Academy of Management Journal
2004, Vol. 47, No. 6, 893 906.
NETWORK MEMORY:
THE INFLUENCE OF PAST AND
CURRENT NETWORKS ON PERFORMANCE
GIUSEPPE SODA
ALESSANDRO USAI
Bocconi University
AKBAR ZAHEER
University of Minnesota
Investigating the efficacy of two alternative network structures, closure and structural
holes, from the contingent perspective of time, we connect past and current social
structures to outcomes. We show that, in the Italian television production industry,
current structural holes rather than past ones, but past closure rather than current
closure, help current network performance. Thus, structural holes and closure are
both valuable, but at different points in time.
In research using social networks, there has for but their meaning as well (Zaheer, Albert, & Za-
some time now been debate on the relative merits heer, 1999: 735). Thus, the relative efficacy of al-
of different network patterns as determinants of ternative network structures can be evaluated in
network outcomes (Burt, 1992; Coleman, 1988). De- terms of the contingency of time. More broadly, we
bate has particularly focused on two network pat-
answer the call to explicitly include a focus on time
terns, network closure and structural holes.1 Fur-
to enhance the quality of organization theory (Els-
ther, a number of researchers have identified the
bach, Sutton, & Whetten, 1999; Goodman, Law-
contingencies under which one structural pattern
rence, Ancona, & Tushman, 2001).
is more beneficial than the other (Ahuja, 2000a;
Our point of departure for this study is the im-
Rowley, Behrens, & Krackhardt, 2000; Podolny,
plicit notion that current outcomes reflect the ef-
2001). However, research has typically examined
fects of enduring patterns of relationships (Powell
the effects of network structure in current networks
& Smith-Doerr, 1994). Relatedly, empirical research
on concurrent outcomes. Given that the strength
on networks over time has been extremely scarce,
and value of ties may diminish, or grow, over time,
as Burt (2000a) lamented. But since networks are
a question can be raised as to the relative influences
constantly changing and evolving (Gulati & Gar-
of past and current network structures on current
giulo, 1999; Madhavan, Koka, & Prescott, 1998;
network outcomes. It is valuable to consider the
Suitor, Welman, & Morgan, 1997), existing struc-
influence of time on networks because time may
tures may not completely explain outcomes. Schol-
affect not only the  phenomena that are observed
ars have suggested that time must pass for relation-
ships to be cemented, strengthened, and become
imbued with trust and affect (Krackhardt, 1992).
All the authors contributed equally. We gratefully ac-
From this perspective, a past network, with its ac-
knowledge the financial support of the Bocconi Univer-
cumulated relational experience, becomes a kind of
sity Department of Organization and Human Resources,
 network memory that cannot be ignored as it may
and Research Committee. We thank Ranjay Gulati, Bill
McEvily, David Krackhardt, Vincenzo Perrone, Harry Sa- project a structural overhang over the present,
pienza, Myles Shaver, Sri Zaheer and, particularly, Pri
much like a shadow of the past. Conversely, as
Shah, for their valuable comments on earlier versions of
relationships dissolve and memories of old obliga-
this article. We also thank Anna Ponti and Tiziana Reina
tions and reciprocity fade with the passage of time,
for their admirable research assistance. All errors are
past networks lose potency, and current ties and
ours.
relationships exert more immediate and compel-
1
Network closure (Coleman, 1988) refers to a pattern
ling effects.
of dense, mutually interconnected ties among the mem-
Moreover, research has suggested that some net-
bers of a network. A network rich in structural holes
work patterns endure better than others. Research
(Burt, 1992) is one in which the different parts of the
on relationship decay has corroborated this notion
network are largely disconnected but bridged by a few
nodes, which have the potential to act as brokers. by showing that nine out of ten  bridge relation-
893
894 Academy of Management Journal December
ships disappear in a year (Burt, 2002). The extent to ucts. They include TV movies, serials, soaps, and
which some network structural patterns age better made-for-TV specials. The industry is both popular
than others and, more to the point, how they relate and economically important, and it possesses high
to network members performance outcomes as symbolic and cultural value (Bourdieu, 1993). It is
they age, thereby become worthwhile issues for made up of specialists from a number of profes-
research. sions musicians, actors, producers, screenwriters,
We studied these issues in the longitudinal con- and directors, among others who manage all the
text of the performance of TV production projects steps in the industry value chain. Like film
in Italy and used the actual market performance of projects, a TV production is the work of a tempo-
these projects as a network outcome. We obtained rary team of specialists who come together for the
data on every one of the 501 projects that were express purpose of creating the production (Miller
produced over a 12-year period (1988 99) for Ital- & Shamsie, 1996). As Hirsch noted,  For a cultural
ian television and constructed networks of all 4,793 product to succeed, networks of relationships
individuals involved in them. We also interviewed among a multitude of professionals must be mobi-
several industry participants and used insights lized, coordinated and managed. . .the formal and
from this phase of our data collection to set up the informal contracts often involve freelance profes-
research context, sharpen the hypotheses, and in- sionals and their associates (2000: 358). Conse-
terpret the results. quently, a TV production is characterized by high
One finding of the study was that past network levels of complexity and by a need for coordination
closure, among project members affects current in managing the temporary networks and directing
project performance but in a curvilinear fashion. resource combinations and recombinations (Bielby
On the other hand, structural holes in a current & Bielby, 1999).
external network more strongly enhanced project Coming to the production process, this industry
network performance than structural holes in a past operates essentially like other cultural industries.
network. In essence, we found that the value of Two fundamental and sometime opposing forces
social capital as closure persisted while that of struc- have to be managed: the creative and the industrial.
tural holes decayed over time. Conversely, structural These domains broadly correspond to nonroutine
holes provided concurrent information and arbitrage and routine activities in a broader organizational
value, but closure needed time for its mutually rein- context. The role of creativity is key, and scholars
forcing relationships to become beneficial. have noted that the long-term survival of firms in
Our study contributes to the literature in a num- such industries depends on renewing their creative
ber of ways. First, by including time as a major resources (Starkey, Barnatt, & Tempest, 2000). Al-
contingency in our implicit comparison of closure though the need for creativity is clear, it is impor-
and structural holes, we help to resolve a central tant to remember that in the end, the imperatives of
debate in the field. Moreover, while the time di- quality, efficiency, and profitability are critical too.
