KNOWLEDGE SHARING IN ORGANIZATIONS: MULTIPLE …faculty.london.edu/lmors/assets/documents/Hansen_Mors_Lovas_05.pdf · KNOWLEDGE SHARING IN ORGANIZATIONS: MULTIPLE NETWORKS, MULTIPLE

KNOWLEDGE SHARING IN ORGANIZATIONS:MULTIPLE NETWORKS, MULTIPLE PHASES

MORTEN T. HANSENINSEAD

MARIE LOUISE MORSMassachusetts Institute of Technology

BJØRN LØVÅSLondon Business School

Different subsets of social networks may explain knowledge sharing outcomes indifferent ways. One subset may counteract another subset, and one subset may explainone outcome but not another. We found support for these arguments in an analysis ofa sample of 121 new-product development teams. Within-team and interunit networkshad different effects on the outcomes of three knowledge-sharing phases: decidingwhether to seek knowledge across subunits, search costs, and costs of transfers. Theseresults suggest that research on knowledge sharing can be advanced by studying howmultiple networks affect various phases of knowledge sharing.

What determines the occurrence and effective-ness of knowledge sharing in an organization?Scholars have examined this question from differ-ent viewpoints, focusing on the problem of trans-ferring tacit and complex knowledge across organ-ization subunits (e.g., Zander & Kogut, 1995), thenature of informal relationships between two par-ties to a transfer (e.g., Gupta & Govindarajan, 2000;Reagans & McEvily, 2003; Tsai, 2002), and theproblem of searching for knowledge (e.g., Ancona &Caldwell, 1992). This growing body of literaturehas shed much light on the various problems un-derlying knowledge sharing in organizations. Yet,because the overarching process of knowledgesharing has been shown to consist of multiplephases—each with their own associated challeng-es—existing research on this question is quite dis-parate and has not shed much light on whethervariables that explain one phase of knowledge shar-ing also explain other phases (Eisenhardt & Santos,2002).

In particular, while some studies have analyzedthe transfer of knowledge from one point to another(e.g., Gupta & Govindarajan, 2000; Szulanski, 1996;Zander & Kogut, 1995), they have excluded thelogically prior phase of searching for knowledge or

have not empirically disentangled the two phasesof search and transfer (Hansen, 1999). Scholarstherefore do not know much about the extent towhich different factors explain search and transfer.Moreover, in the few studies on search for knowl-edge, researchers have tended not to distinguishbetween the decision to seek knowledge and theensuing search process (e.g., Borgatti & Cross, 2003;Hansen, 1999).

These shortcomings in existing research areproblematic. Existing findings may be biased orincomplete to the extent that explanatory proper-ties that have a positive or negative effect on out-comes in one phase, such as transfer, may have theopposite or no effect on outcomes in other phases.There is therefore a need for research that exploresthe extent to which different properties explainoutcomes associated with the various phases ofknowledge sharing.

To address this gap in extant research on knowl-edge sharing, we pose this research question: Doproperties of social networks explain the outcomesof the three phases—deciding to seek knowledge,searching for knowledge, and transferring knowl-edge—in different ways? By social networks, werefer to subsets of established informal relationsthat exist within teams and across subunits in anorganization. We delineate the application of thismultiple network approach to an analysis of onetype of task units, new-product developmentteams, and examine the impacts of teams’ subsetsof social networks on whether they seek knowledge

We thank three anonymous reviewers, Associate Edi-tor Marshall Schminke, Martine Haas, Michael Jacobides,Freek Vermeulen, Olav Sorenson, Julian Birkinshaw, andPhanish Puranam.

Bjørn Løvås died in an accident in July 2005. Hiscoauthors dedicate this article to his memory.

� Academy of Management Journal2005, Vol. 48, No. 5, 776–793.

776

across subunits, their interunit search costs, andthe costs of knowledge transfer.

MULTIPLE NETWORKS, MULTIPLE PHASES

Scholars adopting a relational approach toknowledge sharing have mainly focused on thecharacteristics of established informal relationsthat facilitate or impede the sharing of knowledgein an organization. This body of research has, how-ever, focused on quite different subsets of relations.Some research has focused on analyzing informalrelations within teams and organization subunits(e.g., Levin & Cross, 2004), whereas other studieshave focused on teams’ and individuals’ externalinformal relations (Ancona & Caldwell, 1992; Cum-mings, 2004) or on relations between subunits in anorganization (e.g., Gupta & Govindarajan, 2000).Furthermore, while some research has analyzedproperties of a subunit’s total set of interunit rela-tions (e.g., Hansen, 1999; Tsai, 2001), other studieshave focused exclusively on the properties of dy-adic relations involving one providing and one re-ceiving unit in a transfer event (e.g., Gupta & Gov-indarajan, 2000; Szulanski, 1996).

Apart from a few studies that have begun to ad-dress multiple subsets of relations (e.g., Nohria &Ghoshal, 1997; Schulz, 2001), this body of researchhas not carefully analyzed the potentially differentroles of various subsets of relations, yielding anincomplete understanding of what particular sub-sets of relations affect the decision to seek knowl-edge across subunits, search costs, and costs oftransfer. To address this shortcoming, we develop amultiple network perspective that addresses threesubsets of networks that exist prior to the start of afocal project: established relations between mem-bers of a team located within a focal subsidiary (awithin-team network); a team’s total set of estab-lished relations with those in other subsidiaries,irrespective of whether it transfers knowledge fromthose contacts (an intersubsidiary network); and ateam’s dyadic relations, involving only subsidiariesproviding it with knowledge (a transfer network).The latter two subsets both refer to direct contactsor relations that a team in a focal subsidiary haswith other subsidiaries. The difference betweenthese two subsets is that an intersubsidiary networkincludes all direct contacts that a team has,whereas a transfer network includes only directcontacts with subsidiaries from which it is trans-ferring knowledge.

We argue that these three informal networks af-fect a team’s decision to seek knowledge, searchcosts, and costs of transfers in different ways, andwe explore two overarching theoretical mecha-

nisms for why this may be the case. First, as priorresearch on social networks and communicationpatterns in product development projects hasshown, established informal relations tend to chan-nel actors’ time and energy in the direction of es-tablished contacts—actors interact with thosewhom they have interacted with before (e.g., Gu-lati, 1995; Katz & Allen, 1988). This channelingmechanism implies that different subsets of net-works may channel teams’ time and energy in dif-ferent directions. Second, building on prior re-search that has shown that an actor’s differentsocial relations have different utilities (e.g.,Podolny & Baron, 1997), we argue that the threedifferent subsets of relations serve different pur-poses. Intersubsidiary relations are more beneficialfor search across subsidiaries than are within-teamrelations, whereas dyadic transfer relations aremore useful for transfer than a team’s total set ofintersubsidiary relations. This utility mechanismimplies that properties of the three networks thatwe analyze should have different impacts on theoutcomes of the three phases.

On the basis of these two overarching theoreticalmechanisms, we develop a set of hypotheses,which are depicted in Figure 1. We focus on threerelational variables that have received attention inknowledge-sharing research: the extent of a net-work as given by its size (e.g., Hansen, 2002), thestrength of relations (e.g., Hansen, 1999), and thedegree of perceived competition inherent in rela-tions (e.g., Tsai, 2002).

Deciding Whether to Seek Knowledge acrossSubsidiaries

In the first phase, teams may decide to seekknowledge in other subsidiaries. Assuming that ateam’s decision to seek knowledge results in at-tempts to seek knowledge, an outcome of this phaseis thus whether a focal team in a focal subsidiarycontacts other subsidiaries to inquire about poten-tially useful knowledge for a project. Irrespective ofteams’ varying needs to seek knowledge from sub-sidiaries other than their own, a team’s decision toseek knowledge across subsidiaries may be affectedby established relations that exist both withinteams and across subsidiaries.1

1 We assume that project development tasks are notpurposefully decomposed into subtasks whereby othersubsidiaries provide components to the focal projectand subsidiary, thereby preordaining intersubsidiaryinteractions.

2005 777Hansen, Mors, and Løvås

Within-team network. Product developers mayhave formed enduring relations among themselves,including friendship and advice-seeking informalrelations, that develop over time and exist prior tothe start of a particular product developmentproject (Podolny & Baron, 1997). When developersjoin a common project team, these established re-lations may channel their time and energy towardthe team: when confronted with new project-spe-cific problems, they may seek to solve them byinteracting and sharing knowledge with team mem-bers. Given that teams typically have a finite num-ber of problems to solve, this preference is likely toreduce the need to seek knowledge outside the teams,including knowledge resident in other subsidiaries.

