Quarterly Administrative Science: Collaboration Networks, Structural Holes, and Innovation: A Longitudinal Study

Administrative Science
Quarterly http://asq.sagepub.com/
Collaboration Networks, Structural Holes, and Innovation: A Longitudinal Study

Gautam Ahuja
Administrative Science Quarterly 2000 45: 425
DOI: 10.2307/2667105
The online version of this article can be found at:

http://asq.sagepub.com/content/45/3/425
Published by:
http://www.sagepublications.com
On behalf of:
Samuel Curtis Johnson Graduate School of Management at Cornell University
Additional services and information for Administrative Science Quarterly can be found at:
Email Alerts: http://asq.sagepub.com/cgi/alerts
Subscriptions: http://asq.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://asq.sagepub.com/content/45/3/425.refs.html
>> Version of Record - Sep 1, 2000
What is This?
Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

CollaborationNetworks, To assess the effects of a firm's network of relations on
StructuralHoles, and innovation, this paper elaborates a theoretical framework
that relates three aspects of a firm's ego network-direct
Innovation:A ties, indirect ties, and structural holes (disconnections
LongitudinalStudy between a firm's partners)-to the firm's subsequent
innovation output. It posits that direct and indirect ties
Gautam Ahuja both have a positive impact on innovation but that the
University of Texas at Austin impact of indirect ties is moderated by the number of a
firm's direct ties. Structural holes are proposed to have
both positive and negative influences on subsequent
innovation. Results from a longitudinal study of firms in
the international chemicals industry indicate support for
the predictions on direct and indirect ties, but in the inter-
firm collaboration network, increasing structural holes
has a negative effect on innovation. Among the implica-
tions for interorganizational network theory is that the
optimal structure of interfirm networks depends on the
objectives of the network members.@
Several recent studies have indicatedthat the positions of
firms in interorganizationalnetworks influence firm behavior
and outcomes (e.g., Powell, Koput,and Smith-Doerr,1996;
Walker,Kogut,and Shan, 1997). Because of their facilitative
role in variousinterorganizationalcontexts, network relation-
ships have even been described as network resources
(Gulati,1999). Inspite of the growing consensus that net-
works matter,however, the specific effects of differentele-
ments of networkstructureon organizationalperformance
remainunclear.Inthe social networks literature,a debate has
arisen over the form of networkstructuresthat can appropri-
ately be regardedas beneficial(Walker,Kogut,and Shan,
1997). Accordingto one view, densely embedded networks
with many connections linkingego's alters are facilitativefor
ego, and social structuresare seen as advantageous to the
extent that networks are "closed" (Coleman,1988; Walker,
? 2000 by Cornell University. Kogut,and Shan, 1997). Accordingto an alternateview, how-
0001-8392/00/4503-0425/$3.00. ever, social structuraladvantages derive from the brokerage
0 opportunitiescreated by an open social structure(Burt,
I especially thank Will Mitchell, Mark
1992). Actors can buildrelationshipswith multiplediscon-
Mizruchi, and Anand Swaminathan for nected clusters and use these connections to obtain informa-
their many inputs into this research. Addi- tion and controladvantages over others (Burt,1992). From
tionally, for their research assistance
and/or helpful comments I thank Riitta the perspective of the networktheorist, these differences
Katila,Jennifer Keene, Curba Morris Lam- have different,even contradictory,normativeimplications
pert, and Sai Yayavaram. An earlier ver-
sion of this paper received the West Pub-
(Walker,Kogut,and Shan, 1997). FromColeman's(1988)
lishing Best Paper Award and the standpoint,the optimalsocial structureis one generated by
Sage-Louis Pondy Best Dissertation- buildingdense, interconnectednetworks. FromBurt's(1992)
based Paper Awards from the OMT Divi-
sion of the Academy of Management. position, constructingnetworks consisting of disconnected
The dissertation on which this paper is alters is the optimalstrategy. Clarifyingthe implicationsof
based received the Best Dissertation cohesive versus disconnected networkstructuresfor various
Award from the BPS Division of the Acad-
emy of Management and the Best Disser- organizationaloutcomes is importantto our understandingof
tation Proposal Award from the INFORMS network resources.
College on Organization Science. I grate-
fully acknowledge the financial support of
the University of Texas and the Center for
Relatedly,recent research has led to the importantinsight
International Business Education and that buildingnetworks with large numbers of indirectties
Research at the University of Michigan. I may be an effective way for actors to enjoy the benefits of
especially thank Don Palmer and three networksize without payingthe costs of network mainte-
anonymous reviewers of this journal
whose insightful feedback immeasurably nance associated with directties (Burt,1992). Althoughsuch
improved this paper. Finally, I would like a strategy is undoubtedlyconceptuallyattractive,it appears
to thank Linda Johanson for her editing,
which significantly enhanced the readabili- likelythat its value in a given circumstancewill be contingent
ty and clarity of the final paper. on several factors. Specifically,the relativevalue of direct ver-
425/AdministrativeScience Quarterly,45 (2000): 425-455

sus indirectties is likelyto depend on the degree to which
the benefits providedby direct and indirectties are similarin
magnitudeand content. To the extent that direct ties provide
differenttypes or amounts of benefits, the possibilitiesof
substitutionbetween direct and indirectties may be limited.
Thus, examiningthe content and relativecontributionof
direct and indirectties may also be relevantfrom the per-
spective of designing effective and efficient networks.
Inthis study, I examine the relationshipbetween a firm's
position in the industrynetworkof interfirmcollaborativelink-
ages and its innovationoutput, a significantorganizational
outcome. Scholars in the innovationand interorganizational
learningliteratureshave argued that linkagesand the resul-
tant collaborationnetworks are key vehicles throughwhich
firms obtainaccess to external knowledge (Powell, Koput,
and Smith-Doerr,1996). Examiningthe relationshipbetween
network position and innovationoutput can provideboth an
elucidationof the role of differentelements of networkstruc-
ture in the innovationprocess and an empiricalindicatorof
the effectiveness of knowledge flows throughsuch net-
works. Forthe purposes of this study, I define an interfirm
collaborativelinkageas a voluntaryarrangementbetween
independentorganizationsto share resources. Further,fol-
lowing past research, I make a distinctionbetween collabora-
tive arrangementsthat involvea technologicalcomponent,
such as developing a new technology or sharinga manufac-
turingprocess, and collaborativearrangementsthat are
focused purelyon sharingmarketingassets or brandnames
(Hagedoornand Schakenraad,1994; Singh and Mitchell,
1996). Similarly,I make a distinctionbetween horizontaland
verticallinkages (Stuart,1998; Gulatiand Lawrence, 1999).
Foranalyticclarityand focus, in this paper,I restrictmy atten-
tion to horizontal,technical linkages, i.e., technical linkages
between firms in the same industry.
NETWORKSTRUCTUREAND INNOVATION
OUTPUT
Althoughsociologists have long studied the relationship
between networkstructureand innovation,most research in
this traditionhas largelyfocused on the adoptionor diffusion
of innovations.Even though articles in the popularpress and
academic reportsof the innovation-generationprocess have
consistently used network metaphors, untilrecently,relative-
ly littlework has actuallyused a networkanalyticapproachto
study innovationgeneration. Recently, however, a few pio-
neering studies have explored network structurefrom the
perspective of innovationgeneration (Shan,Walker,and
Kogut,1994; Podolnyand Stuart, 1995; Powell, Koput,and
Smith-Doerr,1996). Forinstance, Podolnyand Stuart(1995)
exploredthe factors that determine whether an innovation
becomes a technologicaldead end or serves as the basis for
subsequent innovations.They found that this outcome was
predictedby the patternof ties in the technological niche of
the innovationas well as by the qualityof the innovationand
the status of the innovator,but they did not directlyexamine
the role of the interfirmnetwork structureas a predictorof
innovationoutput.
426/ASQ, September 2000

Collaboration Networks
Two other studies that have exploredthe nexus between

network structureand innovationperformanceserve as one
proximatepoint of departurefor the currentresearch.1Ina
study of biotechnologystart-ups,Shan, Walker,and Kogut
(1994) predictedand found that one element of a firm's net-
work position, the numberof collaborativerelationshipsit
formed, was positivelyrelatedto its innovationoutput.
Throughblock modeling,they also developed a more sophis-
ticated measure of a firm'snetwork position and found that
this measure was a good predictorof linkageformation,but
they did not explore the possibilitythat elements of a firm's
ego network, other than the numberof directties, might
influence innovationoutput. Inanother study, Powell, Koput,
and Smith-Doerr(1996) traced the formationof interfirm
learningnetworks for biotechnologystart-upsand found that
centralityin such networks is relatedto faster subsequent
growth (in numberof employees) for the start-ups,but they
also did not directlyexamine the impactof network positions
on innovation.
A recent stream of literaturethat has examined the role of
differentnetwork structures in facilitatingoutcomes for net-
work constituents forms the other point of departurefor this
study. In his book, Burt(1992) made a strong case for the
strategic configurationof networks. Accordingto this concep-
tion, designing networks to maximizedisconnections (or
structuralholes) between alters and selecting alters with
many other partners(or many indirectties) are two mecha-
nisms by which actors can develop efficient and effective
networks. This conception, however, raises two issues. First,
as some scholars have noted, the normativeimportance
accorded to structuralholes by this approachis at odds with
other theoreticalperspectives that stress the importanceof
closed social networks (Walker,Kogut,and Shan, 1997). Sec-
ond, the strategy of substitutingindirectties for direct ones
that is endorsed by the effective networks conception pre-
sumes that direct and indirectties offer the same content to
the focal actor.The validityof that assumption may varysig-
nificantlyacross networks.
The two issues raised above have importantimplicationsfor
modelingthe impact of network structureon organizational
outcomes. Forinstance, the debate on structuralholes sug-
gests that an accurate understandingof the role of structural
holes in the collaborationnetwork must account for both
Coleman'sand Burt'svariantsof the argument. Similarly,rec-
ognizingthe possibilitythat even within the same network,
direct and indirectties may vary in their content highlights
the importanceof decomposing the firm'sego network into
I
Other studies that have examined the distinct and separate elements and identifyingthe contents
relationship between collaboration and transmittedthrougheach type of tie.
innovation include Berg, Duncan, and
Friedman (1982) and Hagedoorn and Inthe technologicalcollaborationnetworkthat I studied,
Schakenraad (1994), albeit from slightly
different perspectives. Berg, Duncan, and
interfirmcollaborativelinkagesare associated with two dis-
Friedman (1982) examined the impact of tinct kinds of network benefits. First,they can providethe
research collaboration on research expen- benefit of resource sharing,allowingfirms to combine knowl-
ditures and profitability.Similarly, Hage-
doorn and Schakenraad (1994) related col- edge, skills, and physicalassets. Second, collaborativelink-
laboration to profitability. Neither of these ages can provideaccess to knowledge spillovers,serving as
studies, however, directly examined the informationconduits throughwhich news of technical break-
impact of collaboration on innovative out-
put or used a network perspective. throughs, new insights to problems, or failed approaches
travels from one firmto another.In distinguishingbetween
the resource-sharingand knowledge-spilloverbenefits of col-
laboration,it is importantto distinguishbetween know-how
and information(Kogutand Zander,1992). Know-howentails
accumulatedskills and expertise in some activityand is likely
to includea significanttacit or noncodifiabledimension. Infor-
mation refers primarilyto facts, discrete quantaof informa-
tion that can be transmittedthroughsimple communication
in relativelycomplete form and without loss of integrity
(Kogutand Zander,1992; Szulanski,1996). The resource-shar-
ing benefits of collaborationrelate primarilyto the transfer
and sharingof know-how and physicalassets, while the
knowledge-spilloverbenefits are likelyto involve predomi-
nantlyinformation.
Three aspects of a firm's networkstructureare likelyto be
relevant in connection with the above benefits: (1) the num-
ber of direct ties maintainedby a firm, (2) the numberof indi-
rect ties maintainedby the firm (the firms it can reach in the
networkthroughits partnersand their partners),and (3) the
degree to which a firm'spartnersare linkedto each other
(i.e., whether there are structuralholes in the firm'sego net-
work). Figure1 identifies these three dimensions in the ego
networks of two illustrativefirms. FirmA has three direct
ties, to partnersB, C, and D. FirmA also has nine indirect
ties, the nine firms (Ethrough M) that it can reach throughits
partnersor their partners.Further,its partners,B, C, and D,
are all tied to each other, creatinga closed networkwith no
structuralholes (from FirmA's perspective). In comparison,
Firm1 has more direct ties (Firms2, 3, 4, and 5) but fewer
(onlytwo) indirectties (Firms6 and 7). Finally,its partnersare
unconnected to each other, creatingan open networkwith
several structuralholes (the gaps between partners2, 3, 4,
and 5).
Eachof these three dimensions of a firm's network, its direct
ties, indirectties, and connections between partners,can
influencethe firm's innovationperformance.A firm'sdirect
ties potentiallyprovideboth resource-sharingand knowledge-
spilloverbenefits. Indirectties do not entail formal resource-
sharingbenefits but can provideaccess to knowledge
Figure 1. Illustrationof direct ties, indirect ties, and structural holes in two networks.
E
L 9
_ B 4 god,6F
2
G 6
3
H M 7
Ac1
4
8
J ~~~~~~~~~~~5
D
K