mension has been frequently invoked in the net- Having hit upon a formula that works, TV produc-
work literature, it has been little tested, despite tion projects often replicate successful recipes. Im-
many calls to examine networks in a dynamic con- portantly, the TV production industry in Italy does
text. By empirically comparing the advantages of not follow the rules of the  star system ; the pres-
the two forms of social structure, we subject the ence of a famous actor or director in a TV movie
underlying theories to a powerful test. Second, does not necessarily make a difference to its audi-
rather than focus on how time influences the per- ence. Instead, the production is a collective output
sistence of specific types of ties, we actually test for in which the team plays a critical role.
short-term and long-standing effects of both struc- Although numerous temporary projects are in
tural holes and closure. Third, we demonstrate place at any point in time, project networks are
nonlinear performance effects of social structure, embedded in the much larger network of relations
specifically of closure. Some previous research has among all the specialists working in the industry,
suggested nonlinear effects on performance, but which we refer to as the  external network. An
our focus on the outcomes of network structure, appropriate image is a network of temporary net-
rather than on those of tie strength, represents an works (Giuffre, 1999). The nearly 5,000 individuals
advance. in the industry are interconnected through current
and past working relationships. A good reason to
adopt a network perspective for this industry con-
RESEARCH CONTEXT
text arises from the inherently temporary nature of
Our research context, the Italian television pro- the project organizational form (Faulkner & Ander-
duction industry, produces a wide range of prod- son, 1987). The network of past relations among
2004 Soda, Usai, and Zaheer 895
project members can be expected to exert a partic- at which most information declines in value; win-
ularly potent effect on project outcomes, since or- dows of opportunity may close, and arbitraging
ganizational memory is limited (Walsh & Ungson, possibilities quickly dissipate (Kirzner, 1973). Even
1991). At the same time, current links between and if external conditions remain stable, it is possible
among projects allow specialists to flexibly draw that over time, other nodes in the network may gain
on skills and resources resident in the external access to once-privileged information through the
network that might be valuable in a particular wider network and negate the brokerage advantages
project. To examine the outcomes of closure in of a focal actor. Further, Burt (2002) noted that
project networks and those of structural holes in bridge ties are costlier to maintain than other ties
external networks, we chose the individual special- for two reasons. Not only do fewer people bear the
ist as our unit of analysis, or more accurately, our cost of bridge ties, but also, it is inherently difficult
unit of measurement (Klein, Dansereau, & Hall, to sustain relationships with those unlike oneself
1994). Specialists ties were measured both across and alters connected through bridges are likely to
projects (external network) and within projects (in- be dissimilar. Consequently, Burt argued, bridge
ternal network). Our level of analysis was the TV ties decay faster than other ties (see also Hansen,
production project as represented by the network of 1999).
specialists working on a given project. In our context, all the advantages stemming from
spanning structural holes over a wider, external
network accrue to the level of a project. To the
THEORY AND HYPOTHESES extent specialists from a particular project are con-
nected to other projects in the external network, the
Structural Holes in Past and Current Networks
focal project may benefit because its members may
We begin by reviewing arguments for the perfor- be able to better tap diverse ideas and skills from a
mance advantages of structural holes in general range of other projects. However, leveraging past
before relating those arguments to the contingency structural holes may not be easy because of the
of time. According to theory, actors with networks inherently short-term orientation of the TV produc-
rich in structural holes will gain from their ability tion industry. Like other cultural industries, this
to act as brokers, and these brokerage effects may one is predicated on keeping abreast of current
confer control and power (Burt, 1992). A slightly trends, modes, and social currents. Information di-
different argument draws on the advantages deriv- versity may be of little use if it is based on outdated
ing from superior access to novel and diverse in- cultural and social trends. Another source of infor-
formation opportunities. These include outcomes mation advantage may lie in leveraging new ideas,
such as superior innovation at the organization skills, and techniques across projects; over time,
level (Hargadon & Sutton, 1997) and reducing ego- these elements may become widely diffused and
centric uncertainty (Podolny, 2001), finding a job less distinctive. Furthermore, timing and access
(Granovetter, 1973), getting a promotion (Burt, benefits in cultural, as in other industries, may
1992), and career success generally (Podolny & Bar- have short half-lives, as others may appropriate
ron, 1997) at the individual level of analysis. An- unutilized ideas and opportunities.
other argument for the superior performance out- Thus, overall, time erodes the benefits of diver-
comes of networks rich in structural holes is that sity, brokerage, and timing. Consequently, we sug-
maintaining redundant ties is expensive in terms of gest that it is only current networks in which the
resources such as time and attention (Burt, 1992). presence of structural holes will enhance perfor-
Thus, a network that is composed of nonredundant mance. Formally, we hypothesize:
ties to a greater extent is making better use of scarce
Hypothesis 1. The effect of current structural
resources and is more efficient for timing, access,
holes bridged by project members on current
and information benefits.
project network performance is greater than
When past and current structural holes are com-
that of past structural holes bridged by project
pared, however, a subtly different set of arguments
members.
applies. Recall that the benefits of structural holes
emanate from timing, information, and brokerage.
The passage of time destroys such benefits. It is, for
Closure in Past and Current Project Networks
example, of little value to get a hot stock tip if you
are only going to be able to invest in the stock some We now briefly discuss the relationship between
years down the road. Further, Burt (2000b) implied closure and performance before turning to a com-
that the benefits structural holes confer on a net- parison of the effect of time on the relationship
work are essentially short-lived because of the rate between closure and performance. Some previous
896 Academy of Management Journal December
research has indicated that network characteristics increasingly positive effects of enhanced effi-
exhibit a nonlinear relationship with performance. ciency, trust, and quality start to kick in. Quality
Uzzi (1996) suggested that the relationship of em- and efficiency routines become internalized in a
beddedness, or strong ties, to the likelihood of firm project network and become more and more mutu-
failure is U-shaped; the likelihood of a firm s fail- ally reinforcing. Increasing connectivity within the
ure is greater when the numbers of arm s-length project network yields larger and larger increments
and embedded ties in its network are large and of trust to enhance performance once again. In es-
lower when its network has a mix of both types of sence, our reasoning suggests that medium levels of
ties. At an individual level of analysis, Perry-Smith closure are the worst for performance, yielding nei-
and Shalley (2003) theorized that as the number of ther the benefit of routinized quality nor that of
weak ties increases, creativity increases up to a diversity. Trying to combine the two effects with a
point and then declines. We extend these ideas by medium degree of closure results in a project, in
examining the nonlinear effects of network struc- effect, falling through the cracks. Overall, the neg-
ture, rather than those of tie properties, notably tie ative and the positive effects of network closure
strength. We theorize that the relationship of clo- operate to produce a U-shaped relationship be-
sure with performance is U-shaped. tween closure and performance.
