娀 Academy of Management Journal

2005, Vol. 48, No. 5, 776–793.

KNOWLEDGE SHARING IN ORGANIZATIONS:

MULTIPLE NETWORKS, MULTIPLE PHASES
MORTEN T. HANSEN
INSEAD

MARIE LOUISE MORS

Massachusetts Institute of Technology

BJØRN LØVÅS
London Business School

Different subsets of social networks may explain knowledge sharing outcomes in

different ways. One subset may counteract another subset, and one subset may explain
one outcome but not another. We found support for these arguments in an analysis of
a sample of 121 new-product development teams. Within-team and interunit networks
had different effects on the outcomes of three knowledge-sharing phases: deciding
whether to seek knowledge across subunits, search costs, and costs of transfers. These
results suggest that research on knowledge sharing can be advanced by studying how
multiple networks affect various phases of knowledge sharing.

What determines the occurrence and effective- have not empirically disentangled the two phases
ness of knowledge sharing in an organization? of search and transfer (Hansen, 1999). Scholars
Scholars have examined this question from differ- therefore do not know much about the extent to
ent viewpoints, focusing on the problem of trans- which different factors explain search and transfer.
ferring tacit and complex knowledge across organ- Moreover, in the few studies on search for knowl-
ization subunits (e.g., Zander & Kogut, 1995), the edge, researchers have tended not to distinguish
nature of informal relationships between two par- between the decision to seek knowledge and the
ties to a transfer (e.g., Gupta & Govindarajan, 2000; ensuing search process (e.g., Borgatti & Cross, 2003;
Reagans & McEvily, 2003; Tsai, 2002), and the Hansen, 1999).
problem of searching for knowledge (e.g., Ancona & These shortcomings in existing research are
Caldwell, 1992). This growing body of literature problematic. Existing findings may be biased or
has shed much light on the various problems un-
incomplete to the extent that explanatory proper-
derlying knowledge sharing in organizations. Yet,
ties that have a positive or negative effect on out-
because the overarching process of knowledge
comes in one phase, such as transfer, may have the
sharing has been shown to consist of multiple
opposite or no effect on outcomes in other phases.
phases— each with their own associated challeng-
es— existing research on this question is quite dis- There is therefore a need for research that explores
parate and has not shed much light on whether the extent to which different properties explain
variables that explain one phase of knowledge shar- outcomes associated with the various phases of
ing also explain other phases (Eisenhardt & Santos, knowledge sharing.
2002). To address this gap in extant research on knowl-
In particular, while some studies have analyzed edge sharing, we pose this research question: Do
the transfer of knowledge from one point to another properties of social networks explain the outcomes
(e.g., Gupta & Govindarajan, 2000; Szulanski, 1996; of the three phases— deciding to seek knowledge,
Zander & Kogut, 1995), they have excluded the searching for knowledge, and transferring knowl-
logically prior phase of searching for knowledge or edge—in different ways? By social networks, we
refer to subsets of established informal relations
that exist within teams and across subunits in an
We thank three anonymous reviewers, Associate Edi- organization. We delineate the application of this
tor Marshall Schminke, Martine Haas, Michael Jacobides,
Freek Vermeulen, Olav Sorenson, Julian Birkinshaw, and
multiple network approach to an analysis of one
Phanish Puranam. type of task units, new-product development
Bjørn Løvås died in an accident in July 2005. His teams, and examine the impacts of teams’ subsets
coauthors dedicate this article to his memory. of social networks on whether they seek knowledge
776
2005 Hansen, Mors, and Løvås 777

across subunits, their interunit search costs, and nisms for why this may be the case. First, as prior
the costs of knowledge transfer. research on social networks and communication
patterns in product development projects has
shown, established informal relations tend to chan-
MULTIPLE NETWORKS, MULTIPLE PHASES
nel actors’ time and energy in the direction of es-
Scholars adopting a relational approach to tablished contacts—actors interact with those
knowledge sharing have mainly focused on the whom they have interacted with before (e.g., Gu-
characteristics of established informal relations lati, 1995; Katz & Allen, 1988). This channeling
that facilitate or impede the sharing of knowledge mechanism implies that different subsets of net-
in an organization. This body of research has, how- works may channel teams’ time and energy in dif-
ever, focused on quite different subsets of relations. ferent directions. Second, building on prior re-
Some research has focused on analyzing informal search that has shown that an actor’s different
relations within teams and organization subunits social relations have different utilities (e.g.,
(e.g., Levin & Cross, 2004), whereas other studies Podolny & Baron, 1997), we argue that the three
have focused on teams’ and individuals’ external different subsets of relations serve different pur-
informal relations (Ancona & Caldwell, 1992; Cum- poses. Intersubsidiary relations are more beneficial
mings, 2004) or on relations between subunits in an for search across subsidiaries than are within-team
organization (e.g., Gupta & Govindarajan, 2000). relations, whereas dyadic transfer relations are
Furthermore, while some research has analyzed more useful for transfer than a team’s total set of
properties of a subunit’s total set of interunit rela- intersubsidiary relations. This utility mechanism
tions (e.g., Hansen, 1999; Tsai, 2001), other studies implies that properties of the three networks that
have focused exclusively on the properties of dy- we analyze should have different impacts on the
adic relations involving one providing and one re- outcomes of the three phases.
ceiving unit in a transfer event (e.g., Gupta & Gov- On the basis of these two overarching theoretical
indarajan, 2000; Szulanski, 1996). mechanisms, we develop a set of hypotheses,
Apart from a few studies that have begun to ad- which are depicted in Figure 1. We focus on three
dress multiple subsets of relations (e.g., Nohria & relational variables that have received attention in
Ghoshal, 1997; Schulz, 2001), this body of research knowledge-sharing research: the extent of a net-
has not carefully analyzed the potentially different work as given by its size (e.g., Hansen, 2002), the
roles of various subsets of relations, yielding an strength of relations (e.g., Hansen, 1999), and the
incomplete understanding of what particular sub- degree of perceived competition inherent in rela-
sets of relations affect the decision to seek knowl- tions (e.g., Tsai, 2002).
edge across subunits, search costs, and costs of
transfer. To address this shortcoming, we develop a
multiple network perspective that addresses three Deciding Whether to Seek Knowledge across
subsets of networks that exist prior to the start of a Subsidiaries
focal project: established relations between mem-
bers of a team located within a focal subsidiary (a In the first phase, teams may decide to seek
within-team network); a team’s total set of estab- knowledge in other subsidiaries. Assuming that a
lished relations with those in other subsidiaries, team’s decision to seek knowledge results in at-
irrespective of whether it transfers knowledge from tempts to seek knowledge, an outcome of this phase
those contacts (an intersubsidiary network); and a is thus whether a focal team in a focal subsidiary
team’s dyadic relations, involving only subsidiaries contacts other subsidiaries to inquire about poten-
providing it with knowledge (a transfer network). tially useful knowledge for a project. Irrespective of
The latter two subsets both refer to direct contacts teams’ varying needs to seek knowledge from sub-
or relations that a team in a focal subsidiary has sidiaries other than their own, a team’s decision to
with other subsidiaries. The difference between seek knowledge across subsidiaries may be affected
these two subsets is that an intersubsidiary network by established relations that exist both within
includes all direct contacts that a team has, teams and across subsidiaries.1
whereas a transfer network includes only direct
contacts with subsidiaries from which it is trans-
ferring knowledge. 1
We assume that project development tasks are not
We argue that these three informal networks af- purposefully decomposed into subtasks whereby other
fect a team’s decision to seek knowledge, search subsidiaries provide components to the focal project
costs, and costs of transfers in different ways, and and subsidiary, thereby preordaining intersubsidiary
we explore two overarching theoretical mecha- interactions.
778 Academy of Management Journal October