Existing theory suggests three reasons why thischanneling may occur. First, within-team relationsmay be associated with an in-group bias. Socialpsychologists have studied the tendency of somegroup members to systematically overvalue groupmembers and undervalue nonmembers (e.g.,Brewer, 1979; Tajfel & Turner, 1986). Scholarsstudying research and development (R&D) activi-ties in organizations have reported a similar ten-dency among engineers, known as the “not-invent-ed-here” syndrome, which refers to the bias in agroup of engineers of valuing their own knowledgemore than that of others and henceforth rejecting

knowledge held by others outside their group(Hayes & Clark, 1985; Katz & Allen, 1988). Thesebiases often develop when group members havespent considerable time interacting with one an-other, a scenario that restricts the inflow of newviewpoints and differences and reinforces com-monly held beliefs (Oakes, Haslam, Morrison, &Grace, 1995; Wilder, 1978).

Second, established relations may be associatedwith a team’s inward-looking absorptive capacity(Cohen & Levinthal, 1990). Product developers whointeract regularly with one another may, over time,assimilate one another’s knowledge, thus develop-ing a common knowledge base and a common spe-cialized set of terminologies. Through establishedrelations, they may also develop a capacity for mu-tual problem solving (Uzzi, 1997). These interac-tion benefits in turn make it easier for team mem-bers to absorb one another’s knowledge, leading toa preference to work and share knowledge witheach other as opposed to seeking knowledge resi-dent in subsidiaries other than their own.

Third, established within-team relations maygive team members increased awareness of eachother’s knowledge, including awareness of knowl-edge that may be relevant to project-specific tasks(Austin, 2003; Schulz, 2003). This awareness maylead team members to focus on this body of knowl-

FIGURE 1Effects of Subsets of Social Networks on Outcomes of Three Phases of Knowledge Sharing

778 OctoberAcademy of Management Journal

edge when seeking solutions and further reduce thechances that the team will seek knowledge fromsubsidiaries other than their own.

Members of a team who have a number of ongo-ing established relations with one another prior tothe start of a project may therefore channel theirtime and energy toward the team and not towardother subsidiaries. The density of the within-teamnetwork (i.e., the number of ongoing establishedrelations among team members, divided by the to-tal possible number of such relations)2 and theaverage strength of relations (i.e., the frequency andintensity of interactions) may positively affect thischanneling tendency. The chance of an in-groupbias forming is likely to increase because of theenhanced opportunity for members to jointly em-phasize the value of their own skills and reinforcecommonly held beliefs afforded by extensive, fre-quent, and intense past interactions. Furthermore,teams that have high density and average strengthof within-team relations are likely to have a largercommon knowledge base than teams with low den-sity and strength of relations, increasing the prefer-ence for relying on team members. And finally, themore extensively, frequently, and intensively teammembers have interacted in the past, the morechances they have had to become aware of theextent and type of knowledge held by other teammembers. This should increase their preference forseeking solutions inside the team and not outside.We therefore predict:

Hypothesis 1. The higher the density of a with-in-team network, the less likely the focal teamis to seek knowledge across subsidiaries.

Hypothesis 2. The higher the average strengthof relations in a within-team network, the lesslikely the focal team is to seek knowledgeacross subsidiaries.

Intersubsidiary network. The within-team net-work may channel members’ time and energy in-side their team, while the team’s established rela-tions in its intersubsidiary network may “pull” itoutward. The causal mechanisms that lead teams toseek solutions and knowledge within themselvesmay also explain why some teams tend to lookoutward. First, a team’s intersubsidiary networkmay mitigate the effect of in-group biases. Teamsthat are isolated from outside interactions mayform negative perceptions about others, thereby in-

creasing the risk that an in-group bias is forming(Tajfel & Turner, 1986). The larger a focal team’sintersubsidiary network (i.e., the higher the num-ber of direct intersubsidiary relations), the lowerthis risk should be, as team members have hadmore chances to interact with engineers in othersubsidiaries and thereby have been exposed to dif-ferent views and skills. Moreover, the higher theaverage strength of these established relations, thelower the risk should be, as more frequent andintense interactions will have increased the expo-sure of team members to other views and skills,thereby reducing their negative perceptions of oth-ers and increasing the chances that they will seekknowledge in other subsidiaries.

Second, established intersubsidiary relationsmay also be associated with a focal team’s capacityto absorb knowledge from those contacts, as therehave been a number of interactions in the past inwhich focal team members have had opportunitiesto learn to engage in mutual problem solving anddevelop a common knowledge base with engineersin those subsidiaries (Cohen & Levinthal, 1990;Uzzi, 1997). The larger a team’s intersubsidiary net-work and the more frequent and intense interac-tions have been, the more such opportunities havebeen given to team members, enhancing their ab-sorptive capacity and hence their preference forseeking knowledge from engineers in these othersubsidiaries.

Third, as the size and average strength of a focalteam’s intersubsidiary network increase, the teammembers’ awareness of knowledge held by engi-neers in this network should increase, becauseteam members have had more extensive, frequent,and intense opportunities to learn about thisknowledge (cf. Austin, 2003). With increasedawareness about the knowledge available in othersubsidiaries, including knowledge that is relevantfor completing project-specific tasks, a focal teamshould be more likely to seek knowledge acrosssubsidiaries. In short, we predict:

Hypothesis 3. The larger a focal team’s inter-subsidiary network, the more likely the team isto seek knowledge across subsidiaries.

Hypothesis 4. The higher the average relationstrength in a focal team’s intersubsidiary net-work, the more likely the team is to seek knowl-edge across subsidiaries.

Search Costs

A team in a focal subsidiary that has decided toseek knowledge needs to search for it, an activitythat involves looking for, identifying, and evaluat-

2 The total possible number of relations in a within-team network is defined as N � (N �1), where N is thenumber of team members (Wasserman & Faust, 1994).


ing knowledge resident in other subsidiaries. Dur-ing this phase, teams incur search costs, which inour context are measured as the number of engi-neering-months that a product development teamspends on search.

Because the search phase in our definition pre-cedes the identification of useful knowledge, teammembers may not know ex ante which subsidiariespossess novel and useful knowledge for a particularproject. In such a situation, a team’s total set ofintersubsidiary relations may be useful as it pro-vides an initial search position from which a teamin a focal subsidiary can draw on several intersub-sidiary relations to launch a search for knowledge.

As social network research has shown, however,it is not the sheer number of direct relations in anetwork that provides an advantageous search po-sition but, rather, the extent to which those directrelations provide access to novel knowledge (cf.Burt, 1992; Granovetter, 1973; Hansen, Podolny, &Pfeffer, 2001). In particular, a set of intersubsidiaryrelations with low average strength (i.e., relationsinvolving infrequent and nonintense interactions)is more likely than a set with high average strengthto enable team members to access direct contactsthat possess novel knowledge. A set of relationswith high average strength tends to involve directcontacts that are themselves connected, becausethey are likely to have been introduced to eachother through their common strongly tied actor(Burt, 1992; Granovetter, 1973). Direct contacts thatare themselves connected tend to circulate knowl-edge among one another and thereby possess sim-ilar knowledge, reducing the probability that a fo-cal team will identify novel knowledge whencontacting several of them. Thus, for a givenamount of novel knowledge being sought, a teammay have to make a number of contacts, as each oneis likely not to have much knowledge that differsfrom what the other contacts have. In contrast,when the average strength of the relations in ateam’s intersubsidiary network is low, the team’sdirect contacts are less likely to be connected them-selves and thereby are less likely to possess thesame knowledge. In this situation, the team islikely to identify more novel knowledge for eachcontact it makes and thereby lower search costs fora given amount of novel knowledge sought.3 Wethus predict:

Hypothesis 5. The higher the average relationstrength in a focal team’s intersubsidiary net-work, the higher the focal team’s search costs.

The previous discussion rests on the assumptionthat direct contacts in a focal team’s intersubsidiarynetwork are motivated to help the focal team iden-tify useful knowledge, but this may not be the case.One reason for a lack of motivation to aid search isintersubsidiary competition (Tsai, 2002). A teamand its subsidiary may compete with another sub-sidiary in the sense that both subsidiaries sell prod-ucts to the same external markets and seek to de-velop the same types of products and technologies.In such a situation, one subsidiary’s product devel-opment efforts may constrain the opportunities of acompeting subsidiary to the extent that new prod-ucts and technologies “crowd out” the availablemarket and technology opportunity set for the com-peting subsidiary (Sorenson, 2000).

If subsidiaries that a focal team contacts in itssearch for knowledge perceive that the focal teampresents such a competitive threat, the contactedsubsidiaries may hide what they know (so that thefocal team will not identify useful knowledge), de-clare only parts of their related knowledge (so thatthe team only benefits partially), or fail to point thefocal team to other subsidiaries that may have use-ful knowledge (making search more difficult). Al-though they may be inaccurate, these perceptionsmay nevertheless govern contacted subsidiaries’ re-sponses and are likely to increase a focal team’ssearch costs, as team members may need to expendadditional effort searching to obtain a given amountof useful knowledge. Thus, focal teams whose di-rect contacts perceive high levels of competition aremore likely to have higher search costs than teamswhose direct contacts perceive low levels of compe-tition between them and the focal subsidiary:

Hypothesis 6. The higher the average level ofcompetition perceived by subsidiaries in a fo-cal team’s intersubsidiary network, the higherthe focal team’s search costs.