spillovers. Finally,the degree of connectivitybetween a firm's

partnersinfluences both resource sharingand access to
novel information,albeit in contradictoryways.
Direct Ties, IndirectTies, and Innovation
The numberof directties a firm maintainscan affect its inno-
vative output positivelyby providingthree substantive bene-
fits: knowledge sharing,complementarity,and scale. First,
directties enable knowledge sharing(Berg, Duncan,and
Friedman,1982). When firms collaborateto develop a tech-
nology,the resultantknowledge is availableto all partners.
Thus, each partnercan potentiallyreceive a greater amount
of knowledge from a collaborativeprojectthan it would
obtainfrom a comparableresearch investment made inde-
pendently. Forinstance, if two firms contributean amount of
$X each to a collaborativeR&Deffort, then $2X should be
the amount of collaborativeR&Davailableto each firm, in
additionto any internalR&Ddone by each firm.
Second, collaborationfacilitates bringingtogether comple-
mentaryskills from differentfirms (Richardson,1972; Arora
and Gambardella,1990). Technologyoften demands the
simultaneous use of differentsets of skills and knowledge
bases in the innovationprocess (Aroraand Gambardella,
1990; Powell, Koput,and Smith-Doerr,1996). Developing
multiple,broadcompetencies or maintainingthem in the face
of rapidtechnologicalchanges, however, are difficultfor firms
(Mitchelland Singh, 1996). Transactionalconsiderationsmay
make the purchase of such technologies infeasibleor prohibi-
tive, leaving internaldevelopment and collaborationas the
only viable alternatives(Mitchelland Singh, 1996). Under
such circumstances, collaborationcan enable firms to enjoy
economies of specializationwithout the priorinvestments
entailed by internaldevelopment. By tapping into the devel-
oped competencies of other firms, firms can enhance their
own knowledge base and thereby improvetheir innovation
performance.
A thirdpositive effect of direct ties emerges throughscale
economies in research that arise when largerprojects gener-
ate significantlymore knowledge than smaller projects. Col-
laborationenables firms to take advantage of such scale
economies. If individualfirms have the wherewithalto invest
an amount X in a given research project,then two firms com-
biningresources can potentiallyinvest twice as much. If the
transformationtechnology is characterizedby increasing
returns,such an investment will lead to a more than propor-
tionate returnin terms of innovationoutput, benefiting both
firms significantly.Priorresearch has also shown a positive
impact of interfirmcollaborationon innovationoutput. Ina
study of biotechnologystart-ups,Shan, Walker,and Kogut
(1994) found that the greaterthe numberof collaborativelink-
ages formed by a start-up,the higherthe numberof patents
it obtained.Thus, other things being equal, I suggest:
Hypothesis 1: The more directties that a firm maintains,the

greaterthe firm'ssubsequent innovationoutput.
A firm'scollaborativelinkages can also provideaccess to
knowledge spillovers. Knowledgeflows between firms and
industriesare constituted of both contractualknowledge
transfers and relativelyinformal,uncompensated knowledge
spilloversor leakages (Jaffe, 1986; Bernsteinand Nadiri,
1989; Jaffe, Trajtenberg,and Henderson, 1993). Collaborative
linkages represent arenas of sustained, focused, and relative-
ly intense interaction(Auster,1992). They involve repeated
and regularmeetings between the partners,a focus on speci-
fied objectives, and entail coordination,close contact, and
mutualdependency (Gulatiand Singh, 1998). Sustained inter-
action is characterizedby frequent communication.Focused
interactionimplies that these ties will be used, significantly,
and perhaps predominantly,to communicate on a narrow
range of issues relatingto the objectives and subject of col-
laboration(Rogers and Kincaid,1981). Intense interactionindi-
cates that the partnershave both a great incentive and
opportunityto share information(Granovetter,1973, 1982;
Boorman,1975; Krackhardt, 1992). Giventhese characteris-
tics of the information-exchangeprocess embodied in an
interfirmlinkage,an interfirmlinkageis an importantchannel
of communicationbetween the firmand its direct partners.
An interfirmlinkagecan also be a channel of communication
between the firmand many indirectcontacts (Mizruchi,1989;
Davis, 1991; Haunschild,1993; Gulati,1995). A firm'spart-
ners bringthe knowledge and experience from their interac-
tions with their other partnersto their interactionwith the
focal firm,and vice versa (Gulatiand Garguilo,1999). A firm's
linkagestherefore provideit with access not just to the
knowledge held by its partnersbut also to the knowledge
held by its partner'spartners(Gulatiand Garguilo,1999). The
networkof interfirmlinkagesthus serves as an information
conduit,with each firm connected to the network being both
a recipientand a transmitterof information(Rogers and Kin-
caid, 1981).
The role of the interfirmnetworkas an informationchannel
and facilitatorof knowledge exchange between firms can be
significantin the technologicalcontext (Powell, Koput,and
Smith-Doerr,1996). Innovationis often an information-inten-
sive activityin terms of both informationcollection and infor-
mation processing. Individualfirms can pursue only a limited
numberof technologies and lines of research, but the net-
work can increase a firm'scatchment area for information
and providebenefits in two forms. First,it can serve as an
information-gathering device (Freeman,1991). Firmscan
receive informationon the success and failureof many simul-
taneous research efforts (Rogers and Larsen, 1984). Promis-
ing technologicaltrajectoriesas well as technologicaldead
ends can be broughtto the early notice of a firmthat is
plugged into the network. Second, the networkcan serve as
an information-processingor screening device (Leonard-Bar-
ton, 1984). Eachadditionalnode that a firm has access to can
serve as an information-processingmechanism, absorbing,
sifting, and classifying new technical developments in a man-
ner that goes well beyond the information-processingcapabil-
ities of a single firm. Relevantdevelopments in different
technologies may be broughtto the firm'sattentionthrough
its links,some of whom may specialize in those technologies
or work with partnerswho specialize in them (Freeman,
1982). Alternately,faced with a specific problem,a firmcan
activate its networkto identifythe sources that are likelyto
be well informedabout the specific issue at hand (Freeman,

1982). Otherthings being equal, firms that have many indi-

rect ties, are privyto more informationthan firms whose
reach in the network is more limited,which is likelyto have
an effect on innovation:
Hypothesis 2: The greatera firm'snumberof indirectties, the
greaterthe subsequent innovationoutput of the firm.
The degree to which indirectties benefit the focal firm, how-
ever, is likelyto be contingent on the numberof the focal
firm'sexisting direct ties, such that firms with few direct ties
are likelyto enjoy greater benefits from their indirectties
than firms with many direct ties. Two arguments supportthis
line of reasoning. First,the relativeadditionto knowledge
throughindirectties is likelyto be greaterfor firms with few
directties than for firms with many direct ties. Forfirms with
limitedaccess to the networkthroughdirectties, the infor-
mation providedby indirectties may represent a significant
incrementto the firm'sexisting informationbase, while firms
with many directties are alreadyprivyto a significantpropor-
tion of the knowledge flow of the networkthroughtheir
directties. The additionalaccess to informationprovidedby
their indirectties may then represent only a marginalincre-
ment in their knowledge base.
Second, firms with many direct ties may also be more limited
in their abilityto profitfrom informationfrom their indirect
ties. When a firm'spartnershave many connections, the
informationthat reaches the firmthroughthe networkalso
reaches many others, the other partnersof its partners.
These partnerspotentiallyrepresent competitionfor the firm
in using this information.When informationcirculatesamong
many potentialusers, the alertness, responsiveness, and
flexibilityof individualusers is likelyto determine the benefit
that they obtainfrom it (Zaheerand Zaheer,1997). Firmswith
many directties may be more constrainedin their abilityto
absorb new informationor respond to it as flexiblyas firms
with few direct ties (Glasmeier,1991). Firmswith many
direct ties, being in the thick of things, are less likelyto add
to their knowledge or to absorb as much knowledge through
their indirectties than are firms with few directties, which is
likelyto have an effect on innovation:
Hypothesis 3: The impactof indirectties on a firm'sinnovationout-
put will be moderatedby the level of the firm'sdirectties: the
greaterthe numberof directties, the smallerthe benefit from indi-
rect ties.
Structural Holes and Innovation
Recent research suggests that a thirddimension of a firm's
ego network is also likelyto be importantto innovation:the
degree of connectivity(orthe lack of it) between a firm's
partners(Burt,1992). Accordingto Burt'sstructuralhole the-
ory,ties are redundantto the degree that they lead to the
same actors. Structuralholes are gaps in informationflows
between alters linkedto the same ego but not linkedto each
other. A structuralhole indicates that the people on either
side of the hole have access to differentflows of information
(Hargadonand Sutton, 1997). Ego networks richin structural
holes implyaccess to mutuallyunconnected partnersand,
consequently, to many distinct informationflows. Thus, maxi-
mizingthe structuralholes spanned or minimizingredundan-
cy between partnersis an importantaspect of constructing
an efficient, information-rich
network (Burt,1992).
Fromthe perspective of structuralhole theory, ego networks
in which a firm'spartnershave no linkswith each other are
preferredto networks in which its partnersare densely tied
to each other, but examiningthe impact of a network richin
structuralholes on the resource-sharingbenefits of the net-
work reveals a conclusion that is almost diametricallyoppo-
site to the conclusion reached by relatingknowledge spillover
or informationbenefits to the same network structure.The
resource-sharingbenefits of collaborationarise from firms
combiningtheir skills, sharingtheir knowledge, and conduct-
ing joint projectsto obtain scale economies, all of which pre-
sume the existence of significanttrust between the partners.
Withouttrust and shared norms of behavior,sharingknowl-
edge, combiningskills, and makinglargejoint investments
are likelyto be difficultand unproductivein any context (Cole-
man, 1988). In horizontalnetworks of competitors, the basic
problemof coordinatinginterorganizational relationshipsis
worsened by a heightened threat of opportunisticbehavior
(Gulatiand Singh, 1998). Fromstealing partners'technology
to providingpoorerqualityinvestments on joint projects, to
not fulfillingex ante commitments, alliances offer many
opportunitiesfor cheating. The development of trust and the
reductionof opportunismare then likelyto be necessary pre-
conditionsfor successful resource sharing.
Extensive relationsbetween partnerscan foster the develop-
ment of shared norms of behaviorand explicit interorganiza-
tional knowledge-sharingroutines (Uzzi,1997; Walker,Kogut,
and Shan, 1997; Dyerand Noboeka, 2000). The social con-
straintsassociated with dense, embedded networks can
facilitatelarge relationship-specificinvestments that help
maximizethe benefits from collaboration(Walker,Kogut,and
Shan, 1997). Common partnerscan serve as referralagents
and relayexpectations and responsibilitiesas partof the
process of bringingtogether two hithertounconnected firms
(Gulati,1995; Uzzi, 1997). They can also use their relation-
ships with both partiesto encourage cooperation,reciprocity,
and sharing(Uzzi,1997; Gulati,1999). Deeply embedded net-
works can also foster fine-grainedinformationtransferand
joint problemsolving (Uzzi,1997), two essential components
of successful resource sharing.
Dense ties between partnersare also likelyto help in curbing
opportunism(Coleman,1988; Walker,Kogut,and Shan, 1997;
Rowley, Behrens, and Krackhardt, 2000). In closed networks,
in which ego's partnersare connected to each other, informa-
tion about one actor's opportunisticacts diffuses rapidlyto
other relatedactors, and sanctions for deviant behaviorare
more easily imposed (Walker,Kogut,and Shan, 1997). Fur-
ther, in such a highlyinterconnectedsystem, deviant behav-
ior is less likelyto arise because the threat of reputationloss
with respect to multiplepartnerswill discourage firms from
behavingopportunisticallywith any single partner.By con-
trast, in an open network in which ego's partnersare not

