)
Grant resource difference −0.443
•••
−0.315
•••
−0.429
••
0.150
(0.131) (0.122) (0.179) (0.096)
i’s collaboration centrality squared −0.110
•••
−0.053
•••
−0.135
•••
−0.097
•••
(0.017) (0.012) (0.049) (0.006)
j’s collaboration centrality squared −0.127
•••
−0.075
•••
−0.093
•••
−0.065
•••
(0.016) (0.014) (0.030) (0.005)
Hypothesis 4
Indirect tie 1.577
•••
1.032
•••
1.220
•••
1.247
•••
(0.175) (0.072) (0.125) (0.077)
Control variables
iis untenured −1.633
•••
−1.357
•••
−0.124 −1.671
•••
−1.380
•••
−0.102
(0.139) (0.161) (0.104) (0.130) (0.146) (0.094)
iis clinical faculty −0.754
••
−0.566
•
0.0995 0.016 −0.301 0.277
(0.345) (0.336) (0.242) (0.282) (0.190) (0.200)
iis male 0.189 −0.126 0.177 −0.106
(0.159) (0.149) (0.155) (0.124)
jis untenured −0.635
•••
−0.243
•
−0.232 −0.453
•••
−0.091 −0.629
•••
(0.131) (0.125) (0.170) (0.119) (0.110) (0.134)
(continued)
92 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

members are on the lookout for well-connected individuals, but this is more
important for tie formation than for tie persistence.
Grants are important sources of power within universities (Salancik and
Pfeffer, 1974). We proposed that individuals with grant resources would be
attractive collaborators. In support of this reasoning, we found that individuals
with grants were more likely to form ties. The same effect did not apply for per-
sistence, however, suggesting that grant resources attract people to collabo-
rate, but not necessarily to build persistent ties. The exception is junior faculty
members’ collaborators in the publication network. Our results were consistent
for both publication and grant application ties.
Triadic closure. We hypothesized in H4 that indirect ties positively affect
both the formation and persistence of ties (Simmel, 1950; Coleman, 1990: 318–
319). These indirect ties include collaborators who are at Stanford and other cam-
puses. Our results suggest that triads have a tendency to close, and that indirect
ties are an important means to discover potential partners, but the presence of
an indirect tie between individuals iand jhad no effect on the persistence of ties.
The results were similar for both publication and grant application ties.
Tie inertia. We next theorized in H5 that once a tie forms, it gets a life of its
own (Stinchcombe, 1965). When two persons are in a relationship, they have
more information with which to evaluate that relationship and weigh its future.
With regard to tie inertia, we proposed that, because multiplex ties span sev-
eral different activities, they are more likely to be repeated. Our results sup-
ported this hypothesis: the coefficient for multiplex ties was positive and
significant (net of strength). As a tie changes from being uniplex to being multi-
plex, the probability that it repeats increases. Our results also confirmed our
assertion that strong ties are more likely to be renewed. We separated tie
strength with dummies (baseline is one publication or grant), to see if there is
an optimal number. For publication and grant collaborations, stronger ties
Table 5. (continued)
Publication Tie Grant Application Tie
Model 1: Model 2: Model 3: Model 1B: Model 2B: Model 3B:
Variable
Two-way
clustered
standard
errors
Two-way
clustered
standard
errors
Fixed
individual
effects
Two-way
clustered
standard
errors
Two-way
clustered
standard
errors
Fixed
individual
effects
jis clinical faculty −0.0735 −0.165 −0.148 −0.559
•••
−0.352
•••
−1.138
•••
(0.127) (0.124) (0.282) (0.128) (0.112) (0.203)
jis male −0.327
•••
−0.470
•••
−0.321
•••
−0.429
•••
(0.089) (0.085) (0.087) (0.091)
i’s ethnic dummies Yes Yes Yes Yes
j’s ethnic dummies Yes Yes Yes Yes
Year dummies Yes Yes Yes Yes Yes Yes
Number of observations 5883315 5883315 1455515 5864637 5864637 1306475
•
p<10;
••
p<.05;
•••
p<.01; two-tailed tests.
*Standard errors are in parentheses. Time-hazard dummies are suppressed.
Dahlander and McFarland 93
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Table 6. Tie Persistence Results for Different Estimation Methods*
Publication Tie with Repeated Events Grant Application Tie with Repeated Events
Model 4: Model 5: Model 6: Model 7: Model 4B: Model 5B: Model 6B: Model 7B:
Variable
Two-way
clustered
standard
errors
Two-way
clustered
standard
errors
Two-way
clustered
standard
errors
Fixed
individual
effects
Two-way
clustered
standard
errors
Two-way
clustered
standard
errors
Two-way
clustered
standard
errors
Fixed
individual
effects
Hypothesis 1
Same department 0.417
•••
0.333
••
0.545
••
0.868
•••
0.811
•••
0.835
•••
(0.115) (0.130) (0.215) (0.131) (0.134) (0.134)
Same center 0.293 0.159 −0.184 0.078 0.076 −0.013
(0.224) (0.243) (0.370) (0.264) (0.274) (0.211)
Hypothesis 2a
Same educational
background
−0.053 −0.101 −0.837
•••
−0.009 −0.010 −0.127
(0.128) (0.122) (0.206) (0.128) (0.123) (0.103)
Same gender 0.224 0.219 0.405 0.123 0.119 −0.162
(0.158) (0.166) (0.287) (0.120) (0.128) (0.141)
Same ethnicity −0.140 −0.082 −0.017 0.120 0.078 −0.011
(0.155) (0.173) (0.293) (0.133) (0.137) (0.162)
Same tenure status −0.173 −0.197 −0.099 −0.134 −0.149 −0.169
(0.138) (0.144) (0.185) (0.123) (0.124) (0.108)
Hypothesis 2b
Reference similarity 0.065
•••
0.056
•••
0.039
••
0.080
•••
0.077
•••
0.087
•••
(0.021) (0.020) (0.019) (0.018) (0.018) (0.012)
Reference similarity
squared
−0.002
•••
−0.001
••
−0.002
•••
−0.001
••
−0.001
••
−0.001
•••
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Hypothesis 3
i’s collaboration
centrality
0.479
•••
0.500
•••
0.783
•••
0.966
•••
0.973
•••
1.386
•••
(0.104) (0.097) (0.205) (0.114) (0.113) (0.047)
j’s collaboration
centrality
0.441
•••
0.444
•••
0.552
•••
0.907
•••
0.912
•••
1.408
•••
(0.103) (0.109) (0.182) (0.080) (0.080) (0.052)
Collaboration
centrality difference
−0.311
•••
−0.282
•••
−0.225
•••
−0.894
•••
−0.896
•••
−0.895
•••
(0.067) (0.077) (0.067) (0.147) (0.147) (0.037)
i’s grant resources −0.073 −0.112
•••
−0.056 −0.083 −0.091 0.016
(0.044) (0.044) (0.087) (0.059) (0.057) (0.040)
j’s grant resources 0.351
•
0.157 0.544
••
−0.114 −0.123 −0.111
(0.188) (0.161) (0.278) (0.158) (0.156) (0.139)
Grant resource
difference
−0.383
•
−0.182 −0.115 0.040 0.042 0.191
(0.203) (0.169) (0.240) (0.161) (0.158) (0.121)
i’s collaboration
centrality squared
−0.050
•••
−0.065
•••
−0.088
•••
−0.059
•••
−0.059
•••
−0.079
•••
(0.017) (0.019) (0.033) (0.018) (0.018) (0.006)
j’s collaboration
centrality squared
−0.056
•••
−0.060
•••
−0.050 −0.027
•
−0.028
•
−0.061
•••
(0.021) (0.023) (0.031) (0.015) (0.015) (0.007)
Hypothesis 4
Indirect tie 0.284
•
0.225 −0.308 −0.185 −0.204 0.085
(0.158) (0.158) (0.309) (0.123) (0.152) (0.102)
Hypothesis 5
Tie multiplexity 0.502
•••
0.923
•••
0.505
•••
0.820
•••
(0.166) (0.200) (0.184) (0.144)
Tie strength = 2
shared events
0.872
•••
1.123
•••
0.787
•••
1.056
•••
(0.139) (0.202) (0.134) (0.129)
Tie strength = 3
shared events
0.959
•••
0.018 1.339
•••
1.915
•••
(0.291) (0.395) (0.420) (0.339)
Tie strength = 4
or more
2.070
•••
3.137
•••
0.712 0.361
(0.241) (0.402) (0.450) (0.475)
(continued)
94 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

appear to be more persistent. This finding is consistent with prior work on tie
persistence (Levi Martin and Yeung, 2006), and the results are robust for both
publication and grant application ties.