Transfer Costs

Once a focal team has identified potentially use-ful knowledge, it has to be transferred from theproviding subsidiary to the team, a process thatinvolves modifying, editing, and incorporating theknowledge into the team’s product (Szulanski,1996; Zander & Kogut, 1995). In this phase, teams3 Having a high average relation strength may be espe-

cially problematic for teams pursuing very novelprojects. In the empirical part of our study, we includedan interaction term involving average intersubsidiary re-lation strength and project novelty, but results for this

term were not significant, suggesting that this was not thecase in our data set.


incur transfer costs, which we measure as the num-ber of engineering-months spent on this activity.

While a focal team’s intersubsidiary networkconfers a search position, a transfer event mostlikely does not involve the total set of a team’sintersubsidiary contacts but only those that end upproviding knowledge. Despite having agreed toprovide knowledge, however, providing subsidiar-ies may have various degrees of motivation to aidthe transfer effort, in part depending on the level ofperceived competition between them and the focalsubsidiary. A providing subsidiary may believethat a transfer of knowledge to a team in a compet-ing subsidiary will diminish its own opportunitiesfor developing products and technologies based onthe knowledge being provided to the focal team (cf.Galunic & Eisenhardt, 2001). As a result, the pro-viding subsidiary may “drag its feet” and be moreguarded about what knowledge—and how muchknowledge—it transfers than it would be if no com-petition existed. In this situation, the focal team hasto spend more time negotiating with the providingsubsidiary, increasing the time and effort requiredto transfer knowledge.

The effect of competition on transfer costs pri-marily concerns the extent of competition per-ceived by the providing subsidiary, not by the focalteam or the focal subsidiary. Providing engineerswho perceive that a focal team and subsidiary posea significant competitive threat are likely to bemore guarded and less forthcoming than are pro-viding engineers who do not perceive the sameextent of competition. Stated in a hypothesis:

Hypothesis 7. In a dyadic transfer relationship,the more a providing subsidiary perceives thatit competes with a focal subsidiary, the higherthe focal team’s transfer costs.

Although the extent of perceived competition af-fects the motivation to transfer knowledge to a focalteam, other factors affect the ability of a providingsubsidiary and the focal team to carry out a transfersmoothly. Prior research has demonstrated thattacit knowledge—knowledge that is difficult to ar-ticulate or that can only be acquired through expe-rience—is difficult to transfer smoothly (e.g., Han-sen, 1999). When knowledge is tacit, engineers inthe providing subsidiary will find it more difficultto explain its content and nuances to members ofthe focal team, who in turn may find it difficult tounderstand, thereby making the tasks of modifyingand incorporating the knowledge into the productdifficult. Because of these difficulties, transferringtacit knowledge is likely to be more cumbersome,take a longer time, and thus be more costly than

transferring nontacit knowledge (Zander & Kogut,1995).

As prior research has shown, however, the diffi-culty of transferring knowledge can be alleviated tosome extent if the two parties to a transfer knoweach other well and thus have learned to worktogether (Hansen, 1999; Uzzi, 1997). When two par-ties to a transfer have developed a strong relationprior to the transfer effort, they have likely devel-oped a shared communication frame whereby eachparty has come to understand how the other partyuses subtle phrases and ways of explaining difficultconcepts (Uzzi, 1997). Such strength in a dyadictransfer relation should therefore reduce transfercosts by reducing the time and effort required tounderstand and incorporate knowledge into the fo-cal project. In particular, as the tacitness of theknowledge that a subsidiary provides to a focalteam increases, an existing shared communicationframe afforded by established strong dyadic trans-fer relations is likely to become more important, asthe two parties to a transfer can rely on it to artic-ulate, modify, and incorporate the subtle and im-plicit aspects of the tacit knowledge, thereby reduc-ing transfer costs:

Hypothesis 8. Knowledge tacitness will modifythe main negative effect of dyadic relationshipstrength on a focal team’s transfer costs: theeffect will be stronger when tacitness is highand weaker when tacitness is low.

METHODS AND DATA

We tested the hypotheses using a data set of 121new-product development teams and 41 subsidiar-ies of a large high-technology company. The com-pany, which had annual sales of more than $5billion at the time of the study, was involved indeveloping, manufacturing, and selling a range ofindustrial electronics and other high-technologyproducts and systems. It was structured into 41fairly autonomous subsidiaries that were responsi-ble for their own new-product development, man-ufacturing, and sales and thus had considerablefreedom in choosing which other subsidiaries tocontact to access useful knowledge. For the pur-poses of this study, we considered a new-productdevelopment team in a focal subsidiary to be en-gaged in intersubsidiary search and transfer if itcontacted and obtained knowledge from one ormore of the other 40 subsidiaries. Having negoti-ated access to the company through three seniorcorporate R&D managers, we visited 14 subsidiariesand conducted open-ended interviews with morethan 30 project engineers and managers to better


understand the context and to develop survey in-struments to test our hypotheses.

To select product development teams, we usedthe firm’s databases on projects to develop a list ofprojects that the subsidiaries undertook and iden-tified 147 projects. We administered three differentsurvey instruments to create the variables: a net-work survey distributed to the R&D managers ineach of the 41 subsidiaries, asking about relationsacross the subsidiaries (a 100 percent response ratewas achieved); a survey distributed to the 147project managers of each of the product develop-ment teams included in the study, asking aboutsources of knowledge for the project (an 82 percentresponse rate); and a survey distributed to 510 in-dividual members of the projects, asking abouttheir own personal relations with colleagues (a 51percent response rate).

The final sample included 121 projects that tookplace in 27 different subsidiaries. We used the in-formation on projects from the databases to analyzeresponse rates but found no differences betweenthe final 121 projects and the 26 projects for whichthe survey was not returned in terms of number ofengineers, budget, and project age. In addition, wetook several steps to minimize potential commonmethod biases (Podsakoff & Organ, 1986). This con-cern was reduced substantially because we reliedon three different sets of respondents. Furthermore,as recommended by prior research, we focused onbehavioral measures and not on perceptual ones,which are prone to common method biases(Gatignon, Tushman, Smith, & Anderson, 2002).

Dependent Variables

Sought knowledge. We asked each project man-ager to indicate whether team members had soughtknowledge in any of the other subsidiaries exceptthe focal one. Project managers were able to answerthis question reliably, as they kept detailed logs onhow engineers on the project spent their time. Fif-ty-five project managers (45%) reported that theirteams had sought knowledge in at least one of theother 40 subsidiaries. We coded this dependentvariable, sought knowledge, as 1 if a team hadsought knowledge from another and 0 otherwise.

Search costs. The preliminary interviews in-formed us that one of the most salient search costsfor these projects was engineering-months spentlooking for, identifying, and evaluating knowledgefrom other subsidiaries and that project managerskept track of the number of engineering-monthsspent on this activity. The company defined anengineering-month as the equivalent of a productdeveloper’s full-time work for one month. We

asked the project managers, “How many engineer-ing-months did the team spend searching for (look-ing for, identifying, and evaluating) the technicaladvice, software and hardware that the teamwanted from other subsidiaries?” The responsesranged from 0.1 to 15 (or from approximately 1 to22 percent of total engineering-months for aproject), with a mean of 4.39 (or 5 percent of totalengineering-months). To compute the variablesearch costs, we used the absolute number of engi-neering-months spent on search.

Transfer costs. To measure transfer costs, weasked the project managers, “How many engineer-ing-months did the team spend modifying, editingand incorporating the technical advice, softwareand hardware that came from the other subsidiar-ies?” Project managers kept a close eye on engineer-ing-months spent on this activity and thus couldprovide reliable answers to this question. The re-sponses ranged from 0.1 to 40 (or approximately 1to 57 percent of total project engineering-months),with a mean of 7.94 (or 10 percent of total engineer-ing-months). To compute the variable transfercosts, we used the absolute number of engineering-months spent on transfer. Five projects did notreport transfer costs, and we thus had to drop thesefrom the analysis (there were no significant differ-ences between these five and the other projects).

These search and transfer costs variables do notspecify the benefits of using knowledge from othersubsidiaries. To control for the benefits that theteams obtained by spending time searching andtransferring, we entered a control variable denotingthe amount of knowledge obtained from other sub-sidiaries (described in the section about control vari-ables). This way we measured search and transfercosts given a certain amount of knowledge obtained.