linkedto each other, the possibilityof opportunisticactions is

greater.
The contradictoryeffects of connections between partners
thus prompttwo competing predictionswith respect to the
relationshipbetween structuralholes and innovation.Many
structuralholes in ego's networkwill increase ego's access
to diverse informationand, hence, enhance innovationout-
put. Conversely,ego networks with fewer structuralholes
might promote trust generationand reduce opportunism,
leadingto more productivecollaborationfrom the perspective
of resource sharing.
Hypothesis 4a: The greaterthe structuralholes spanned by a firm,
the greaterthe firm'ssubsequent innovationoutput.
Hypothesis 4b: The greaterthe structuralholes spanned by a firm,

the less the firm'ssubsequent innovationoutput.
METHODS
I chose to conduct my research in the chemicals industryfor
several reasons. First,technologicalcollaborationhas been
and continues to be a significantfeature of this industry.Sec-
ond, patents are a meaningfulmeasure of innovationin this
industry.The linkbetween patents and innovationis likelyto
be stronger in industriesin which patents providefirms with
fairlystrong protectionfor their proprietaryknowledge. Prior
research indicates that the chemicals industryis one in which
patents are generallyregardedto be effective and used wide-
ly and consistently, relativeto most other industries(Levinet
al., 1987). 1 tested the hypotheses on a longitudinaldata set
comprisingthe linkageand patentingactivities of 97 leading
firms from the chemicals industryin Western Europe,Japan,
and the United States. The sample was selected to include
the largest chemicals firms in these three areas, which con-
stitute the core of the global chemicals industry,to ensure
the availabilityand reliabilityof data. Informationon the key
variable,collaborativelinkages, is extremely difficultto obtain
for smallerfirms over an extended time period. Past network
studies on alliances have used a similarstrategy of focusing
on the leadingfirms in an industry(Gulati,1995; Gualtiand
Garguilo,1999).
Innovationoutput, the dependent variable,was measured
throughthe patentingfrequency of each firm,the numberof
patents received in a given year. Patents are an important
measure of innovationoutput because they are directlyrelat-
ed to inventiveness, they represent an externallyvalidated
measure of technologicalnovelty (Griliches,1990), and they
confer propertyrightson the assignee and therefore have
economic significance (Kamienand Schwartz, 1982; Scherer
and Ross, 1990). Further,empiricalstudies have shown that
patents are closely relatedto measures such as new prod-
ucts (Comanorand Scherer, 1969), innovationand invention
counts (Kleinecht,1982; Basberg, 1983; Achilladelis,
Schwarzkopf,and Cines, 1987), and sales growth (Scherer,
1965). Expertratingsof corporatetechnologicalstrength.have
also been found to be highlycorrelatedwith the numberof
patents held by corporations(Narin,Noma, and Perry,1987).
The use of patents as a measure of innovativeoutput has
some limitations,too. Some inventionsare not patentable,
others are not patented for strategic reasons. Further,firms
may differ in their patentingpropensity(Cohenand Levin,
1989; Griliches,1990). Research, and the logic of appropri-
ability,indicatethat the degree to which these factors are a
problemvaries significantlyacross industries(Levinet al.,
1987; Cohen and Levin,1989). This insight providesa partial
solutionto the problem,in that an appropriateresearch
design can be used to controlfor interindustrydifferences in
patentingpropensity(Basberg, 1987). Limitingthe study to a
single industrialsector in which patents are a meaningfulindi-
cator of innovationminimizessuch problems, as the factors
that affect patentingpropensityare likelyto be stable within
such a context (Basberg, 1987; Cohen and Levin,1989;
Griliches,1990). Because, even withinan industry,firms
might differ in patentingpropensityfor unobserved reasons, I
treated this as a problemof unobserved heterogeneity and
controlledfor such variationsthroughmy statisticalapproach.
I identifiedthe leadingfirms in the chemicals industryfrom
lists that are publishedannuallyby trade journalssuch as
ChemicalWeek and C&ENews. To minimizesurvivorbias, I
selected the sample from the lists at the beginningof the
study period. Inthese publishedlists subsidiarieswere often
listed separatelyfrom parentfirms. Froman originalsample
of approximately120 firms, after matchingsubsidiarieswith
their parentfirms, a sample of 107 firms remained.Forten of
these firms, reliablepatent data or covariatedata could not
be obtained, and they were droppedfrom the analysis. The
remainingfirms includeall the key players in the industry
over the study period.The panel is unbalanced,as some of
the firms were acquiredby other firms or restructuredso as
to make comparisondifficultbeyond a particularyear. A full
list of the sample firms is availablefrom the author.
Data
I obtainedyearly patent counts, collaborationdata, and firm-
attributedata for the firms in the sample. The panel used for
the analysis includes collaborativeactivityfor the period
1981-1991 and patentingactivityfor the period 1982-1992,
reflectinga one-year lag between collaborationand patenting.
I used U.S. patent data for all firms, includingthe foreign
firms in the sample, to maintainconsistency, reliability,and
comparability,as patentingsystems across nations differ in
the applicationof standards,system of grantingpatents, and
value of protectiongranted.The U.S. represents one of the
largest marketsfor chemicals, and firms desirous of commer-
cializingtheir inventionswould patent in the United States if
they were to patent anywhere at all. This observationis sup-
ported by statistics from the U.S. Patent Office, which indi-
cate that almost half of all U.S. patents are issued to foreign
entities. Priorresearch using patent data on international
samples has followed a similarstrategy of using U.S. patent
data for internationalfirms (Stuartand Podolny,1996; Stuart,
1998).
To obtain patent counts for each firm, I prepareda list for
each firm in the sample of all its divisions, subsidiaries,and

joint ventures, using Who Owns Whom (UnitedStates, Unit-

ed Kingdomand Ireland,ContinentalEurope,and Asia edi-
tions) and the Directory of Corporate Affiliations. Thereafter, I
traced each firm'shistorythroughthe study periodto
account for any name changes and reorganizationsand to
obtain informationon the timings of events such as the
foundingand dissolutionof jointventures. This master list
was used to identifyall patents issued to the sample firms.
The list of chemicals patents owned by these firms was
derivedfrom the above master list using the technology-class
informationon the patents. The U.S. patent system classifies
the technology domain into 400 broadclasses and several
hundredthousand subclasses nested withinthe classes.
Patent examiners assign each patent to a primarytechnology
class. The Patent Manualwas used to identifythe technology
classes correspondingto chemicals. Using the identified
technology classes, I separated the chemicals patents of the
sample firms from other patents they obtained. Finally,I com-
puted the patent count for each firmfor each year by assign-
ing to that firmall chemicals patents issued solely to the firm
or to its subsidiariesand half of the chemicals patents issued
jointlyto the firm and a partneror issued to jointventures of
the firm.This was done to avoid spurious inflationof patent
counts throughdouble countingof jointlyheld patents.
I obtained data on collaborationthroughdetailed archival
research on the chemicals and materialssector, using three
main types of data sources to identifylinkageactivity:(1)
electromagneticdatabases, includingboth general business
news media such as the Dow Jones News RetrievalText
Indexand sector-specific databases such as Metadex, (2)
general business printmedia, such as the Frostand Sullivan
Predicasts Index(U.S., International,and Europe),as well as
industry-specificpublications,such as ChemicalWeek and
Plastics Technology,and, (3) government publicationsand
consultant reportsfor the chemicals industry.The data collec-
tion and coding exercise for the entire data set involved
studying over 130,000 electronic news stories and dozens of
text works. The full data set includes details on corporatecol-
laborativeactions across all functionalareas in the chemicals
and materialssector. Forthe currentstudy, I used a subset of
the data coveringthe technical collaborationactivities of the
sample firms over the period 1981 to 1991. The linkages
used here include268 jointventures, in which the collaborat-
ing firms formed a new organizationalentity, and 152 joint
research or technology-sharingagreements, in which no new
organizationwas formed.
In previous studies, lackof availabledata on linkagedissolu-
tion has meant that a distinctionhas not usuallybeen made
between linkagesformed by the firmand the linkages main-
tained by the firmat any point in time. Thus, I attempted to
recorddissolutionor continuityinformationfor all linkages.
This also helped to ensure that linkagesthat were announced
but subsequently did not materializewere identifiedand
removed from the data. This effort producedmixed results.
Forthe jointventures in the sample, the exercise was quite
successful. For 191 of the 268 jointventures, I was able to
establish either the date of dissolutionor the survivalof the
jointventure beyond 1991, the concludingyear of this study.
Forseveral of the remainingjointventures I was able to
establish continuityof operations untilsome date between
the date of foundingand 1991. This occurredwhen the last
informationavailableabout the jointventure was dated prior
to 1991 but did not refer to dissolutionactivity.I treated all
jointventures for which I did not have a recordof dissolution
as continuingto exist until 1991 for two reasons. First,my
success at identifyingdissolutions in the majorityof cases
led me to believe that, at least for this sample of firms, joint
venture dissolutiontends to be reported.Hence, the absence
of a confirmingreportof dissolutionwas best interpretedas
an indicatorof continuingoperations. Second, in many of the
cases, trade and news reports indicatedongoing operations
or specific activitiesat these jointventures for several years
after founding.The fact that other news about these ven-
tures was being reportedmade it seem likelythat their disso-
lutionwould also be reported.Assuming continuityin the
absence of news of dissolutionseemed to be the more accu-
rate assumption to make about these ventures.
The situationwas quite differentfor research agreements
and technology development and sharingarrangementsnot
involvingthe formationof a separate entity. Forsuch agree-
ments, I coded dissolution based on the tenure specified in
the formationannouncement or on a formalnotice of conclu-
sion of the research, when available.For long-term(multi-
year) or general programsof research, one of the two above
conditionswas often the case. Inthe majorityof cases, how-
ever, I was unable to establish formaldissolutions. In such
cases, I presumed the agreement to exist untilthe last year
in which it was documented or untilthe year after the year it
was founded, whichever was later.The assumption that such
agreements have a short life relativeto jointventures is con-
sistent with the specific and short-termnatureof their objec-
tives in most cases. Forresearch agreements, there were
also~cases in which the existence or ongoing activities of a
collaborationwere discussed but the foundingof the collabo-
rationitself was not reportedor indicated.Insuch cases, I
treated the collaborationas havingbeen founded in the year
immediatelypriorto the year in which it was first
documented.
Financialfigures and employment data came from COMPUS-
TAT,Worldscope(severalvolumes), trade publications,com-
pany annualreports,Japan Companyhandbooksand Daiwa
Instituteresearch guides. Forall firms, financialdata were
converted to constant (1985) U.S. dollarsto ensure standard-
izationwithinthe sample.
Model Estimation and Econometric Issues
Model specification. The dependent variable,innovationout-
put as represented by patent counts, is a count variableand
takes only non-negativeinteger values. The linearregression
model assumes homoskedastic, normallydistributederrors.
Because these assumptions are violated with count variables,
a Poisson regression approachis more appropriate(Haus-
man, Hall,and Griliches,1984). To account for unobserved
heterogeneity,the possibilitythat observationallyequivalent

firms may differon unmeasuredcharacteristics,I used the

panel Poisson approach(Hausman,Hall,and Griliches,1984)
and estimated randomeffects Poisson models. Inthe ran-
dom effects Poisson model, an additionaleffect, p*,is includ-
ed in the Poisson specificationto reflect firm-specifichetero-
geneity:
E(Pjt/Xl1) = eXit3+,.i
This firm-specificeffect permits observations of the same