Outcome success. By distinguishing between tie formation and tie persis-
tence, we were able to evaluate the success of the event that brought two indi-
viduals together. Hypothesis 6 proposed that successful publications would lead
to repeated collaboration because participants learn from the successes and fail-
ures of their past collaboration (March, 1999), but we found no support for this
relation, as all coefficients are non-significant (and we tried various polynomial
forms). The effect of prior collaboration success is not a factor in tie persistence.
This association is likely ambiguous because success can reinforce a tie and/or
lead others to invite people into new collaborations. The latter likely reduces
their chance of repeating any single collaboration, while the former increases it.
Robustness Checks
We conducted several robustness checks to strengthen our inferences. First,
we varied the precision of our name-matching procedures for publications
Table 6. (continued)
Publication Tie with Repeated Events Grant Application Tie with Repeated Events
Model 4: Model 5: Model 6: Model 7: Model 4B: Model 5B: Model 6B: Model 7B:
Variable
Two-way
clustered
standard
errors
Two-way
clustered
standard
errors
Two-way
clustered
standard
errors
Fixed
individual
effects
Two-way
clustered
standard
errors
Two-way
clustered
standard
errors
Two-way
clustered
standard
errors
Fixed
individual
effects
Hypothesis 6
Cites per year −0.147 −0.225 −0.025 0.014
(0.106) (0.144) (0.023) (0.030)
Control variables
iis untenured 0.451
•••
0.273
•
0.218 −0.007 0.072 −0.159 −0.201 −0.029
(0.127) (0.146) (0.146) (0.212) (0.142) (0.175) (0.179) (0.138)
iis clinical faculty 0.264 0.482
••
0.419
••
1.183
•••
0.282 −0.325 −0.326 −0.140
(0.276) (0.199) (0.184) (0.382) (0.383) (0.243) (0.248) (0.333)
iis male 0.110 −0.057 −0.122 −0.141 −0.321 −0.297
(0.235) (0.255) (0.180) (0.142) (0.207) (0.214)
jis untenured 0.408
•
0.447
••
0.415
•
0.458 −0.141 0.028 0.022 −0.003
(0.212) (0.214) (0.233) (0.379) (0.134) (0.154) (0.153) (0.215)
jis clinical faculty 0.042 0.062 0.031 −0.278 −0.361
•••
−0.124 −0.139 −0.752
•
(0.133) (0.125) (0.127) (0.645) (0.138) (0.139) (0.139) (0.411)
jis male 0.091 −0.108 −0.190 −0.102 −0.371
•••
−0.380
•••
(0.160) (0.156) (0.173) (0.091) (0.138) (0.144)
i’s ethnic dummies Yes Yes Yes Yes Yes Yes
j’s ethnic dummies Yes Yes Yes Yes Yes Yes
Year dummies Yes Yes Yes Yes Yes Yes Yes Yes
Number of
observations
6430 6430 6430 4407 25158 25158 25158 20517
•
p<10;
••
p<.05;
•••
p<.01; two-tailed tests.
*Standard errors are in parentheses. Time hazard dummies are suppressed. The results account for fact that tie
persistence is conditioned upon a tie being formed by including the inverse Mills ratio.
Dahlander and McFarland 95
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

(Newman, 2001) and assessed the results, without finding any significant dif-
ferences (see Appendix B). Second, we estimated our models with comple-
mentary log-logistic regressions, without any substantial change in the results.
Third, we tackled a methodological issue that arises when converting affiliation
networks to one-mode data. Specifically, when large teams of individuals colla-
borate together, the conversion to one-mode network data suggests there is a
fully connected clique even though it is likely that all of these people did not
work together very closely. For instance, norms in physics are such that teams
sometimes involve dozens of researchers. It is unclear how much the n’th indi-
vidual contributes. Our approach was to reconstruct networks by ignoring
those events with more than 15 faculty members. Some work compares team
assembly mechanisms with randomly assembled teams that could have
formed but did not (see, e.g., Ruef, Aldrich, and Carter, 2003; Ingram and
Torfason, 2010), and as another robustness check, we therefore separated the
analysis into collaborations formed between two faculty members versus colla-
borations formed among three or more. Had we missed something fundamen-
tal when converting the two-mode data to one-mode data, we would expect
differences between the analysis of dyads and larger teams. Instead, all of our
robustness checks rendered remarkably consistent results, which strength-
ened our inferences. All of our alternative estimations are available upon
request from the first author. Some of these precautions also have implications
for future research, which we elaborate on in the discussion.