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

Within-team network. Product developers may knowledge held by others outside their group
have formed enduring relations among themselves, (Hayes & Clark, 1985; Katz & Allen, 1988). These
including friendship and advice-seeking informal biases often develop when group members have
relations, that develop over time and exist prior to spent considerable time interacting with one an-
the start of a particular product development other, a scenario that restricts the inflow of new
project (Podolny & Baron, 1997). When developers viewpoints and differences and reinforces com-
join a common project team, these established re- monly held beliefs (Oakes, Haslam, Morrison, &
lations may channel their time and energy toward Grace, 1995; Wilder, 1978).
the team: when confronted with new project-spe- Second, established relations may be associated
cific problems, they may seek to solve them by with a team’s inward-looking absorptive capacity
interacting and sharing knowledge with team mem- (Cohen & Levinthal, 1990). Product developers who
bers. Given that teams typically have a finite num- interact regularly with one another may, over time,
ber of problems to solve, this preference is likely to assimilate one another’s knowledge, thus develop-
reduce the need to seek knowledge outside the teams, ing a common knowledge base and a common spe-
including knowledge resident in other subsidiaries. cialized set of terminologies. Through established
Existing theory suggests three reasons why this relations, they may also develop a capacity for mu-
channeling may occur. First, within-team relations tual problem solving (Uzzi, 1997). These interac-
may be associated with an in-group bias. Social tion benefits in turn make it easier for team mem-
psychologists have studied the tendency of some bers to absorb one another’s knowledge, leading to
group members to systematically overvalue group a preference to work and share knowledge with
members and undervalue nonmembers (e.g., each other as opposed to seeking knowledge resi-
Brewer, 1979; Tajfel & Turner, 1986). Scholars dent in subsidiaries other than their own.
studying research and development (R&D) activi- Third, established within-team relations may
ties in organizations have reported a similar ten- give team members increased awareness of each
dency among engineers, known as the “not-invent- other’s knowledge, including awareness of knowl-
ed-here” syndrome, which refers to the bias in a edge that may be relevant to project-specific tasks
group of engineers of valuing their own knowledge (Austin, 2003; Schulz, 2003). This awareness may
more than that of others and henceforth rejecting lead team members to focus on this body of knowl-
2005 Hansen, Mors, and Løvås 779

edge when seeking solutions and further reduce the creasing the risk that an in-group bias is forming
chances that the team will seek knowledge from (Tajfel & Turner, 1986). The larger a focal team’s
subsidiaries other than their own. intersubsidiary network (i.e., the higher the num-
Members of a team who have a number of ongo- ber of direct intersubsidiary relations), the lower
ing established relations with one another prior to this risk should be, as team members have had
the start of a project may therefore channel their more chances to interact with engineers in other
time and energy toward the team and not toward subsidiaries and thereby have been exposed to dif-
other subsidiaries. The density of the within-team ferent views and skills. Moreover, the higher the
network (i.e., the number of ongoing established average strength of these established relations, the
relations among team members, divided by the to- lower the risk should be, as more frequent and
tal possible number of such relations)2 and the intense interactions will have increased the expo-
average strength of relations (i.e., the frequency and sure of team members to other views and skills,
intensity of interactions) may positively affect this thereby reducing their negative perceptions of oth-
channeling tendency. The chance of an in-group ers and increasing the chances that they will seek
bias forming is likely to increase because of the knowledge in other subsidiaries.
enhanced opportunity for members to jointly em- Second, established intersubsidiary relations
phasize the value of their own skills and reinforce may also be associated with a focal team’s capacity
commonly held beliefs afforded by extensive, fre- to absorb knowledge from those contacts, as there
quent, and intense past interactions. Furthermore, have been a number of interactions in the past in
teams that have high density and average strength which focal team members have had opportunities
of within-team relations are likely to have a larger to learn to engage in mutual problem solving and
common knowledge base than teams with low den- develop a common knowledge base with engineers
sity and strength of relations, increasing the prefer- in those subsidiaries (Cohen & Levinthal, 1990;
ence for relying on team members. And finally, the Uzzi, 1997). The larger a team’s intersubsidiary net-
more extensively, frequently, and intensively team work and the more frequent and intense interac-
members have interacted in the past, the more tions have been, the more such opportunities have
chances they have had to become aware of the been given to team members, enhancing their ab-
extent and type of knowledge held by other team sorptive capacity and hence their preference for
members. This should increase their preference for seeking knowledge from engineers in these other
seeking solutions inside the team and not outside. subsidiaries.
We therefore predict: Third, as the size and average strength of a focal
team’s intersubsidiary network increase, the team
Hypothesis 1. The higher the density of a with-
members’ awareness of knowledge held by engi-
in-team network, the less likely the focal team
neers in this network should increase, because
is to seek knowledge across subsidiaries.
team members have had more extensive, frequent,
Hypothesis 2. The higher the average strength and intense opportunities to learn about this
of relations in a within-team network, the less knowledge (cf. Austin, 2003). With increased
likely the focal team is to seek knowledge awareness about the knowledge available in other
across subsidiaries. subsidiaries, including knowledge that is relevant
for completing project-specific tasks, a focal team
Intersubsidiary network. The within-team net- should be more likely to seek knowledge across
work may channel members’ time and energy in- subsidiaries. In short, we predict:
side their team, while the team’s established rela-
Hypothesis 3. The larger a focal team’s inter-
tions in its intersubsidiary network may “pull” it
subsidiary network, the more likely the team is
outward. The causal mechanisms that lead teams to
to seek knowledge across subsidiaries.
seek solutions and knowledge within themselves
may also explain why some teams tend to look Hypothesis 4. The higher the average relation
outward. First, a team’s intersubsidiary network strength in a focal team’s intersubsidiary net-
may mitigate the effect of in-group biases. Teams work, the more likely the team is to seek knowl-
that are isolated from outside interactions may edge across subsidiaries.
form negative perceptions about others, thereby in-
Search Costs
2
The total possible number of relations in a within- A team in a focal subsidiary that has decided to
team network is defined as N ⫻ (N ⫺1), where N is the seek knowledge needs to search for it, an activity
number of team members (Wasserman & Faust, 1994). that involves looking for, identifying, and evaluat-
780 Academy of Management Journal October