firmto be correlatedacross periods and thus buildsserial
correlationdirectlyinto the model. Further,the Riis assumed
to be drawnfrom the gamma distribution.This specification
of ahimplies that the varianceto mean ratiois no longer
unity,as is assumed in the regularPoisson model but,
instead, becomes (1 + xaXt), in which axis the reciprocalof
the standarddeviationof the heterogeneity distribution.Thus,
in this model, the ratioof the varianceto the mean is permit-
ted to grow with the mean (Hausman,Hall,and Griliches,
1984). The model estimates axfrom the observed data and
thus directlycaptures any overdispersion.
Hausman,Hall,and Griliches(1984) also provideda fixed-
effects estimator for count data that handles unobserved het-
erogeneity by computingwithin-firmestimates of the coeffi-
cients. Inthis approach,only the variationwithin a firm
across time is used to estimate the regression coefficients.
Thus, unobserved variationsbetween firms are not problem-
atic because between-firmvariationis not used in the com-
putationsof the estimates. Inthis paper,for robustness, I
used both fixed effects and randomeffects to estimate the
models.
Measures
Dependent Variable. I measured Patentsit, as the number of
successful patent applications,or granted patents, for firm i
in year t. The majorityof patent applicationsare examined
and ruledupon withintwo to three years of application.The
granted patent carriesthe date of the originalapplication.I
used this date to assign a granted patent to the particular
year in which it was originallyappliedfor. Forinstance, a
patent appliedfor in 1986 but granted in 1988 is considered a
1986 patent. This procedurepermittedconsistency in the
treatment of all patents and controlledfor differences in
delays that may occur in grantingpatents after the application
is filed. Because patents are likelyto correspondto activity
immediatelyprecedingthe patent application,I used a one-
year lead with respect to key influences, such as R&Dand
linkages.Thus, the patent count for 1986 is regressed
against the 1985 values of other covariates such as R&Dand
directties.
Independent Variables
Direct and indirect ties. To obtaina count of the direct ties
maintainedby a firm in any year, I counted the numberof
direct partnersof the focal firm,or its degree centralityin the
network. Hypothesis 1 predicteda positive impactof this
variableon patentingfrequency. I used three alternativemea-
sures to capturea firm's reach in the networkthroughits
indirectties. The first variablewas a simple count of indirect
ties (indirectties, count). Foreach firm, I computed the num-
ber of other firms in the networkthat it was tied to at path
distances of two or greater,which thus excluded direct ties.
But this simple count of indirectties does not account for the
weakening or decay in tie strength between firms that are
connected by increasinglylarge path distances. For instance,
this measure counts both two-step-distantties (firmsthat are
linkedto the focal firmthroughonly one intermediaryfirm
when using the shortest path between the two firms)and
five-step-distantties (firmsthat are linkedto the focal firm
throughfour intermediaryfirms when using the shortest path
between the two firms)as the same. Yet it is probablethat
as the shortest paths connecting two firms grow longer,the
likelihoodof informationtransmission between them
decreases. Burt(1991) provideda frequency decay measure
that accounts for this decline in tie strength across progres-
sively distant ties. This measure (indirectties, distance-
weighted count)attaches weights of the form 1 - [fi/(N+1)1
to each tie, where f1is the total numberof nodes that can be
reached up to and includingthe path distance i, and N is the
total numberof firms that can be reached by the focal firm in
any numberof steps. The argumentfor this weighting
scheme is that the rate at which the strength of a relation
decreases with the increasinglength of its corresponding
path distance should varywith the social structurein which it
occurs (Burt,1991). The largerthe numberof firms to which
the focal firm must devote its networktime and energy, the
weaker the relationshipthat it can sustain with any individual
firm.Thus, decay in relationshipstrength is relatedto the
numberof other firms reached at each path distance. For
example, for a firmwith 3 direct ties, 5 two-step ties, and 7
three-step ties (here N = 15, i.e., 3 + 5 + 7), the frequency
decay formula will attach weights of 1 - (3/16) = 13/16 to
each direct tie, 1 - 8/16 = 8/16 to each two-step tie, and 1 -
15/16 = 1/16 to each three-step tie. Thus, ties at progressive-
ly longer path distances receive progressivelysmaller
weights. The total numberof indirectties weighted by their
path distances can now be computed easily. Inthis illustra-
tion, the weighted count of indirectties for this hypothetical
firm is 5(8/16) + 7(1/16) = 47/16.
I devised a thirdmeasure (indirectties, distance and informa-
tion weighted count)that is a refinement of Burt'sfrequency-
decay measure. Burt'smeasure accounts for lowered proba-
bilitiesof informationtransmissionacross longer path
distances but implicitlyassumes that all nodes generate the
same amount of new information.Ina technology network,
some firms may create more knowledge than others and
could hence be the source of more information.Accordingly,
each node could be weighted by the amount of new informa-
tion it generates. The numberof new patents created by a
firm providesat least a crude measure of the variations
across firms in their new knowledge creationcapabilities.
Hence, for each node (firm),I used the numberof patents
appliedfor by the firm in that periodas the weight for that
node in the computation.Operationally,for each firm, I multi-

plied the vector of Burt'sfrequency-decayweighted path dis-

tances by the vector of patent counts to compute this new
variable.Forexample, if the 5 two-step ties of the hypotheti-
cal firm described above produced 1, 2, 4, 6, and 8 patents,
respectively,while the 7 three-step ties produced2, 4, 6, 8,
10, 0, and 5 patents, respectively,then the patent-weighted
measure of indirectties would have a value of 12 and 11/16
1(8/16)(1 + 2 + 4 + 6 + 8) + (1/16)(2 + 4 + 6 + 8 + 10 + 0 +
5)] for this firm.
Direct ties x Indirectties represents the interactionbetween
the two priorvariables.Hypothesis 3 predicteda negative
impact of this variableon patentingfrequency.There are
three versions of this variable,based on the three measures
of indirectties. I used the ratioof nonredundantcontacts to
total contacts for the ith firm (Burt,1991) to measure the
structuralholes in the ego networkof a firm.This measure is
computed as
I / Ci,
[1j[1-1q PiqMjql
where pjqis the proportionof i's relationsinvested in the con-

nection with contact q, mjqis the marginalstrength of the
relationshipbetween contact j and contact q, and Ciis the
total numberof contacts for firm i. Highervalues on this
index reflect firms whose ego networks are richin structural
holes, i.e., the firms' partnersare not connected to each
other. If all of a firm'spartnersare unconnected to each
other, the index takes a value of 1, indicatingthat none of the
firm'scontacts are redundant.Connections between a firm's
partners imply a higher lqqpiqmjqand thus a lower value for
this index, reflectinghigherredundancyand fewer structural
holes. Forfirms without any partners,the index is set to 0. I
used STRUCTURE (Burt,1991) to compute this measure.
Control variables. R&Dexpendituresare likelyto be a signif-
icant determinantof innovativeoutcomes. I collected R&D
data from COMPUSTAT, Worldscope,DIRAnalyst'sguides,
Japan Companyhandbooks,industryand company journals,
and annualreports.When R&Ddata were not availablefor
some periods, I used a regression imputationprocedure(Lit-
tle and Rubin,1987) to impute missing values for this vari-
able and complete the data. Ina few cases, this imputation
procedureled to negative or improbablevalues for R&D,so I
estimated R&Dindividuallyusing data from the most recent
availableperiods for that firm. Since the dependent variable,
patents, includes only chemicals patents, an appropriatecon-
trol would be to includeonly the R&Dexpenditures on chem-
icals-relatedbusinesses ratherthan corporateR&D.Unfortu-
nately, business-level research expendituresare not
commonly reported.As an approximation,I obtained the ratio
of chemicals sales to total sales for each firmand appliedit
to the corporateR&Dfigures to obtain chemicals R&D.I
used the naturallog of chemicals R&Dexpendituresas a
controlvariablein all models.
439/ASQ,September 2000
It is conventionalto controlfor firm-sizeeffects in analyses of
innovativeproductivity(Cohenand Levin,1989).1 used the
naturallog of numberof chemicals employees as a measure
of firmsize. Numberof employees was obtainedfrom COM-
PUSTAT,Worldscope,Japan Companyhandbooks,and indus-
try and company journalsand reports.As with R&D,to obtain
the numberof employees in the chemicals businesses of the
firm, I multipliedthe total numberof employees by the ratio
of chemicals sales to total sales.
Arguments have been made for both positive and negative
impacts of diversificationon innovationperformance(Cohen
and Levin,1989).1 do not make any predictionon the sign of
this effect but controlfor its influence by includingthe vari-
able diversification,entropy.The followingformulawas used
to calculatethe measure: 1P. x In(1/Pd), where Pi is defined
as the percentage of firm sales in business segment j and In
(1/P.)is the weight for each segment j (Palepu,1985).
Firmscan vary in their area of strategic focus within an indus-
try,and differentindustrysegments can offer differing
degrees of opportunityto innovate.Thus, some firms may be
active in relativelyrichertechnologicaldomains than other
firms. To capturethe firm-specificdifferences in areas of
strategic focus, I constructed a measure of technological
opportunity.Foreach firm, I identifiedthe technologicalclass-
es in which it was active in any year. I then identifiedthe
numberof total patents in that set of classes by the U. S.
Patent Office in that year. Thereafter,using the firm'sown
distributionof patentingeffort across classes as weights, I
computed a weighted indicatorof the relativerichness of the
firm'sspecific environment.The weights reflect the fact that
the firm'sown efforts across those classes were not distrib-
uted equally.As an illustration,say that FirmA has patented
in technologicalclasses 1, 2, and 3 in 1983 and obtained 5,
20, and 25 patents in these classes, respectively. I obtained
the total patents in these classes from the U.S. Patent data-
base in that year, as 400, 1000, and 600. The value of the
technological-opportunity variablefor this firm-yearobserva-
tion would then be 400(5/50) + 1000(20/50) + 600(25/50) =
740. Essentially,high values of this variableindicatethat the
firmwas involvedin technology segments that offered rela-
tively higheropportunityto innovatethan other segments.
Firmsconductingresearch in multiplegeographic regions
may enjoy access to more diverse knowledge environments,
which may influence innovationoutput. To controlfor the
geographic breadthof a firm'sresearch efforts, I computed
the variableinternationalresearch presence as the Herfindahl
index of the firm's patentingacross nations in that year. The
U.S. Patent database provides informationon the physical
locationof the inventorat the time of the invention.Based on
this, it is possible to construct an indicatorof the distribution
of a firm's inventiveefforts across nations. The Herfindahl
index is computed as INi2, where N1is the proportionof the
firm'spatents in nation i. A firmwith a research presence dis-
tributedacross several nations will have a lower score on this
index than one whose research efforts are concentrated in a
single nation. Forinstance, a firmwith 100 patents split
equallyover five nations will have a Herfindahlindex score of