In our context, high closure implies that a project We now turn to the effect of closure on project
team is composed of highly mutually intercon- performance over time. Research points to the su-
nected specialists who share imeanings, trust, and perior ability of closure to deal with the passage of
routines. Furthermore, common network links be- time. Krackhardt (1998) found that closed net-
come the conduits for the communication of cul- works, specifically, ones with triadic, or Simme-
tural norms, interpretations, and perceptions lian, ties, were more enduring and stronger than
(Ibarra & Andrews, 1993). These are important at- networks that did not maintain such ties. Burt
tributes of task teams in cultural industries since a (2000a) showed in a sample of bankers that bridge
quality symbolic product needs a clear and consis- decay occurred less among mutually connected
tent identity that relies heavily on shared under- people. The mutual relationships that are charac-
standings (Podolny & Baron, 1997), shared codes, teristic of closure appear to provide stability to past
and shared language (Nahapiet & Ghoshal, 1998), closure. Consequently, the structures that under-
and a collective mind (Weick & Roberts, 1993). In gird closure are more likely to influence outcomes
general, as Coleman s (1988) theory of social capital down the road.
based on closure suggests, actors in a dense and Further, the routines and operating procedures
highly interconnected network begin to develop that get established through history also facilitate
common routines, abstain from antisocial and op- an efficient flow through the social structure of
portunistic behavior, and create shared meanings, such resources. In addition, the probability of un-
understandings, and trust. Common mental models ethical behavior is lowered by the rapid dissemina-
help to improve the access to and flow of informa- tion of information about ethical transgression in
tion (Gnyawali & Madhavan, 2001). When closure dense networks. Such networks also promote the
is high, these multiple factors tend to reinforce alignment of individuals actions with a group s
each other and, combined with efficient routines, goals and priorities and develop norms of behavior
have a synergistic, positive effect on performance. that are likely to endure over time (Brass, Butter-
Low closure, however, can enhance performance field, & Skaggs, 1998). Past structure that arises
as well. Limited mutual connections between from repeated ties through time becomes a reposi-
members of a project network allow them to freely tory of meaning and identity for the individual in a
express points of view since  groupthink, with its network (White, 1992). Such meaning and identity
deleterious effects on creative ideas and thought is reinforced through time to cast a long shadow
(Janis, 1972), may not have set in. Further, lower into the present. Put differently, it is only over a
social pressures from low closure may encourage period of time that actors in a network have oppor-
entrepreneurial behavior and innovation, as im- tunities to interact with one another to develop
plied by Portes and Sensenbrenner s (1993 sugges- trust and observe others behaviors long enough to
tion that high closure creates social norms through create norms for future behavior, and in general
conformity, constraining and restricting individual build up social capital through mutual obligations
creativity and expression (see also Amabile, 1996). that can be reciprocated later (Blau, 1964). Taken
As closure increases from a low level, the nega- together, these arguments imply that at high levels
tive effects of constrained creativity and innovation of closure over time, better understood, more trust-
come into play, and performance drops. As the worthy, and more efficient behaviors and routines
degree of closure becomes higher still, however, the come into play to enhance project performance.
2004 Soda, Usai, and Zaheer 897
At the same time, low closure in the past has a each year of the study period. For example, for
stronger effect on creativity and project perfor- productions broadcast in 1995, we used past net-
mance than low closure in the present. The reason work data for 1988 94; for those of 1996, the net-
is that ties from the past, even if few, constrain work data were for 1989 95; and so on.
creativity more than few current ties. With increas- We chose 1995 as the cutoff between the present
ing closure, the negative constraining effect of past and the past for two reasons. This choice repre-
ties on creativity is stronger still, resulting in a sented both a standard median split of the data and
greater deleterious impact on performance at me- a period of stabilization of the network of special-
dium levels of closure, before the positive effects ists. On plotting the number of specialists over
from the efficient routines of past high closure time, we found that the curve flattened out around
overcome the negative. Put differently, both low 1994 96. The 249 projects involved a total of 4,793
closure and high closure from the past exert stron- different specialists for whom individual-level net-
ger effects on project performance than current low work measures were calculated and aggregated to
and high closure. Overall, we conclude that the the level of the project.
relationship between closure and performance is a A key characteristic of the study is that even
U-shaped one, and in the context of the contin- though the final analysis was at the level of the
gency of time, is stronger for past closure than it is project, all the relational measures were computed
for current. Thus, we have: by aggregating individual-level measures to the
project level. The network part of the study in-
Hypothesis 2. The curvilinear (U-shaped) ef-
volved the use of 4,793 4,793 matrices represent-
fect of past internal network closure on current
ing past and present ties among all the specialists
project network performance is greater than
involved in the production projects over the period
that of current internal network closure.
of the study.
METHODS
Model Lag Structure
In order to test our hypotheses, we developed a
We adopted a standard multiple regression
unique data set of TV productions in Italy for the
model. However, as our study was longitudinal, the
period 1988 99. TV productions are the result of
model had a lagged structure that took into account
project work; as discussed earlier, each TV produc-
the fact that some of the variables were measured
tion team is a small and temporary network embed-
over seven-year moving windows. The lag structure
ded in a bigger web of past and present ties.
adopted for those variables (past closure and past
structural holes) took the following form:
Data
7
All data used in statistical analyses are from ar- yt 0 1xt 2 xt i . . . t .

i 1
chival public sources and in particular from an
annual publication of Italian public television
No official data on TV productions were available
(RAI) that contains information about the perfor-
prior to 1988. In order to take into account the
mance of each  product in terms of its audience.
possible decay of relations over time (within the
Our data set includes all TV productions (TV mov-
seven-year window), we included a control vari-
ies, serials, and so forth) produced and broadcast
able for the age of the relations in the final model.
