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

some time now been debate on the relative merits
of different network patterns as determinants of
network outcomes (Burt, 1992; Coleman, 1988). De-
bate has particularly focused on two network pat-
terns, network closure and structural holes.

1

Fur-

ther, a number of researchers have identified the
contingencies under which one structural pattern
is more beneficial than the other (Ahuja, 2000a;
Rowley, Behrens, & Krackhardt, 2000; Podolny,
2001). However, research has typically examined
the effects of network structure in current networks
on concurrent outcomes. Given that the strength
and value of ties may diminish, or grow, over time,
a question can be raised as to the relative influences
of past and current network structures on current
network outcomes. It is valuable to consider the
influence of time on networks because time may
affect not only the “phenomena that are observed

but their meaning as well” (Zaheer, Albert, & Za-
heer, 1999: 735). Thus, the relative efficacy of al-
ternative network structures can be evaluated in
terms of the contingency of time. More broadly, we
answer the call to explicitly include a focus on time
to enhance the quality of organization theory (Els-
bach, Sutton, & Whetten, 1999; Goodman, Law-
rence, Ancona, & Tushman, 2001).

Our point of departure for this study is the im-

plicit notion that current outcomes reflect the ef-
fects of enduring patterns of relationships (Powell
& Smith-Doerr, 1994). Relatedly, empirical research
on networks over time has been extremely scarce,
as Burt (2000a) lamented. But since networks are
constantly changing and evolving (Gulati & Gar-
giulo, 1999; Madhavan, Koka, & Prescott, 1998;
Suitor, Welman, & Morgan, 1997), existing struc-
tures may not completely explain outcomes. Schol-
ars have suggested that time must pass for relation-
ships to be cemented, strengthened, and become
imbued with trust and affect (Krackhardt, 1992).
From this perspective, a past network, with its ac-
cumulated relational experience, becomes a kind of
“network memory” that cannot be ignored as it may
project a structural overhang over the present,
much like a shadow of the past. Conversely, as
relationships dissolve and memories of old obliga-
tions and reciprocity fade with the passage of time,
past networks lose potency, and current ties and
relationships exert more immediate and compel-
ling effects.

Moreover, research has suggested that some net-

work patterns endure better than others. Research
on relationship decay has corroborated this notion
by showing that nine out of ten “bridge” relation-

All the authors contributed equally. We gratefully ac-

knowledge the financial support of the Bocconi Univer-
sity Department of Organization and Human Resources,
and Research Committee. We thank Ranjay Gulati, Bill
McEvily, David Krackhardt, Vincenzo Perrone, Harry Sa-
pienza, Myles Shaver, Sri Zaheer and, particularly, Pri
Shah, for their valuable comments on earlier versions of
this article. We also thank Anna Ponti and Tiziana Reina
for their admirable research assistance. All errors are
ours.

1

Network closure (Coleman, 1988) refers to a pattern

of dense, mutually interconnected ties among the mem-
bers of a network. A network rich in structural holes
(Burt, 1992) is one in which the different parts of the
network are largely disconnected but bridged by a few
nodes, which have the potential to act as brokers.

娀 Academy of Management Journal
2004, Vol. 47, No. 6, 893–906.

893

ships disappear in a year (Burt, 2002). The extent to
which some network structural patterns age better
than others and, more to the point, how they relate
to network members’ performance outcomes as
they age, thereby become worthwhile issues for
research.

We studied these issues in the longitudinal con-

text of the performance of TV production projects
in Italy and used the actual market performance of
these projects as a network outcome. We obtained
data on every one of the 501 projects that were
produced over a 12-year period (1988 –99) for Ital-
ian television and constructed networks of all 4,793
individuals involved in them. We also interviewed
several industry participants and used insights
from this phase of our data collection to set up the
research context, sharpen the hypotheses, and in-
terpret the results.

One finding of the study was that past network

closure, among project members affects current
project performance but in a curvilinear fashion.
On the other hand, structural holes in a current
external network more strongly enhanced project
network performance than structural holes in a past
network. In essence, we found that the value of
social capital as closure persisted while that of struc-
tural holes decayed over time. Conversely, structural
holes provided concurrent information and arbitrage
value, but closure needed time for its mutually rein-
forcing relationships to become beneficial.

Our study contributes to the literature in a num-

ber of ways. First, by including time as a major
contingency in our implicit comparison of closure
and structural holes, we help to resolve a central
debate in the field. Moreover, while the time di-
mension has been frequently invoked in the net-
work literature, it has been little tested, despite
many calls to examine networks in a dynamic con-
text. By empirically comparing the advantages of
the two forms of social structure, we subject the
underlying theories to a powerful test. Second,
rather than focus on how time influences the per-
sistence of specific types of ties, we actually test for
short-term and long-standing effects of both struc-
tural holes and closure. Third, we demonstrate
nonlinear performance effects of social structure,
specifically of closure. Some previous research has
suggested nonlinear effects on performance, but
our focus on the outcomes of network structure,
rather than on those of tie strength, represents an
advance.

RESEARCH CONTEXT

Our research context, the Italian television pro-

duction industry, produces a wide range of prod-

ucts. They include TV movies, serials, soaps, and
made-for-TV specials. The industry is both popular
and economically important, and it possesses high
symbolic and cultural value (Bourdieu, 1993). It is
made up of specialists from a number of profes-
sions—musicians, actors, producers, screenwriters,
and directors, among others—who manage all the
steps in the industry value chain. Like film
projects, a TV production is the work of a tempo-
rary team of specialists who come together for the
express purpose of creating the production (Miller
& Shamsie, 1996). As Hirsch noted, “For a cultural
product to succeed, networks of relationships
among a multitude of professionals must be mobi-
lized, coordinated and managed. . .the formal and
informal contracts often involve freelance profes-
sionals and their associates” (2000: 358). Conse-
quently, a TV production is characterized by high
levels of complexity and by a need for coordination
in managing the temporary networks and directing
resource combinations and recombinations (Bielby
& Bielby, 1999).

Coming to the production process, this industry

operates essentially like other cultural industries.
Two fundamental and sometime opposing forces
have to be managed: the creative and the industrial.
These domains broadly correspond to nonroutine
and routine activities in a broader organizational
context. The role of creativity is key, and scholars
have noted that the long-term survival of firms in
such industries depends on renewing their creative
resources (Starkey, Barnatt, & Tempest, 2000). Al-
though the need for creativity is clear, it is impor-
tant to remember that in the end, the imperatives of
quality, efficiency, and profitability are critical too.
Having hit upon a formula that works, TV produc-
tion projects often replicate successful recipes. Im-
portantly, the TV production industry in Italy does
not follow the rules of the “star system”; the pres-
ence of a famous actor or director in a TV movie
does not necessarily make a difference to its audi-
ence. Instead, the production is a collective output
in which the team plays a critical role.