5 (0.20)2= 0.20. Anotherfirmwith the same numberof

patents, but active in only one nation,will have a Herfindahl
index score of 1.
It is possible that in an interfirmtechnology networkthe
structural-holesmeasure might capturethe technical diversity
of the skill bases in a firm'salliancenetwork ratherthan the
social structuraleffects postulated here. If a firm'spartners
are active in widely divergenttechnologicalareas, they may
be unconnected to each other and, hence, generate structur-
al holes in the focal firm'snetwork.At the same time, such
diversityin the partnerbase may make successful collabora-
tion unlikelyfor largelytechnical reasons, such as absorptive
capacity(Cohenand Levinthal,1989; Laneand Lubatkin,
1998; Stuart,1998). If so, then the structural-holesmeasure
might reflect the negative effects of this technologicaldis-
tance between partnersratherthan social structuraleffects.
To controlfor this possibility,I created a variable,technologi-
cal distance between partners,to capturethe degree of
diversityof a firm's partners.To compute the technological
distance between partners,I used the approachsuggested
by Jaffe (1986). First,I used the distributionof a firm'sprior
patentingacross the patent classes providedby the United
States Patent and TrademarkOffice (USPTO),to construct a
vector representingeach firm'sposition in technology space.
Inthis vector, each USPTOtechnology class represents a dis-
tinct dimension, and for any firm,the proportionof the firm's
patents that fall withinthe technology class is the value of
the correspondingelement in this vector. For instance, if a
firm had obtained 20 percent of its patents in the Kthtechnol-
ogy class, then the Kthelement of the vector for this firm
would have a value of 0.20. After representingall the firms in
this technology space, for each firm, I calculatedthe Euclid-
ean distance between all pairsof its partnersand took the
average of these distances as the value of the variable,tech-
nologicaldistance between partners.If partnersare techno-
logicallydistant from each other, in that they have very differ-
ent technologicalbackgroundsas represented by their prior
patentingfocus, then this variableshould have a relatively
high value. Ifthey focus on the same technology classes and
have very similarpatentingprofilesacross the classes, then
the value of this variablewill be relativelylow. To illustrate,if
FirmA has 10 percent of its patents in Class A, 40 percent in
Class B, 50 percent in Class C, and 0 percent in Class D,
while FirmB has 40 percent of its patents in Class A, 0 per-
cent in Class B, 10 percent in Class C, and 50 percent in
Class D, then the technologicaldistance between the two
firms can be computed as the Euclideandistance between
the vectors: 0.1, 0.4, 0.5, 0 and 0.4, 0, 0.10, 0.5, i.e., the
square root of the sum [(0.1 - 0.4)2 + (0.4 - 0.0)2 + (0.5 -
0.1)2 + (0.0 -0 .5)2]. Computingthis distance between all of
a firm'spartnersand takingthe average of these distances
providesthe value of this variablefor a given firm-year.
I also includedvariablesto controlfor the profitabilityand liq-
was capturedthrougha returnon
uidityof firms. Profitability
assets variable,while liquiditywas represented throughthe
currentratio(ratioof currentassets to currentliabilities).Over
time, innovationrates can increase or decrease for all firms. I
controlledfor such periodeffects by includinga series of
dummies for every year from 1981 to 1990, 1991 being the
omitted category. I also includedcontrolvariablesfor the
nationalityof the firms;Japan and USAwere dummy vari-
ables coded to equal 1 for Japanese and Americanfirms,
respectively. Europeanfirms constituted the omitted
category.
RESULTS
Table1, which providesdescriptivestatistics for the linkage
networkover time, shows that the mean numberof direct
ties grew steadily over the periodof the study, reachinga
peak in 1990 and then decliningin 1991. The overalldensity
of the network indicates the proportionof potentialnetwork
ties that are actuallyrealizedand reflects the same trend,
peakingat 5.7 percent. Thus, the network is relativelysparse,
with less than 6 percent of potentialconnections actually
being realized.The distributionof ties is capturedby the two
remainingvariablesin table 1. The percentage of isolates indi-
cates the firms that maintainno ties at all. The percentage of
such firms steadily declines from 26 percent in 1981 to 9 per-
cent in 1989 before increasingto 11 percent by 1991. Thus,
in most years, over 80 percent of the firms had at least one
direct tie. The network centralizationmeasure indicatesthe
degree to which a single actor dominates the network. If link-
ages are distributedequallyamong all nodes, this index has a
low value. It reaches zero when all firms have the same num-
ber of ties. Highvalues on this variableindicatethat linkage
activityis centered in one leadingfirm,and relativelyfew link-
ages occur between other firms. The observed values indi-
cate that network centralizationis moderate, reachinga peak
of around22 percent in the mid-eighties.Thus, it appears
that even though most firms are linkedto the network, some
are significantlymore active than others.
Table2 providesdescriptivestatistics and correlationsfor all
variablesfor the 996 observations in the sample. Even
though the sample represents the prominentfirms in the
industry,there is considerablevarianceon all the key vari-
ables, such as patents, R&D,directties, and log employees.
The three measures for indirectties are relativelyhighlycor-
Table 1
Descriptive Statistics on the Linkage Network
Network Network
Mean density S. D. centralization %
Year degree (%) Degree Min. Max. (%) Isolates
1981 3.34 2.9 3.64 0.0 18 14 26

1982 3.77 3.4 4.19 0.0 25 18 22
1983 4.23 3.7 4.52 0.0 27 19 20
1984 4.31 3.9 4.53 0.0 27 18 14
1985 4.64 4.1 4.69 0.0 29 21 13
1986 5.11 4.3 5.13 0.0 31 22 14
1987 5.38 4.6 5.25 0.0 28 20 11
1988 5.86 5.2 5.53 0.0 28 16 11
1989 6.19 5.4 5.88 0.0 34 19 9
1990 6.70 5.7 6.64 0.0 34 19 11
1991 6.43 5.6 6.40 0.0 29 18 14

related with each other, as would be expected. Among the

independentvariables,the measure for structuralholes is
correlatedwith the three measures for indirectties.2
Intable 3, I reportthe results of the regression analyses
using the random-effectsPoisson estimators. Model 1 pre-
sents the base model with only the controlvariables.Model
2 adds the directties variableto the specification. Models 3a
through3c add the three measures of indirectties, respec-
tively. Models 4a through4c add the interactionterms, Direct
ties x Indirectties (three measures), and models 5a through
5c add the structuralholes variablesto complete the specifi-
cation. I use the complete specification(models 5a to 5c) to
discuss the results.
The results supportthe predictionsfor all four hypotheses.
The coefficient of directties is positive and significant,sup-
portinghypothesis 1, which predicteda positive impact of
direct ties on firm innovationoutput. The indirectties coeffi-
cient (allthree measures) is positive and significant,support-
ing hypothesis 2's predictionof a positive relationship
between indirectties and firm innovationoutput. Hypothesis
3, predictinga negative impactof the interactionbetween
directties and indirectties on the innovationoutput of a firm
is supported,the negative and significantcoefficient indicat-
ing that havinga highernumberof directties reduces the
impactof indirectties. Finally,hypothesis 4 proposed com-
peting predictionsfor the effect of structuralholes on firm
2 innovationoutput. The data indicate, in supportof Coleman's
For the interaction variable, the correla- position,that havingmany structuralholes is associated with
tions between the component terms and reduced innovationoutput.
the interaction terms were high. I took
several steps to address this issue. For
instance, I also reran the analyses, after Two aspects of the above results are worth probingfurther.
centering both component variables on First,providingsome quantitativeindicationof the interaction
their means priorto constructing the
interaction term, for all three versions of
effect could help in interpretingthe results. Second, examin-
the Direct ties x Indirect ties variable ing the relativemagnitudeof the effects of direct and indirect
(Cronbach, 1987; Jacquard, Turrisi,and ties is of intrinsicinterest. To illustratethe interactioneffect
Wan, 1990). This transformation reduced
the higher correlation between the com- first, suppose that a firm is at the mean level of directties (5)
ponent terms and the interaction term in and has 20 indirectties. Forthis firm, indirectties increase
all cases. For the transformed (mean-
deviated) variables the correlations the patentingrate by a multiplierof 1.03 (= exp[0.051*2 -
between the component terms and the 0.007*5*2]). Now, consider anotherfirmthat is also at the
interaction terms were -.20, and -.92 same level of indirectties (20) but has 6 direct ties. Forthis
(versus -.11 and .99 earlier) for Direct ties
x Indirect ties, count, -.57 and -.67 (ver- firm, indirectties raise the patentingrate by a multiplierof
sus .22 and .94 earlier) for Direct ties x 1.016 (= exp[0.051*2 - 0.007*6*2]). Thus, having a higher
Indirect ties, distance weighted count,
and -.50 and -.60 (versus .32 and .89 ear- level of direct ties reduces the benefit from indirectties.
lier) for Direct ties x Indirect ties, dis-
tance and information weighted count. To compare the relativestrength of the direct and indirecttie
Additionally, since some of the variables effects, we can examine the impactof a one-standard-devia-
are meaningful only for firms with two or
more ties (structural holes, technological tion increase in each on the patentingoutput of a firm. Con-
distance between partners), I also reran sider a firm (as above) that has 5 direct ties and 20 indirect
the analysis using only the 744 observa-
tions that represent firm-years with two
ties. Forsuch a firm,a one-standard-deviationincrease in
or more linkages. Since firms with no ties directties increases the patentingrate by 23 percent
have zeroes on all network variables
(direct ties, indirect ties, structural holes),
[5.4(0.057 - 0.007*2) = 0.231. Forthe same firma one-stan-
omitting these observations leads to a dard-deviationincrease in indirectties increases the patenting
sample with lower correlations between rate by 4 percent [2.746(0.051 - 0.007*5) = 0.041.Thus, the
the network variables. Finally,as a further
cross-check against collinearity, I estimat- coefficient on the indirect-tiesvariablesuggests that indirect
ed the models on subsamples after ran- ties do contributeto innovationoutput;however, the magni-
domly omitting observations to check the
stability of the estimated coefficients.
tude of this contributionis significantlysmallerthan the con-
I report these results below. tributionmade by directties.
Table 2
Means, Standard Deviations, and Correlations (N = 996)
Mean S. D. 1 2 3 4
1. Patents1t 51.18 82.67

2. Direct tiesit1l 5.27 5.40 .30
3. Indirect ties/10, counti11 6.37 2.75 .11 .28
4. Indirect ties, dist. wtd. t-1 19.42 8.92 .04 .05 .96
5. Indirect ties, dist. & info. wtd. t-* 12.25 6.50 -.03 .06 .88 .88
6. Direct ties X Indirect ties/1 0, counti 1 37.70 36.25 .31 .99 .35 .11
7. Direct ties X Indirect ties, dist. wtd. It-1 104.4 84.15 .32 .94 .44 .22
8. Direct ties X Indirect ties, dist. & info. wtd. * 66.54 58.04 .25 .89 .43 .21
9. Structural holes1t 1 .79 .32 .14 .31 .86 .82
10. Diversification, entropyit1 1 .29 .33 .36 .18 .10 .07
11. International research presencei 1 .86 .20 -.24 -.00 -.03 -.04
12. Technological opportunityi1 15.68 6.02 .17 .24 .18 .10
13.Return on assetsit 1 .03 .03 .15 -.07 -.02 -.01
14. Current ration1 1.58 .62 .23 -.18 -.19 -.15
15.Japan .43 .50 -.28 .27 .14 .06
16. USA .26 .44 .17 -.23 -.10 -.05
17. R&Djt1 3.48 1.56 .66 .39 .18 .08
18. Firm sizers1 1.94 1.30 .65 .22 .05 .01
19.Tech. Distance between partnersit 1 .28 .21 .10 .40 .50 .43
20.Year 1981 .09 .28 -.03 -.1 1 -.18 -.13
21.Year 1982 .09 .29 -.04 -.08 -.1 1 -.07
22.Year 1983 .09 .29 -.04 -.06 -.13 -.07
23.Year 1984 .09 .29 -.02 -.05 .01 .05
24. Year 1985 .10 .29 -.02 -.03 .02 .02
25.Year 1986 .09 .29 -.01 .00 -.05 -.08
26.Year 1987 .09 .29 .02 .02 .09 .08
27.Year 1988 .09 .28 .05 .05 .13 .13
28.Year 1989 .09 .29 .04 .07 .14 .11
29.Year 1990 .09 .28 .03 .11 .09 .01
5 6 7 8 9 10 11 12 13
6. Direct ties X Indirect ties/1 0, counts 1 .13

7. Direct ties X Indirect ties, dist. wtd. it-1 .22 .97
8. Direct ties X Indirect ties, dist. & info. wtd. it .32 .94 .95
9. Structural holesit 1 .71 .33 .40 .37
10. Diversification, entropyit1 .00 .19 .21 .16 .13
11. International research presencet,1 -.03 -.02 -.04 -.03 -.03 -.20
12.Technological opportunityit1 .20 .27 .27 .33 .09 .07 .17
13.Return on assetst,1 .01 -.04 -.04 -.03 -.03 -.1 1 -.14 .01
14. Current ratioit1 -.14 -.18 -.20 -.20 -.16 .03 -.16 .02 .39
15.Japan .10 .27 .27 .28 .06 -.10 .48 .26 -.38
16. USA -.07 -.23 -.25 -.24 -.08 -.13 .12 -.01 .42
17. R&Dit. .07 .42 .43 .40 .20 .24 -.37 .27 .22
18. Firm sizeit 1 -.08 .23 .25 .18 .10 .35 -.52 .00 .21
19.Tech. Distance between partnerst,1 .33 .42 .51 .44 .43 .19 .02 .14 -.18
20.Year 1981 -.17 -.14 -.14 -.16 -.12 -.01 .06 -.10 -.02
21.Year 1982 -.15 -.11 -.1 1 -.14 -.09 .02 .02 -.08 -.15
22. Year 1983 -.20 -.09 -.07 -.14 -.07 .02 .00 -.16 -.06
23. Year 1984 -.13 -.05 -.04 -.13 -.01 .00 .01 -.1 1 .06
24.Year 1985 -.05 -.03 -.03 -.07 .01 .00 .03 -.08 -.07
25.Year 1986 -.09 -.02 -.04 -.05 .00 .02 -.01 -.04 -.01
26.Year 1987 .06 .04 .05 .03 .06 -.01 .00 .00 .08
27. Year 1988 .22 .09 .12 .16 .07 -.01 -.01 .10 .16
28.Year 1989 .28 .11 .12 .21 .07 .00 -.04 .18 .11
29. Year 1990 .18 .13 .09 .19 .06 .00 -.03 .19 .03
Several of the controlvariableresults are also significant