by any of the six national TV channels in Italy in
the period 1988 99. The few TV productions that
were either produced or delivered over multiple
Variables
years were dropped from the sample; only the first
production or broadcast is included. Dependent variable: Project performance. The
The overall data set contains information on all size of the audience that watched a show, or its
501 television productions created over that pe- number of viewers, is fairly unanimously consid-
riod. To assess the effect of the past network on ered the measure of any TV production s perfor-
current network performance, for each production mance. In Italy, audience data are collected by only
we used data from the seven years preceding the one independent institution, Auditel, which sells
year of its broadcast. Consequently, current perfor- the service to broadcasters. From 1986, Auditel has
mance and networks were assessed for the 249 been monitoring the Italian broadcasting market
productions broadcast in the 1995 99 period. The using a panel of 5,101 families and more than
seven-year window was moved five times, once for 14,000 individuals, stratified by various areas resi-
898 Academy of Management Journal December
dential populations. Since audience numbers are Past internal network closure. This measure de-
highly skewed, we used the natural logarithm of the fined closure as the density of past ties among the
number of people that watched each TV production members of the network for each actual TV produc-
in our sample as our measure of performance and our tion project. We computed the measure by applying
dependent variable. a block density procedure to the entire network of
past collaboration among the 4,793 specialists in-
volved in the study. We first computed five square
Independent Variables
4,793 4,793 matrixes of past ties corresponding
Figure 1 illustrates the relational contexts in which to each of the five moving seven-year windows (for
closure and structural holes were computed. We example, the window was 1988 94 for 1995 pro-
computed all network measures with the Social Net- ductions, 1989 95 for the 1996 productions, and so
work Analysis software package UCINET VI (Borgatti, on). In those matrixes, the cell xij represented the
Everett, & Freeman, 2002). As we explained before, number of projects on which specialists i and j had
current independent variables were measured at time worked together in the past seven years.
t, just as our performance measure was, and past Next, we computed a binary matrix sized 4,793
independent variables were measured for the interval 249 with information about which of the 4,793
t 1 through 7, our time window. specialists worked on which of the 249 TV produc-
FIGURE 1
Internal and External Relationsa
a
The relational contexts in which closure and structural holes were calculated are illustrated. The circles represent the boundaries of
the project networks. The lines between individuals within a project network indicate that those specialists worked together on projects
other than the focal project (the internal network). The lines outside the project represent individuals tied together by working on projects
other than the focal project (the external network). For graphical clarity, we represent the low-structural-holes quadrants with no relations
external to the given project network. Focal projects are shown with solid circles. Other concurrent and past projects are shown with solid
circles.
2004 Soda, Usai, and Zaheer 899
tions in the period 1995 99. In this matrix, xij was Demaree, & Wolf, 1984). This approach is con-
coded 1 if professional i had worked on TV pro- sistent with the general conception of structural
duction j. For example, production number 3 of holes as a form of capital that can be accumulated
1995 is a vector (column) in which the value 1 in in a group.
the cell indicates that specialist number 2,300 is a Current structural holes. We computed this
part of it, and 0 if not. We applied this vector to the measure using the same procedure adopted for
4,793 4,793 matrix of 1988 94 collaborations to computing the past structural holes measure but
ascertain the number of times the specialists had used the network of current ties as the input for the
worked together in the past. Thereafter, we calcu- constraint index procedure.
lated the density of network ties among the current
project members on the basis of past relations.
Control Variables
Current internal network closure. We calcu-
lated current internal network closure in the same Besides the network characteristics of project
way as past internal network closure, using the teams for which we hypothesized effects, a number
network of current ties as the input for the block- of other factors may reasonably affect the success of
density procedure instead of the network of past a television production. Control variables were
ties. The networks of current ties are represented by used to account for these factors. First, we discuss
square matrixes sized 4,793 4,793 in which xij is our network controls and then our industry-
the number of projects on which specialists i and j specific controls.
were currently working together. Network controls. The size of a project could
Past structural holes. We measured past project affect the dependent variable. We measured the
structural holes as the average of individual mem- size of the project network by counting the number
bers constraint in the broader network of past ties. of different professionals involved in a project. We
Network constraint  increases with the extent to also included a control for the average age of
which an individual s network is directly or indi- project network relations, given the importance of
rectly concentrated in a single contact. A network the time dimension in our analysis. Time was im-
concentrated in one contact means fewer structural portant because some projects are characterized by
holes, and so less social capital (Burt, Hogarth, & relatively old ties, while others include ties based
Michaud, 2000: 135). We adopted Burt s (1992) on more recent collaborations. The measure was a
measure of constraint to compute structural holes. weighted average (ra) of the age (n) of relations,
The index C measures the extent to which all of i s calculated using the amount of relational activity in
network is directly or indirectly invested in his or year t n, with age (ranging from 7 to 1) as the
her relationship with contact j. Thus, weight. To assess the sensitivity of alternative de-
cay functions, we ran the previously calculated
2
average age (ra) using different transformations as
cij pij pij qpiqpqj


alternative controls. We calculated log (ra), (ra)1/2,
and 1/(ra)2. None of our results changed. Further,
for q i, j, where pij is the proportion of i s rela- we evaluated the sensitivity of the results for alter-
tions invested in contact j and the total in paren- native time windows.
theses is the proportion of i s relations that are Finally, we included past project network cen-
directly or indirectly invested in the connection trality and current project network centrality in the
with contact j. We summed the cij across contacts j final model. These variables measured the degree
to get the network constraint index C for each spe- centrality (Freeman, 1979) of the project network
cialist in the past network, excluding the ties within the entire set of past and current 4,793
among current project members. We then averaged 4,793 collaboration networks and were computed
each specialist s individual score at the level of the by aggregating the degree centrality scores of
project network to obtain the structural holes score project members, with respect to relations that
at the group level. We multiplied the value of con- members maintained outside their project team
straint by 1 in order to capture structural holes (their external network). There were two main rea-
(the  opposite of constraint). sons for including centrality as a control. First, our
Aggregating structural holes to the project level expectation from considerable previous research
by averaging individual scores reflects our view was that centrality is related to performance (Tsai &
that structural holes are a  configurational prop- Ghoshal, 1998). Second, when testing for the effect of
erty of a team or project rather than a  shared structural holes, it is appropriate to put in a degree
property, as are cultural or perceived dimensions centrality control to ensure that the observed relation-
(Klein & Kozlowsi, 2000), or  judgments (James, ship between structural holes and performance is not
TABLE 1
Means, Standard Deviations, and Correlations
Variable Mean s.d. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1. 1995 0.11 0.31
2. 1996 0.17 0.37 .16*
3. 1997 0.24 0.42 .20** .25**
4. 1998 0.21 0.40 .18** .23** .28**
5. Number of episodes 8.94 25.12 .07 .01 .01 .02
6. TV movies 0.31 0.46 .04 .07 .00 .06 .21**
7. Prime time 0.87 0.33 .03 .01 .03 .01 .42** .08
8. Major channel 0.61 0.48 .02 .09 .06 .07 .13* .08 .03
9. Average age of project 2.68 0.61 .52** .11 .01 .26** .13* .09 .12 .09
relations
10. Size of project network 25.37 7.69 .04 .24** .03 .09 .17** .16** .11 .22** .08
11. Past project network 0.01 0.01 .18** .11 .03 .17** .09 .11 .06 .15* .17** .08
centrality
12. Current project network 0.01 0.00 .35** .20** .23** .18** .09 .02 .12 .12 .25** .17** .25**
centrality
13. Past structural holes 0.18 0.09 .18** .07 .09 .58** .03 .10 .03 .01 .15* .05 .03 .16**
14. Current structural holes 0.12 0.04 .07 .36** .09 .06 .04 .19** .24** .14* .04 .72** .15* .24** .15*
15. Current internal network 0.05 0.08 .11 .04 .03 .10 .06 .14* .06 .21** .06 .14 .11 .24** .22** .06
closure
16. Squared current internal 0.01 0.04 .05 .09 .03 .07 .01 .09 .03 .12 .02 .09 .01 .02 .10 .25** .83**
network closure
17. Past internal network 0.25 0.50 .00 .11 .07 .01 .08 .16* .33** .13* .02 .17** .12 .07 .42** .19** .31** .24**
closure
18. Squared past internal 0.25 1.17 .00 .13* .07 .00 .03 .11 .28** .12 .11 .24** .01 .09 .30** .26** .31** .27** .89**
network closure
19. Network performance 8.37 0.60 .12 .05 .05 .01 .25** .13* .55** .46** .25** .07 .04 .06 .03 .22** .15* .07 .13* .05
* p .05
** p .01
2004 Soda, Usai, and Zaheer 901
spurious, a result of the relationship between degree terms testing the curvilinear effect of past network
centrality and performance. closure. The change in explained variance is again
Past project network centrality was computed as significant (p .05), with 58 percent explained by
the average number of relations existing between this model. We report on the tests of the specific
any project member and any other specialist in the hypotheses below.