Within-Team Network Variables

Within-team network density. To test Hypothe-sis 1, we used responses from the individual teammember survey. Following accepted proceduresfrom the social network literature for soliciting aperson’s network contacts (e.g., Podolny & Baron,1997), we asked each team member three questions:“Looking back over the last year, are there anypersons in your subsidiary: (i) from whom youregularly sought information and advice to helpyour project work, (ii) to whom you would go on aregular basis to get buy-in for your work, or (iii)with whom you interacted informally as a friend?”Each respondent entered the names of individualsdescribed by either of these three choices, and wethen counted the number of contacts these individu-als had with other members of the respondent’s team.


In addition, to avoid recording relations that werecreated after the start of a focal project, we asked eachrespondent how long each relation had been in exis-tence and excluded relations that commenced afterthe start of the focal project. We computed within-team network density as the number of existing rela-tions divided by the number of possible asymmetricrelations, which is given by N � (N � 1), where N isthe number of team members (Wasserman & Faust,1994). This measure ranges from 0 (“no relationsexist”) to 1 (“all possible relations exist”).4

Within-team relation strength. To test Hypoth-esis 2, we asked team members two questions:“How frequently have you interacted with this per-son over the past year?” (1, “once a day,” to 7,“once every 3 months”) and “How close are you tothis person? (1, “very close,” to 7, “distant”). Wereverse-scored the two dimensions and took theaverage of these two dimensions to create averagewithin-team relation strength. The Cronbach alphafor this scale was .81.

Intersubsidiary Network Variables

Intersubsidiary network size. To test Hypothe-sis 3, we constructed a measure of the number ofintersubsidiary relations. We asked the subsidiaryR&D managers, “Over the past two years, are thereany units [i.e., subsidiaries] from whom your unit[i.e., focal subsidiary] regularly sought technicaland/or market-related input?” The question wasfollowed by a list of the 41 subsidiaries included inthe study, and the manager could then check theappropriate subsidiaries. We asked the respondentsto indicate the age of each relation and excludedthose that had commenced after the start of focalprojects. These relations thus vary among theprojects within a subsidiary. Finally, we assignedthe appropriate intersubsidiary relations to eachfocal project and computed the number of estab-lished relations with other subsidiaries that existedprior to the start of the focal project, labeling thevariable intersubsidiary network size.

Intersubsidiary relation strength. We asked thesubsidiary R&D managers to answer the followingfor each of the relations he or she listed on thesurvey: “How frequently do people in your [subsid-iary] interact with this [subsidiary]?” (1, “once aday,” to 7, “once every 3 months”) and “How closeis the working relationship between your [subsid-iary] and this [subsidiary]?” (1, “very close, practi-

cally like being in the same work group,” to 7, “dis-tant, like an arm’s length delivery of the input”). Wereverse-scored and averaged the two items, whichwere highly correlated (r � .83; � � .92). To testHypotheses 4 and 5, we computed an average for eachteam, average intersubsidiary relation strength.

Intersubsidiary perceived competition. To mea-sure the level of perceived competition betweensubsidiaries, we asked each subsidiary R&D man-ager to indicate, for each relationship he or shelisted on the survey, “How would you describe thecompetitive nature between your [subsidiary] andthis [subsidiary]?” (1, “non-competitive: never com-pete for markets and technologies,” to 7, “competi-tive: frequently compete for markets and technolo-gies”). These perceptions are asymmetric: an R&Dmanager in subsidiary A may perceive subsidiary Bas a competitive threat to A, but the R&D manager insubsidiary B may not perceive subsidiary A as athreat to B. Because all 41 R&D managers completedthe survey and indicated the level of perceived com-petition with subsidiaries with which they had rela-tions, we were able to obtain information on howother subsidiaries perceived a focal subsidiary.

We used this information to construct two mea-sures of perceived competition in each team’s in-tersubsidiary network. First, to test Hypothesis 6,we computed the average value of direct contacts’responses to the above question. This variable, di-rect contacts’ average perceived competition, mea-sured how all subsidiaries with which a focal teamhad established relations perceived the subsidiaryto which the team belonged. Second, we computedfocal subsidiary’s average perceived competition—that is, how each subsidiary’s R&D manager per-ceived the level of competition with direct contacts.

Transfer Network Variables

Perceived competition in transfer. To test Hy-pothesis 7, we followed the same procedures butlimited the measure to only those subsidiaries thatprovided knowledge to a focal team. The averagevalue of their responses to the question about per-ceived competition was the variable providers’ per-ceived competition.

Dyadic transfer relation strength and tacitnessof knowledge. We limited this relational variableto subsidiaries that provided knowledge to a focalteam. Using the scale for intersubsidiary relationstrength described above, we constructed the mea-sure dyadic transfer relation strength (0, “no estab-lished relation,” to 7, “very strong relation”). Forthis measure, we subtracted the mean from thevalues to reduce the correlation with the interac-tion effect. To measure the tacitness of transferred

4 Because this information was missing for eightteams, we created a dummy variable that took on a valueof 1 for missing information and 0 otherwise.


knowledge, we asked each project manager to indi-cate, for the knowledge transferred from each pro-viding subsidiary: (1) “How well documented wasthe knowledge?” (1, “It was very well docu-mented,” to 7, “It was not well documented”); (2)“Was all of this knowledge explained to your teamin writing?” (1, “All of it was,” to 7, “None of itwas”); and (3) “What type of knowledge came fromthis unit?” (1, “mainly reports, manuals, docu-ments, self-explanatory software, etc.,” to 7,“mainly personal practical know-how, tricks of thetrade”). To compute the variable, we averaged theresponses and then subtracted the mean from thevalues. The Cronbach alpha for this variable was.81. To test Hypothesis 8, we interacted the tacit-ness variable with the variable denoting thestrength of dyadic transfer relations (tacitness �transfer relation strength).

Control Variables

Need to seek knowledge. We entered three vari-ables to control for teams’ need or inclination toseek knowledge outside their own subsidiary. First,a team engaged in tasks that depart significantlyfrom its subsidiary’s knowledge base may be moremotivated to search outside than teams that candraw on their subsidiaries’ existing knowledge. Wemeasured the extent to which each team drew onits subsidiary’s knowledge base via project manag-ers’ logs of the sources of the software and hard-ware used in their projects. The variable, existingware, was the fraction of all the software and hard-ware used in a project that had already been devel-oped and existed in the subsidiary to which theproject belonged.

Second, we used a two-item scale to captureproject novelty, the extent to which each projectstrayed from its subsidiary’s technological andmarket expertise. Each project manager was asked,“Prior to this project, how much experience didyour subsidiary have with the technologies andtechnical competencies that the project required?”and “Prior to this project, how much experiencedid your subsidiary have with the market for whichthe product was developed?” (1, “none of the re-quired expertise,” to 7, “all of the required exper-tise”). Answers to the two questions were averagedand then reversed the scale so that a high valueindicated a relatively novel project (� � .66).

Third, because prior research has shown thatteam members’ tenure negatively affects the prob-ability of their seeking outside help (e.g., Katz &Allen, 1988), we included team tenure, a variabledenoting the average number of years individualteam members had been with the company. Be-

cause this information was incomplete for 23 out ofthe 121 projects, we also included a dummy vari-able that took on a value of 1 if this information wasincomplete and 0 otherwise.

Search and transfer controls. Because it is pos-sible that projects with many engineering-monthsavailable have more capacity to search, we usedtwo control variables: a project’s total number of en-gineering-months (in hundreds of months), as esti-mated at the beginning of the project (project totalengineering-months) and the budget (in millions ofdollars) at the start of the project (project budget).

We also wanted to control for the amount ofknowledge sought in search and the amount ofknowledge obtained during transfer. As we men-tioned in the section defining search and transfercosts, knowledge obtained is a proximate measurefor the benefits of transferring knowledge. Al-though we did not have measures of the amount ofknowledge sought, we created a measure of theamount of knowledge obtained from other subsid-iaries. Each project manager was asked to indicatethe fraction of all his or her project’s hardware andsoftware that came from other subsidiaries (0, “noknowledge obtained,” to 1, “all knowledge camefrom other subsidiaries”). In field interviews withproject managers, we discovered that they keptclose track of the amount of knowledge obtained,as this indicated the savings they could accrueand thus affected the amount of engineeringhours required to complete the project. For soft-ware, for example, they would often keep trackof lines of code that were provided by other sub-sidiaries. To compute this variable, we took theweighted average of the two fractions—one forhardware and one for software—and multiplied itby the total number of a project’s budgeted engi-neering months. This variable, amount of knowl-edge, measured the absolute amount of engineeringmonths represented by the knowledge obtainedfrom other subsidiaries.5

Finally, we also controlled for each team’s capac-ity to absorb the knowledge obtained from othersubsidiaries. Because larger teams may have more

5 Although we posit that the amount of knowledgeobtained predicts search costs, it is reasonable to suspectthat search costs, which may be an indication of searcheffort, also predict the amount of knowledge obtained. Tocontrol for this endogeneity problem, we analyzed twosimultaneous equations, one specified as search costs �f (amount knowledge, all other independent variables),and one specified as amount knowledge � f (search costs,some other predictor variables). This test did not alterour results, and we are therefore reasonably confidentthat this type of endogeneity is not salient in our models.


skills and capacity for assimilating knowledge, weentered the variable team size, measuring the num-ber of engineers spending at least 50 percent oftheir time working on a focal project (team size).For a second measure, absorptive capacity, weasked each project manager, “Prior to the project,did your project team have the expertise required toassimilate the knowledge that came from the othersubsidiaries?” (1, “had very little of that expertise,”to 7, “had very much of that expertise”).