(models 5a, 5b, and 5c). Technologicaldistance between part-
ners was negative and significant,supportingthe argument
that absorptivecapacityissues are likelyto be importantin
the context of technology alliances (Laneand Lubatkin,1998;
Stuart, 1998). R&Dand firmsize are both positivelyassociat-

Table 2 (continued)
14 15 16 17 18 19 20 21 22
15.Japan -.51
16. USA .28 -.51
17. R&Dit1 .20 -.40 .17
18.Firm sizeit 1 .22 -.64 .26 .80
19.Tech. Distance between partnersit 1 -.31 .23 -.21 .17 .11
20.Year 1981 .00 .01 .00 -.12 .00 -.02
21.Year 1982 -.03 .00 .00 -.1 1 .00 -.01 -.10
22.Year 1983 -.04 -.01 .00 -.08 .00 .02 -.1 0 -.1 0
23.Year 1984 -.04 .00 .01 -.07 .01 .02 -.10 -.10 -.10
24. Year 1985 -.04 -.01 .00 -.05 .01 .04 -.10 -.10 -. 10
25. Year 1986 -.03 .00 -.01 .02 .01 .00 -.10 -.10 -. 10
26.Year 1987 .05 .00 .00 .05 -.01 -.01 -.10 -.10 -. 10
27.Year 1988 .03 .01 .00 .08 -.01 -.03 -.10 -.10 -.10
28.Year 1989 .05 .00 -.01 .08 -.01 -.01 -.1 0 -.1 0 -.1 0
29.Year 1990 .02 .01 .00 .10 .00 .01 -.1 0 -.1 0 -.10
23 24 25 26 27 28
24.Year 1985 -.10
25.Year 1986 -.10 -.10
26.Year 1987 -.10 -.10 -.10
27.Year 1988 -.10 -.10 -.10 -.10
28.Year 1989 -.10 -.10 -.10 -.10 -.10
29.Year 1990 -.10 -.10 -.10 -.10 -.10 -.10
*
Measured in 100's.
ed with patentingfrequency,The estimated coefficient, how-

ever, is significantlyless than unity in all three specifications.
If the regressor variablesare in log form, as it is for these
two variables,then the coefficients of the Poisson specifica-
tion can also be interpretedas elasticities of the regressor
variablewith respect to the dependent variable.Here, the
positive but less than unitycoefficient on R&Dand firm size
indicates that patentingfrequency increases with R&D
expendituresand firm size, but it does so less than propor-
tionately.This is broadlyconsistent with past research (Acs
3 and Audretsch, 1988, 1991).
I also conducted several supplementary
analyses to evaluate the robustness of Being a diversifiedfirm is negativelyassociated with patent-
the results, including reestimating the
models after centering the direct ties and ing frequency in this research. Priorresearch on the impact
indirect ties variables on their means prior of diversificationon innovativeactivityhas been mixed, with
to creating the interaction terms, using a
Poisson fixed-effect specification instead
studies showing both a positive and a negative impactof
of the random effects reported here, and diversificationon innovation(Cohenand Levin,1989). Two
repeating the analyses with a sample of broadarguments relate diversificationto innovativeoutput.
firms with two or more linkages only. The
results (available from the author) were Diversificationcan encourage innovationby providinga stimu-
very similar to the reported results. To fur- lus of multipleknowledge bases within a single firm, leading
ther investigate the possibilities of
collinearity problems, I randomly omitted to cross-fertilizationof ideas. Diversificationcan also imply
observations from the full sample to cre- increased bureaucratization and operationalcontrols within
ate 300 subsamples, each of which had
between 650 and 900 observations. I
firms and inhibitinnovation.The results of this research sup-
then reestimated these models on all 300 port the latterinterpretation,that being active in multiple
subsamples. A warning sign of collinearity businesses is associated with a negative effect on patenting.
problems is that omitting even a few
observations can cause significant
changes in the coefficient estimates and Among the other controlvariables,currentratio,the measure
drive them to insignificance or cause the of liquidity,was positivelyassociated with patenting.The
coefficients to be significant but reversed
in sign (Greene, 1997: 420). Even though
estimated alpha coefficient is positive and significantlydiffer-
in half the samples more than one-third of ent from 0. This indicates that there were significantfirm-
the observations were omitted, these
sensitivity analyses provided strong sup-
level unobserved effects in the data that were capturedby
port for the reported results. the heterogeneity parameters
Q- O O CNYN
C1N C- O t L t "t- OCD C 0rL0 M CN O r-
U 0) CD 't O O 1 O LOO- LO OIO O O - (Cr-
O O O0 CJ O O' O CY) O ( CN LO O0
C.)(oD
LO 0'r-ttO NG O CIJ O t O
O O t O
O -
O r- 00 O 0 C4,l 00r-Cs4 O
Cs CY)CY CO CD
4
U
C) ( OOOO
CY) OD O OOOO
CNi C CY) m Cm O OO C
LO
SO II S 0 S _ ____I So 0
S
0 0 0 I
O M M
SN t C1 0 _II 0 0 C1
O O-- O 00
m CY r- LO 0 OLOO O O C14 r- Or- O"t O OO O O CN O r- M M 0 r-- O S-O -- O
U) r- O O O O 0 CN O CN M M M CN CN
OCD00 00 C- CCi O M M C O 0 L Ct 0 t M O t O
LCO0 I oXtQ O
CD- O
OGG O
O r N GtO A CN M ~M
~~~~~~~~~~~~~~ - 00 CN -
OOO OOO CNGOON
CY) _~ *O CY *- 1- - -- 1- 1- - 1 -- I'- 1 r-- I'- .N ' .- 1-
CN CD0 NO G O 0 0 C0 NO M ) C N
co- - - 1 - 1 - 1 - 1 - 1 - LD - I - -~ -
CO ID o
|~L 00 00 00 1OC)- N1 t N t M N r-- N CN Y"t M) MY
(
CDY) G O o~~
rs
w
C
o
0
O O OO
0
O CN
CN
t
M
o t t I-
CN M M M
00 0 0
o
-o O
CN
0 nCY N~CDC~
C) O D0 0-G O 0OG r- G M 0 r--r- - mr.-
-G.
o o o o o o o o o o~~~~~~~~~~~~~~~~~~ o
o
o
o
CN
rs CN CN M M M
o CD)-- L C D b o C C o mo m
CY) G O
rD bO rO G O
CYN O CY)OO r-- 000 0
O~~~~~~~~~ 0 0 O O O O O O O O cN O O cN CY CY CY O
O
N
.s~~~~~~~~~~~ @
v X~~~~~0
OD O CD O0 0 M "t n MD M CDOM
CD4 OD O
O
0 @ ~~~~~~~~~~~~~~~~I M It
O O
M
O
M M S~ M
Dt
SI' SS00 0
"t
MO
C C C
oo
v ~O )csbCD NS
CD- CD CD) 00 G N t QO) CDn- O-G
o 1~~~~~~~0
LU O CD O O 00 t t O O O O n O S S b b O O C
~~~~~~1-
1 ' '
C'l1-1'
E c>_ _ l- l~~~~~~~~~~~~~~~~~~~~~~~~~~~~~-
l-E l -
0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~-0
0~~~~~~~~~~~~~~~~~~~0
0.~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ L
0-
LU~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~LQ
E
0) I I az z0 E z O c o ? o A
o~~~~~~~~~~~~~~o )
aBZ a) +- DoI
U) a), U) U) U) CD CDCl +-,

E
J o~CD 1 0 07 0E E xE xE +~C)-Q
~~~~~~~~~~~~~~~~~~x
?0 @
*? @ a @
CD
0
? E o
=3 ~~~~~~~CD
C l .
O5
-l 0 E) -,Q) Cl) l)- ,aa " - C
> u C) a, *? C) C) C)
APaUD - CD
C D C

r
o nC4
-
I-
t
C4
00 OCD0
C4
MD;;S
r- C4
e -r 0 -r- I eA
O- C 00
-1 -
0 MD0 C'4r- I
C1 C14 r C4 C4 r-
a
C4 M
O2
)
O O O O O O O D CD CD
O CO C C C 00 CD
CD 00 O
M M CDM
t Cj O 00t M M 0, M M' o C CY)
U):7 OOm mm O O O O Ot OmO O 1-0 am O OO00

- r- 0)1- -) - - ~1- N- - 1-- *- *- r-* 10 *-
~~~~~~~~~~
0
w C~)Q
F
N C*
CDo F M
L
N C)10
i
N C)10
C*
O
C1) N
AO
C1) CD
C M
N
CN
n
S O r-1
C, MN
0
C)CCN C 1 Ne
CN C
10 0 0) N N 0)1 )r- -10Q*r--
CNi
CD C)-) O CCDN
0 0 0 N 10 0)- N ? 0 0 -S
O
C)
Q IB O,
CD)QN~-N'- MO NNN'-'O'-'C
Om NM mm MO )N0)'-CN
NM MO Oa
M N 10O N1
OO OO 0
O S S~IJ. J S. S
e L)O CN "t A.CN -1LO)N r- CN t CN CD
) CN
-1 M N
A ~~~~ )
N CN N M CN -
CY eD . . . . r--DeDn
t O. . 0. 0 0 .0 . 0 .0 . Cs
0. 0 . .Cs
0.0. .0 . .
oC N C N CNNC NONC
NSC
CY)~~~~~~~~~~~~~~~~~~~~~~~~~~~a
O CD O O O
M L N O O- O O O Lo
CY) n 77 S
0 00N s cnoO
s
CY)
0
'1
Y CY) "t
O O
N
O
cN LO LO
O
CY
O
r- LO 0'
O O
r-
O
C
y)7o 00 c
O4 O O O- O Ot LO O CY CY r-
C
C C C C
L~~~~~ci
CD
N 00 N
N ~~CD C s C s CDstC snC sbC
_ _ I _ I LO
_-- I _ I _ _ _ _ _ It_ _ E
I._0
CD
-0
(~~C
Nb Q Cl)
NCNnCNo
Noo
C:)
=3 a)
0 CD:
a)
O CY) '* I)00 CD

O rO 0 CD
) 00
_) 0) 0) 0) 0) 0) 0) 00) 0) 0)
a) CD
vV V*CD
v
Ql .-_ D
CDv coCDv coCDv coCDv coCDv coCDv coCDv coCDv co cCD -
CD
2 a) 00U*

DISCUSSION AND CONCLUSION
This study examined the impact of three aspects of a firm's
ego network-direct ties, indirect ties and structural holes-
on the innovation output of the firm. The theoretical frame-
work suggested that the three aspects of network structure
play different roles in the innovation process. According to
this framework, direct ties serve as sources of resources and
information, indirect ties serve primarilyas sources of infor-
mation, and structural holes between partners serve two
contradictory roles. They expand the diversity of information
that the firm has access to but also increase the firm's expo-
sure to potential malfeasance. In this study, I predicted, and
found, that direct and indirect ties influence innovation output
positively, but the impact of indirect ties is moderated by the
firm's level of direct ties. Finally, I presented competing pre-
dictions about the effect of structural holes in the focal firm's
network and found that, at least in this interfirm collaboration
network, increasing structural holes decreases innovation
output. The findings have some important theoretical implica-
tions.
This study was motivated by two theoretical puzzles and
their implications for firms in interorganizational networks.
First, I sought to evaluate the idea that building networks
with large numbers of indirect ties may be an effective way
for actors to enjoy the benefits of network size without pay-
ing the costs of network maintenance associated with direct
ties (Burt, 1992). Second, I sought to understand the degree
to which closed or open networks could be appropriately
regarded as the normative ideal (Coleman, 1988; Burt, 1992;
Walker, Kogut, and Shan, 1997). The arguments and conclu-
sions of this study shed some light on both of these issues.
The results of this study both vindicate and qualify the pre-
scription to use indirect ties as an efficient and effective way
of maximizing network benefits. In an interfirm technology
linkage network, a firm's indirect ties serve as a mechanism
for knowledge spillovers and contribute positively and signifi-
cantly to its innovation output. Given that, unlike direct ties,
indirect ties entail relatively low or no maintenance costs for
the firm, these benefits are extremely welcome. Thus, the
results provide support for the basic premise that network
effectiveness can be enhanced through indirect ties (Burt,
1992). But the paper also suggests that caution is required
before interpreting these results as a mandate to build large
networks of indirect ties. The arguments and findings of this
paper draw attention to three factors that need to be consid-
ered before embarking on a strategy of substituting indirect
for direct ties.
First, this study highlights the fact that even within the same
network, direct ties and indirect ties can differ significantly in
the nature or content of benefits that they provide to the
focal actor. Although I did not directly measure the contents
of direct and indirect ties, I argued that in the interfirm link-
age network these ties differ in the nature of benefits
offered: direct ties provide resource-sharing and information-
spillover benefits, but indirect ties provide only the latter.
Clearly, under these circumstances, the degree of substitu-