entire network of past ties (the seven-year windows In our first hypothesis, we posited that the effect
of past relations). Current project network central- of current structural holes on performance would
ity was computed as the average number of rela- be greater than that of past structural holes. The
tions existing between any of a project s members hypothesis was supported, as the results of a beta
and any other specialist in the entire network of difference test indicate. Table 3 reports these re-
current ties. sults. The difference in the coefficients of the vari-
Industry controls. Not all TV channels have the ables for current and past structural holes was sig-
same potential for reaching high audience levels. In nificant (t 3.48, p .001).
particular, the Italian broadcasting market has been Our second hypothesis was that the U-shaped
traditionally led by the two major channels of RAI effect of past internal network closure on project
(the state-owned television corporation) and Me- performance would be greater than the U-shaped
diaset (the major private competitor). Accordingly, effect of current internal network closure. We
we included a dummy variable for major channel, tested the hypothesis in two, albeit related, ways:
set to 1 when a production was shown on either First, we examined the sign and the significance of
RAI Uno or Canale5, and 0 otherwise. Another both the linear and the squared terms of the pair of
element that affects the audience potential of any terms that comprised the past and the current U-
television program is the time slot in which the shaped effects. A negative and significant coeffi-
program is broadcast. We therefore also included a cient for the linear term coupled with a positive
dummy control variable for prime time. Further, and significant coefficient for the squared term
TV productions have various characteristics; differ- would support a U-shaped curvilinear effect. Cur-
ent formats exist (TV movies, soaps, sitcoms, and rent internal network closure was not statistically
so on), and numbers of episodes differ. To control significant ( linear term 0.04, n.s.; squared term
for such task characteristics, we computed two ad- 0.08, n.s.), although the signs were in the expected
ditional variables: the number of episodes (the direction, but past internal network closure was
number actually broadcast) and a dummy variable significant ( linear term 0.27, p .05; squared term
for TV movie (1, TV movie, 0 otherwise). Finally, 0.29, p .01). Further, we employed stepwise
we controlled for periodicity effects by including a regression to test for the increment in explained
series of dummies corresponding to the years 1995, variance with each pair of variables. The increment
1996, 1997, and 1998, with 1999 being the omitted was significant for past internal network closure
category. terms (F 3.94, p .05) but was not significant
with current internal closure terms added to the
model (F 0.50, n.s.). Consequently, both tests
RESULTS
indicated support for Hypothesis 2.
Table 1 is a matrix of the correlations, means, and Our control variables for the years 1995 99 were
standard deviations of the variables that we used in not statistically significant in the final model, nor
our analysis. was number of episodes. However, as expected, the
Table 2 presents our results; we employed ordi- major channel dummy was strongly significant, as
nary least squares (OLS) regression methods. was the one for prime time. The negative and sig-
We begin with model 1, containing the industry nificant TV movie dummy suggests that the movie
control variables, which together explain 53 per- genre on TV is less popular than are other formats.
cent of the variance (adjusted R2) in our dependent Our network control variables were also not signif-
variable (audience). In model 2 we add network icant in the final model, with the exception of the
controls, the size of the project network, past and average age of network relations, which suggests
current network centrality, and the average age of older network relations have a positive effect on
project network relations. This model explains 55 performance. To evaluate the sensitivity of our sev-
percent of the variance, and the change in variance en-year time windows, we also estimated our mod-
explained is statistically significant (p .001). els using six- and five-year time windows. Our
Model 3 introduces the past and the current struc- results remained unchanged when we used six-
tural holes variables, testing our Hypothesis 1. The year windows. In the estimation using five-year
change in explained variance is significant (p windows, our squared term for past closure was
.05). Our final model, model 5, further includes the still significant, while the past closure term itself
902 Academy of Management Journal December
TABLE 2
Results of Regression Analysis for Project Network Performancea
Variables Model 1 Model 2 Model 3 Model 4 Model 5
Control
1995 .12* .23 .02 .01 .04
1996 .01 .05 .05 .04 .02
1997 .04 .01 .01 .01 .01
1998 .06 .09 .09 .04 .06
Number of episodes .00 .01 .01 .00 .01
TV movies .12** .13** .12** .11* .11*
Prime time .53*** .52*** .49*** .49*** .48***
Major channel .47*** .47*** .48*** .47*** .48***
Network control
Size of project network .09 .08 .09 .12 .10
Past project network centrality .07 .09 .09 .07
Current project network .02 .07 .07 .07
centrality
Average age of project network .18*** .15** .14** .11
relations
Hypothesized
Current structural holes .24*** .28*** .29***
Past structural holes .04 .05 .02
Current internal network .02 .04
closure
Squared current internal .07 .08
network closure
Past internal network closure .27*
Squared past internal network .29**
closure
F 32.30*** 26.40*** 24.00*** 21.00*** 19.60***
R2 .55 .57 .59 .59 .61
Adjusted R2 .53 .55 .57 .56 .58
R2 .02*** .02** .02 .01**
a
Standardized regression coefficients are shown. n 249.