ing knowledge resident in other subsidiaries. Dur- Hypothesis 5. The higher the average relation
ing this phase, teams incur search costs, which in strength in a focal team’s intersubsidiary net-
our context are measured as the number of engi- work, the higher the focal team’s search costs.
neering-months that a product development team
The previous discussion rests on the assumption
spends on search.
that direct contacts in a focal team’s intersubsidiary
Because the search phase in our definition pre-
network are motivated to help the focal team iden-
cedes the identification of useful knowledge, team
tify useful knowledge, but this may not be the case.
members may not know ex ante which subsidiaries
One reason for a lack of motivation to aid search is
possess novel and useful knowledge for a particular
intersubsidiary competition (Tsai, 2002). A team
project. In such a situation, a team’s total set of
and its subsidiary may compete with another sub-
intersubsidiary relations may be useful as it pro-
sidiary in the sense that both subsidiaries sell prod-
vides an initial search position from which a team
ucts to the same external markets and seek to de-
in a focal subsidiary can draw on several intersub-
velop the same types of products and technologies.
sidiary relations to launch a search for knowledge.
In such a situation, one subsidiary’s product devel-
As social network research has shown, however,
opment efforts may constrain the opportunities of a
it is not the sheer number of direct relations in a
competing subsidiary to the extent that new prod-
network that provides an advantageous search po-
ucts and technologies “crowd out” the available
sition but, rather, the extent to which those direct
market and technology opportunity set for the com-
relations provide access to novel knowledge (cf.
peting subsidiary (Sorenson, 2000).
Burt, 1992; Granovetter, 1973; Hansen, Podolny, &
If subsidiaries that a focal team contacts in its
Pfeffer, 2001). In particular, a set of intersubsidiary
search for knowledge perceive that the focal team
relations with low average strength (i.e., relations
presents such a competitive threat, the contacted
involving infrequent and nonintense interactions)
subsidiaries may hide what they know (so that the
is more likely than a set with high average strength
focal team will not identify useful knowledge), de-
to enable team members to access direct contacts
clare only parts of their related knowledge (so that
that possess novel knowledge. A set of relations
the team only benefits partially), or fail to point the
with high average strength tends to involve direct
focal team to other subsidiaries that may have use-
contacts that are themselves connected, because
ful knowledge (making search more difficult). Al-
they are likely to have been introduced to each
though they may be inaccurate, these perceptions
other through their common strongly tied actor
may nevertheless govern contacted subsidiaries’ re-
(Burt, 1992; Granovetter, 1973). Direct contacts that
sponses and are likely to increase a focal team’s
are themselves connected tend to circulate knowl-
search costs, as team members may need to expend
edge among one another and thereby possess sim-
additional effort searching to obtain a given amount
ilar knowledge, reducing the probability that a fo-
of useful knowledge. Thus, focal teams whose di-
cal team will identify novel knowledge when
rect contacts perceive high levels of competition are
contacting several of them. Thus, for a given
more likely to have higher search costs than teams
amount of novel knowledge being sought, a team
whose direct contacts perceive low levels of compe-
may have to make a number of contacts, as each one
tition between them and the focal subsidiary:
is likely not to have much knowledge that differs
from what the other contacts have. In contrast, Hypothesis 6. The higher the average level of
when the average strength of the relations in a competition perceived by subsidiaries in a fo-
team’s intersubsidiary network is low, the team’s cal team’s intersubsidiary network, the higher
direct contacts are less likely to be connected them- the focal team’s search costs.
selves and thereby are less likely to possess the
same knowledge. In this situation, the team is
Transfer Costs
likely to identify more novel knowledge for each
contact it makes and thereby lower search costs for Once a focal team has identified potentially use-
a given amount of novel knowledge sought.3 We ful knowledge, it has to be transferred from the
thus predict: providing subsidiary to the team, a process that
involves modifying, editing, and incorporating the
knowledge into the team’s product (Szulanski,
3
Having a high average relation strength may be espe- 1996; Zander & Kogut, 1995). In this phase, teams
cially problematic for teams pursuing very novel
projects. In the empirical part of our study, we included
an interaction term involving average intersubsidiary re- term were not significant, suggesting that this was not the
lation strength and project novelty, but results for this case in our data set.
2005 Hansen, Mors, and Løvås 781