tion possible between direct and indirect ties is limited. More

generally, this finding suggests that the value of a strategy of
substituting indirect ties for direct ties will vary significantly
across networks. In any network, an analysis of the substan-
tive benefit provided by each kind of tie must be conducted
before a network reconfiguration is attempted.
A second aspect, closely related to the previous one, is that
even when direct and indirect ties provide the same kind of
benefit, the magnitude of the benefits provided by indirect
ties may be significantly different from those provided by
direct ties. The results of this paper suggest that the actual
magnitude of benefits from indirect ties is relatively low.
Although this result could be peculiar to this setting, one
argument suggests that this result may actually be more
widely applicable. This conclusion is based on one key
insight: in many networks, indirect ties simultaneously play
two different roles vis-6-vis the focal actor. On the one hand,
they are resources that extend the actor's reach in the net-
work and improve his or her access to information. On the
other hand, in many networks, such indirect ties are also
competitors of the focal actor in terms of using such informa-
tion. To illustrate this dual aspect of indirect ties, figure 2
(adapted from Burt, 1992) shows a firm X that builds a direct
tie to a partner (1) who has three other partners (2, 3, and 4).
These three indirect ties (firms 2, 3, and 4) are now potential-
ly providers of information to the focal firm (X), and news of
new technical developments arising in one of these firms can
make its way to the focal firm through the common partner
(1). Thus, adding the indirect ties has extended the focal
firm's information reach in the network significantly. Yet mov-
ing our focus away from the focal firm and onto the indirect
ties themselves draws attention to another, less benign
Figure 2. Illustrationof indirect ties in a network as both resources and

competitors (adapted from Burt, 1992: 20).
15
14 1
10 16
FIRMX
8~~~~~~~~~~~~~~
649/ASQ,September 2000

aspect of this network. These indirect ties, linked to the focal
firm through the common partner, are also linked to each
other through the same common partner. Information that
arises in one of these indirect ties, say Firm 2, reaches the
focal firm, but it also reaches the other indirect ties, Firms 3
and 4. If the same information can be used profitably by the
other firms, and one firm's use of it precludes its most fruit-
ful use by another, then the network benefits that accrue to
the focal firm are likely to be much smaller than might other-
wise be anticipated by a simple consideration of its expanded
reach. More generally, this argument suggests that the
degree to which indirect ties provide benefits of greater infor-
mational reach will vary by the nature of the information and
the network. To the extent that the sources of information in
many networks are also potential users of similar informa-
tion, competition to use the information within the clusters in
which it originates can reduce the benefits that ego can
expect from even an effectively configured network.
The negative interaction between direct and indirect ties sug-
gests a third reason to be careful in terms of evaluating the
impact of indirect ties. Although individually higher numbers
of direct ties and indirect ties are both beneficial, having
many indirect and many direct ties is not necessarily better.
Between the more limited addition to their knowledge base
through their indirect ties and their more constrained ability
to absorb and act on the information, actors with many direct
ties may be unable to profit from their indirect ties as can
actors with fewer direct ties. Thus, in addition to being limit-
ed in magnitude, the value of indirect ties is also likely to be
contingent on the number of a firm's direct ties (see also
Burt, 1997). This conclusion is likely to apply in particularto
networks such as the one described above, in which many
actors can potentially use the same information. In such net-
works, alertness, responsiveness, and flexibility are likely to
be important in terms of profitably using information obtained
through network ties (Zaheer and Zaheer, 1997).
The above arguments suggest several mechanisms that
potentially limit the benefits of indirect ties, but my objective
in presenting these arguments is not to indicate that indirect
ties are inferior to direct ties or vice versa. Rather, it is to
draw attention to a broader conclusion: whether direct ties
are more productive than indirect ties depends on the con-
text being studied, and the effects of ties, whether direct or
indirect, are likely to be contingent on several factors. The
nature and content of the ties, the type of outcome being
studied, and the broader network structure within which a tie
is embedded are all likely to influence the value of a tie.
Although, in this study, indirect ties provided relatively less
significant benefits than direct ties, that conclusion is unlikely
to be universally true. For instance, in Bian's (1997) analysis
of job searches in China, direct ties provide only an interme-
diation benefit by connecting potential employees with job-
granting officials, while the indirect tie in the form of the job-
granting official provides the more substantive benefit of an
actual job.
The arguments and findings on structural holes further rein-
force the basic conclusion that the impact of different net-

work attributesand positions can only be understood relative

to a particularcontext. The strategy of matchingthe type of
benefit (resource sharingversus informationspillovers)with
the form of social structure(a closed versus an open net-
work) by itself draws attention to the contradictoryeffects of
network structureon innovationoutput. The results of the
statisticalanalyses furthercontributeto illuminatingthe
debate on the appropriateform of facilitativesocial struc-
tures. In interorganizational collaborations,it appears that the
benefits of increasingtrust, developing and improvingcollab-
orationroutines,and reducingopportunismthat are provided
by a group of cohesive interconnectedpartnersoutweigh the
disadvantages of not havingthe informationaldiversitythat is
providedby havingmany structuralholes in a firm'snetwork.
Reconcilingthis result with that of an earlierstudy on struc-
turalholes and innovationis useful. Inan interestingprocess
study of innovation,Hargadonand Sutton (1997) demonstrat-
ed how a firmexploits its position as the spanner of structur-
al holes to develop new products. On the surface, the results
of that study appearto conflictwith the findings reported
here, but a key difference between the networkcontext they
depict and the one studied here is relevant. In Hargadonand
Sutton's (1997) study, the focal firm is a product-development
consultingfirmthat bridges structuralholes between clients
in differentindustries.Here, the network consists of collabo-
rativelinkages between firms in the same industry.Thus, the
natureof ties between firms varies significantlyfor the two
networks. Collaborationand resource sharingbetween com-
petitors, two salient features of this network,are not the
issue in Hargadonand Sutton's network. Rather,the key prin-
ciple there is brokerage.By contrast, in the collaboration
between competitors that is studied here, developing norms
of cooperationis likelyto be especially important,hence the
benefits of interconnected,closed networks. Again,the basic
conclusion that emerges from the above comparisonof
results between this study and Hargadonand Sutton's (1997)
study is that whether structuralholes are good, bad, or irrele-
vant is liableto be a functionof the context. When develop-
ing a collaborativemilieuand overcomingopportunismare
essential to success, closed networks are likelyto be more
beneficial.When speedy access to diverse informationis
essential, structuralholes are likelyto be advantageous.
My finalpoint concerns the implicationsof the contingency
arguments highlightedabove for the broader,developing
literatureon network resources and social capital(Adlerand
Kwon, 1999). Althoughthe facilitativerole of networks has
led to their identificationas network resources or social capi-
tal (Burt,1997; Gulati,1999), and networkattributeshave
been associated with several distinct benefits, such as trust,
information,and power, scholars have been unableto agree
on the form of social structuresthat constitute social capital
or network resources. Forinstance, cohesion theorists have
presented densely interconnectednetworks as the normative
ideal (Coleman,1988). Conversely,others have emphasized
the benefits of structural-hole-richnetworks (Burt,1992).
Scholars in a thirdtraditionhave argued that a network of
partnersexclusively tied to a focal actor is to be preferredto
one in which the focal actor's partnershave many other part-
ners (Cook and Emerson, 1978; Brass and Burkhardt, 1992).
For the actor seeking to develop social capital, these posi-
tions suggest a confusing panoply of choices. At one level,
the arguments of this paper add further complexity to this
problem by highlighting the fact that each of these social
structural choices in fact entails a significant trade-off
between two potentially beneficial network outcomes.
Densely interconnected networks enable trust but limit the
inflow of diverse and fresh insights. Structural-hole-rich net-
works provide informational benefits but inhibit trust develop-
ment. Partners exclusively tied to an actor provide power
benefits, but it is partners with many other partners that pro-
vide the indirect ties that enhance his or her informational
reach within the network. At another level, however, the con-
clusions of this study suggest a path out of this dilemma.
The arguments and results from this study suggest that the
debate about the appropriate form of social capital may be
profitably informed by the extension of an established princi-
ple of organization design to the network arena: the optimal
structural design is contingent on the actions that the struc-
ture seeks to facilitate (Lawrence and Lorsch, 1967). What
constitutes an enabling social structure for one set of actions
may well be disabling for others (Podolny and Baron, 1997).
Thus, the form taken by social capital is likely to be contin-
gent on what actors seek to enable through it. Under the
appropriate circumstances, exclusive, cohesive, and non-
redundant connections can all constitute social capital. A net-
work composed of relationships with partners with few ties
to others would facilitate control over exchange partners
(Cook and Emerson, 1978; Brass and Burkhardt, 1992). Such
a network might be the objective for a firm seeking power
over its buyers or suppliers (Porter, 1980). A network com-
posed of partners with many interlocking and redundant ties
would facilitate the development of trust and cooperation
(Granovetter, 1985; Coleman, 1988; Portes and Sensenbren-
ner, 1993). Such a network may be useful from the firm's
perspective when it and its partners are faced with a com-
mon external threat, for instance, adverse political or legisla-
tive actions, or in the context of standard setting in high-tech-
nology industries (Oliver, 1990; Kogut, Walker, and Kim,
1995). Finally,a network of many non-overlapping ties would
provide information benefits (Burt, 1992). Such a network
would be ideal for an organization whose primary business
entails the brokerage of information or technology (Hargadon
and Sutton, 1997). Identifying the benefit sought from a
social structure is therefore likely to be critical in identifying
the form of social structure that is most likely to be facilita-
tive. What this study has shown is that there is no simple,
universal answer.

REFERENCES
Achilladelis, B., A. Schwarzkopf, Bian, Y. Cronbach, L.