DISCUSSION
We began this paper with a simple observation. A large body of research attri-
butes positive outcomes to collaborations (Brass et al., 2004). In response,
research has begun to explore where new ties come from, but it has paid sig-
nificantly less attention to how those ties persist. Although some research has
examined tie formation and persistence between organizations (Gulati and
Gargiulo, 1999; Broschak, 2004; Powell et al., 2005) and even nations (Ingram
and Torfason, 2010), there has been considerably less research on how intraor-
ganizational collaborations form and persist. To address this gap, we focused
on intraorganizational collaborations occurring within universities as faculty pub-
lish articles and apply for grants. We developed a range of different explana-
tions for the formation and persistence of these task relationships, and we
were able to empirically test and compare them using a unique dataset on
faculty collaborations. While it has been suggested that similar factors shape
tie formation and persistence (McPherson, Smith-Lovin, and Cook, 2001), our
contribution has been to identify multiple reasons why this is not necessarily
the case. Some cues relevant to persistence are not available for tie formation.
Tie formation and tie persistence are only partly characterized by the same fea-
tures and with different magnitudes and patterns of association.
Tie formation is the result of multiple factors: shared organizational foci, sta-
tus homophily (same ascribed and achieved traits), value homophily (intellectual
similarity), and characteristics of cumulative advantage (centrality and resource
richness). Tie persistence results from some of the same factors, such as
shared organizational foci and value homophily, but they have less relevance.
Moreover, tie persistence does not correspond with status homophily or cumula-
tive advantage. People do not look to individual traits and grant resources when
96 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

deciding whether collaborations are worth repeating or not. Instead, they focus
on their experiences in the dyad. If the dyad entails sunk costs and layered social
obligations (strength of tie, multiplexity), they are more likely to sustain it. This
pattern of results helps explain the difference between processes of tie forma-
tion and tie persistence. Collaborations form in a context in which people
approach a broad assortment of unfamiliar potential partners and select those
who are proximate, have identical traits and similar knowledge, and who exhibit
a degree of social success and evidence of interpersonal trust. That is, tie forma-
tion is mostly a function of opportunity and preference. By contrast, collabora-
tions persist in a context of familiar partners, and they are sustained when the
individuals are somewhat proximate (not too far apart), have similar knowledge
(but not identical), and a shared sense of dyadic history. That is, tie persistence
is more a function of obligation and complementary experience than opportunity
and preference selection. With formation, the results identified a context that is
more uncertain because there is less firsthand information. With persistence,
the results identified a context that is more certain and in which people make
substantive reflections on experience from the tie itself (strength and multiplex-
ity). People tend to stick to the ties they have formed, for better or worse, espe-
cially stronger ties that are multiplex and span multiple types of association.
These findings are important when we consider the effects or returns that
new and persistent ties have on salient outcomes in organizations. In supple-
mental analyses, we found that persistent ties had greater returns on the rate
of productivity and quality of performance than did new ties. That said, it is self-
evident that tie formation and tie persistence are important aspects of organiza-
tional life. A successful portfolio of ties likely includes a mix of new and persis-
tent ties, but research has been slow to specify their differences, and in turn,
the different ways to manage their occurrence. Our baseline expectation was
that ties are likely to be repeated over time (Kollock, 1994). But as Larson (1992:
99) noted, ‘‘the relative stability of . . . ties should not obscure their inherent vul-
nerabilities,‘‘ and ‘‘subsequent research should include analyses of how and
why these organizational forms disappear.’’ This article reveals that not only is
tie persistence valuable, it is problematic and worth study on its own.
Our results are partly a function of the pool of potentially repeatable ties
being different from the overall pool of potential ties (March and March, 1977).
When forming collaborations, individuals winnow down the pool of potential
ties, resulting in a narrow pool of collaborations that one decides whether to
sustain or not. By implication this means that some individuals are locked out
of fruitful collaborations, while others become less open to collaborations with
attractive potential partners (Uzzi, 1996). For instance, we saw that two individ-
uals will choose to collaborate based on propinquity and homophily. This means
that the remaining pool of ties that have the potential to be repeated will be
more proximate and homogenous to begin with. We also saw that people tend
to select individuals who are central and well connected. This means that the
pool of repeatable ties is not only more local and homogenous but is also
geared toward established members. Both of these factors make it inherently
difficult for newcomers to establish ties. Young faculty members will likely
struggle to find partners because the most productive ones are already too
busy with their current collaborations. Academics therefore face a trade-off
between attaching themselves to those perceived as desirable partners and
attaching themselves to those who are available (Gould, 2002).
Dahlander and McFarland 97
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

A similar implication follows for interdisciplinary collaborations. Studies in the
sociology of science have documented a sharp rise in research collaborations
across the disciplines in recent decades (e.g., Wuchty, Jones, and Uzzi, 2007).
Many have even propounded the benefits of highly collaborative team science
(see Rawlings and McFarland, 2011, for a review). Despite the virtues of inter-
disciplinary collaborations, many of them fail to form, and those that do form
often fall apart. Our results show that interdisciplinary collaborations are often
distant ties involving different types of academics. To form and sustain these
ties, pairs of colleagues must interact frequently to share knowledge. And even
then, the collaboration will need to be layered with multiple work activities
(grants, student training, co-teaching, and publishing) and experience some suc-
cess if it is to continue. Should this be accomplished, sustained interdisciplinary
collaborations may have their proclaimed returns (Stokols et al., 2008). This is
perhaps why interdisciplinary centers may be useful organizational means of
corralling faculty and promoting continued distant collaborations.
Our findings on tie formation and tie persistence also have implications for
how managers can generate persistent ties in their firms. Mixers, speed dates,
and special forums have been explored as methods to encourage previously
unconnected actors to affiliate (Ingram and Morris, 2007). The study by Ingram
and Morris showed the inherent difficulty of establishing new ties because indi-
viduals fall back on existing relationships. Given the returns of persistent ties,
our study suggests that organizations should encourage tie persistence by
adopting activities that look less like mixers and more like team-building exer-
cises. Extended retreats at off-site locations, ropes courses, and the like are all
designed to encourage communication and interdependence, get past initial
surface problems, forge trust, and develop dyadic and team identities
(Hackman and Katz, 2010). The literature on teams frequently draws on classic
social psychology to describe how stable, cohesive teams can form: to become
stable teams, groups need time and interaction to work out surface problems
and come to agreed-upon goals; they need to perform activities that require
interdependent roles, afford collective rewards, and form a shared identity; and
the interactions need to encourage members to learn how to coordinate their
individual abilities so as to maximize team performance (Forsyth, 1990: 104–
105). Many of the work activities we list concern shared endeavors (e.g., co-
writing or advising), and our analyses of persistent ties highlight the effects of
repeated interaction (strong ties) and complementarity (multiplexity and value
homophily). The social psychological literature on teams, however, is more
focused on establishing affective bonds than instrumental, trusted collabora-
tions that are mutually fulfilling. As such, retreats for collaborators may need to
include activities in which participants learn about different sides of their col-
leagues and thereby generate a more varied sense of value in one another. It
may require asking collaborators to perform an assortment of different types of
collaborative activities (e.g., designing a program, a party, or a research project).