Although numerous temporary projects are in

place at any point in time, project networks are
embedded in the much larger network of relations
among all the specialists working in the industry,
which we refer to as the “external network.” An
appropriate image is a network of temporary net-
works (Giuffre, 1999). The nearly 5,000 individuals
in the industry are interconnected through current
and past working relationships. A good reason to
adopt a network perspective for this industry con-
text arises from the inherently temporary nature of
the project organizational form (Faulkner & Ander-
son, 1987). The network of past relations among

894

December

Academy of Management Journal

project members can be expected to exert a partic-
ularly potent effect on project outcomes, since or-
ganizational memory is limited (Walsh & Ungson,
1991). At the same time, current links between and
among projects allow specialists to flexibly draw
on skills and resources resident in the external
network that might be valuable in a particular
project. To examine the outcomes of closure in
project networks and those of structural holes in
external networks, we chose the individual special-
ist as our unit of analysis, or more accurately, our
unit of measurement (Klein, Dansereau, & Hall,
1994). Specialists’ ties were measured both across
projects (external network) and within projects (in-
ternal network). Our level of analysis was the TV
production project as represented by the network of
specialists working on a given project.

THEORY AND HYPOTHESES

Structural Holes in Past and Current Networks

We begin by reviewing arguments for the perfor-

mance advantages of structural holes in general
before relating those arguments to the contingency
of time. According to theory, actors with networks
rich in structural holes will gain from their ability
to act as brokers, and these brokerage effects may
confer control and power (Burt, 1992). A slightly
different argument draws on the advantages deriv-
ing from superior access to novel and diverse in-
formation opportunities. These include outcomes
such as superior innovation at the organization
level (Hargadon & Sutton, 1997) and reducing ego-
centric uncertainty (Podolny, 2001), finding a job
(Granovetter, 1973), getting a promotion (Burt,
1992), and career success generally (Podolny & Bar-
ron, 1997) at the individual level of analysis. An-
other argument for the superior performance out-
comes of networks rich in structural holes is that
maintaining redundant ties is expensive in terms of
resources such as time and attention (Burt, 1992).
Thus, a network that is composed of nonredundant
ties to a greater extent is making better use of scarce
resources and is more efficient for timing, access,
and information benefits.

When past and current structural holes are com-

pared, however, a subtly different set of arguments
applies. Recall that the benefits of structural holes
emanate from timing, information, and brokerage.
The passage of time destroys such benefits. It is, for
example, of little value to get a hot stock tip if you
are only going to be able to invest in the stock some
years down the road. Further, Burt (2000b) implied
that the benefits structural holes confer on a net-
work are essentially short-lived because of the rate

at which most information declines in value; win-
dows of opportunity may close, and arbitraging
possibilities quickly dissipate (Kirzner, 1973). Even
if external conditions remain stable, it is possible
that over time, other nodes in the network may gain
access to once-privileged information through the
wider network and negate the brokerage advantages
of a focal actor. Further, Burt (2002) noted that
bridge ties are costlier to maintain than other ties
for two reasons. Not only do fewer people bear the
cost of bridge ties, but also, it is inherently difficult
to sustain relationships with those unlike oneself—
and alters connected through bridges are likely to
be dissimilar. Consequently, Burt argued, bridge
ties decay faster than other ties (see also Hansen,
1999).

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
extent specialists from a particular project are con-
nected to other projects in the external network, the
focal project may benefit because its members may
be able to better tap diverse ideas and skills from a
range of other projects. However, leveraging past
structural holes may not be easy because of the
inherently short-term orientation of the TV produc-
tion industry. Like other cultural industries, this
one is predicated on keeping abreast of current
trends, modes, and social currents. Information di-
versity may be of little use if it is based on outdated
cultural and social trends. Another source of infor-
mation advantage may lie in leveraging new ideas,
skills, and techniques across projects; over time,
these elements may become widely diffused and
less distinctive. Furthermore, timing and access
benefits in cultural, as in other industries, may
have short half-lives, as others may appropriate
unutilized ideas and opportunities.

Thus, overall, time erodes the benefits of diver-

sity, brokerage, and timing. Consequently, we sug-
gest that it is only current networks in which the
presence of structural holes will enhance perfor-
mance. Formally, we hypothesize:

Hypothesis 1. The effect of current structural
holes bridged by project members on current
project network performance is greater than
that of past structural holes bridged by project
members.

Closure in Past and Current Project Networks

We now briefly discuss the relationship between

closure and performance before turning to a com-
parison of the effect of time on the relationship
between closure and performance. Some previous

2004

895

Soda, Usai, and Zaheer

research has indicated that network characteristics
exhibit a nonlinear relationship with performance.
Uzzi (1996) suggested that the relationship of em-
beddedness, or strong ties, to the likelihood of firm
failure is U-shaped; the likelihood of a firm’s fail-
ure is greater when the numbers of arm’s-length
and embedded ties in its network are large and
lower when its network has a mix of both types of
ties. At an individual level of analysis, Perry-Smith
and Shalley (2003) theorized that as the number of
weak ties increases, creativity increases up to a
point and then declines. We extend these ideas by
examining the nonlinear effects of network struc-
ture, rather than those of tie properties, notably tie
strength. We theorize that the relationship of clo-
sure with performance is U-shaped.

In our context, high closure implies that a project

team is composed of highly mutually intercon-
nected specialists who share imeanings, trust, and
routines. Furthermore, common network links be-
come the conduits for the communication of cul-
tural norms, interpretations, and perceptions
(Ibarra & Andrews, 1993). These are important at-
tributes of task teams in cultural industries since a
quality symbolic product needs a clear and consis-
tent identity that relies heavily on shared under-
standings (Podolny & Baron, 1997), shared codes,
and shared language (Nahapiet & Ghoshal, 1998),
and a collective mind (Weick & Roberts, 1993). In
general, as Coleman’s (1988) theory of social capital
based on closure suggests, actors in a dense and
highly interconnected network begin to develop
common routines, abstain from antisocial and op-
portunistic behavior, and create shared meanings,
understandings, and trust. Common mental models
help to improve the access to and flow of informa-
tion (Gnyawali & Madhavan, 2001). When closure
is high, these multiple factors tend to reinforce
each other and, combined with efficient routines,
have a synergistic, positive effect on performance.