and M. Cines 1997 "Bringing strong ties back in: 1987 "Statistical tests for modera-
1987 "A study of innovation in the Indirect ties, network bridges, tor variables: Flaws in analy-
pesticide industry: Analysis of and job searches in China." sis recently proposed." Psy-
the innovation record of an American Sociological chological Bulletin, 102:
industrial sector." Research Review, 62: 366-385. 41 4-41 7.
Policy, 16: 175-212.
Boorman, S. Davis, G. F.
Acs, Z., and D. B. Audretsch 1975 "A combinatorial optimization 1991 "Agents without principles?
1988 "Innovation in large and small model for transmission of job The spread of the poison pill
firms: An empirical analysis." information through contact through the intercorporate
American Economic Review, networks." Bell Journal of network." Administrative Sci-
78: 678-690. Economics, 6: 216-249. ence Quarterly, 36: 583-613.
1991 R&D, firm size and innova-
tive activity." In Z. J. Acs, and Brass, D. J., and M. E. Burkhardt Dyer, J. H., and K. Noboeka
D. B. Audretsch (eds.), Inno- 1992 "Centrality and power in orga- 2000 "Creating and managing a
vation and Technological nizations." In N. Nohria and high performance knowledge-
R. Eccles (eds.), Networks sharing network: The Toyota
Change: An International
and Organizations: 191-215. case." Strategic Management
Comparison: 39-59. Ann
Arbor: University of Michigan Boston: HarvardBusiness Journal, 21: 345-368.
Press. School Press.
Freeman, C.
Adler, R, and S. Kwon Burt, R. S. 1982 The Economics of Industrial
1999 "Social capital: The good, the 1991 STRUCTURE.Version 4.2. Innovation. Cambridge, MA:
New York:Center for the MIT Press.
bad, the ugly." Working
paper, Department of Man- Social Sciences, Columbia 1991 "Networks of innovators: A
University. synthesis of research
agement, University of South-
ern California. 1992 Structural Holes: The Social issues." Research Policy, 20:
Structure of Competition. 499-514.
Arora, A., and A. Gambardella Cambridge, MA: Harvard Uni-
1990 "Complementarity and exter- versity Press. Glasmeier, A.
nal linkages: The strategies of 1997 "The contingent value of 1991 "Technological discontinuities
large firms in biotechnology." social capital." Administrative and flexible production net-
works: The case of Switzer-
Journal of Industrial Econom- Science Quarterly, 42:
land and the world watch
ics, 38: 361-379. 339-365.
industry." Research Policy,
Auster, E. R. Cohen, W. M., and R. C. Levin 20: 469-485.
1992 "The relationship of industry 1989 "Empiricalstudies of innova-
evolution to patterns of tech- tion and market structure." In Granovetter, M.
1973 "The strength of weak ties."
nological linkages, joint ven- R. Schmalensee and R. D.
American Journal of Sociolo-
tures, and direct investment Willig (eds.), Handbook of
gy, 78: 1360-1380.
between U.S. and Japan." Industrial Organization:
1059-1107. New York:North- 1982 "The strength of weak ties: A
Management Science, 38:
network theory revisited." In
778-792. Holland.
P. Marsden, and N. Lin (eds.),
Basberg, B. L. Cohen, W. M., and D. A. Levinthal Social Structure and Network
1983 "Foreign patenting in the U.S. 1989 "Innovation and learning: The Analysis: 103-130. Beverly
as a technology indicator: The two faces of R&D." Econom- Hills, CA: Sage.
case of Norway." Research ic Journal, 99: 569-596. 1985 "Economic action and social
Policy, 12: 227-237. structure: The problem of
1987 "Patents and the measure- Coleman, J. S. embeddedness." American
ment of technological change: 1988 "Social capital in the creation Journal of Sociology, 91:
A survey of the literature." of human capital." American 481-510.
Research Policy, 16: 131-141. Journal of Sociology, 94:
S95-S1 20. Greene, W. H.
Berg, S., J. Duncan, and R Fried- 1997 Econometric Analysis, 3rd ed.
man Comanor, W. S., and F. M. New York:Macmillan.
1982 Joint Venture Strategies and Scherer
Corporate Innovation. Cam-
1969 "Patent statistics as a mea- Griliches, Z.
sure of technical change." 1990 "Patent statistics as econom-
bridge, MA: Oelgeschlager,
Journal of Political Economy, ic indicators: A survey." Jour-
Gunn and Hain.
77: 392-398. nal of Economic Literature,
Bernstein, J., and M. Nadiri 27: 1661-1707.
1989 "Research and development Cook, K. S., and R. M. Emerson
and intra-industryspillovers: 1978 "Power, equity and commit- Gulati, R.
An empirical application of ment in exchange networks." 1995 "Social structure and alliance
American Sociological formation patterns: A longitu-
dynamic duality." Review of
Review, 43: 712-739. dinal analysis." Administrative
Economic Studies, 56:
Science Quarterly, 40:
249-269.
619-652.

1999 "Network location and learn- Jaffe, A. B., M. Trajtenberg, and Mitchell, W., and K. Singh
ing: The influence of network R. Henderson 1996 "Survival of businesses using
resources and firm capabili- 1993 "Geographic localization of collaborative relationships to
ties on alliance formation." knowledge spillovers as evi- commercialize complex
Strategic Management Jour- denced by patent citations." goods." Strategic Manage-
nal, 20: 397-420. Quarterly Journal of Econom- ment Journal, 17: 169-195.
ics, 108: 577-598.
Gulati, R., and M. Garguilo Mizruchi, M. S.
1999 "Where do networks come Kamien, M. I., and N. L. Schwartz 1989 "Similarityof political behav-
from?" American Journal of 1982 Market Structure and Innova- ior among large American cor-
Sociology, 104: 1439-1493. tion. New York:Cambridge porations." American Journal
University Press. of Sociology, 95: 401-424.
Gulati, R., and P. Lawrence
1999 "Organizing vertical networks: Kleinecht, A. Narin, F., E. Noma, and R. Perry
A design perspective." Paper 1982 Patenting in the Netherlands: 1987 "Patents as indicators of cor-
presented at SMJ Special A Cross-section Test on the porate technological
Issue Conference on Strate- Industry Life Cycle. Paris: strength." Research Policy,
gic Networks, Evanston, IL. OECD. 16: 143-155.
Gulati, R., and H. Singh Kogut, B., G. Walker, and D.-J. Oliver, C.
1998 "The architecture of coopera- Kim 1990 "Determinants of interorgani-
tion: Managing coordination 1995 "Cooperation and entry induc- zational relationships: Integra-
costs and appropriation con- tion as an extension of tech- tion and future directions."
cerns in strategic alliances." nological rivalry."Research Academy of Management
Administrative Science Quar- Policy, 24: 77-95. Review, 15: 241-265.
terly, 43: 781-814.
Kogut, B., and U. Zander Palepu, K.
Hagedoorn, J., and J. Schaken- 1992 "Knowledge of the firm, com- 1985 "Diversification strategy, prof-
raad binative capabilities, and the it performance and the
1994 "The effect of strategic tech- replication of technology." entropy measure." Strategic
nology alliances on company Organization Science, 3: Management Journal, 6:
performance." Strategic Man- 383-397. 239-255.
agement Journal, 15:
291-309. Krackhardt, D. Podolny, J. M., and J. N. Baron
1992 "The strength of strong ties: 1997 "Resources and relationships:
Hargadon, A., and R. I. Sutton The importance of philos in Social networks and mobility
1997 "Technology brokering and organizations." In N. Nohria in the workplace." American
innovation in a product devel- and R. Eccles (eds.), Net- Sociological Review, 62:
opment firm." Administrative works and Organizations: 673-693.
Science Quarterly, 42: 216-239. Boston: Harvard
716-749. Business School Press. Podolny, J. M., and T. E. Stuart
1995 "A role-based ecology of
Haunschild, P R. Lane, P. J., and M. Lubatkin technological change." Ameri-
1993 "Interorganizationalimitation: 1998 "Relative absorptive capacity can Journal of Sociology, 100:
The impact of interlocks on and interorganizational learn- 1224-1260.
corporate acquisition activity." ing." Strategic Management
Porter, M. E.
Administrative Science Quar- Journal, 19: 461-477.
1980 Competitive Strategy: Tech-
terly, 38: 564-592.
Lawrence, P. R., and J. W. Lorsch niques for Analyzing Indus-
Hausman, J., B. Hall, and Z. 1967 "Differentiation and integra- tries and Competitors. New
Griliches tion in complex organiza- York:Free Press.
1984 "Econometric models for tions." Administrative Science
Portes, A., and J. Sensenbrenner
count data with an application Quarterly, 12: 1-47.
to the patents-R&D relation- 1993 "Embeddedness and immi-
Leonard-Barton, D. gration: Notes on the social
ship." Econometrica, 52:
1984 "Inter-personal communica- determinants of economic
909-938.
tion patterns among Swedish action." American Journal of
Jacquard, J., R. Turrisi, and C. K. and Boston-area entrepre- Sociology, 98: 1320-1350.
Wan neurs." Research Policy, 13:
1990 Interaction Effects in Multiple 101-114. Powell, W. W., K. W. Koput, and L.
Smith-Doerr
Regression. Newbury Park,
CA: Sage. Levin, R. C., A. K. Klevorick, R. R. 1996 "Interorganizationalcollabora-
Nelson, and S. G. Winter tion and the locus of innova-
Jaffe, A. B. 1987 "Appropriatingthe returns tion: Networks of learning in
1986 "Technological opportunity from research and develop- biotechnology." Administra-
and spillovers of R&D: Evi- ment." Brookings Papers on tive Science Quarterly, 41:
dence from firms' patents, Economic Activity, 3: 116-145.
profits, and market value." 783-820.
American Economic Review, Richardson, G. B.
76: 984-1001. Little, R. J. A., and D. B. Rubin 1972 "The organization of indus-
1987 Statistical Analysis with Miss- try." Economic Journal, 82:
ing Data. New York:Wiley. 883-896.

Rogers, E. M., and D. L. Kincaid Shan, W., G. Walker, and B. Kogut Szulanski, G.
1981 Communication Networks: 1994 "Interfirmcooperation and 1996 "Exploring internal stickiness:
Toward a New Paradigm for startup innovation in the Impediments to the transfer
Research. New York:Free biotechnology industry." of best practice within the
Press. Strategic Management Jour- firm." Strategic Management
nal, 15: 387-394. Journal, 17: Winter Special
Rogers, E. M., and J. K. Larsen Issue: 27-43.
1984 Silicon Valley Fever: Growth Singh, K., and W. Mitchell
of High Technology Culture. 1996 "Precarious collaboration: Uzzi, B.
New York:Basic Books. Business survival after part- 1997 "Social structure and compe-
ners shut down or form new tition in interfirm networks:
Rowley, T., D. Behrens, and D. The paradox of embedded-
partnerships." Strategic Man-
Krackhardt
agement Journal, 17: Evolu- ness." Administrative Science
2000 "Redundant governance
tionary Perspectives on Strat- Quarterly, 42: 35-67.
structures: An analysis of egy Supplement: 99-115.
structural and relational Walker, G., B. Kogut, and W. Shan
embeddedness in the steel Stuart, T. E. 1997 "Social capital, structural
and semiconductor indus- 1998 "Network positions and holes and the formation of an
tries." Strategic Management propensities to collaborate: industry network." Organiza-
Journal, 21: 369-386. An investigation of strategic tion Science, 8: 109-125.
alliance formation in a high-
Scherer, F. M. Zaheer, A., and S. Zaheer
technology industry." Admin-
1965 "Firmsize, market structure, 1997 "Catching the wave: Alert-
istrative Science Quarterly,
opportunity and the output of 43: 668-698.
ness, responsiveness, and
patented inventions." Ameri- market influence in global
can Economic Review, 55: Stuart, T. E., and J. M. Podolny electronic networks." Man-
1097-1125. 1996 "Local search and the evolu- agement Science, 43:
tion of technological capabili- 1493-1509.
Scherer, F. M., and D. Ross ties." Strategic Management
1990 Industrial Market Structure
Journal, 17: Evolutionary Per-
and Economic Performance.
spectives on Strategy Supple-
Chicago: Rand McNally. ment: 21-38.


Quarterly Administrative Science: Collaboration Networks, Structural Holes, and Innovation: A Longitudinal Study

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Quarterly Administrative Science: Collaboration Networks, Structural Holes, and Innovation: A Longitudinal Study

Загружено:

Авторское право:

Доступные форматы

Administrative Science

Collaboration Networks, Structural Holes, and Innovation: A Longitudinal Study

The online version of this article can be found at:

Samuel Curtis Johnson Graduate School of Management at Cornell University

Email Alerts: http://asq.sagepub.com/cgi/alerts

>> Version of Record - Sep 1, 2000

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

Two other studies that have exploredthe nexus between

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

spillovers. Finally,the degree of connectivitybetween a firm's

Hypothesis 1: The more directties that a firm maintains,the

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

1982). Otherthings being equal, firms that have many indi-

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

linkedto each other, the possibilityof opportunisticactions is

Hypothesis 4b: The greaterthe structuralholes spanned by a firm,

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

joint ventures, using Who Owns Whom (UnitedStates, Unit-

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

firms may differon unmeasuredcharacteristics,I used the

This firm-specificeffect permits observations of the same

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

plied the vector of Burt'sfrequency-decayweighted path dis-

where pjqis the proportionof i's relationsinvested in the con-

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

5 (0.20)2= 0.20. Anotherfirmwith the same numberof

1981 3.34 2.9 3.64 0.0 18 14 26

442/ASQ, September 2000

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

related with each other, as would be expected. Among the

1. Patents1t 51.18 82.67

6. Direct ties X Indirect ties/1 0, counts 1 .13

Several of the controlvariableresults are also significant

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

ed with patentingfrequency,The estimated coefficient, how-

CY) _~ *O CY *- 1- - -- 1- 1- - 1 -- I'- 1 r-- I'- .N ' .- 1-

U) a), U) U) U) CD CDCl +-,

446/ASQ, September 2000

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

U):7 OOm mm O O O O Ot OmO O 1-0 am O OO00

O CY) '* I)00 CD

447/ASQ, September 2000

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

tion possible between direct and indirect ties is limited. More

Figure 2. Illustrationof indirect ties in a network as both resources and

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

work attributesand positions can only be understood relative

452/ASQ, September 2000

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

Achilladelis, B., A. Schwarzkopf, Bian, Y. Cronbach, L.

453/ASQ, September 2000

454/ASQ, September 2000

Downloaded from asq.sagepub.com at ILLINOIS STATE UNIV on December 5, 2012

455/ASQ, September 2000

Вам также может понравиться

CY) _~ O CY - 1- - -- 1- 1- - 1 -- I'- 1 r-- I'- .N ' .- 1-