One might also find ways to reward collaborations that often go unnoticed or
unrewarded. In some sense, interdisciplinary centers do this by bringing visibi-
lity to collaborations that may span groups and do not get the recognition of dis-
ciplinary journals.
98 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Limitations and Future Studies
We have presented a thorough effort to analyze the formation and persistence
of faculty collaborations. While this focus limits our empirical claims to intraor-
ganizational task relationships, we believe they extend readily to informal and
unplanned work relations that emerge in the context of organizations. The net-
work of informal work relations has long been found to correspond with per-
sonal ties like friendship, so we expect our findings will be consistent when
extended to faculty friendship relations within the university and possibly
friendships in firms more generally (Krackhardt, 1992). We leave it to future
work, however, to reveal whether the same mechanisms of relational forma-
tion and persistence hold for these subtly different types of ties.
We have attempted to be thorough in our analyses and to consider multiple
alternative explanations. But like all research, this paper entails certain empirical
limitations. In some cases, these limitations point toward fruitful future lines of
research. For example, our work examines collaborations at a single university
from which we acquired rich, longitudinal data on multiple networks. Like prior
work on this topic (see Burt, 2000, 2002), we were forced to draw boundaries
around a network and sacrificed a degree of generalizability for a greater rich-
ness of detail. Regardless, we believe the factors elucidated are more broadly
generalizable in the presence of the following scope conditions: (1) individuals
can choose with whom to create a tie, (2) there are no norms that prevent indi-
viduals from having ties with more than one other individual, and (3) there are
tangible or intangible benefits participants can draw from ties. These scope
conditions are present for many types of ties in organizations but do not extend
to all social ties. Marriage, for example, does not fulfill these conditions
because most Western societies strongly support monogamous relationships.
By focusing on a single university we were unable to take into account the
fact that the rate of tie formation and tie persistence differs across organiza-
tions and types of ties (e.g., research universities create more publications and
coauthors at a faster rate than teaching universities), but this potential concern
only moves the general cumulative survival function of ties up and down. In
supplementary analyses, we conducted similar analyses of the formation and
persistence of a different type of tie—dissertation collaborations—with remark-
ably similar results. Future work would do well to study our models of tie for-
mation and persistence for multiple universities and firms when such data
become available.
Our focus on the repetition of coauthoring papers may lead us to miss
instances of failed ties, through which individuals repeatedly try to publish their
papers and fail. But we also analyzed grant applications (failed and successful),
and these show very consistent results with those on publication, lending sup-
port to the notion that published papers afford a visible trail of research colla-
borations we can follow. That said, future work could elaborate more on the
asymmetries of collaboration, examining the conditions under which some
scholars decline to collaborate with someone else.
Our analyses concern dyadic ties at the expense of collaboration in teams.
Although most collaborations are pairwise, many entail three or more partners.
Some research has moved in the direction of analyzing team assembly mechan-
isms (e.g., Ruef, Aldrich, and Carter, 2003; Ingram and Torfason, 2010), and
there should be many opportunities to analyze this in the context of research
Dahlander and McFarland 99
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

collaborations. Our robustness checks comparing dyads with larger teams sug-
gest that these are similar, but using the team as the unit of analysis can pro-
duce additional theoretical leverage in future work. In future research, authors
could focus specifically on teamwork and consider the different ways in which
tie persistence operates in teams, given that the ‘‘certainty’’ generated from a
collective may have different effects than the quality or products of dyadic ties.
A potential limitation of our network approach is that structural perspectives
overlook the effects of individual personalities. Collecting psychological infor-
mation on over three thousand people over fifteen years may be a desirable
direction for future research, but we found it difficult to get decent response
rates from faculty. Therefore we sought to account for personality effects using
fixed individual effects. Psychologists who study personality consider traits as
relatively stable over time. Costa and McCrae (1992), for example, showed that
extroversion can be considered a trait that remains stable over time, with test-
retest correlations of about .8. As such, our models that include fixed individual
effects account for much, albeit not all, of these traits. While we controlled for
unobserved individual heterogeneity, future research could theorize directly
about individual differences. For instance, extroverted individuals are likely to
form more ties, but are their collaborations also more persistent? In this stream
of research, Mehra, Kilduff, and Brass (2001) gained important ground in under-
standing how high and low self-monitors can be integrated to understand struc-
tural advantages within organizations. There is room to develop these ideas
even further, but to some extent, testing them will remain a difficult task
because studies of tie formation and tie persistence require longitudinal data.
Although we measured tie strength and multiplexity, it is practically impossi-
ble to capture all factors affecting tie formation and persistence. It is thus
important to consider the possibility of omitted variables. One area for future
research is to theorize about relational histories and how conflict events and
setbacks are dealt with in collaborations. This would likely require detailed eth-
nographic information about collaborations over time (see, e.g., Owen-Smith,
2001). More cultural perspectives would likely afford a more agent-centric and
interactional view of tie formation and persistence than the structural one we
provide (Emirbayer and Goodwin, 1994), though we do not believe they will be
antithetical. In the aggregate, these moment-to-moment moves should align
with structural conditions.
A limitation of our work is the difficulty of capturing the content of ties and
interactional instances salient to their formation and persistence. For instance,
it is plausible that rare violations of trust and confidence undermine a tie’s per-
sistence. In this study, we were unable to capture behaviors that color relation-
ships or the interactions that guide the course of a tie’s history. A fruitful venue
for future empirical research is to examine how interactions within a dyad
affect the formation and persistence of ties. This can also reveal the relative
effect of interactions as compared with structural factors (Butts, 2008).
Some work has begun to integrate cultural and structural perspectives of
networks (Ruef, 2002; Lizardo, 2006). Such studies unearth how the benefits
associated with the structure and content of networks are contingent on the
culture in which they are embedded (Xiao and Tsui, 2007; Morris, Podolny, and
Sullivan, 2008). The most salient cultural dimension in our context is the disci-
pline, which affects the questions people ask and how research is pursued and
evaluated. While we accounted for disciplinary effects in our models using
100 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

fixed individual effects, future work could theorize more explicitly about these
differences.
There are several promising areas for future research. Kilduff, Tsai, and
Hanke (2006: 1039) proposed that ‘‘the social network . . . exists as layers upon
layer of relations, built up over time and space in the cognitions of members.’’