Statistical Approach

Because search and transfer costs were only rel-evant for those projects that sought and foundknowledge, we excluded from the analysis ofsearch and transfer costs those projects that did notseek knowledge.6 Thus, although the analysis ofwhether teams sought knowledge included all 121projects on which we obtained data, the analyses ofsearch and transfer costs include only 55 and 50projects, respectively. To ensure that the smallersets of data for search and transfer costs were largeenough for us to detect the expected effects, wecalculated the statistical power and the effect size,and both tests showed that we had adequate samplesize (Aiken & West, 1991; Cohen, 1988).7

In addition, because multiple project observa-tions from the same subsidiary might not be inde-pendent, we used robust standard errors that cor-rected for subsidiary-specific clustering and usedthe robust clustering procedure as implemented inStata 7.0 for all our models (Moulton, 1986; Rogers,1993). We used logistic regression analysis tomodel whether a team in a focal subsidiary soughtknowledge in one or more of the other subsidiariesand ordinary least squares (OLS) regression analy-sis to model the number of engineering-months

spent on search and transfer. We performed tests toensure that the OLS specification was appropriatefor our data. The Kolmogorov-Smirnov test of nor-mality revealed that the residuals were normallydistributed in the OLS models. For an alternativespecification, we also ran truncated regressionmodels wherein the distributions of the search andtransfer cost variables were presumed to be trun-cated at zero as there were no negative values.These models did not reveal any different effectsand are thus not reported here.

RESULTS

Table 1 reports descriptive statistics, and Table 2reports results pertaining to the three dependentvariables. Models 1–3 in Table 2 report results onteams deciding whether to seek knowledge acrosssubsidiaries. Model 2 shows that the within-teamnetwork density variable has a negative and signif-icant effect on whether a team sought knowledgeacross subsidiaries, lending support to Hypothesis1. As model 2 reveals, the effect for the averagewithin-team relation strength variable is also nega-tive but is only significant at the .10 level. To in-vestigate further, we separated the two underlyingdimensions of closeness and frequency underlyingrelation strength, as shown in model 3. The fre-quency dimension is negative and significant at the.01 level, lending support to Hypothesis 2.

Model 2 also reveals that intersubsidiary networksize has a positive effect on whether teams soughtknowledge across subsidiaries, lending support toHypothesis 3. However, average intersubsidiary re-lation strength is not significant in model 2, andthus Hypothesis 4 is not supported.

Figure 2a reveals that the supported effects are offairly substantial magnitudes.8 For example, whenwithin-team network density increases from 0.1 to0.3, the likelihood that teams sought knowledgedecreases from 0.62 to 0.31. In contrast, when in-tersubsidiary network size increases from 1 to 5,the likelihood of seeking knowledge increases from0.28 to 0.59.

Models 4 and 5 in Table 2 report the results forsearch costs. As model 5 indicates, average inter-

6 To control for this potential selection bias, we alsoran a two-part logit model (Manning, Duan, & Rogers,1987). In the first part, the dependent variable was whetherteams sought knowledge. From this model, we calculatedthe predicted probability of each of the 121 teams seekingknowledge, p (seek knowledge), and then entered this vari-able in the second part of the models for search and transfercosts. The results of this test revealed no selection bias.

7 For a sample to be considered large enough to showexpected effects, the statistical power of the analysisshould be at least 0.8 (Aiken & West, 1991). In our anal-yses, statistical power was at a minimum 0.9 in all of thetested models. Furthermore, the effect sizes were greaterthan 0.35 and thus big enough to be considered largeeffects (Cohen, 1988). We therefore do not believe thatthere is any problem with the small samples for searchand transfer costs.

8 We computed the lines in Figures 2a and 2b byincluding the coefficient for the intercept and evaluatingall other variables in a model at their mean values. On thebasis of model 2, the two lines in Figure 2a are computedas follows:

Prob. � expy/(1 � expy�6.34 � within-team network density).

Prob. � expy/(1 � expy�0.32 � intersubsidiary network size).


subsidiary relation strength has a positive and sig-nificant effect on search costs, supporting Hypoth-esis 5. The variables for the competition effects onsearch costs were also entered in model 5, and theyshow that the extent of competition perceived bydirect contacts has a positive and significant effect.That is, if subsidiaries with which a team had es-tablished relations perceived the relationships tobe competitive, the team’s search costs increased.These findings support Hypothesis 6.

It is also interesting to note that the two powerfulpredictors of whether a team sought knowledge—

within-team network density and intersubsidiarynetwork size—are not significant predictors of searchcosts. Those factors that explain whether a teamsought knowledge do not explain the costs of search.

The variables pertaining to Hypotheses 7 and 8,which concern transfer costs, were entered in mod-els 6 and 7 in Table 2. The coefficient for theproviding subsidiaries’ perceived level of competi-tion (providers’ perceived competition) is signifi-cant and positive, indicating that a team’s transfercosts increased to the extent that the providingsubsidiaries saw the subsidiary in which the team

TABLE 1Descriptive Statistics for Sought Knowledge, Search, and Transer Costsa

Variables Mean s.d. Minimum Maximum 1 2 3 4 5 6 7

1. Sought knowledge 0.45 0.50 0.00 1.002. Search costsb 4.39 5.17 0.10 15.003. Transfer costsc 7.94 9.90 0.10 40.00 .70***4. Existing ware 0.45 0.31 0.00 1.00 �.40*** �.05 �.165. Project novelty 3.12 1.55 1.00 7.00 .20* .05 .06 �.40***6. Budget 1.85 4.66 0.09 45.20 .20* .39** .50*** �.17† �.017. Project total

engineering-months0.80 2.31 0.10 24.00 .20* .32* .50*** �.20* .01 .90***

8. Team tenure 14.13 5.47 0.10 31.00 �.04 �.26† �.15 .01 �.04 �.03 �.18*9. Team size 5.40 2.39 2.00 11.00 .07 .16 �.04 �.03 �.03 .10 .05

10. Amount of knowledgeobtainedb

25.00 53.4 0.43 306.00 .41** .50*** �.25† .02 .80*** .90***

11. Absorptive capacityb 4.37 1.87 1.00 7.00 �.22 �.01 �.01 �.22 .10 .1912. Tacitness of

transferredknowledgec, d

3.26 1.16 1.33 6.00 .33* .17 .11 �.15 .08 .07

13. Within-team networkdensity

0.09 0.14 0.00 1.00 �.20* .06 �.19 .07 �.14 �.07 �.14

14. Average within-teamrelation strength

2.55 1.46 1.00 5.50 �.17† .11 �.17 �.03 �.10 �.05 �.04

15. Intersubsidiarynetwork size

5.60 2.95 1.00 14.00 .27** .17 .17 .09 .21* .26** .18†

16. Averageintersubsidiaryrelation strength

4.09 1.26 2.18 7.00 .06 .09 �.04 �.30*** �.04 �.05 �.05

17. Direct contacts’average perceivedcompetitionb

3.06 1.21 1.00 5.00 .26† .18 .27* �.28* .01 �.06

18. Focal subsidiaries’average perceivedcompetition

2.58 .97 1.00 6.00 .09 �.06 .02 �.09 �.10 �.15 �.11

19. Providers’ perceivedcompetitionc

1.95 1.79 1.00 6.30 .02 .14 .12 �.18 �.01 .09

20. Dyadic transferrelation strengthc, d

3.84 2.38 0.00 7.00 �.07 �.10 .02 �.28* .10 .16

21. Tacitness � transferrelation strengthc

11.57 9.92 0.00 39.69 .12 �.25† �.26† �.03 �.01 .03

a Unless otherwise stated, n � 121.b n � 55 as only 55 teams entered the search process.c n � 50 as five observations with missing information on transfer costs are omitted.d Mean-deviated (precentered mean, s.d., minimum, and maximum are shown).