Low closure, however, can enhance performance

as well. Limited mutual connections between
members of a project network allow them to freely
express points of view since “groupthink,” with its
deleterious effects on creative ideas and thought
(Janis, 1972), may not have set in. Further, lower
social pressures from low closure may encourage
entrepreneurial behavior and innovation, as im-
plied by Portes and Sensenbrenner’s (1993 sugges-
tion that high closure creates social norms through
conformity, constraining and restricting individual
creativity and expression (see also Amabile, 1996).

As closure increases from a low level, the nega-

tive effects of constrained creativity and innovation
come into play, and performance drops. As the
degree of closure becomes higher still, however, the

increasingly positive effects of enhanced effi-
ciency, trust, and quality start to kick in. Quality
and efficiency routines become internalized in a
project network and become more and more mutu-
ally reinforcing. Increasing connectivity within the
project network yields larger and larger increments
of trust to enhance performance once again. In es-
sence, our reasoning suggests that medium levels of
closure are the worst for performance, yielding nei-
ther the benefit of routinized quality nor that of
diversity. Trying to combine the two effects with a
medium degree of closure results in a project, in
effect, falling through the cracks. Overall, the neg-
ative and the positive effects of network closure
operate to produce a U-shaped relationship be-
tween closure and performance.

We now turn to the effect of closure on project

performance over time. Research points to the su-
perior ability of closure to deal with the passage of
time. Krackhardt (1998) found that closed net-
works, specifically, ones with triadic, or Simme-
lian, ties, were more enduring and stronger than
networks that did not maintain such ties. Burt
(2000a) showed in a sample of bankers that bridge
decay occurred less among mutually connected
people. The mutual relationships that are charac-
teristic of closure appear to provide stability to past
closure. Consequently, the structures that under-
gird closure are more likely to influence outcomes
down the road.

Further, the routines and operating procedures

that get established through history also facilitate
an efficient flow through the social structure of
such resources. In addition, the probability of un-
ethical behavior is lowered by the rapid dissemina-
tion of information about ethical transgression in
dense networks. Such networks also promote the
alignment of individuals’ actions with a group’s
goals and priorities and develop norms of behavior
that are likely to endure over time (Brass, Butter-
field, & Skaggs, 1998). Past structure that arises
from repeated ties through time becomes a reposi-
tory of meaning and identity for the individual in a
network (White, 1992). Such meaning and identity
is reinforced through time to cast a long shadow
into the present. Put differently, it is only over a
period of time that actors in a network have oppor-
tunities to interact with one another to develop
trust and observe others’ behaviors long enough to
create norms for future behavior, and in general
build up social capital through mutual obligations
that can be reciprocated later (Blau, 1964). Taken
together, these arguments imply that at high levels
of closure over time, better understood, more trust-
worthy, and more efficient behaviors and routines
come into play to enhance project performance.

896

December

Academy of Management Journal

At the same time, low closure in the past has a

stronger effect on creativity and project perfor-
mance than low closure in the present. The reason
is that ties from the past, even if few, constrain
creativity more than few current ties. With increas-
ing closure, the negative constraining effect of past
ties on creativity is stronger still, resulting in a
greater deleterious impact on performance at me-
dium levels of closure, before the positive effects
from the efficient routines of past high closure
overcome the negative. Put differently, both low
closure and high closure from the past exert stron-
ger effects on project performance than current low
and high closure. Overall, we conclude that the
relationship between closure and performance is a
U-shaped one, and in the context of the contin-
gency of time, is stronger for past closure than it is
for current. Thus, we have:

Hypothesis 2. The curvilinear (U-shaped) ef-
fect of past internal network closure on current
project network performance is greater than
that of current internal network closure.

METHODS

In order to test our hypotheses, we developed a

unique data set of TV productions in Italy for the
period 1988 –99. TV productions are the result of
project work; as discussed earlier, each TV produc-
tion team is a small and temporary network embed-
ded in a bigger web of past and present ties.

Data

All data used in statistical analyses are from ar-

chival public sources and in particular from an
annual publication of Italian public television
(RAI) that contains information about the perfor-
mance of each “product” in terms of its audience.
Our data set includes all TV productions (TV mov-
ies, serials, and so forth) produced and broadcast
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
years were dropped from the sample; only the first
production or broadcast is included.

The overall data set contains information on all

501 television productions created over that pe-
riod. To assess the effect of the past network on
current network performance, for each production
we used data from the seven years preceding the
year of its broadcast. Consequently, current perfor-
mance and networks were assessed for the 249
productions broadcast in the 1995–99 period. The
seven-year window was moved five times, once for

each year of the study period. For example, for
productions broadcast in 1995, we used past net-
work data for 1988 –94; for those of 1996, the net-
work data were for 1989 –95; and so on.

We chose 1995 as the cutoff between the present

and the past for two reasons. This choice repre-
sented both a standard median split of the data and
a period of stabilization of the network of special-
ists. On plotting the number of specialists over
time, we found that the curve flattened out around
1994 –96. The 249 projects involved a total of 4,793
different specialists for whom individual-level net-
work measures were calculated and aggregated to
the level of the project.

A key characteristic of the study is that even

though the final analysis was at the level of the
project, all the relational measures were computed
by aggregating individual-level measures to the
project level. The network part of the study in-
volved the use of 4,793

⫻ 4,793 matrices represent-

ing past and present ties among all the specialists
involved in the production projects over the period
of the study.

Model Lag Structure

We adopted a standard multiple regression

model. However, as our study was longitudinal, the
model had a lagged structure that took into account
the fact that some of the variables were measured
over seven-year moving windows. The lag structure
adopted for those variables (past closure and past
structural holes) took the following form:

y

t

⫽

␤

0

⫹

␤

1

x

t

⫹

␤

2

冘

i

⫽1

7

x

t

⫺i

⫹ . . . ⫹ ␧

t

.

No official data on TV productions were available
prior to 1988. In order to take into account the
possible decay of relations over time (within the
seven-year window), we included a control vari-
able for the age of the relations in the final model.