Indeed, individuals who have left the network can still influence it by becoming
examples for the remaining members of what can be achieved. Future studies
would benefit from investigating what happens when individuals who were
previously part of the same network are separated. What is the rate of decay
of these ties compared with others? Shared organizational foci had a strong
effect in our study, and there are reasons to believe that ties spanning universi-
ties are fragile. Such ties transcend institutional boundaries and require greater
efforts of coordination and communications, placing them at higher risk of dis-
solution. In a world characterized by high mobility between universities, this
appears to be an important facet for future work to consider.
Another area for future research may reside in the interaction between tie
formation and tie persistence. We have shown that tie formation and tie persis-
tence emanate from different factors, but our analysis downplayed the interac-
tion between these two processes. For instance, in a context in which
potential new collaborators are widely available, the obligation to maintain old
ties may not be as strong. In contrast, when chances for new collaboration are
low, old ties may become more salient. Tie formation and tie persistence are
conceptually separated in our empirical analysis because an analysis of persis-
tence requires that the tie has come into existence. A fruitful area for future
research would be to theorize and analyze the mutual constraints between tie
formation and tie persistence. This would be particularly fruitful in contexts with
great variation in the availability of potential partners, as this would affect the
interaction between tie formation and persistence.
We have focused on the persistence of ties, but persistent ties are not an
organizational panacea. A strategy built on sustaining existing ties can deter
people from forming new collaborations that are more beneficial. In fact, when
it is difficult to locate attractive potential collaborators, people will stay in their
current collaborations even when there are better available matches (Cohen,
March, and Olsen, 1972). Hence, it is likely that most organizations will require
a mixture of efforts aimed at exploring new ties and further exploiting the
potential of current ones (March, 1991).
Finally, this paper concerns intraorganizational task relationships, such as
work collaborations. In many instances, these collaborations reflect the infor-
mal organization of a firm (and invisible colleges for universities) and therefore
resemble personal ties to a great extent. In fact, supplemental work in this
paper showed that the more multiplex these collaborations, the more they are
recognized as close contacts. Hence, many of the mechanisms identified here
likely apply to personal relationships, but we leave it to future work to carefully
disentangle how the mechanisms of informal work relations differ from those
of personal relations and whether those mechanisms foster the persistence of
the collaborative ties that become critical organizational resources.
Acknowledgments
Associate Editor Martin Ruef, the anonymous reviewers, Matt Bothner, David Diehl,
Sasha Goodman, Adam Kleinbaum, Jim March, Woody Powell, Craig Rawlings, Catalina
Dahlander and McFarland 101
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Stefanescu-Cuntze, Dan Wang, Karl Wennberg, Mimir Project members, and the
Stanford Networks and Organizations Workshop provided useful comments and ideas.
Linda Johanson provided invaluable editorial assistance. These data were collected by
the Mimir Project conducted at Stanford University by Daniel A. McFarland, Dan
Jurafsky, Chris Manning, and Woody Powell. This article is based on work supported by
the Office of the President at Stanford University and the National Science Foundation
under Grant No. 0835614. Any opinions, findings, conclusions, and recommendations
expressed in this article are those of the authors and do not necessarily reflect the views
of Stanford University or the National Science Foundation.
REFERENCES
Abbott, A.
1988 The System of Professions: An Essay on the Division of Expert Labor. Chicago:
University of Chicago Press.
Abbott, A.
2001 Chaos of Disciplines. Chicago: University of Chicago Press.
Ahuja, G.
2000 ‘‘Collaboration networks, structural holes and innovation: A longitudinal study.’’
Administrative Science Quarterly, 45: 425–455.
Azoulay, P., W. Ding, and T. Stuart
2009 ‘‘The effect of academic patenting on the rate, quality, and direction of (public)
research output.’’ Journal of Industrial Economics, 57: 637–676.
Azoulay, P., J. Graff Zivin, and J. Wang
2010 ‘‘Superstar extinction.’’ Quarterly Journal of Economics, 125: 549–589.
Baker, W. E., R. R. Faulkner, and G. A. Fisher
1998 ‘‘Hazards of the market: The continuity and dissolution of interorganizational
market relationships.’’ American Sociological Review, 63: 147–177.
Baldassari, D., and M. Diani
2007 ‘‘The integrative power of civic networks.’’ American Journal of Sociology, 113:
735–780.
Basalla, G.
1988 The Evolution of Technology. Cambridge: Cambridge University Press.
Blau, P. M.
1964 Exchange and Power in Social Life. New York: Wiley.
Blau, P. M.
1973 The Organization of Academic Work. New York: Wiley.
Bothner, M., J. M. Podolny, and E. B. Smith
2011 ‘‘Organizing contests for status: The Matthew vs. the Mark effect.’’ Manage-
ment Science, 57: 439–457.
Brass, D. J., J. Galaskiewicz, H. R. Greve, and W. Tsai
2004 ‘‘Taking stock of networks and organizations: A multilevel perspective.’’ Acad-
emy of Management Journal, 47: 795–817.
Broschak, J. P.
2004 ‘‘Managers’ mobility and market interface: The effect of managers’ career
mobility on the dissolution of market ties.’’ Administrative Science Quarterly, 49:
608–640.
Burt, R. S.
1980 ‘‘Models of network structure.’’ Annual Review of Sociology, 6: 79–141.
Burt, R. S.
1992 Structural Holes: The Social Structure of Competition. Cambridge, MA: Harvard
University Press.
Burt, R. S.
2000 ‘‘Decay functions.’’ Social Networks, 22: 1–28.
102 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Burt, R. S.
2002 ‘‘Bridge decay.’’ Social Networks, 24: 333–363.
Butts, C. T.
2008 ‘‘A relational event framework for social action.’’ Sociological Methodology, 38:
155–200.
Cameron, A. C., J. B. Gelbach, and D. L. Miller
2011 ‘‘Robust inference with multi-way clustering.’’ Journal of Business and Eco-
nomic Statistics, 29 (2): 238–249.
Cohen, M. D., J. G. March, and J. P. Olsen
1972 ‘‘A garbage can model of organizational choice.’’ Administrative Science Quar-
terly, 17 (1): 1–25.
Coleman, J. C.
1974 Power and the Structure of Society. New York: Norton.
Coleman, J. C.
1988 ‘‘Social capital in the creation of human capital.’’ American Journal of Sociology,
94: 95–120.
Coleman, J. C.
1990 Foundations of Social Theory. Cambridge, MA: Belknap Press of Harvard Uni-
versity Press.
Coser, L. A.
1974 Greedy Institutions: Patterns of Undivided Commitment. New York: Free Press.
Costa, P. T., and R. R. McCrae
1992 ‘‘Normal personality assessment in clinical practice: The NEO personality inven-
tory.’’ Psychological Assessment, 4: 5–13.
De Solla Price, D.