† p � .10* p � .05

** p � .01*** p � .001


was located as a competitor. This result supportsHypothesis 7. Figure 2b shows the magnitude of theeffect: for example, when providers’ perceivedcompetition increases from 1 to 4 on the seven-point scale, a focal team’s transfer costs go up by4.8 engineering-months, which is a substantialjump, given that these teams spent an average 8engineering-months on transfer.

Interestingly, as model 7 shows, the other twocompetition variables are not significant in thetransfer models. Whether a focal subsidiary per-ceives relationships to be competitive (focal sub-

sidiary’s average perceived competition) has noimpact on transfer costs, nor does the level ofcompetition perceived by the total set of directcontacts in a team’s intersubsidiary network.Thus, the competition variables affect the threedependent variables in different ways: while theyhave no effect on whether a team sought knowl-edge, only direct contacts’ perceptions of compe-tition increased search costs, whereas only pro-viders’ perceptions of competition increasedtransfer costs.

Model 7 in Table 2 also reveals that the interac-

8 9 10 11 12 13 14 15 16 17 18 19 20

�.30**�.10 .13

.29† �.08 .10�.11 .01 .18 �.18

.28** �.30** .01 �.01 �.13

�.18† .27** .06 �.01 �.04 .40***

�.21* .17† .20 �.18 �.10 �.08 �.11

.06 �.05 �.18 .14 .04 .07 .27** �.25**

.11 �.07 �.09 �.09 .27* .03 �.17† .04 �.08

.02 .10 �.19 .14 .07 .02 .08 .15 .04 .20

�.08 .11 .14 .30* .11 �.36* �.11 �.15 �.04 �.13 .50***

.16 .25† .14 .26† .17 �.13 .14 �.15 .20 �.29* .15 .40***

�.23 .02 .01 .18 .05 .30* .10 �.18 .35** �.10 �.07 �.06 �.15


TA

BL

E2

Res

ult

sfr

omR

egre

ssio

nA

nal

ysis

a

Dep

end

ent

Var

iabl

es

Mod

el1:

Sou

ght

Kn

owle

dge

Mod

el2:

Sou

ght

Kn

owle

dge

Mod

el3:

Sou

ght

Kn

owle

dge

Mod

el4:

Sea

rch

Cos

ts

Mod

el5:

Sea

rch

Cos

ts

Mod

el6:

Tra

nsf

erC

osts

Mod

el7:

Tra

nsf

erC

osts

Inte

rcep

t0.

11(1

.21)

0.52

(2.7

6)0.

03(2

.81)

6.37

(3.4

1)�

13.1

0(1

0.3)

7.27

(5.1

4)�

12.0

6(1

0.97

)

Con

trol

vari

able

sE

xist

ing

war

e�

2.76

(1.0

4**)

�3.

97(1

.50*

*)�

4.16

(1.5

3**)

0.62

(3.2

0)3.

16(4

.56)

�1.

28(3

.48)

�8.

17(7

.22)

Pro

ject

nov

elty

0.02

(0.1

5)�

0.18

(0.2

1)�

0.13

(0.2

1)0.

28(0

.55)

0.93

(0.4

2*)

0.30

(1.1

5)0.

71(1

.16)

Bu

dge

t0.

13(0

.00)

�0.

03(0

.00)

�0.

02(0

.00)

0.54

(0.0

0*)

0.35

(0.0

0)0.

71(0

.00)

0.41

(0.0

0)P

roje

ctto

tal

engi

nee

rin

g-m

onth

s0.

43(0

.01)

0.62

(0.0

0)0.

74(0

.00)

�0.

77(0

.01)

�0.

27(0

.01)

�1.

82(0

.02)

0.10

(0.0

1)T

eam

ten

ure

0.04

(0.0

1)0.

08(0

.04†

)0.

10(0

.04†

)�

0.19

(0.1

3)�

0.23

(0.1

4†)

�0.

14(0

.15)

�0.

09(0

.20)

Tea

msi

ze�

0.03

(0.1

1)�

0.06

(0.1

6)�

0.04

(0.1

6)0.

09(0

.25)

0.34

(0.4

1)�

0.25

(0.3

2)�

0.73

(0.6

9)A

mou

nt

know

led

ge0.

02(0

.04)

0.03

(0.0

4)0.

13(0

.16)

0.03

(0.1

1)A

bsor

pti

veca

pac

ity

�0.

46(0

.43)

�0.

39(0

.49)

0.13

(0.4

8)1.

10(0

.73)

Tac

itn

ess

oftr

ansf

erre

dkn

owle

dge

b1.

59(1

.10)

Wit

hin

-tea

mn

etw

ork

Wit

hin

-tea

mn

etw

ork

den

sity

�6.

34(3

.13*

)�

6.34

(3.1

3*)

2.52

(10.

97)

�13

.39

(16.

70)

Ave

rage

wit

hin

-tea

mre

lati

onst

ren

gth

�0.

47(0

.29†

)0.

31(

1.70

)0.

48(1

.20)

Ave

rage

wit

hin

-tea

mcl

osen

ess

0.11

(0.2

0)A

vera

gew

ith

in-t

eam

freq

uen

cy�

0.83

(0.3

0**)

Inte

rsu

bsid

iary

net

wor

kIn

ters

ubs

idia

ryn

etw

ork

size

0.32

(0.1

0***

)0.

32(0

.10*

**)

0.21

(0.2

8)0.

91(0

.61)

Ave

rage

inte

rsu

bsid

iary

rela

tion

stre

ngt

h0.

08(0

.23)

0.05

(0.2

5)1.

60(0

.75*

)1.

71(1

.58)

Dir

ect

con

tact

s’av

erag

ep

erce

ived

com

pet

itio

n1.

86(0

.43*

**)

1.80

(1.4

3)F

ocal

subs

idia

ries

’av

erag

ep

erce

ived

com

pet

itio

n0.

17(0

.30)

0.21

(0.3

0)�

0.06

(0.3

8)�

0.51

(1.3

3)

Tra

nsf

ern

etw

ork

Pro

vid

ers’

per

ceiv

edco

mp

etit

ion

1.70

(0.8

3*)

Dya

dic

tran

sfer

rela

tion

stre

ngt

hb

�1.

78(0

.67*

)T

acit

nes

s�

tran

sfer

rela

tion

stre

ngt

h�

1.48

(0.5

0**)

Typ

eof

anal

ysis

Log

itL

ogit

Log

itO

LS

OL

SO

LS

OL

Sn

121

121

121

5555

50c

50c

Lik

elih

ood

-rat

ioch

i-sq

uar

ete

std/F

22.3

0***

5.01

*12

.37*

**23

0.41

***

5.03

**69

.49*

**P

seu

do-

R2/R

20.

180.

310.

340.

280.

460.

290.

61

aD

um

my

vari

able

sd

enot

ing

mis

sin

gin

form

atio

non

tota

len

gin

eeri

ng-

mon

ths

and

team

ten

ure

are

not

show

n.

All

mod

els

show

nw

ith

robu

stst

and

ard

erro

rs.

bM

ean

-dev

iate

d.

cF

ive

obse

rvat

ion

sar

eom

itte

dow

ing

tom

issi

ng

info

rmat

ion

ontr

ansf

erco

sts.

dC

omp

ared

wit

hp

revi

ous

mod

el.

†p

�.1

0*

p�

.05

**p

�.0

1**

*p

�.0

01T

wo-

tail

edte

sts

for

vari

able

coef

fici

ents

.

tion effect including the tacitness and dyadic trans-fer relation strength variables is negative and sig-nificant, supporting Hypothesis 8. As the tacitnessof the knowledge being transferred increased, anincrease in the strength of the established relationbetween a team and the providing subsidiaries re-duced the number of engineering-months spent ontransfer. As Figure 2b shows, having strong estab-lished relations was especially beneficial whenhighly tacit knowledge was involved.9

Finally, model 7 in Table 2 also shows that prop-erties of the within-team network that affectedwhether teams sought knowledge (that is, the den-sity and strength of relations) and the properties ofthe intersubsidiary network that affected searchcosts (i.e., contacts’ perceived competition and

9 The lines in Figure 2b that involve tacitness werecomputed with results from model 7 in Table 2, evaluat-ing the variables except dyadic transfer relation strength

and the tacitness of transferred knowledge at their meanvalues: transfer costs � �1.78 � dyadic transfer relationstrength � 1.59 � tacitness � 1.48 � tacitness � transferrelation strength, where “tacit low” means that tacitnessis –0.5 standard deviations below the mean (i.e., �.74),and “tacit high” means that tacitness is �1 s.d. above themean (i.e., �1.47).

FIGURE 2aPlot of Effects for Sought Knowledge

FIGURE 2bPlot of Effects for Transfer Costs


strength) have no significant effects on transfercosts.