Variables

Dependent variable: Project performance. The

size of the audience that watched a show, or its
number of viewers, is fairly unanimously consid-
ered the measure of any TV production’s perfor-
mance. In Italy, audience data are collected by only
one independent institution, Auditel, which sells
the service to broadcasters. From 1986, Auditel has
been monitoring the Italian broadcasting market
using a panel of 5,101 families and more than
14,000 individuals, stratified by various areas’ resi-

2004

897

Soda, Usai, and Zaheer

dential populations. Since audience numbers are
highly skewed, we used the natural logarithm of the
number of people that watched each TV production
in our sample as our measure of performance and our
dependent variable.

Independent Variables

Figure 1 illustrates the relational contexts in which

closure and structural holes were computed. We
computed all network measures with the Social Net-
work Analysis software package UCINET VI (Borgatti,
Everett, & Freeman, 2002). As we explained before,
current independent variables were measured at time
t, just as our performance measure was, and past
independent variables were measured for the interval
t

⫺ 1 through ⫺7, our time window.

Past internal network closure. This measure de-

fined closure as the density of past ties among the
members of the network for each actual TV produc-
tion project. We computed the measure by applying
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
4,793

⫻ 4,793 matrixes of past ties corresponding

to each of the five moving seven-year windows (for
example, the window was 1988 –94 for 1995 pro-
ductions, 1989 –95 for the 1996 productions, and so
on). In those matrixes, the cell x

ij

represented the

number of projects on which specialists i and j had
worked together in the past seven years.

Next, we computed a binary matrix sized 4,793

⫻

249 with information about which of the 4,793
specialists worked on which of the 249 TV produc-

FIGURE 1

Internal and External Relations

a

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.

898

December

Academy of Management Journal

tions in the period 1995–99. In this matrix, x

ij

was

coded 1 if professional i had worked on TV pro-
duction j. For example, production number 3 of
1995 is a vector (column) in which the value 1 in
the cell indicates that specialist number 2,300 is a
part of it, and 0 if not. We applied this vector to the
4,793

⫻ 4,793 matrix of 1988–94 collaborations to

ascertain the number of times the specialists had
worked together in the past. Thereafter, we calcu-
lated the density of network ties among the current
project members on the basis of past relations.

Current internal network closure. We calcu-

lated current internal network closure in the same
way as past internal network closure, using the
network of current ties as the input for the block-
density procedure instead of the network of past
ties. The networks of current ties are represented by
square matrixes sized 4,793

⫻ 4,793 in which x

ij

is

the number of projects on which specialists i and j
were currently working together.

Past structural holes. We measured past project

structural holes as the average of individual mem-
bers’ constraint in the broader network of past ties.
Network constraint “increases with the extent to
which an individual’s network is directly or indi-
rectly concentrated in a single contact. A network
concentrated in one contact means fewer structural
holes, and so less social capital” (Burt, Hogarth, &
Michaud, 2000: 135). We adopted Burt’s (1992)
measure of constraint to compute structural holes.
The index C measures the extent to which all of i’s
network is directly or indirectly invested in his or
her relationship with contact j. Thus,

c

ij

⫽ p

ij

⫹ p

ij

冉

冘

q

p

iq

p

qj

冊

2

for q

⫽ i, j, where p

ij

is the proportion of i’s rela-

tions invested in contact j and the total in paren-
theses is the proportion of i’s relations that are
directly or indirectly invested in the connection
with contact j. We summed the c

ij

across contacts j

to get the network constraint index C for each spe-
cialist in the past network, excluding the ties
among current project members. We then averaged
each specialist’s individual score at the level of the
project network to obtain the structural holes score
at the group level. We multiplied the value of con-
straint by

⫺1 in order to capture structural holes

(the “opposite” of constraint).

Aggregating structural holes to the project level

by averaging individual scores reflects our view
that structural holes are a “configurational prop-
erty” of a team or project rather than a “shared
property,” as are cultural or perceived dimensions
(Klein & Kozlowsi, 2000), or “judgments” (James,

Demaree, & Wolf, 1984). This approach is con-
sistent with the general conception of structural
holes as a form of capital that can be accumulated
in a group.

Current structural holes. We computed this

measure using the same procedure adopted for
computing the past structural holes measure but
used the network of current ties as the input for the
constraint index procedure.

Control Variables

Besides the network characteristics of project

teams for which we hypothesized effects, a number
of other factors may reasonably affect the success of
a television production. Control variables were
used to account for these factors. First, we discuss
our network controls and then our industry-
specific controls.

Network controls. The size of a project could

affect the dependent variable. We measured the
size of the project network by counting the number
of different professionals involved in a project. We
also included a control for the average age of
project network relations, given the importance of
the time dimension in our analysis. Time was im-
portant because some projects are characterized by
relatively old ties, while others include ties based
on more recent collaborations. The measure was a
weighted average (r

a

) of the age (n) of relations,

calculated using the amount of relational activity in
year t

⫺ n, with age (ranging from 7 to 1) as the

weight. To assess the sensitivity of alternative de-
cay functions, we ran the previously calculated
average age (r

a

) using different transformations as

alternative controls. We calculated log (r

a

), (r

a

)

1/2

,

and 1/(r

a

)

2

. None of our results changed. Further,

we evaluated the sensitivity of the results for alter-
native time windows.

Finally, we included past project network cen-

trality and current project network centrality in the
final model. These variables measured the degree
centrality (Freeman, 1979) of the project network
within the entire set of past and current 4,793

⫻

4,793 collaboration networks and were computed
by aggregating the degree centrality scores of
project members, with respect to relations that
members maintained outside their project team
(their external network). There were two main rea-
sons for including centrality as a control. First, our
expectation from considerable previous research
was that centrality is related to performance (Tsai &
Ghoshal, 1998). Second, when testing for the effect of
structural holes, it is appropriate to put in a degree
centrality control to ensure that the observed relation-
ship between structural holes and performance is not