1963 Little Science, Big Science. New York and London: Columbia University Press.
Emirbayer, M., and J. Goodwin
1994 ‘‘Network analysis, culture, and the problem of agency.’’ American Journal of
Sociology, 99: 1411–1454.
Feld, S.
1981 ‘‘The focused organization of social ties.’’ American Journal of Sociology, 86:
1015–1035.
Festinger, L., S. Schacter, and K. Bach
1950 Social Pressure in Informal Groups. New York: Harper.
Fischer, C. S.
1982 To Dwell Among Friends. Chicago: University of Chicago Press.
Forsyth, D. R.
1990 Group Dynamics. Pacific Grove, CA: Brooks/Cole Publishing.
Friedkin, N. E.
1978 ‘‘University social structure and social networks among scientists.’’ American
Journal of Sociology, 83: 1444–1465.
Gould, R. V.
1991 ‘‘Multiple networks and mobilization in the Paris Commune, 1871.’’ American
Sociological Review, 56: 716–729.
Gould, R. V.
2002 ‘‘The origins of status hierarchies: A formal theory and empirical test.’’ Ameri-
can Journal of Sociology, 107: 1143–1178.
Granovetter, M.
1973 ‘‘The strength of weak ties.’’ American Journal of Sociology, 78: 1360–1380.
Granovetter, M.
1983 ‘‘The strength of weak ties: A network theory revisited.’’ Sociological Theory, 1:
201–233.
Gulati, R., and M. Gargiulo
1999 ‘‘Where do interorganizational networks come from?’’ American Journal of
Sociology, 104: 1439–1493.
Dahlander and McFarland 103
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Hackman, J. R., and N. Katz
2010 ‘‘Group behavior and performance.’’ In S. T. Fiske, D. T. Gilbert, and G. Lindzey
(eds.), Handbook of Social Psychology, 2: 1208–1251. Hoboken, NJ: Wiley.
Hallinan, M. T., and R. Williams
1989 ‘‘Interracial friendship choice in secondary schools.’’ American Sociological
Review, 54: 67–78.
Ibarra, H.
1992 ‘‘Homophily and differential returns: Sex differences in network structure and
access in an advertising firm.’’ Administrative Science Quarterly, 37: 422–447.
Ibarra, H.
1993 ‘‘Network centrality, power, and innovation involvement: Determinants of tech-
nical and administrative roles.’’ Academy of Management Journal, 36: 471–501.
Ingram, P., and M. W. Morris
2007 ‘‘Do people mix at mixers? Structure, homophily, and the ‘life of the party’.’’
Administrative Science Quarterly, 52: 558–585.
Ingram, P., and M. T. Torfason
2010 ‘‘Organizing the in-between: The population dynamics of network-weaving orga-
nizations in the global interstate network.’’ Administrative Science Quarterly, 55:
577–605.
Jackson, M. O.
2008 Social and Economic Networks. Princeton, NJ: Princeton University Press.
Jacobs, J., and S. Frickel
2009 ‘‘Interdisciplinarity: A critical assessment.’’ Annual Review of Sociology, 35: 43–65.
Kagan, J.
2009 The Three Cultures: Natural Sciences, Social Sciences, and the Humanities in
the 21st Century. New York: Cambridge University Press.
Katz, R.
1982 ‘‘The effects of group longevity on project communication and performance.’’
Administrative Science Quarterly, 27: 81–104.
Kilduff, M., W. Tsai, and R. Hanke
2006 ‘‘A paradigm too far? A dynamic stability consideration of the social network
research program.’’ Academy of Management Review, 31: 1031–1046.
Kleinbaum, A. M., T. E. Stuart, and M. L. Tushman
2013 ‘‘Discretion within the constraints of opportunity: Gender homophily and struc-
ture in a formal organization’’ Organization Science (forthcoming).
Kollock, P.
1994 ‘‘The emergence of exchange structures: An experimental study of uncertainty,
commitment, and trust.’’ American Journal of Sociology, 100: 313–345.
Koput, K. W.
1997 ‘‘A chaotic model of innovative search: Some answers, many questions.’’ Orga-
nization Science, 8: 528–542.
Krackhardt, D.
1992 ‘‘The strength of strong ties: The importance of philos in organizations. ‘‘ In N.
Nohria and R. Eccles (eds.), Networks and Organizations: Structure, Form, and
Action: 216–239. Boston: Harvard Business School.
Krackhardt, D.
1998 ‘‘Simmelian ties: Super strong and sticky.’’ In R. M. Kramer and M. A. Neale
(eds.), Power and Influence in Organizations: 21–38. Thousand Oaks, CA: Sage.
Krackhardt, D.
1999 ‘‘The ties that torture: Simmelian tie analysis in organizations.’’ In S. B. Andrews
and D. Knoke (eds.), Research in the Sociology of Organizations, 16: 183–210. Green-
wich, CT: JAI Press.
Lamont, M.
2009 How Professors Think: Inside the Curious World of Academic Judgment. Cam-
bridge, MA: Harvard University Press.
104 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Larson, A.
1992 ‘‘Network dyads in entrepreneurial settings: A study of the governance of
exchange relationships.’’ Administrative Science Quarterly, 37: 76–104.
Lazarsfeld, P. F., and R. K. Merton
1954 ‘‘Friendship as a social process: A substantive and methodological analysis.’’ In
M. Berger, T. Able, and C. Page (eds.), Freedom and Control in Modern Society: 18–
66. New York: Octagon.
Levi Martin, J., and K-T. Yeung
2006 ‘‘Persistence of close personal ties over a 12-year period.’’ Social Networks, 28:
331–362.
Levitt, B., and J. G. March
1988 ‘‘Organizational learning.’’ Annual Review of Sociology, 14: 319–340.
Lizardo, O.
2006 ‘‘How cultural tastes shape personal networks.’’ American Sociological Review,
71: 778–807.
Lomi, A., and P. Pattison
2006 ‘‘Manufacturing relations: An empirical study of the organization of production
across multiple networks.’’ Organization Science, 17: 313–332.
March, J. C., and J. G. March
1977 ‘‘Almost random careers: The Wisconsin school superintendancy, 1942–1978.’’
Administrative Science Quarterly, 22: 377–409.
March, J. G.
1991 ‘‘Exploration and exploitation in organizational learning.’’ Organization Science,
2: 71–87.
March, J. G.
1999 ‘‘A learning perspective on the network dynamics of institutional integration.’’
In M. Egeberg and P. Lægreid (eds.), Organizing Political Institutions: 129–155. Oslo:
Scandinavian University Press.
March, J. G., and H. A. Simon
1958 Organizations. New York: Wiley & Sons.
Marsden, P. V., and K. E. Campbell
1984 ‘‘Measuring tie strength.’’ Social Forces, 63: 482–501.