incur transfer costs, which we measure as the num- transferring nontacit knowledge (Zander & Kogut,
ber of engineering-months spent on this activity. 1995).
While a focal team’s intersubsidiary network As prior research has shown, however, the diffi-
confers a search position, a transfer event most culty of transferring knowledge can be alleviated to
likely does not involve the total set of a team’s some extent if the two parties to a transfer know
intersubsidiary contacts but only those that end up each other well and thus have learned to work
providing knowledge. Despite having agreed to together (Hansen, 1999; Uzzi, 1997). When two par-
provide knowledge, however, providing subsidiar- ties to a transfer have developed a strong relation
ies may have various degrees of motivation to aid prior to the transfer effort, they have likely devel-
the transfer effort, in part depending on the level of oped a shared communication frame whereby each
perceived competition between them and the focal party has come to understand how the other party
subsidiary. A providing subsidiary may believe uses subtle phrases and ways of explaining difficult
that a transfer of knowledge to a team in a compet- concepts (Uzzi, 1997). Such strength in a dyadic
ing subsidiary will diminish its own opportunities transfer relation should therefore reduce transfer
for developing products and technologies based on costs by reducing the time and effort required to
the knowledge being provided to the focal team (cf. understand and incorporate knowledge into the fo-
Galunic & Eisenhardt, 2001). As a result, the pro- cal project. In particular, as the tacitness of the
viding subsidiary may “drag its feet” and be more knowledge that a subsidiary provides to a focal
guarded about what knowledge—and how much team increases, an existing shared communication
knowledge—it transfers than it would be if no com- frame afforded by established strong dyadic trans-
petition existed. In this situation, the focal team has fer relations is likely to become more important, as
to spend more time negotiating with the providing the two parties to a transfer can rely on it to artic-
subsidiary, increasing the time and effort required ulate, modify, and incorporate the subtle and im-
to transfer knowledge. plicit aspects of the tacit knowledge, thereby reduc-
The effect of competition on transfer costs pri- ing transfer costs:
marily concerns the extent of competition per-
ceived by the providing subsidiary, not by the focal Hypothesis 8. Knowledge tacitness will modify
team or the focal subsidiary. Providing engineers the main negative effect of dyadic relationship
who perceive that a focal team and subsidiary pose strength on a focal team’s transfer costs: the
a significant competitive threat are likely to be effect will be stronger when tacitness is high
more guarded and less forthcoming than are pro- and weaker when tacitness is low.
viding engineers who do not perceive the same
extent of competition. Stated in a hypothesis:
METHODS AND DATA
Hypothesis 7. In a dyadic transfer relationship, We tested the hypotheses using a data set of 121
the more a providing subsidiary perceives that new-product development teams and 41 subsidiar-
it competes with a focal subsidiary, the higher ies of a large high-technology company. The com-
the focal team’s transfer costs. pany, which had annual sales of more than $5
billion at the time of the study, was involved in
Although the extent of perceived competition af- developing, manufacturing, and selling a range of
fects the motivation to transfer knowledge to a focal industrial electronics and other high-technology
team, other factors affect the ability of a providing products and systems. It was structured into 41
subsidiary and the focal team to carry out a transfer fairly autonomous subsidiaries that were responsi-
smoothly. Prior research has demonstrated that ble for their own new-product development, man-
tacit knowledge— knowledge that is difficult to ar- ufacturing, and sales and thus had considerable
ticulate or that can only be acquired through expe- freedom in choosing which other subsidiaries to
rience—is difficult to transfer smoothly (e.g., Han- contact to access useful knowledge. For the pur-
sen, 1999). When knowledge is tacit, engineers in poses of this study, we considered a new-product
the providing subsidiary will find it more difficult development team in a focal subsidiary to be en-
to explain its content and nuances to members of gaged in intersubsidiary search and transfer if it
the focal team, who in turn may find it difficult to contacted and obtained knowledge from one or
understand, thereby making the tasks of modifying more of the other 40 subsidiaries. Having negoti-
and incorporating the knowledge into the product ated access to the company through three senior
difficult. Because of these difficulties, transferring corporate R&D managers, we visited 14 subsidiaries
tacit knowledge is likely to be more cumbersome, and conducted open-ended interviews with more
take a longer time, and thus be more costly than than 30 project engineers and managers to better
782 Academy of Management Journal October

understand the context and to develop survey in- asked the project managers, “How many engineer-
struments to test our hypotheses. ing-months did the team spend searching for (look-
To select product development teams, we used ing for, identifying, and evaluating) the technical
the firm’s databases on projects to develop a list of advice, software and hardware that the team
projects that the subsidiaries undertook and iden- wanted from other subsidiaries?” The responses
tified 147 projects. We administered three different ranged from 0.1 to 15 (or from approximately 1 to
survey instruments to create the variables: a net- 22 percent of total engineering-months for a
work survey distributed to the R&D managers in project), with a mean of 4.39 (or 5 percent of total
each of the 41 subsidiaries, asking about relations engineering-months). To compute the variable
across the subsidiaries (a 100 percent response rate search costs, we used the absolute number of engi-
was achieved); a survey distributed to the 147 neering-months spent on search.
project managers of each of the product develop- Transfer costs. To measure transfer costs, we
ment teams included in the study, asking about asked the project managers, “How many engineer-
sources of knowledge for the project (an 82 percent ing-months did the team spend modifying, editing
response rate); and a survey distributed to 510 in- and incorporating the technical advice, software
dividual members of the projects, asking about and hardware that came from the other subsidiar-
their own personal relations with colleagues (a 51 ies?” Project managers kept a close eye on engineer-
percent response rate). ing-months spent on this activity and thus could
The final sample included 121 projects that took provide reliable answers to this question. The re-
place in 27 different subsidiaries. We used the in- sponses ranged from 0.1 to 40 (or approximately 1
formation on projects from the databases to analyze to 57 percent of total project engineering-months),
response rates but found no differences between with a mean of 7.94 (or 10 percent of total engineer-
the final 121 projects and the 26 projects for which ing-months). To compute the variable transfer
the survey was not returned in terms of number of costs, we used the absolute number of engineering-
engineers, budget, and project age. In addition, we months spent on transfer. Five projects did not
took several steps to minimize potential common report transfer costs, and we thus had to drop these
method biases (Podsakoff & Organ, 1986). This con- from the analysis (there were no significant differ-
cern was reduced substantially because we relied ences between these five and the other projects).
on three different sets of respondents. Furthermore, These search and transfer costs variables do not
as recommended by prior research, we focused on specify the benefits of using knowledge from other
behavioral measures and not on perceptual ones, subsidiaries. To control for the benefits that the
which are prone to common method biases teams obtained by spending time searching and
(Gatignon, Tushman, Smith, & Anderson, 2002). transferring, we entered a control variable denoting
the amount of knowledge obtained from other sub-
sidiaries (described in the section about control vari-
Dependent Variables
ables). This way we measured search and transfer
Sought knowledge. We asked each project man- costs given a certain amount of knowledge obtained.
ager to indicate whether team members had sought
knowledge in any of the other subsidiaries except
Within-Team Network Variables
the focal one. Project managers were able to answer
this question reliably, as they kept detailed logs on Within-team network density. To test Hypothe-
how engineers on the project spent their time. Fif- sis 1, we used responses from the individual team
ty-five project managers (45%) reported that their member survey. Following accepted procedures
teams had sought knowledge in at least one of the from the social network literature for soliciting a
other 40 subsidiaries. We coded this dependent person’s network contacts (e.g., Podolny & Baron,
variable, sought knowledge, as 1 if a team had 1997), we asked each team member three questions:
sought knowledge from another and 0 otherwise. “Looking back over the last year, are there any
Search costs. The preliminary interviews in- persons in your subsidiary: (i) from whom you
formed us that one of the most salient search costs regularly sought information and advice to help
for these projects was engineering-months spent your project work, (ii) to whom you would go on a
looking for, identifying, and evaluating knowledge regular basis to get buy-in for your work, or (iii)
from other subsidiaries and that project managers with whom you interacted informally as a friend?”
kept track of the number of engineering-months Each respondent entered the names of individuals
spent on this activity. The company defined an described by either of these three choices, and we
engineering-month as the equivalent of a product then counted the number of contacts these individu-
developer’s full-time work for one month. We als had with other members of the respondent’s team.
2005 Hansen, Mors, and Løvås 783