2004

899

Soda, Usai, and Zaheer

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

relations

2.68

0.61

.52**

.11

⫺.01

⫺.26** ⫺.13*

.09

.12

.09

10. Size of project network

25.37

7.69

⫺.04

⫺.24** ⫺.03

.09

.17**

⫺.16**

.11

⫺.22** ⫺.08

11. Past project network

centrality

0.01

0.01

⫺.18** ⫺.11

⫺.03

⫺.17** ⫺.09

.11

.06

.15*

⫺.17** ⫺.08

12. Current project network

centrality

0.01

0.00

⫺.35** ⫺.20**

.23**

⫺.18** ⫺.09

.02

.12

.12

⫺.25** ⫺.17** ⫺.25**

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

closure

0.05

0.08

⫺.11

.04

.03

⫺.10

⫺.06

⫺.14*

.06

.21**

.06

.14

.11

.24**

.22**

.06

16. Squared current internal

network closure

0.01

0.04

⫺.05

.09

⫺.03

⫺.07

⫺.01

⫺.09

.03

.12

.02

.09

.01

.02

⫺.10

⫺.25** .83**

17. Past internal network

closure

0.25

0.50

.00

.11

⫺.07

.01

.08

⫺.16* ⫺.33**

.13*

.02

⫺.17**

.12

⫺.07

⫺.42** ⫺.19** .31** .24**

18. Squared past internal

network closure

0.25

1.17

.00

.13*

⫺.07

⫺.00

.03

⫺.11

⫺.28**

.12

.11

⫺.24**

.01

⫺.09

⫺.30** ⫺.26** .31** .27** .89**

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

spurious, a result of the relationship between degree
centrality and performance.

Past project network centrality was computed as

the average number of relations existing between
any project member and any other specialist in the
entire network of past ties (the seven-year windows
of past relations). Current project network central-
ity was computed as the average number of rela-
tions existing between any of a project’s members
and any other specialist in the entire network of
current ties.

Industry controls. Not all TV channels have the

same potential for reaching high audience levels. In
particular, the Italian broadcasting market has been
traditionally led by the two major channels of RAI
(the state-owned television corporation) and Me-
diaset (the major private competitor). Accordingly,
we included a dummy variable for major channel,
set to 1 when a production was shown on either
RAI Uno or Canale5, and 0 otherwise. Another
element that affects the audience potential of any
television program is the time slot in which the
program is broadcast. We therefore also included a
dummy control variable for prime time. Further,
TV productions have various characteristics; differ-
ent formats exist (TV movies, soaps, sitcoms, and
so on), and numbers of episodes differ. To control
for such task characteristics, we computed two ad-
ditional variables: the number of episodes (the
number actually broadcast) and a dummy variable
for TV movie (1, TV movie, 0 otherwise). Finally,
we controlled for periodicity effects by including a
series of dummies corresponding to the years 1995,
1996, 1997, and 1998, with 1999 being the omitted
category.

RESULTS

Table 1 is a matrix of the correlations, means, and

standard deviations of the variables that we used in
our analysis.

Table 2 presents our results; we employed ordi-

nary least squares (OLS) regression methods.

We begin with model 1, containing the industry

control variables, which together explain 53 per-
cent of the variance (adjusted R

2

) in our dependent

variable (audience). In model 2 we add network
controls, the size of the project network, past and
current network centrality, and the average age of
project network relations. This model explains 55
percent of the variance, and the change in variance
explained is statistically significant (p

⬍ .001).

Model 3 introduces the past and the current struc-
tural holes variables, testing our Hypothesis 1. The
change in explained variance is significant (p

⬍

.05). Our final model, model 5, further includes the

terms testing the curvilinear effect of past network
closure. The change in explained variance is again
significant (p

⬍ .05), with 58 percent explained by

this model. We report on the tests of the specific
hypotheses below.

In our first hypothesis, we posited that the effect

of current structural holes on performance would
be greater than that of past structural holes. The
hypothesis was supported, as the results of a beta
difference test indicate. Table 3 reports these re-
sults. The difference in the coefficients of the vari-
ables for current and past structural holes was sig-
nificant (t

⫽ 3.48, p ⬍ .001).

Our second hypothesis was that the U-shaped

effect of past internal network closure on project
performance would be greater than the U-shaped
effect of current internal network closure. We
tested the hypothesis in two, albeit related, ways:
First, we examined the sign and the significance of
both the linear and the squared terms of the pair of
terms that comprised the past and the current U-
shaped effects. A negative and significant coeffi-
cient for the linear term coupled with a positive
and significant coefficient for the squared term
would support a U-shaped curvilinear effect. Cur-
rent internal network closure was not statistically
significant (

␤

linear term

⫽ ⫺0.04, n.s.;

␤

squared term

⫽

0.08, n.s.), although the signs were in the expected
direction, but past internal network closure was
significant (

␤

linear term

⫽ ⫺0.27, p ⬍ .05;

␤

squared term

⫽ 0.29, p ⬍ .01). Further, we employed stepwise
regression to test for the increment in explained
variance with each pair of variables. The increment
was significant for past internal network closure
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

indicated support for Hypothesis 2.

Our control variables for the years 1995–99 were

not statistically significant in the final model, nor
was number of episodes. However, as expected, the
major channel dummy was strongly significant, as
was the one for prime time. The negative and sig-
nificant TV movie dummy suggests that the movie
genre on TV is less popular than are other formats.
Our network control variables were also not signif-
icant in the final model, with the exception of the
average age of network relations, which suggests
older network relations have a positive effect on
performance. To evaluate the sensitivity of our sev-
en-year time windows, we also estimated our mod-
els using six- and five-year time windows. Our
results remained unchanged when we used six-
year windows. In the estimation using five-year
windows, our squared term for past closure was
still significant, while the past closure term itself

2004

901

Soda, Usai, and Zaheer

became just nonsignificant, though with the right
sign. With a three-year window, the effects of the
past turn nonsignificant, as we would expect given
our theory.

To control for the possible influence of multiple

links between a few project network members, we
first computed the standard deviation of degree
centrality scores within each group in the past and
included it in the final model together with the
main effect and the squared term of past closure; it
was not significant (

␤ ⫽ 0.01, p ⫽ .92). Second, we

tested a dichotomized measure of internal density
in which relations were recoded as 1 and absent
relations as 0. We found that this measure was
significant and had the same sign as the scaled
measure we used, although the significance of the
effect was lower (

␤ ⫽ -0.11, p ⫽.05), suggesting that

the intensity of relations among a few members was
not biasing our results.

DISCUSSION

Although the network perspective presents a

conception of social context as an antithesis to the

TABLE 2

Results of Regression Analysis for Project Network Performance

a

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
centrality

.02

⫺.07

⫺.07

⫺.07

Average age of project network
relations

.18***

.15**

.14**

.11

†

Hypothesized

Current structural holes

.24***

.28***

.29***

Past structural holes

.04

.05

.02

Current internal network
closure

⫺.02

⫺.04

Squared current internal
network closure

.07

.08

Past internal network closure

⫺.27*

Squared past internal network
closure

.29**

F

32.30***

26.40***

24.00***

21.00***

19.60***

R

2

.55

.57

.59

.59

.61

Adjusted R

2

.53

.55

.57

.56

.58

⌬R

2

.02***

.02**

.02

.01**

a

Standardized regression coefficients are shown. n

⫽ 249.