p .10
* p .05
** p .01
*** p .001
TABLE 3 To control for the possible influence of multiple
Results of Beta Difference Testsa links between a few project network members, we
first computed the standard deviation of degree
Past Structural Current Structural
centrality scores within each group in the past and
Statistic Holes Holes
included it in the final model together with the
main effect and the squared term of past closure; it
0.04 0.24
b 0.13 4.53 was not significant ( 0.01, p .92). Second, we
s.d. (bi) 0.44 1.25
tested a dichotomized measure of internal density
Covariance (b1 b2) 0.07
in which relations were recoded as 1 and absent
Difference between 4.41
relations as 0. We found that this measure was
coefficients
significant and had the same sign as the scaled
t 3.48***
measure we used, although the significance of the
a
t (bi bj)/ s(bi bj) for i j, where s(bi bj) [s2(bi)
effect was lower ( -0.11, p .05), suggesting that
s2(bj) 2cov(bi, bj)]1/2.
the intensity of relations among a few members was
*** p .001
not biasing our results.
became just nonsignificant, though with the right
DISCUSSION
sign. With a three-year window, the effects of the
past turn nonsignificant, as we would expect given Although the network perspective presents a
our theory. conception of social context as an antithesis to the
2004 Soda, Usai, and Zaheer 903
atomistic view of social and organizational action, ship also represents an advance. Specifically, in the
we argue that the network perspective should also context of interfirm ties, prior research has exam-
address the temporal and historical relational con- ined the relationship between embeddedness and
text in which actors are embedded. Some prelimi- performance, where embeddedness is defined as tie
nary evidence suggests that different network pat- strength (Uzzi, 1996, 1997). Our posited relation-
terns age differently and that some past structures ship between past closure and performance relies
may exert stronger effects on performance than cur- on structural properties of networks rather than on
rent ones. Time may modify the nature of the flow tie strength, and it is U-shaped rather than inverted
through a network to benefit one or the other type U-shaped, thereby presenting intriguing prospects
of structure more and change the nature of the for future research. Further, although we found that
relationship between structure and performance. In current closure did not affect performance, this re-
this way, time can be seen as a contingent factor sult may not necessarily have been due to low
shaping the efficacy of alternative network struc- creativity, as we theorized. High current closure
tures. This is an important question for the field as may yet be associated with high creativity, since
the relative advantages of structural holes and clo- the social pressures causing conformity may not
sure as alternative sources of social capital have have had enough time to limit creativity. In such a
been the subject of major theoretical and empirical case, we speculate, inefficient routines might more
debate (Adler & Kwon, 2002). than counterbalance high creativity, which would
Our first finding is consistent with our prediction explain our findings.
that the effect of bridge ties is temporary. Rather Another important contribution of our work is in
than examining the stability of bridges per se, we addressing the enduring debate in the field on the
investigated whether a bridge in the past still influ- relative value of structural holes and closure. We
ences present outcomes or, in other words, whether help move discussion along by examining these
the effect of this form of social structure on out- social structures with a focus on the contingency of
comes persists over time. We found that it does not. time. Our findings suggest that rather than viewing
Neither bridge ties nor their outcomes can be ex- structural holes and closure as conflicting alterna-
pected to endure long because alters will eventu- tive forms of social structure, they can usefully be
ally react to limit their own vulnerability derived seen as complementary from the perspective of
from structural constraint by closing up structural time. In particular, we found that the best-perform-
holes or by gaining access to information through ing projects were those with high past closure and
the wider network to negate any brokerage advan- high current structural holes, which also suggests
tage. On the other hand, we found that structural some normative implications of our study. Managers
holes in the present help performance, as we ex- should compose teams considering both past working
pected. Thus, although bridge ties work well ini- relationships among team members and simulta-
tially, they turn ineffective in the long term. neous links to other projects for optimal performance.
An important aspect of our contribution here is Coming to the generalizability of our empirical
that, rather than focusing on the durability of ties, findings and the boundary conditions of our the-
we evaluate the outcomes of past and current ties, ory, we note that our research context is composed
or the extent to which network structures add value of temporary network organizations in a cultural
over time. One implication of these findings is that industry, which implies that tacit knowledge and
the cost of creating and maintaining bridge ties has intellectual capital are critical. Learning takes place
to be evaluated in the context of their short-term in an interactive and relational context. Despite
advantages coupled with their lack of longer-term these contextual characteristics, the conditions of
benefits. To the extent that relations are an invest- our research setting might be more generalizable
ment of time and resources, the payback from than they appear at first sight. Many industries are
bridge ties will need to be large and quick to justify increasingly moving toward rapid change, instability,
making an investment in them. and uncertainty (Eisenhardt, 1989), and in such set-
Even though longitudinal work in the network tings, organizations tend to adopt flexible forms of
domain is rare, some research in the broader liter- organization and team-based work practices (Illnich,
ature and theory supports the general notion that D Aveni, & Lewin, 1996). Moreover, numerous indus-
closure is long-lasting. We argued for a U-shaped tries, such as consulting, are known to rely on tem-
relationship between past network closure and per- porary project teams, and most group projects in or-
formance, a result we found, confirming that clo- ganizations are temporary rather than permanent. We
sure has long-lasting benefits; we did not find any view our findings as being broadly applicable in such
relationship between current network closure and settings as well. The applicability of our theory may
performance. Our finding of a U-shaped relation- be limited, however, when teams produce standard-
904 Academy of Management Journal December
ized outputs, when the most valuable learning is in- surement, subsequently aggregating individual-level
ternally derived, and when interdependence among data to the project level, the value of structural holes
team members is low. for a network as a whole may not necessarily be
convergent with its value for an individual broker.
In fact, a broker might distort or hoard information,
Limitations and Directions for Future Research
reducing overall network efficiency (Baker & Iyer,
Our study, like others, has some limitations. 1992; Stasser & Titus, 1985).