In addition, to avoid recording relations that were cally like being in the same work group,” to 7, “dis-
created after the start of a focal project, we asked each tant, like an arm’s length delivery of the input”). We
respondent how long each relation had been in exis- reverse-scored and averaged the two items, which
tence and excluded relations that commenced after were highly correlated (r ⫽ .83; ␣ ⫽ .92). To test
the start of the focal project. We computed within- Hypotheses 4 and 5, we computed an average for each
team network density as the number of existing rela- team, average intersubsidiary relation strength.
tions divided by the number of possible asymmetric Intersubsidiary perceived competition. To mea-
relations, which is given by N ⫻ (N ⫺ 1), where N is sure the level of perceived competition between
the number of team members (Wasserman & Faust, subsidiaries, we asked each subsidiary R&D man-
1994). This measure ranges from 0 (“no relations ager to indicate, for each relationship he or she
exist”) to 1 (“all possible relations exist”).4 listed on the survey, “How would you describe the
Within-team relation strength. To test Hypoth- competitive nature between your [subsidiary] and
esis 2, we asked team members two questions: this [subsidiary]?” (1, “non-competitive: never com-
“How frequently have you interacted with this per- pete for markets and technologies,” to 7, “competi-
son over the past year?” (1, “once a day,” to 7, tive: frequently compete for markets and technolo-
“once every 3 months”) and “How close are you to gies”). These perceptions are asymmetric: an R&D
this person? (1, “very close,” to 7, “distant”). We manager in subsidiary A may perceive subsidiary B
reverse-scored the two dimensions and took the as a competitive threat to A, but the R&D manager in
average of these two dimensions to create average subsidiary B may not perceive subsidiary A as a
within-team relation strength. The Cronbach alpha threat to B. Because all 41 R&D managers completed
for this scale was .81. the survey and indicated the level of perceived com-
petition with subsidiaries with which they had rela-
tions, we were able to obtain information on how
Intersubsidiary Network Variables
other subsidiaries perceived a focal subsidiary.
Intersubsidiary network size. To test Hypothe- We used this information to construct two mea-
sis 3, we constructed a measure of the number of sures of perceived competition in each team’s in-
intersubsidiary relations. We asked the subsidiary tersubsidiary network. First, to test Hypothesis 6,
R&D managers, “Over the past two years, are there we computed the average value of direct contacts’
any units [i.e., subsidiaries] from whom your unit responses to the above question. This variable, di-
[i.e., focal subsidiary] regularly sought technical rect contacts’ average perceived competition, mea-
and/or market-related input?” The question was sured how all subsidiaries with which a focal team
followed by a list of the 41 subsidiaries included in had established relations perceived the subsidiary
the study, and the manager could then check the to which the team belonged. Second, we computed
appropriate subsidiaries. We asked the respondents focal subsidiary’s average perceived competition—
to indicate the age of each relation and excluded that is, how each subsidiary’s R&D manager per-
those that had commenced after the start of focal ceived the level of competition with direct contacts.
projects. These relations thus vary among the
projects within a subsidiary. Finally, we assigned
Transfer Network Variables
the appropriate intersubsidiary relations to each
focal project and computed the number of estab- Perceived competition in transfer. To test Hy-
lished relations with other subsidiaries that existed pothesis 7, we followed the same procedures but
prior to the start of the focal project, labeling the limited the measure to only those subsidiaries that
variable intersubsidiary network size. provided knowledge to a focal team. The average
Intersubsidiary relation strength. We asked the value of their responses to the question about per-
subsidiary R&D managers to answer the following ceived competition was the variable providers’ per-
for each of the relations he or she listed on the ceived competition.
survey: “How frequently do people in your [subsid- Dyadic transfer relation strength and tacitness
iary] interact with this [subsidiary]?” (1, “once a of knowledge. We limited this relational variable
day,” to 7, “once every 3 months”) and “How close to subsidiaries that provided knowledge to a focal
is the working relationship between your [subsid- team. Using the scale for intersubsidiary relation
iary] and this [subsidiary]?” (1, “very close, practi- strength described above, we constructed the mea-
sure dyadic transfer relation strength (0, “no estab-
lished relation,” to 7, “very strong relation”). For
4
Because this information was missing for eight this measure, we subtracted the mean from the
teams, we created a dummy variable that took on a value values to reduce the correlation with the interac-
of 1 for missing information and 0 otherwise. tion effect. To measure the tacitness of transferred
784 Academy of Management Journal October