†

p

⬍ .10

* p

⬍ .05

** p

⬍ .01

*** p

⬍ .001

TABLE 3

Results of Beta Difference Tests

a

Statistic

Past Structural

Holes

Current Structural

Holes

␤

0.04

0.24

b

0.13

4.53

s.d. (b

i

)

0.44

1.25

Covariance (b

1

b

2

)

0.07

Difference between

coefficients

4.41

t

3.48***

a

t

⫽ (b

i

⫺ b

j

)/ s(b

i

⫺ b

j

) for i

⫽ j, where s(b

i

⫺ b

j

)

⫽ [s

2

(b

i

)

⫹

s

2

(b

j

)

⫺ 2cov(b

i

, b

j

)]

1/2

.

*** p

⬍ .001

902

December

Academy of Management Journal

atomistic view of social and organizational action,
we argue that the network perspective should also
address the temporal and historical relational con-
text in which actors are embedded. Some prelimi-
nary evidence suggests that different network pat-
terns age differently and that some past structures
may exert stronger effects on performance than cur-
rent ones. Time may modify the nature of the flow
through a network to benefit one or the other type
of structure more and change the nature of the
relationship between structure and performance. In
this way, time can be seen as a contingent factor
shaping the efficacy of alternative network struc-
tures. This is an important question for the field as
the relative advantages of structural holes and clo-
sure as alternative sources of social capital have
been the subject of major theoretical and empirical
debate (Adler & Kwon, 2002).

Our first finding is consistent with our prediction

that the effect of bridge ties is temporary. Rather
than examining the stability of bridges per se, we
investigated whether a bridge in the past still influ-
ences present outcomes or, in other words, whether
the effect of this form of social structure on out-
comes persists over time. We found that it does not.
Neither bridge ties nor their outcomes can be ex-
pected to endure long because alters will eventu-
ally react to limit their own vulnerability derived
from structural constraint by closing up structural
holes or by gaining access to information through
the wider network to negate any brokerage advan-
tage. On the other hand, we found that structural
holes in the present help performance, as we ex-
pected. Thus, although bridge ties work well ini-
tially, they turn ineffective in the long term.

An important aspect of our contribution here is

that, rather than focusing on the durability of ties,
we evaluate the outcomes of past and current ties,
or the extent to which network structures add value
over time. One implication of these findings is that
the cost of creating and maintaining bridge ties has
to be evaluated in the context of their short-term
advantages coupled with their lack of longer-term
benefits. To the extent that relations are an invest-
ment of time and resources, the payback from
bridge ties will need to be large and quick to justify
making an investment in them.

Even though longitudinal work in the network

domain is rare, some research in the broader liter-
ature and theory supports the general notion that
closure is long-lasting. We argued for a U-shaped
relationship between past network closure and per-
formance, a result we found, confirming that clo-
sure has long-lasting benefits; we did not find any
relationship between current network closure and
performance. Our finding of a U-shaped relation-

ship also represents an advance. Specifically, in the
context of interfirm ties, prior research has exam-
ined the relationship between embeddedness and
performance, where embeddedness is defined as tie
strength (Uzzi, 1996, 1997). Our posited relation-
ship between past closure and performance relies
on structural properties of networks rather than on
tie strength, and it is U-shaped rather than inverted
U-shaped, thereby presenting intriguing prospects
for future research. Further, although we found that
current closure did not affect performance, this re-
sult may not necessarily have been due to low
creativity, as we theorized. High current closure
may yet be associated with high creativity, since
the social pressures causing conformity may not
have had enough time to limit creativity. In such a
case, we speculate, inefficient routines might more
than counterbalance high creativity, which would
explain our findings.

Another important contribution of our work is in

addressing the enduring debate in the field on the
relative value of structural holes and closure. We
help move discussion along by examining these
social structures with a focus on the contingency of
time. Our findings suggest that rather than viewing
structural holes and closure as conflicting alterna-
tive forms of social structure, they can usefully be
seen as complementary from the perspective of
time. In particular, we found that the best-perform-
ing projects were those with high past closure and
high current structural holes, which also suggests
some normative implications of our study. Managers
should compose teams considering both past working
relationships among team members and simulta-
neous links to other projects for optimal performance.

Coming to the generalizability of our empirical

findings and the boundary conditions of our the-
ory, we note that our research context is composed
of temporary network organizations in a cultural
industry, which implies that tacit knowledge and
intellectual capital are critical. Learning takes place
in an interactive and relational context. Despite
these contextual characteristics, the conditions of
our research setting might be more generalizable
than they appear at first sight. Many industries are
increasingly moving toward rapid change, instability,
and uncertainty (Eisenhardt, 1989), and in such set-
tings, organizations tend to adopt flexible forms of
organization and team-based work practices (Illnich,
D’Aveni, & Lewin, 1996). Moreover, numerous indus-
tries, such as consulting, are known to rely on tem-
porary project teams, and most group projects in or-
ganizations are temporary rather than permanent. We
view our findings as being broadly applicable in such
settings as well. The applicability of our theory may
be limited, however, when teams produce standard-

2004

903

Soda, Usai, and Zaheer

ized outputs, when the most valuable learning is in-
ternally derived, and when interdependence among
team members is low.

Limitations and Directions for Future Research

Our study, like others, has some limitations.

First, we did not formally estimate a decay function
for our major network variables, although we eval-
uated the sensitivity of different functions for our
control for the age of relations as well as the sensi-
tivity of differently sized time windows. To fully
understand the role of time, it would help to pre-
cisely calibrate the points at which closure or struc-
tural holes starts helping or hurting performance.
We could conceivably have done so with our exist-
ing data, but in our formulation we treated the past
as an accumulation of past structure and relational
experience. In the spirit of creating midrange the-
ory (Merton, 1968), we took a first step in linking
network structure and performance via the lens of
time. We leave for future research a detailed explo-
ration of decay in the network performance effect.
Further research may try to tease out whether it is
the ties that decay, or the structure, or the effect of
the structure on outcomes. Relatedly, although our
direct research question involves comparing social
structures over time, our model may contain some
endogeneity in the sense that past ties are precur-
sors to current ties (Gulati, 1995). However, when
we introduced past and current closure sequen-
tially into the regression models, we found that the
past independently affected the dependent vari-
able, and current closure did not. Future research
may explore the effect of past and current struc-
tures on outcomes in a manner that more directly
addresses any possible endogeneity.