First, we did not formally estimate a decay function It is also possible that, although high closure in
for our major network variables, although we eval- an internal network produces groupthink and re-
uated the sensitivity of different functions for our duces novelty, a context of both internal and exter-
control for the age of relations as well as the sensi- nal networks may compensate for this effect with
tivity of differently sized time windows. To fully bridges to diverse external networks. When we
understand the role of time, it would help to pre- checked empirically for such compensation in our
cisely calibrate the points at which closure or struc- data using an interaction term, however, we found
tural holes starts helping or hurting performance. none. Despite this empirical result, future research
We could conceivably have done so with our exist- should explore this important possibility.
ing data, but in our formulation we treated the past
as an accumulation of past structure and relational
Concluding Remarks
experience. In the spirit of creating midrange the-
ory (Merton, 1968), we took a first step in linking
We investigated the efficacy of alternative net-
network structure and performance via the lens of
work structures from the perspective of time by
time. We leave for future research a detailed explo-
posing the question, How durable are the benefits
ration of decay in the network performance effect.
of different forms of social capital? In doing so, we
Further research may try to tease out whether it is
also spoke to the debate about the relative values of
the ties that decay, or the structure, or the effect of
structural holes and closure, and we contributed to
the structure on outcomes. Relatedly, although our
its resolution by viewing these social structures
direct research question involves comparing social
with the contingency of time. We showed that, in
structures over time, our model may contain some
the Italian television production industry, current
endogeneity in the sense that past ties are precur-
structural holes rather than past ones, but past clo-
sors to current ties (Gulati, 1995). However, when
sure rather than current closure, helped current
we introduced past and current closure sequen-
performance. Our findings suggest that network
tially into the regression models, we found that the
closure casts a long shadow, while brokerage struc-
past independently affected the dependent vari-
tures have only short-lived effects. Thus, closure
able, and current closure did not. Future research
and structural holes are both valuable, but at dif-
may explore the effect of past and current struc-
ferent points in time.
tures on outcomes in a manner that more directly
addresses any possible endogeneity.
Second, although we were extremely compre-
REFERENCES
hensive with regard to the inclusion of the TV
Adler, P. S., & Kwon, S. 2002. Social capital: Prospects
productions and the specialists involved in them,
for a new concept. Academy of Management Re-
we did not distinguish among the different roles
view, 27: 17 40.
that the specialists might play and how role differ-
Ahuja, G. 2000a. Collaboration networks, structural
entiation might influence a network and its out-
holes, and innovation: A longitudinal study. Admin-
comes. It would be useful for future research to
istrative Science Quarterly, 45: 425 455.
include consideration of roles in networks as well
Ahuja, G. 2000b. The duality of collaboration: Induce-
as the genres of TV productions. Future research
ments and opportunities in the formation of inter-
might also investigate other network variables,
firm linkages. Strategic Management Journal, 21:
such as centrality, as antecedents of performance
317 343.
over time (Tsai, 2002).
Amabile, T. M. 1996. Creativity in context. Boulder, CO:
A third limitation is the aggregation method we
Westview.
used in computing the structural holes score for the
project teams. We averaged the constraint scores of Baker, W. E., & Iyer, A. 1992. Information networks and
market behavior. Journal of Mathematical Sociol-
the team members, perhaps not capturing the di-
ogy, 16: 305 332.
versity of contacts at the project level. More gener-
ally, although we conceptualized and measured
Bielby, W.T., & Bielby, D. D. 1999. Organizational medi-
structural holes with individuals as the unit of mea- ation of project-based labor markets: Talent agencies
2004 Soda, Usai, and Zaheer 905
and the careers of screenwriters. American Socio- Granovetter, M. 1973. The strength of weak ties. Ameri-
logical Review, 64: 64  85. can Journal of Sociology, 78: 1360 1380.
Blau, P. M. 1964. Exchange and power in social life. Gulati, R. 1995. Social structure and alliance formation
New York: Wiley. pattern: A longitudinal analysis. Administrative
Science Quarterly, 40: 619  642.
Borgatti, S. P., Everett, M. G., & Freeman, L. C. 2002.
Gulati, R., & Gargiulo, M. 1999. Where do interorganiza-
UCINET 6 for Windows. Harvard, MA: Analytic
tional networks come from? American Journal of
Technologies.
Sociology, 104: 1439 1493.
Bourdieu, P. 1993. The field of cultural production:
Hansen, M. T. 1999. The search-transfer problem: The
Essays on art and literature. New York: Columbia
role of weak ties in sharing knowledge across orga-
University Press.
nizational subunits. Administrative Science Quar-
Brass, D. J., Butterfield, K. D., & Skaggs, B. C. 1998.
terly, 44: 82 111.
Relationships and unethical behavior: A social net-
Hargadon, A., & Sutton, R. I. 1997. Technology brokering
work perspective. Academy of Management Re-
and innovation in a product development firm. Ad-
view, 23: 14  31.
ministrative Science Quarterly, 42: 716 749.
Burt, R. S. 1992. Structural holes: The social structure
Hirsch, P. M. 2000. Cultural industries revisited. Orga-
of competition. Cambridge, MA: Harvard University
nization Science, 11: 356  361.
Press.
Ibarra, H., & Andrews, S. B. 1993. Power, social influ-
Burt, R. S. 2000a. Decay functions. Social Networks, 22:
ence, and sense making: Effects of network centrality
1 28.
and proximity on employee perceptions. Adminis-
Burt, R. S. 2000b. The network structure of social capital.
trative Science Quarterly, 38: 277 303.
In R. I. Sutton & B. M. Staw (Eds.), Research in
Illnich, A. Y., D Aveni, R., & Lewin, A. Y. 1995. New
organizational behavior, vol. 22: 345 423. Green-
organizational forms and strategies for managing in
wich, CT: JAI Press.
hypercompetitive environments. Organization Sci-
Burt, R. S. 2002. Bridge decay. Social Networks, 24:
ence, 3: 211 220.
333 363.
James, L. R, Demaree, R. G, & Wolf, G. 1984. Estimating
Burt, R. S., Horgarth, R., & Michaud, C. 2000. The social
within-group interrater reliability with and without
capital of French and American managers. Organi-
response bias. Journal of Applied Psychology, 69:
zation Science, 11: 123 147.
85 98.
Coleman, J. S. 1988. Social capital in the creation of
Janis, I. L. 1972. Victims of groupthink. Boston: Hough-
human capital. American Journal of Sociology,
ton Mifflin.
94(supplement): 95 120.
Kirzner, I. M. 1973. Competition & entrepreneurship.
Eisenhardt, K. M. 1989. Making fast decisions in high-
Chicago: University of Chicago Press.
velocity environments. Academy of Management
Klein, K. J., Dansereau, F., & Hall, R. J. 1994. Levels issues
Journal, 32: 53 576.
in theory development, data collection, and analysis.