McPherson, J. M., and L. Smith-Lovin
1987 ‘‘Homophily in voluntary organizations: Status distance and the composition of
face to face groups.’’ American Sociological Review, 52: 370–379.
McPherson, M., L. Smith-Lovin, and J. Cook
2001 ‘‘Birds of a feather: Homophily in social networks.’’ Annual Review of Sociol-
ogy, 27: 415–444.
Mehra, A., M. Kilduff, and D. J. Brass
2001 ‘‘The social networks of high and low self-monitors: Implications for workplace
performance.’’ Administrative Science Quarterly, 46: 121–146.
Merton, R. K.
1973 The Sociology of Science: Theoretical and Empirical Investigations. Chicago:
University of Chicago Press.
Merton, R. K.
1976 Sociological Ambivalence and Other Essays. New York: Free Press.
Morris, M. W., J. Podolny, and B. Sullivan
2008 ‘‘Culture and coworker relations: Interpersonal patterns in American, Chinese,
German and Spanish divisions of a global bank.’’ Organization Science, 19: 517–532.
Newman, M. E. J.
2001 ‘‘The structure of scientific collaboration networks.’’ Proceedings of the
National Academy of Sciences, 98: 404–409.
Newman, M. E. J.
2008 ‘‘The physics of networks.’’ Physics Today, November: 33–38.
Dahlander and McFarland 105
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Oettl, A.
2012 ‘‘Reconceptualizing stars: Scientist helpfulness and peer performance.’’ Man-
agement Science, 58: 1122–1140.
Owen-Smith, J.
2001 ‘‘Managing laboratory work through skepticism: Processes of evaluation and
control.’’ American Sociological Review, 66: 427–452.
Padgett, J. F., and C. K. Ansell
1993 ‘‘Robust action and the rise of the Medici, 1400–1434.’’ American Journal of
Sociology, 98: 1259–1319.
Padgett, J. F., and P. D. McLean
2006 ‘‘Organizational invention and elite transformation: The birth of partnership sys-
tems in Renaissance Florence.’’ American Journal of Sociology, 111: 1463–1568.
Palmer, D.
1983 ‘‘Broken ties: Interlocking directorates and intercorporate coordination.’’ Admin-
istrative Science Quarterly, 28: 40–55.
Peterson, M. W.
2007 ‘‘The study of colleges and universities as organizations.’’ In P. J. Gumport
(ed.), Sociology of Higher Education: Contributions and Their Contexts: 147–186. Bal-
timore, MD: Johns Hopkins University Press.
Pfeffer, J., and W. L. Moore
1980 ‘‘Power in university budgeting: A replication and extension.’’ Administrative
Science Quarterly, 25: 637–653.
Polidoro, F., G. Ahuja, and W. Mitchell
2011 ‘‘When the social structure overshadows competitive incentives: The effects of
network embeddedness on joint venture dissolution.’’ Academy of Management
Journal, 34: 203–223.
Powell, W. W., D. R. White, K. W. Koput, and J. Owen-Smith
2005 ‘‘Network dynamics and field evolution: The growth of inter-organizational colla-
boration in the life sciences.’’ American Journal of Sociology, 110: 1132–1205.
Rawlings, C. M., and D. A. McFarland
2011 ‘‘Influence flows in the academy: Using affiliation networks to assess peer
effects among researchers.’’ Social Science Research, 40: 1001–1017.
Reagans, R.
2011 ‘‘Close encounters: Analyzing how social similarity and propinquity contribute to
strong network connections.’’ Organization Science, 22: 835–849.
Reagans, R., and B. McEvily
2003 ‘‘Network structure and knowledge transfer: The effects of cohesion and
range.’’ Administrative Science Quarterly, 48: 240–267.
Robins, G., P. Pattison, Y. Kalish, and D. Lusher
2007 ‘‘An introduction to exponential random graph (p*) models for social networks.’’
Social Networks, 29: 173–191.
Ruef, M.
2002 ‘‘Strong ties, weak ties, and islands: Structural and cultural predictors of organi-
zational innovation.’’ Industrial and Corporate Change, 11: 427–449.
Ruef, M., H. E. Aldrich, and N. M. Carter
2003 ‘‘The structure of founding teams: Homophily, strong ties, and isolation among
U.S. entrepreneurs.’’ American Sociological Review, 68: 195–222.
Salancik, G. R., and J. Pfeffer
1974 ‘‘The bases and use of power in organizational decision making: The case of a
university.’’ Administrative Science Quarterly, 19: 453–473.
Seabright, M., D. Levinthal, and M. Fichman
1992 ‘‘Role of individual attachments in the dissolution of inter-organizational relation-
ships.’’ Academy of Management Journal, 35: 122–160.
106 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

Simmel, G.
1950 The Sociology of Georg Simmel. Glencoe, IL: Free Press.
Simon, H. A.
1971 ‘‘Designing organizations for an information-rich world.’’ In M. Greenberger
(ed.), Computers, Communication, and the Public Interest: 37–52. Baltimore, MD:
Johns Hopkins University Press.
Singer, J. D., and J. B. Willett
1991 ‘‘Modeling the days of our lives: Using survival analysis when designing and
analyzing longitudinal studies of duration and timing of events.’’ Psychological Bulle-
tin, 110: 268–290.
Singer, J. D., and J. B. Willett
1993 ‘‘It’s about time: Using discrete-time survival analysis to study duration and the
timing of events.’’ Journal of Educational Statistics, 18: 155–195.
Singer, J. D., and J. B. Willett
2003 Applied Longitudinal Data Analysis: Modeling Change and Event Occurence.
Oxford: Oxford University Press.
Snijders, T. A. B.
2001 ‘‘The statistical evaluation of social network dynamics.’’ Sociological Methodol-
ogy, 31: 361–395.
Stinchcombe, A. L.
1965 ‘‘Social structure and organization.’’ In J. G. March (ed.), Handbook of Organiza-
tions: 142–193. Chicago: Rand McNally.
Stokols, D., K. L. Hall, B. K. Taylor, and R. P. Moser
2008 ‘‘The science of team science: Overview of the field and introduction to the
supplement.’’ American Journal of Preventive Medicine, 35 (2S): S77–S89.
Uzzi, B.
1996 ‘‘The sources and consequences of embeddedness for the economic perfor-
mance of organizations: The network effect.’’ American Sociological Review, 61:
674–698.
Uzzi, B.
1997 ‘‘Social structure and competition in interfirm networks: The paradox of
embeddedness.’’ Administrative Science Quarterly, 42: 35–67.
Waldinger, F.
2012 ‘‘Peer effects in science: Evidence from the dismissal of scientists in Nazi Ger-
many.’’ Review of Economic Studies, 79: 838–861.