Second, although we were extremely compre-

hensive with regard to the inclusion of the TV
productions and the specialists involved in them,
we did not distinguish among the different roles
that the specialists might play and how role differ-
entiation might influence a network and its out-
comes. It would be useful for future research to
include consideration of roles in networks as well
as the genres of TV productions. Future research
might also investigate other network variables,
such as centrality, as antecedents of performance
over time (Tsai, 2002).

A third limitation is the aggregation method we

used in computing the structural holes score for the
project teams. We averaged the constraint scores of
the team members, perhaps not capturing the di-
versity of contacts at the project level. More gener-
ally, although we conceptualized and measured
structural holes with individuals as the unit of mea-

surement, subsequently aggregating individual-level
data to the project level, the value of structural holes
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,
reducing overall network efficiency (Baker & Iyer,
1992; Stasser & Titus, 1985).

It is also possible that, although high closure in

an internal network produces groupthink and re-
duces novelty, a context of both internal and exter-
nal networks may compensate for this effect with
bridges to diverse external networks. When we
checked empirically for such compensation in our
data using an interaction term, however, we found
none. Despite this empirical result, future research
should explore this important possibility.

Concluding Remarks

We investigated the efficacy of alternative net-

work structures from the perspective of time by
posing the question, How durable are the benefits
of different forms of social capital? In doing so, we
also spoke to the debate about the relative values of
structural holes and closure, and we contributed to
its resolution by viewing these social structures
with the contingency of time. We showed that, in
the Italian television production industry, current
structural holes rather than past ones, but past clo-
sure rather than current closure, helped current
performance. Our findings suggest that network
closure casts a long shadow, while brokerage struc-
tures have only short-lived effects. Thus, closure
and structural holes are both valuable, but at dif-
ferent points in time.

REFERENCES

Adler, P. S., & Kwon, S. 2002. Social capital: Prospects

for a new concept. Academy of Management Re-
view, 27: 17– 40.

Ahuja, G. 2000a. Collaboration networks, structural

holes, and innovation: A longitudinal study. Admin-
istrative Science Quarterly, 45: 425– 455.

Ahuja, G. 2000b. The duality of collaboration: Induce-

ments and opportunities in the formation of inter-
firm linkages. Strategic Management Journal, 21:
317–343.

Amabile, T. M. 1996. Creativity in context. Boulder, CO:

Westview.

Baker, W. E., & Iyer, A. 1992. Information networks and

market behavior. Journal of Mathematical Sociol-
ogy, 16: 305–332.

Bielby, W.T., & Bielby, D. D. 1999. Organizational medi-

ation of project-based labor markets: Talent agencies

904

December

Academy of Management Journal

and the careers of screenwriters. American Socio-
logical Review, 64: 64 – 85.

Blau, P. M. 1964. Exchange and power in social life.

New York: Wiley.

Borgatti, S. P., Everett, M. G., & Freeman, L. C. 2002.

UCINET 6 for Windows. Harvard, MA: Analytic
Technologies.

Bourdieu, P. 1993. The field of cultural production:

Essays on art and literature. New York: Columbia
University Press.

Brass, D. J., Butterfield, K. D., & Skaggs, B. C. 1998.

Relationships and unethical behavior: A social net-
work perspective. Academy of Management Re-
view, 23: 14 –31.

Burt, R. S. 1992. Structural holes: The social structure

of competition. Cambridge, MA: Harvard University
Press.

Burt, R. S. 2000a. Decay functions. Social Networks, 22:

1–28.

Burt, R. S. 2000b. The network structure of social capital.

In R. I. Sutton & B. M. Staw (Eds.), Research in
organizational behavior, vol. 22: 345– 423. Green-
wich, CT: JAI Press.

Burt, R. S. 2002. Bridge decay. Social Networks, 24:

333–363.

Burt, R. S., Horgarth, R., & Michaud, C. 2000. The social

capital of French and American managers. Organi-
zation Science, 11: 123–147.

Coleman, J. S. 1988. Social capital in the creation of

human capital. American Journal of Sociology,
94(supplement): 95–120.

Eisenhardt, K. M. 1989. Making fast decisions in high-

velocity environments. Academy of Management
Journal, 32: 53–576.

Elsbach, K. D., Sutton, R. I., & Whetten, D. A. 1999.

Perspectives on developing management theory,
circa 1999: Moving from shrill monologues to (rela-
tively) tame dialogues. Academy of Management
Review, 26: 627– 633.

Faulkner, R. R., & Anderson, A. B. 1987. Short-term

projects and emergent careers: Evidence from Holly-
wood. American Journal of Sociology, 92: 879 –909.

Freeman L. C. 1979. Centrality in social networks: Con-

ceptual clarification. Social Networks, 1: 215–239.

Giuffre, K. 1999. Sandpiles of opportunity: Success in the

art world. Social Forces, 77: 815– 832.

Goodman, P. S., Lawrence, B. S., Ancona, D. G., & Tush-

man, M. L. 2001. Introduction to special issue on
time and organizational research. Academy of Man-
agement Review, 26: 507–511.

Gnyawali, D. R., & Madhavan, R. 2001. Cooperative net-

works and competitive dynamics: A structural em-
beddedness perspective. Academy of Management
Review, 26: 431– 445.

Granovetter, M. 1973. The strength of weak ties. Ameri-

can Journal of Sociology, 78: 1360 –1380.

Gulati, R. 1995. Social structure and alliance formation

pattern: A longitudinal analysis. Administrative
Science Quarterly, 40: 619 – 642.

Gulati, R., & Gargiulo, M. 1999. Where do interorganiza-

tional networks come from? American Journal of
Sociology, 104: 1439 –1493.

Hansen, M. T. 1999. The search-transfer problem: The

role of weak ties in sharing knowledge across orga-
nizational subunits. Administrative Science Quar-
terly, 44: 82–111.

Hargadon, A., & Sutton, R. I. 1997. Technology brokering

and innovation in a product development firm. Ad-
ministrative Science Quarterly, 42: 716 –749.