Elsbach, K. D., Sutton, R. I., & Whetten, D. A. 1999.
Academy of Management Review, 19: 195 229.
Perspectives on developing management theory,
Klein, K. J., & Kozlowski, S. W. 2000. From micro to
circa 1999: Moving from shrill monologues to (rela-
meso: Critical steps in conceptualizing and conduct-
tively) tame dialogues. Academy of Management
ing multilevel research. Organizational Research
Review, 26: 627 633.
Methods, 3: 211 236.
Faulkner, R. R., & Anderson, A. B. 1987. Short-term
Krackhardt, D. 1992. The strength of strong ties: The
projects and emergent careers: Evidence from Holly-
importance of philos in organizations. In N. Nohria &
wood. American Journal of Sociology, 92: 879  909.
R. G. Eccles (Eds.), Networks and organizations:
Freeman L. C. 1979. Centrality in social networks: Con-
Structure, form, and action: 216 239. Boston: Har-
ceptual clarification. Social Networks, 1: 215 239.
vard Business School Press.
Giuffre, K. 1999. Sandpiles of opportunity: Success in the
Krackhardt, D. 1998. Simmelian ties: Super strong and
art world. Social Forces, 77: 815 832.
sticky. In R. M. Kramer & M. A. Neale (Eds.), Power
and influence in organizations: 21 38. Thousand
Goodman, P. S., Lawrence, B. S., Ancona, D. G., & Tush-
Oaks, CA: Sage.
man, M. L. 2001. Introduction to special issue on
time and organizational research. Academy of Man- Madhavan, R., Koka, B. R., & Prescott, J. E. 1998. Net-
agement Review, 26: 507 511.
works in transition: How industry events (re)shape
interfirm relationships. Strategic Management
Gnyawali, D. R., & Madhavan, R. 2001. Cooperative net-
Journal, 19: 439  459.
works and competitive dynamics: A structural em-
beddedness perspective. Academy of Management Merton, R. K. 1968. Social theory and social structure.
Review, 26: 431 445. New York: Free Press.
906 Academy of Management Journal December
Miller, D., & Shamsie, J. 1996. The resource-based view creation: The role of intrafirm networks. Academy
of the firm in two environments: The Hollywood of Management Journal, 41: 464  476.
film studios from 1936 to 1965. Academy of Man-
Uzzi, B. 1996. The sources and consequences of embed-
agement Journal, 39: 519  543.
dedness for the economic performance of organiza-
tions. American Sociological Review, 61: 674 698.
Morgan, G. 1989. Creative organization theory. Beverly
Hills, CA: Sage
Uzzi, B. 1997. Social structure and competition in inter-
firm networks: The paradox of embeddedness. Ad-
Nahapiet, J., & Ghoshal, S. 1998. Social capital, intellec-
ministrative Science Quarterly, 42: 35 67.
tual capital, and the organizational advantage. Acad-
emy of Management Review, 23: 242 266.
Walsh, J. P., & Ungson, G. R. 1991. Organizational mem-
ory. Academy of Management Review, 16: 57 91.
Perry-Smith, J. E., & Shalley, C. E. 2003. The social side of
creativity: A static and dynamic social network per-
Weick, K. E., & Roberts, K. H. 1993. Collective mind in
spective. Academy of Management Review, 28:
organizations: Heedful interrelating on flight decks.
89  106.
Administrative Science Quarterly, 38: 357 381.
Podolny, J. M. 2001. Networks as the pipes and prisms of
White, H. C. 1992. Agency as control in formal networks.
the market. American Journal of Sociology, 107:
In N. Nohria & R. G. Eccles (Eds.), Networks and
33 60.
organizations: Structure, form, and action: 92 117.
Boston: Harvard Business School Press.
Podolny, J. M., & Baron, J. N. 1997. Resources and rela-
tionships: Social networks and mobility in the work-
Zaheer, S., Albert, S., & Zaheer, A. 1999. Time scales and
place. American Sociological Review, 62: 673 693.
organization theory. Academy of Management Re-
view, 24: 725 741.
Portes, A., & Sensenbrenner, J. 1993. Embeddedness and
immigration: Notes on the social determinants of
economic social action. American Journal of Soci-
ology, 98: 1320 1350.
Giuseppe Soda ([e-mail]; Ph.D,
Powell, W. W., & Smith-Doerr, L. 1994. Networks and
Bocconi University) is an associate professor of organi-
economic life. In N. J. Smelser & R. Swedberg (Eds.),
zation theory at Bocconi University of Milan. He also is
The handbook of economic sociology: 368  402.
the director of the Department of Organization and Hu-
Princeton, NJ: Princeton University Press.
man Resource Management at Bocconi University School
Rowley, T., Behrens, D., & Krackhardt, D. 2000. Redun-
of Management. He has been a visiting assistant professor
dant governance structures: An analysis of structural
at Carnegie Mellon University. His research interests and
and relational embeddedness in the steel and semi-
publications include micro and macro networks, organi-
conductor industries. Strategic Management Jour-
zational design, and new forms of organization.
nal, 21: 369  286.
Alessandro Usai ([e-mail]) earned
Starkey, K., Barnatt, C., & Tempest, S. 2000. Beyond
his Ph.D. at Bologna University; he has been a visiting
networks and hierarchies: Latent organizations in
scholar at the University of Minnesota and an assistant
the UK television industry. Organization Science,
professor at Bocconi University of Milan. He is currently
11(3): 299  305.
the managing director of Cinecittá Holding, the major
Stasser, G., & Titus, W. 1985. Pooling of unshared infor- Italian film production organization. His research inter-
mation in group decision-making: Biased informa- ests include network analysis and studies of performance
tion sampling during discussion. Journal of Person- in cultural industries.
ality and Social Psychology, 48: 1467 1478.
Akbar Zaheer ([e-mail]; Ph.D., Massa-
Suitor, J. J., Welman, B., & Morgan, D. L. 1997. It s about
chusetts Institute of Technology) is the Curtis L. Carlson
time: How, why, and when networks change. Social Professor of Strategic Management and Organization and
Networks, 19: 1 7. the director of the Strategic Management Research Center
at the Carlson School of Management, University of Min-
Tsai, W. 2002. Social structure of  coopetition within a
nesota. Besides networks, his research interests include
multiunit organization: Coordination, competition,
trust in organizational contexts, strategic alliances, and
and intraorganizational knowledge sharing. Organi-
mergers and acquisitions.
zation Science, 13(2): 179  190.
Tsai, W., & Ghoshal, S. 1998. Social capital and value