DISCUSSION AND CONCLUSION

The main finding of this study is that teams’different subsets of networks to a large extent af-fected the outcomes of the three phases of decidingto seek knowledge, incurring search costs, and in-curring transfer costs in different ways. The within-team network variables denoting the density andfrequency of relations reduced the probability ofseeking knowledge across subsidiaries but did notaffect search and transfer costs. The intersubsidiarynetwork variable capturing network size increasedthe probability of seeking knowledge, whereas in-tersubsidiary relation strength and degree of per-ceived competition increased search costs but didnot affect transfer costs. And the variable denotingthe degree of perceived competition in a team’stransfer network increased transfer costs, whereasrelation strength in a transfer network reducedtransfer costs when transfers involved highly tacitknowledge. These effects would have been maskedhad we not distinguished between the three phasesof knowledge sharing and decomposed teams’ net-works into three distinct subsets.

Before we discuss the implications of these find-ings, it is worth acknowledging some limitations ofour study. Because we studied only one company,we cannot claim that these results would necessar-ily hold for all organizations. In particular, thecompany we studied relied quite extensively oninformal relations among employees to accomplishwork, implying that relational properties played anespecially important role in knowledge sharinghere. Other mechanisms for knowledge sharing, in-cluding reliance on headquarters’ coordination andtransfer prices between subsidiaries, played rela-tively small roles in this company, yet they may beimportant mechanisms in different settings.

The company we studied was also characterizedby a fairly cooperative culture in which new prod-uct developers frequently helped one another.Teams in other companies that actively encouragecompetition may experience outright rejection ofrequests for knowledge and strong resistance totransfers, which will raise search and transfer costsover what the teams in our setting incurred. Thisbias may be in a conservative direction, however:even though we studied an organization with afairly cooperative culture, we still found strong ef-fects for the competition variables, which may havean even more pronounced effect in organizationswith highly competitive cultures.

Although these limitations suggest that some

caution is needed in interpreting the findings, ourstudy provides new insights into the process ofknowledge sharing in organizations.

Emphasizing the Role of Competition inKnowledge Sharing

An important finding in our study is the positiveand significant effect of perceived competition onboth search and transfer costs. Although scholarshave noted that employees often hoard knowledge(e.g., Davenport & Prusak, 1998), the mechanismscontributing to such reduced motivation to provideknowledge have not been elucidated. Our findingssuggest that hoarding may in part be a result ofperceived competition between users and potentialproviders of knowledge and that such competitionis likely to vary across pairs of providers and users.The findings suggest that employees are not neces-sarily predisposed to hoarding knowledge; rather,their tendency to do so can be explained by inter-unit patterns of competition in an organization.

Furthermore, our results indicate that it matterswho perceives competition: for search and transfercosts, it is the perceptions of the direct contacts ina focal actor’s network and not the perception ofthe focal actor that matter. This pattern makes in-tuitive sense: If direct contacts perceive the search-ing actor to be a competitor, they are likely to beless cooperative and are more likely to hoard theirknowledge or even refuse to cooperate at all, irre-spective of the perceptions of the focal actor.

One implication of our findings is that researchon knowledge sharing needs to address the possi-bly negative aspects of competition in interunitrelations (cf. Tsai, 2002). This implication opens upseveral avenues for new research. For example,while we treated the extent of perceived competi-tion as an exogenous variable, it may be endoge-nous to the process of knowledge sharing. Twounits that share knowledge may, over time, begin todevelop similar products and services for similarmarkets because they increasingly rely on the sameknowledge base and ideas. This convergence, inturn, may make the relationship more competitive,triggering a process of avoiding competitive rela-tions and forming new interunit relations to ex-change knowledge (cf. Galunic & Eisenhardt, 2001).Knowledge sharing and the formation of, and pos-sibly dissolution of, interunit relations thus co-evolve in such a competitive landscape (Koza &Lewin, 1999). Although our data were cross-sec-tional and hence we could not disentangle theseintertemporal processes, in subsequent studies re-searchers could use longitudinal data to analyzewhen knowledge sharing increases perceived com-


petition in interunit relations and how perceivedcompetition affects the formation and dissolutionof such relations.

Multiple Networks, Multiple Phases ofKnowledge Sharing

The most important implication of our findingsis that research on knowledge sharing needs tofully incorporate the phase level of knowledgesharing in an organization and the subset level ofsocial networks in order to advance a robust theoryof knowledge sharing. We are not the first to con-ceive of knowledge sharing as consisting of multi-ple phases, yet our approach here represents abroader view than other perspectives. For example,although Szulanski (1996) considered various stepsof knowledge transfers, including initiation, imple-mentation, ramp-up, and integration of transfers,we go beyond transfer by also including the “frontend” of knowledge sharing—the decision to seekknowledge in the first place, and the searchprocess.

Despite the attention researchers have paid todifferent subsets of networks, such as within-teamand extrateam subsets, little research compares dif-ferent subsets of networks and analyzes their dif-ferent impacts on various phases of knowledgesharing. Our results suggest this scarcity has im-peded further insights into how and why actorsshare knowledge across subunits in an organiza-tion. To illustrate, consider existing work on thestrength of relations and knowledge transfers. Re-searchers have shown that transfers of knowl-edge—especially tacit knowledge—between a pro-viding and receiving unit is facilitated to the extentthat the two parties have a strong established rela-tion (Hansen, 1999; Szulanski, 1996; Uzzi, 1997).From this research one might conclude that highaverage relationship strength will improve knowl-edge sharing in an organization. Our results indi-cate that this hypothesis is incorrect and that itdepends on the particular phase of the knowledge-sharing process being studied and which subsets ofnetworks are being considered. Specifically, ourresults revealed that high average relationshipstrength in within-team networks led to fewerknowledge seeking attempts across subsidiariesand, thus, fewer intersubsidiary knowledge-sharingevents. High average relation strength in the inter-subsidiary network was also linked to highersearch costs than low average relation strength,whereas our results validated the prediction thatstrong relations in a transfer situation facilitate thetransfer of tacit knowledge. These findings providea more nuanced perspective on the effects of rela-

tion strength on knowledge sharing that would notemerge from studying the transfer process only(measured by the number of engineering monthsspent on search). These findings also validate andrefine results reported by Hansen (1999): while thatstudy advanced the argument that weak interunitties facilitate search but undermine the transfer oftacit knowledge, it had no direct measures of searchand transfer outcomes but only an analysis ofproject task outcomes in the form of completiontime. Results reported here show that weak ties(i.e., low average relation strength) benefited searchby reducing search costs, while strong ties helpedtransfer by reducing transfer costs (that is, engineermonths spent on transfer), verifying Hansen’s(1999) earlier argument by measuring search andtransfer costs separately.

Our results also showed that the network size hasdifferent impacts at the within-team and intersub-sidiary levels of analysis. Specifically, whereas thedensity of relations inside a team decreased thelikelihood that it would seek knowledge, the size ofa team’s network across subsidiaries increased thislikelihood. That is, established relations at theteam level countervailed established relations atthe subsidiary level. This finding casts new light onresearch that has examined the effects of relationalproperties on the extent of interunit knowledgesharing and communication (Gupta & Govindara-jan, 2000; Szulanski, 1996; Zander & Kogut, 1995).It is not necessarily the absolute number of estab-lished interunit relations that explains why organ-ization members share knowledge but, rather, therelative number of within-team and interunit rela-tions. For example, even though some teams in ourdata set had very high numbers of intersubsidiaryrelations (e.g., nine, which is more than one stan-dard deviation above the mean level), a very highwithin-team network density (e.g., 0.60) would off-set the positive effect of such a high number ofintersubsidiary relations on the probability of seek-ing knowledge across subsidiaries.

This emphasis on the relative numbers of within-team and interunit relations raises the issue of howthey relate to each other. Because we did not ana-lyze the formation of relations, we cannot predictwhether the existence of within-team relationscrowds out the interunit ones (and vice versa), al-though this may not have been the case in our dataset, as the correlation between within-team net-work density and intersubsidiary network size waslow (–.08). Given that the time and energy that canbe devoted to building and maintaining relation-ships may be constrained, however, we can specu-late that within-team relations may come at theexpense of the development and maintenance of


interunit relations. If managers see developing andmaintaining relations among members within aunit, such as a team, as a good practice and encour-age it (cf. Ghoshal, Korine, & Szulanski, 1994), sucha management practice may inadvertently nega-tively affect knowledge sharing across the organi-zation to the extent that it leads to fewer interunitrelations and thus fewer knowledge-sharing events(cf. Ghoshal & Bartlett, 1990).

In conclusion, our study shows that the problemof knowledge sharing can be usefully conceptual-ized as consisting of three phases that to a largeextent are affected by an actor’s different subsets ofsocial networks in an organization. As the variouseffects of the degree of perceived competition, re-lation strength, and network size illustrate, disen-tangling interunit knowledge sharing into the threedistinct phases of deciding to seek knowledge,searching, and transferring can yield significantnew insights into how network variables govern theoccurrence and effectiveness of knowledge sharingin organizations. By adopting a multiple networkperspective on knowledge sharing, researchers cangain a fine-grained understanding of how differentsubsets of an actor’s relations in an organizationcontribute to knowledge-sharing behaviors andoutcomes.