Hirsch, P. M. 2000. Cultural industries revisited. Orga-

nization Science, 11: 356 –361.

Ibarra, H., & Andrews, S. B. 1993. Power, social influ-

ence, and sense making: Effects of network centrality
and proximity on employee perceptions. Adminis-
trative Science Quarterly, 38: 277–303.

Illnich, A. Y., D’Aveni, R., & Lewin, A. Y. 1995. New

organizational forms and strategies for managing in
hypercompetitive environments. Organization Sci-
ence, 3: 211–220.

James, L. R, Demaree, R. G, & Wolf, G. 1984. Estimating

within-group interrater reliability with and without
response bias. Journal of Applied Psychology, 69:
85–98.

Janis, I. L. 1972. Victims of groupthink. Boston: Hough-

ton Mifflin.

Kirzner, I. M. 1973. Competition & entrepreneurship.

Chicago: University of Chicago Press.

Klein, K. J., Dansereau, F., & Hall, R. J. 1994. Levels issues

in theory development, data collection, and analysis.
Academy of Management Review, 19: 195–229.

Klein, K. J., & Kozlowski, S. W. 2000. From micro to

meso: Critical steps in conceptualizing and conduct-
ing multilevel research. Organizational Research
Methods, 3: 211–236.

Krackhardt, D. 1992. The strength of strong ties: The

importance of philos in organizations. In N. Nohria &
R. G. Eccles (Eds.), Networks and organizations:
Structure, form, and action: 216 –239. Boston: Har-
vard Business School Press.

Krackhardt, D. 1998. Simmelian ties: Super strong and

sticky. In R. M. Kramer & M. A. Neale (Eds.), Power
and influence in organizations: 21–38. Thousand
Oaks, CA: Sage.

Madhavan, R., Koka, B. R., & Prescott, J. E. 1998. Net-

works in transition: How industry events (re)shape
interfirm

relationships.

Strategic

Management

Journal, 19: 439 – 459.

Merton, R. K. 1968. Social theory and social structure.

New York: Free Press.

2004

905

Soda, Usai, and Zaheer

Miller, D., & Shamsie, J. 1996. The resource-based view

of the firm in two environments: The Hollywood
film studios from 1936 to 1965. Academy of Man-
agement Journal, 39: 519 –543.

Morgan, G. 1989. Creative organization theory. Beverly

Hills, CA: Sage

Nahapiet, J., & Ghoshal, S. 1998. Social capital, intellec-

tual capital, and the organizational advantage. Acad-
emy of Management Review, 23: 242–266.

Perry-Smith, J. E., & Shalley, C. E. 2003. The social side of

creativity: A static and dynamic social network per-
spective. Academy of Management Review, 28:
89 –106.

Podolny, J. M. 2001. Networks as the pipes and prisms of

the market. American Journal of Sociology, 107:
33– 60.

Podolny, J. M., & Baron, J. N. 1997. Resources and rela-

tionships: Social networks and mobility in the work-
place. American Sociological Review, 62: 673– 693.

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.

Powell, W. W., & Smith-Doerr, L. 1994. Networks and

economic life. In N. J. Smelser & R. Swedberg (Eds.),
The handbook of economic sociology: 368 – 402.
Princeton, NJ: Princeton University Press.

Rowley, T., Behrens, D., & Krackhardt, D. 2000. Redun-

dant governance structures: An analysis of structural
and relational embeddedness in the steel and semi-
conductor industries. Strategic Management Jour-
nal, 21: 369 –286.

Starkey, K., Barnatt, C., & Tempest, S. 2000. Beyond

networks and hierarchies: Latent organizations in
the UK television industry. Organization Science,
11(3): 299 –305.

Stasser, G., & Titus, W. 1985. Pooling of unshared infor-

mation in group decision-making: Biased informa-
tion sampling during discussion. Journal of Person-
ality and Social Psychology, 48: 1467–1478.

Suitor, J. J., Welman, B., & Morgan, D. L. 1997. It’s about

time: How, why, and when networks change. Social
Networks, 19: 1–7.

Tsai, W. 2002. Social structure of “coopetition” within a

multiunit organization: Coordination, competition,
and intraorganizational knowledge sharing. Organi-
zation Science, 13(2): 179 –190.

Tsai, W., & Ghoshal, S. 1998. Social capital and value

creation: The role of intrafirm networks. Academy
of Management Journal, 41: 464 – 476.

Uzzi, B. 1996. The sources and consequences of embed-

dedness for the economic performance of organiza-
tions. American Sociological Review, 61: 674 – 698.

Uzzi, B. 1997. Social structure and competition in inter-

firm networks: The paradox of embeddedness. Ad-
ministrative Science Quarterly, 42: 35– 67.

Walsh, J. P., & Ungson, G. R. 1991. Organizational mem-

ory. Academy of Management Review, 16: 57–91.

Weick, K. E., & Roberts, K. H. 1993. Collective mind in

organizations: Heedful interrelating on flight decks.
Administrative Science Quarterly, 38: 357–381.

White, H. C. 1992. Agency as control in formal networks.

In N. Nohria & R. G. Eccles (Eds.), Networks and
organizations: Structure, form, and action: 92–117.
Boston: Harvard Business School Press.

Zaheer, S., Albert, S., & Zaheer, A. 1999. Time scales and

organization theory. Academy of Management Re-
view, 24: 725–741.

Giuseppe Soda ([e-mail]; Ph.D,
Bocconi University) is an associate professor of organi-
zation theory at Bocconi University of Milan. He also is
the director of the Department of Organization and Hu-
man Resource Management at Bocconi University School
of Management. He has been a visiting assistant professor
at Carnegie Mellon University. His research interests and
publications include micro and macro networks, organi-
zational design, and new forms of organization.

Alessandro Usai ([e-mail]) earned
his Ph.D. at Bologna University; he has been a visiting
scholar at the University of Minnesota and an assistant
professor at Bocconi University of Milan. He is currently
the managing director of Cinecitta´ Holding, the major
Italian film production organization. His research inter-
ests include network analysis and studies of performance
in cultural industries.

Akbar Zaheer ([e-mail]; Ph.D., Massa-
chusetts Institute of Technology) is the Curtis L. Carlson
Professor of Strategic Management and Organization and
the director of the Strategic Management Research Center
at the Carlson School of Management, University of Min-
nesota. Besides networks, his research interests include
trust in organizational contexts, strategic alliances, and
mergers and acquisitions.

906

December

Academy of Management Journal