knowledge, we asked each project manager to indi- cause this information was incomplete for 23 out of
cate, for the knowledge transferred from each pro- the 121 projects, we also included a dummy vari-
viding subsidiary: (1) “How well documented was able that took on a value of 1 if this information was
the knowledge?” (1, “It was very well docu- incomplete and 0 otherwise.
mented,” to 7, “It was not well documented”); (2) Search and transfer controls. Because it is pos-
“Was all of this knowledge explained to your team sible that projects with many engineering-months
in writing?” (1, “All of it was,” to 7, “None of it available have more capacity to search, we used
was”); and (3) “What type of knowledge came from two control variables: a project’s total number of en-
this unit?” (1, “mainly reports, manuals, docu- gineering-months (in hundreds of months), as esti-
ments, self-explanatory software, etc.,” to 7, mated at the beginning of the project (project total
“mainly personal practical know-how, tricks of the engineering-months) and the budget (in millions of
trade”). To compute the variable, we averaged the dollars) at the start of the project (project budget).
responses and then subtracted the mean from the We also wanted to control for the amount of
values. The Cronbach alpha for this variable was knowledge sought in search and the amount of
.81. To test Hypothesis 8, we interacted the tacit- knowledge obtained during transfer. As we men-
ness variable with the variable denoting the tioned in the section defining search and transfer
strength of dyadic transfer relations (tacitness ⫻ costs, knowledge obtained is a proximate measure
transfer relation strength). for the benefits of transferring knowledge. Al-
though we did not have measures of the amount of
knowledge sought, we created a measure of the
Control Variables
amount of knowledge obtained from other subsid-
Need to seek knowledge. We entered three vari- iaries. Each project manager was asked to indicate
ables to control for teams’ need or inclination to the fraction of all his or her project’s hardware and
seek knowledge outside their own subsidiary. First, software that came from other subsidiaries (0, “no
a team engaged in tasks that depart significantly knowledge obtained,” to 1, “all knowledge came
from its subsidiary’s knowledge base may be more from other subsidiaries”). In field interviews with
motivated to search outside than teams that can project managers, we discovered that they kept
draw on their subsidiaries’ existing knowledge. We close track of the amount of knowledge obtained,
measured the extent to which each team drew on as this indicated the savings they could accrue
its subsidiary’s knowledge base via project manag- and thus affected the amount of engineering
ers’ logs of the sources of the software and hard- hours required to complete the project. For soft-
ware used in their projects. The variable, existing ware, for example, they would often keep track
ware, was the fraction of all the software and hard- of lines of code that were provided by other sub-
ware used in a project that had already been devel- sidiaries. To compute this variable, we took the
oped and existed in the subsidiary to which the weighted average of the two fractions— one for
project belonged. hardware and one for software—and multiplied it
Second, we used a two-item scale to capture by the total number of a project’s budgeted engi-
project novelty, the extent to which each project neering months. This variable, amount of knowl-
strayed from its subsidiary’s technological and edge, measured the absolute amount of engineering
market expertise. Each project manager was asked, months represented by the knowledge obtained
“Prior to this project, how much experience did from other subsidiaries.5
your subsidiary have with the technologies and Finally, we also controlled for each team’s capac-
technical competencies that the project required?” ity to absorb the knowledge obtained from other
and “Prior to this project, how much experience subsidiaries. Because larger teams may have more
did your subsidiary have with the market for which
the product was developed?” (1, “none of the re-
5
quired expertise,” to 7, “all of the required exper- Although we posit that the amount of knowledge
tise”). Answers to the two questions were averaged obtained predicts search costs, it is reasonable to suspect
and then reversed the scale so that a high value that search costs, which may be an indication of search
effort, also predict the amount of knowledge obtained. To
indicated a relatively novel project (␣ ⫽ .66).
control for this endogeneity problem, we analyzed two
Third, because prior research has shown that simultaneous equations, one specified as search costs ⫽
team members’ tenure negatively affects the prob- f (amount knowledge, all other independent variables),
ability of their seeking outside help (e.g., Katz & and one specified as amount knowledge ⫽ f (search costs,
Allen, 1988), we included team tenure, a variable some other predictor variables). This test did not alter
denoting the average number of years individual our results, and we are therefore reasonably confident
team members had been with the company. Be- that this type of endogeneity is not salient in our models.
2005 Hansen, Mors, and Løvås 785

skills and capacity for assimilating knowledge, we spent on search and transfer. We performed tests to
entered the variable team size, measuring the num- ensure that the OLS specification was appropriate
ber of engineers spending at least 50 percent of for our data. The Kolmogorov-Smirnov test of nor-
their time working on a focal project (team size). mality revealed that the residuals were normally
For a second measure, absorptive capacity, we distributed in the OLS models. For an alternative
asked each project manager, “Prior to the project, specification, we also ran truncated regression
did your project team have the expertise required to models wherein the distributions of the search and
assimilate the knowledge that came from the other transfer cost variables were presumed to be trun-
subsidiaries?” (1, “had very little of that expertise,” cated at zero as there were no negative values.
to 7, “had very much of that expertise”). These models did not reveal any different effects
and are thus not reported here.
Statistical Approach
RESULTS
Because search and transfer costs were only rel-
evant for those projects that sought and found Table 1 reports descriptive statistics, and Table 2
knowledge, we excluded from the analysis of reports results pertaining to the three dependent
search and transfer costs those projects that did not variables. Models 1–3 in Table 2 report results on
seek knowledge.6 Thus, although the analysis of teams deciding whether to seek knowledge across
whether teams sought knowledge included all 121 subsidiaries. Model 2 shows that the within-team
projects on which we obtained data, the analyses of network density variable has a negative and signif-
search and transfer costs include only 55 and 50 icant effect on whether a team sought knowledge
projects, respectively. To ensure that the smaller across subsidiaries, lending support to Hypothesis
sets of data for search and transfer costs were large 1. As model 2 reveals, the effect for the average
enough for us to detect the expected effects, we within-team relation strength variable is also nega-
calculated the statistical power and the effect size, tive but is only significant at the .10 level. To in-
and both tests showed that we had adequate sample vestigate further, we separated the two underlying
size (Aiken & West, 1991; Cohen, 1988).7 dimensions of closeness and frequency underlying
In addition, because multiple project observa- relation strength, as shown in model 3. The fre-
tions from the same subsidiary might not be inde- quency dimension is negative and significant at the
pendent, we used robust standard errors that cor- .01 level, lending support to Hypothesis 2.
rected for subsidiary-specific clustering and used Model 2 also reveals that intersubsidiary network
the robust clustering procedure as implemented in size has a positive effect on whether teams sought
Stata 7.0 for all our models (Moulton, 1986; Rogers, knowledge across subsidiaries, lending support to
1993). We used logistic regression analysis to Hypothesis 3. However, average intersubsidiary re-
model whether a team in a focal subsidiary sought lation strength is not significant in model 2, and
knowledge in one or more of the other subsidiaries thus Hypothesis 4 is not supported.
and ordinary least squares (OLS) regression analy- Figure 2a reveals that the supported effects are of
sis to model the number of engineering-months fairly substantial magnitudes.8 For example, when
within-team network density increases from 0.1 to
0.3, the likelihood that teams sought knowledge
6
To control for this potential selection bias, we also decreases from 0.62 to 0.31. In contrast, when in-
ran a two-part logit model (Manning, Duan, & Rogers, tersubsidiary network size increases from 1 to 5,
1987). In the first part, the dependent variable was whether the likelihood of seeking knowledge increases from
teams sought knowledge. From this model, we calculated 0.28 to 0.59.
the predicted probability of each of the 121 teams seeking Models 4 and 5 in Table 2 report the results for
knowledge, p (seek knowledge), and then entered this vari- search costs. As model 5 indicates, average inter-
able in the second part of the models for search and transfer
costs. The results of this test revealed no selection bias.
7
For a sample to be considered large enough to show 8
We computed the lines in Figures 2a and 2b by
expected effects, the statistical power of the analysis
including the coefficient for the intercept and evaluating
should be at least 0.8 (Aiken & West, 1991). In our anal-
all other variables in a model at their mean values. On the
yses, statistical power was at a minimum 0.9 in all of the
basis of model 2, the two lines in Figure 2a are computed
tested models. Furthermore, the effect sizes were greater
as follows:
than 0.35 and thus big enough to be considered large
effects (Cohen, 1988). We therefore do not believe that Prob. ⫽ expy /(1 ⫹ expy⫺6.34 ⫻ within-team network density ).
there is any problem with the small samples for search
and transfer costs. Prob. ⫽ expy /(1 ⫹ expy⫹0.32 ⫻ intersubsidiary network size ).
786 Academy of Management Journal October