REFERENCES

Aiken, L. S., & West, S. G. 1991. Multiple regression:Testing and interpreting interactions. ThousandOaks, CA: Sage.

Ancona, D. G., & Caldwell, D. F. 1992. Bridging theboundary: External activity and performance in or-ganizational teams. Administrative Science Quar-terly, 37: 634–665.

Austin, J. R. 2003. Transactive memory in organizationalgroups: The effects of content, consensus, specializa-tion, and accuracy on group performance. Journal ofApplied Psychology, 88: 866–878.

Borgatti, S. P., & Cross, R. 2003. A relational view ofinformation seeking and learning in social networks.Management Science, 49: 432–445.

Brewer, M. B. 1979. Ingroup bias in the minimal inter-group situation: A cognitive motivational analysis.Psychological Bulletin, 86: 307–324.

Burt, R. S. 1992. Structural holes: The social structureof competition. Cambridge, MA: Harvard UniversityPress.

Cohen, J. 1988. Statistical power analysis for the be-havioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum.

Cohen, W., & Levinthal, D. 1990. Absorptive capacity: Anew perspective on learning and innovation. Ad-ministrative Science Quarterly, 35: 128–152.

Cummings, J. N. 2004. Work groups, structural diversity,and knowledge sharing in a global organization.Management Science, 50: 352–364.

Davenport, T., & Prusak, L. 1998. Working knowledge:How organizations manage what they know. Bos-ton: Harvard Business School Press.

Eisenhardt, K. M., & Santos, F. M. 2002. Knowledge-based view: A new theory of strategy? In A. Petti-grew, H. Thomas, & R. Whittington (Eds.), Hand-book of strategy and management: 139–164.London: Sage.

Galunic, D. C., & Eisenhardt K. M. 2001. Architecturalinnovation and modular corporate forms. Academyof Management Journal, 44: 1129–1149.

Gatignon, H., Tushman, M. L., Smith, W., & Anderson, P.2002. A structural approach to assessing innovation:Construct development of innovation locus, type,and characteristics. Management Science, 48:1103–1122.

Ghoshal, S., & Bartlett, C. A. 1990. The multinationalcorporation as an interorganizational network.Academy of Management Review, 15: 603–625.

Ghoshal, S., Korine, H., & Szulanski, G. 1994. Interunitcommunication in multinational corporations. Man-agement Science, 40: 96–110.

Granovetter, M. S. 1973. The strength of weak ties. Amer-ican Journal of Sociology, 6: 1360–1380.

Gulati, R. 1995. Does familiarity breed trust? The impli-cations of repeated ties for contractual choice inalliances. Academy of Management Journal, 38:85–112.

Gupta, A. K., & Govindarajan, V. 2000. Knowledge flowswithin multinational corporations. Strategic Man-agement Journal, 21: 473–496.

Hansen, M. T. 1999. The search-transfer problem: Therole of weak ties in sharing knowledge across organ-ization subunits. Administrative Science Quarterly,44: 82–111.

Hansen, M. T. 2002. Knowledge networks: Explainingeffective knowledge sharing in multiunit companies.Organization Science, 13: 232–248.

Hansen, M. T., Podolny, J. M. & Pfeffer, J. 2001. So manyties, so little time: A task contingency perspective onthe value of corporate social capital in organizations.In K. S. Cook & J. Hagan (Eds.), Research in thesociology of organizations, vol. 18: 21–57. Green-wich, CT: JAI Press.

Hayes, R. H., & Clark, K. B. 1985. Exploring the sourcesof productivity differences at the factory level.New York: Wiley.

Katz, R., & Allen, T. J. 1988. Investigating the not in-vented here (NIH) syndrome: A look at the perfor-mance, tenure, and communication patterns of 50R&D project groups. In M. L. Tushman & W. L. Moore(Eds.), Readings in the management of innovation


(2nd ed.): 293–309. New York: Ballinger/Harper &Row.

Koza, M. P., & Lewin, A. Y. 1999. The co-evolution ofnetwork alliances: A longitudinal analysis of an in-ternational professional service network. Organiza-tion Science, 10: 638–653.

Levin, D., & Cross, R. 2004. The strength of weak ties youcan trust: The mediating role of trust in effectiveknowledge transfer. Management Science, 50:1477–1490.

Manning, W. G., Duan, N., & Rogers, W. H. 1987. MonteCarlo evidence on the choice between sample selec-tion and two-part models. Journal of Econometrics,35: 59–82.

Moulton, B. R. 1986. Random group effects and the pre-cision of regression estimates. Journal of Economet-rics, 32: 385–397.

Nohria, N., & Ghoshal, S. 1997. The differentiated net-work: Organizing multinational corporations forvalue creation. San Francisco: Jossey-Bass.

Oakes, P. J., Haslam, S. A., Morrison, B., & Grace, D. 1995.Becoming an in-group: Reexamining the impact offamiliarity on perceptions of group homogeneity.Social Psychology Quarterly, 58: 52–61.

Podolny, J. M., & Baron, J. N. 1997. Relationships andresources: Social networks and mobility in the work-place. American Sociological Review, 62: 673–693.

Podsakoff, P. M., & Organ, D. W. 1986. Self-reports inorganizational research: Problems and prospects.Journal of Management, 12: 531–544.

Reagans, R., & McEvily, B. 2003. Network structure andknowledge transfer: The effects of cohesion andrange. Administrative Science Quarterly, 48: 240–267.

Rogers, W. 1993. Regression standard errors in clusteredsamples. Stata Technical Bulletin, 13: 19–23.

Schulz, M. 2001. The uncertain relevance of newness:Organizational learning and knowledge flows. Acad-emy of Management Journal, 44: 661–681.

Schulz, M. 2003. Pathways of relevance: Exploring in-flows of knowledge into subunits of multinationalcorporations. Organization Science, 14: 440–459.

Sorenson, O. 2000. The effect of population level learn-ing on market entry: The American automobile in-dustry. Social Science Research, 29: 307–326.

Szulanski, G. 1996. Exploring internal stickiness: Imped-iments to the transfer of best practice within thefirm. Strategic Management Journal 17(winter spe-cial issue): 27–43.

Tajfel, H., & Turner, J. C. 1986. The social identity theory

of intergroup behavior. In S. Worchel & W. G. Austin(Eds.), Psychology of intergroup relations (2nd ed.):7–24. Nelson Hall.

Tsai, W. 2001. Knowledge transfer in intraorganizationalnetworks: Effects of network position and absorptivecapacity on business unit innovation and perfor-mance. Academy of Management Journal, 44: 996–1004.

Tsai, W. 2002. Social structure of “coopetition” within amultiunit organization: Coordination, competition,and intraorganizational knowledge sharing. Organi-zation Science, 13(2): 179–190.

Uzzi, B. 1997. Social structure and competition in inter-firm networks: The paradox of embeddedness. Ad-ministrative Science Quarterly, 42: 35–67.

Wasserman, S., & Faust, K. 1994. Social network anal-ysis: Methods and applications. New York: Cam-bridge University Press.

Wilder, D. A. 1978. Reduction of intergroup discrimina-tion through individuation of the out-group. Journalof Personality and Social Psychology, 36: 1361–1374.

Zander, U., & Kogut, B. 1995. Knowledge and the speedof transfer and imitation of organizational capabili-ties: An empirical test. Organization Science, 6:76–92.

Morten T. Hansen ([email protected]) is a pro-fessor at INSEAD, France. He was formerly an associateprofessor at Harvard Business School and received hisPh.D. from the Graduate School of Business at StanfordUniversity. His research has been concerned with therole of knowledge in organizations and, specifically, withthe problems of searching for and transferring knowledgeacross dispersed subunits in an organization.

Marie Louise Mors is a postdoctoral fellow at the MITSloan School of Management. She received her Ph.D.from INSEAD. Her research interests include social net-works and processes of knowledge creation and sharingin global organizations.

Bjørn Løvås was an assistant professor of strategic andinternational management at the London BusinessSchool. He received his Ph.D. from INSEAD. His researchand teaching focused on strategies for innovation andgrowth in knowledge and talent intensive industries,with emphasis on how firms manage and compete on theinformal relationships of their members.


Documents

KNOWLEDGE SHARING IN ORGANIZATIONS: MULTIPLE …faculty.london.edu/lmors/assets/documents/Hansen_Mors_Lovas_05.pdf · KNOWLEDGE SHARING IN ORGANIZATIONS: MULTIPLE NETWORKS, MULTIPLE