Wang, D., X. Shi, D. A. McFarland, and J. Leskovec
2012 ‘‘Measurement error in network data: A re-classification.’’ Social Networks, 34:
396–409.
Wasserman, S., and K. Faust
1994 Social Network Analysis: Methods and Applications. Cambridge: Cambridge
University Press.
Weingart, P.
2000 ‘‘Interdisciplinarity: The paradoxical discourse.’’ In P. Weingart and N. Stehr
(eds.), Practicing Interdisciplinarity: 25–42. Toronto: University of Toronto Press.
Wuchty, S., B. F. Jones, and B. Uzzi
2007 ‘‘The increasing dominance of teams in production of knowledge.’’ Science,
316: 1036–1039.
Xiao, Z., and A. Tsui
2007 ‘‘When brokers may not work: The cultural contingency of social capital in Chi-
nese high-tech firms.’’ Administrative Science Quarterly, 52: 1–31.
Zuckerman, H., and R. K. Merton
1972 ‘‘Age, aging, and age structure in science.’’ In M. White Riley, M. Johnson, and
A. Foner (eds.), Aging and Society, vol. 3: A Theory of Age Stratification: 292–356.
New York: Russell Sage Foundation.
Dahlander and McFarland 107
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

APPENDIX A: Description of Schools and Their Associated Departments
School Description Departments
Earth Sciences The Earth Sciences house
departments that focus on
studying the planet Earth.
Scholars conduct research on the
environment, oceans and climate,
fresh water, and geology.
Applied Earth Sciences
Geology and Environmental Sciences
Geology
Geophysics
Petroleum Engineering
Education The School of Education researches
educational practice and policy.
The school’s faculty includes
scholars from social science
disciplines as well as schools of
education.
No departments, but three academic programs:
Psychological Studies in Education
Social Sciences, Policy and Practice
Curriculum Studies, and Teacher Education
Engineering The Engineering School conducts
research on different types of
engineering. It also offers
entrepreneurship research through
the Stanford Technology Ventures
Program.
Aero/Astro
BioEngineering
Chemical Engineering
Civil Engineering
Computer Science
EES&OR
Electrical Engineering
Engineering Eco Systems
Industrial Engineering
Management Science & Engineering
Materials Science Engineering
Mechanical Engineering
Operations Research
Graduate School of Business The school’s faculty includes
scholars with advanced degrees
from business schools and social
science disciplines.
No departments, but seven academic areas:
Accounting
Economics
Finance
Marketing
Operations, Information, and Technology
Organizational Behavior
Political Economy
School of Humanities and Sciences
(Humanities)
Art
Asian Languages
Classics
Comparative Literature
Drama
English
French & Italian
German
History
Music
Philosophy
Religious Studies
Slavic Studies
Spanish and Portuguese
School of Humanities and Sciences
(Sciences)
Applied Physics
Biological Sciences
Chemistry
Food Research
Mathematics
Physics
Progress in Human Biology
Statistics
(continued)
108 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

APPENDIX B: Name Matching Procedure
Faculty members were linked to ISI publications through an ordered list of heuristic rules
for name matching. We ourselves were involved in the matching, but we also used
research assistants. Accurate name matching is difficult because listed names often
entail only a last name and first initial, and common names like ‘‘Smith, M’’ can match
different persons. While statistical or machine-learning models have been used for auto-
mated clustering, we opted for the transparency and reliability of hand-written rules and
manual review of difficult cases. We used a dozen heuristic rules for matching including
restricting the search to ISI publications when at least one author had a Stanford mailing
address; dropping all authors whose last name did not exactly match a Stanford faculty
member; exact matches on first authors’ last name, first initial, second initial, and whose
APPENDIX A: (continued)
School Description Departments
School of Humanities and Sciences
(Social Science)
Anthropological Sciences
Communication
Cultural Anthropology
Economics
Linguistics
Political Science
Psychology
Sociology
Law The great majority of faculty has
advanced degrees from law
schools.
No departments
Medicine The School of Medicine is the
largest at Stanford. It houses
almost 800 faculty and more than
1000 postdoctoral students as
well as M.D.s. Medical students
can also gain clinical experience at
Stanford Hospital and Clinics and
at Lucile Packard Children’s
Hospital.
Anesthesia
Biochemistry
Cardiothoracic Surgery
Cell Biology
Comparative Medicine
Radiology
Dermatology
Developmental Biology
Functional Restoration
Genetics
Gyn & Obstetrics
Health Research and Policy
Medicine
Microbiology and Immunology
Molecular & Cellular Physiology
Neurobiology
Neurology
Neurosurgery
Ophthalmology
Orthopedic Surgery
Otolaryngology & HNS
Pathology
Pediatrics
Pharmacology
Psychiatry
Radiation Oncology
Surgery
Urology
Dahlander and McFarland 109
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from

institution was Stanford; matching on last name and first initial for uncommon names,
and so forth. To assess the accuracy of name matching, we randomly sampled 20 cases
within each decision heuristic and manually assessed whether each match was correct
by downloading the article and checking faculty members’ CVs. In particular, we
counted the number of true positives, false positives, true negatives, and false nega-
tives. True positives are instances in which we believed the author of an article fit a
Stanford faculty member and the match was done correctly, whereas false positives are
cases in which the attribution to a faculty member was incorrect. True negatives are
cases in which we correctly dropped a case that did not match a Stanford faculty mem-
ber, and false negatives are cases in which we incorrectly dropped a case. From these
counts, we derived two metrics for name-matching accuracy in the author database as a
whole: precision and recall. Our precision and recall measures are significantly better
than other papers that use ISI data and articulate their methods. We also reran our analy-
sis using different precision and recall measures, with very similar results.
Authors’ Biographies
Linus Dahlander is an associate professor at ESMT European School of Management
and Technology, Schlossplatz 1, 10178 Berlin, Germany (e-mail: linus.dahlander@esmt
.org). His main fields of study are networks, communities, and innovation. He studies
how interpersonal networks unfold when people can choose their collaboration partners
and the consequences for innovation outcomes. He received a Ph.D. from Chalmers
University of Technology and was a postdoctoral fellow at Stanford University.
Daniel A. McFarland is an associate professor of education, sociology and organiza-
tional behavior at Stanford University, 520 Galvez Mall, Stanford, CA 94305 (e-mail:
mcfarland@stanford.edu). He specializes in the study of social dynamics and is currently
engaged in several projects concerning the coevolution of social networks and cultural
systems. He received a Ph.D. in sociology at the University of Chicago and teaches
graduate courses in social networks, organizations, sociology of knowledge, and micro-
sociology.
110 Administrative Science Quarterly 58 (2013)
at INFO.CENTRE,EUROPEAN SCHOOL on February 9, 2013asq.sagepub.comDownloaded from