TABLE 1
Descriptive 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 1.00

2. Search costsb 4.39 5.17 0.10 15.00
3. Transfer costsc 7.94 9.90 0.10 40.00 .70***
4. Existing ware 0.45 0.31 0.00 1.00 ⫺.40*** ⫺.05 ⫺.16
5. 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† ⫺.01
7. Project total 0.80 2.31 0.10 24.00 .20* .32* .50*** ⫺.20* .01 .90***
engineering-months
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 knowledge 25.00 53.4 0.43 306.00 .41** .50*** ⫺.25† .02 .80*** .90***
obtainedb
11. Absorptive capacityb 4.37 1.87 1.00 7.00 ⫺.22 ⫺.01 ⫺.01 ⫺.22 .10 .19
12. Tacitness of 3.26 1.16 1.33 6.00 .33* .17 .11 ⫺.15 .08 .07
transferred
knowledgec, d
13. Within-team network 0.09 0.14 0.00 1.00 ⫺.20* .06 ⫺.19 .07 ⫺.14 ⫺.07 ⫺.14
density
14. Average within-team 2.55 1.46 1.00 5.50 ⫺.17† .11 ⫺.17 ⫺.03 ⫺.10 ⫺.05 ⫺.04
relation strength
15. Intersubsidiary 5.60 2.95 1.00 14.00 .27** .17 .17 .09 .21* .26** .18†
network size
16. Average 4.09 1.26 2.18 7.00 .06 .09 ⫺.04 ⫺.30*** ⫺.04 ⫺.05 ⫺.05
intersubsidiary
relation strength
17. Direct contacts’ 3.06 1.21 1.00 5.00 .26† .18 .27* ⫺.28* .01 ⫺.06
average perceived
competitionb
18. Focal subsidiaries’ 2.58 .97 1.00 6.00 .09 ⫺.06 .02 ⫺.09 ⫺.10 ⫺.15 ⫺.11
average perceived
competition
19. Providers’ perceived 1.95 1.79 1.00 6.30 .02 .14 .12 ⫺.18 ⫺.01 .09
competitionc
20. Dyadic transfer 3.84 2.38 0.00 7.00 ⫺.07 ⫺.10 .02 ⫺.28* .10 .16
relation strengthc, d
21. Tacitness ⫻ transfer 11.57 9.92 0.00 39.69 .12 ⫺.25† ⫺.26† ⫺.03 ⫺.01 .03
relation strengthc

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

subsidiary relation strength has a positive and sig- within-team network density and intersubsidiary
nificant effect on search costs, supporting Hypoth- network size—are not significant predictors of search
esis 5. The variables for the competition effects on costs. Those factors that explain whether a team
search costs were also entered in model 5, and they sought knowledge do not explain the costs of search.
show that the extent of competition perceived by The variables pertaining to Hypotheses 7 and 8,
direct contacts has a positive and significant effect. which concern transfer costs, were entered in mod-
That is, if subsidiaries with which a team had es- els 6 and 7 in Table 2. The coefficient for the
tablished relations perceived the relationships to providing subsidiaries’ perceived level of competi-
be competitive, the team’s search costs increased. tion (providers’ perceived competition) is signifi-
These findings support Hypothesis 6. cant and positive, indicating that a team’s transfer
It is also interesting to note that the two powerful costs increased to the extent that the providing
predictors of whether a team sought knowledge— subsidiaries saw the subsidiary in which the team
2005 Hansen, Mors, and Løvås 787

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

was located as a competitor. This result supports sidiary’s average perceived competition) has no
Hypothesis 7. Figure 2b shows the magnitude of the impact on transfer costs, nor does the level of
effect: for example, when providers’ perceived competition perceived by the total set of direct
competition increases from 1 to 4 on the seven- contacts in a team’s intersubsidiary network.
point scale, a focal team’s transfer costs go up by Thus, the competition variables affect the three
4.8 engineering-months, which is a substantial dependent variables in different ways: while they
jump, given that these teams spent an average 8 have no effect on whether a team sought knowl-
engineering-months on transfer. edge, only direct contacts’ perceptions of compe-
Interestingly, as model 7 shows, the other two tition increased search costs, whereas only pro-
competition variables are not significant in the viders’ perceptions of competition increased
transfer models. Whether a focal subsidiary per- transfer costs.
ceives relationships to be competitive (focal sub- Model 7 in Table 2 also reveals that the interac-
 TABLE 2
Results from Regression Analysisa

Model 1: Model 2: Model 3: Model 4: Model 5: Model 6: Model 7:

Sought Sought Sought Search Search Transfer Transfer
Dependent Variables Knowledge Knowledge Knowledge Costs Costs Costs Costs

Intercept 0.11 (1.21) 0.52 (2.76) 0.03 (2.81) 6.37 (3.41) ⫺13.10 (10.3) 7.27 (5.14) ⫺12.06 (10.97)
Control variables
Existing ware ⫺2.76 (1.04**) ⫺3.97 (1.50**) ⫺4.16 (1.53**) 0.62 (3.20) 3.16 (4.56) ⫺1.28 (3.48) ⫺8.17 (7.22)
Project novelty 0.02 (0.15) ⫺0.18 (0.21) ⫺0.13 (0.21) 0.28 (0.55) 0.93 (0.42*) 0.30 (1.15) 0.71 (1.16)
Budget 0.13 (0.00) ⫺0.03 (0.00) ⫺0.02 (0.00) 0.54 (0.00*) 0.35 (0.00) 0.71 (0.00) 0.41 (0.00)
Project total engineering-months 0.43 (0.01) 0.62 (0.00) 0.74 (0.00) ⫺0.77 (0.01) ⫺0.27 (0.01) ⫺1.82 (0.02) 0.10 (0.01)
Team tenure 0.04 (0.01) 0.08 (0.04†) 0.10 (0.04†) ⫺0.19 (0.13) ⫺0.23 (0.14†) ⫺0.14 (0.15) ⫺0.09 (0.20)
Team size ⫺0.03 (0.11) ⫺0.06 (0.16) ⫺0.04 (0.16) 0.09 (0.25) 0.34 (0.41) ⫺0.25 (0.32) ⫺0.73 (0.69)
Amount knowledge 0.02 (0.04) 0.03 (0.04) 0.13 (0.16) 0.03 (0.11)
Absorptive capacity ⫺0.46 (0.43) ⫺0.39 (0.49) 0.13 (0.48) 1.10 (0.73)
Tacitness of transferred knowledgeb 1.59 (1.10)
Within-team network
Within-team network density ⫺6.34 (3.13*) ⫺6.34 (3.13*) 2.52 (10.97) ⫺13.39 (16.70)
Average within-team relation strength ⫺0.47 (0.29†) 0.31 ( 1.70) 0.48 (1.20)
Average within-team closeness 0.11 (0.20)
Average within-team frequency ⫺0.83 (0.30**)
Intersubsidiary network
Intersubsidiary network size 0.32 (0.10***) 0.32 (0.10***) 0.21 (0.28) 0.91 (0.61)
Average intersubsidiary relation strength 0.08 (0.23) 0.05 (0.25) 1.60 (0.75*) 1.71 (1.58)
Direct contacts’ average perceived competition 1.86 (0.43***) 1.80 (1.43)
Focal subsidiaries’ average perceived 0.17 (0.30) 0.21 (0.30) ⫺0.06 (0.38) ⫺0.51 (1.33)
competition
Transfer network
Providers’ perceived competition 1.70 (0.83*)
Dyadic transfer relation strengthb ⫺1.78 (0.67*)
Tacitness ⫻ transfer relation strength ⫺1.48 (0.50**)

Type of analysis Logit Logit Logit OLS OLS OLS OLS

n 121 121 121 55 55 50c 50c
Likelihood-ratio chi-square testd/F 22.30*** 5.01* 12.37*** 230.41*** 5.03** 69.49***
Pseudo-R2/R2 0.18 0.31 0.34 0.28 0.46 0.29 0.61

a
Dummy variables denoting missing information on total engineering-months and team tenure are not shown. All models shown with robust standard errors.
b
Mean-deviated.
c
Five observations are omitted owing to missing information on transfer costs.
d
Compared with previous model.
†
p ⬍ .10
* p ⬍ .05
** p ⬍ .01
*** p ⬍ .001
Two-tailed tests for variable coefficients.
2005 Hansen, Mors, and Løvås 789

FIGURE 2a
Plot of Effects for Sought Knowledge

FIGURE 2b
Plot of Effects for Transfer Costs

tion effect including the tacitness and dyadic trans- Finally, model 7 in Table 2 also shows that prop-
fer relation strength variables is negative and sig- erties of the within-team network that affected
nificant, supporting Hypothesis 8. As the tacitness whether teams sought knowledge (that is, the den-
of the knowledge being transferred increased, an sity and strength of relations) and the properties of
increase in the strength of the established relation the intersubsidiary network that affected search
between a team and the providing subsidiaries re- costs (i.e., contacts’ perceived competition and
duced the number of engineering-months spent on
transfer. As Figure 2b shows, having strong estab-
lished relations was especially beneficial when and the tacitness of transferred knowledge at their mean
highly tacit knowledge was involved.9 values: transfer costs ⫽ ⫺1.78 ⫻ dyadic transfer relation
strength ⫹ 1.59 ⫻ tacitness ⫺ 1.48 ⫻ tacitness ⫻ transfer
relation strength, where “tacit low” means that tacitness
9
The lines in Figure 2b that involve tacitness were is –0.5 standard deviations below the mean (i.e., ⫺.74),
computed with results from model 7 in Table 2, evaluat- and “tacit high” means that tacitness is ⫹1 s.d. above the
ing the variables except dyadic transfer relation strength mean (i.e., ⫹1.47).
790 Academy of Management Journal October

strength) have no significant effects on transfer caution is needed in interpreting the findings, our
costs. study provides new insights into the process of
knowledge sharing in organizations.
DISCUSSION AND CONCLUSION
Emphasizing the Role of Competition in
The main finding of this study is that teams’
Knowledge Sharing
different subsets of networks to a large extent af-
fected the outcomes of the three phases of deciding An important finding in our study is the positive
to seek knowledge, incurring search costs, and in- and significant effect of perceived competition on
curring transfer costs in different ways. The within- both search and transfer costs. Although scholars
team network variables denoting the density and have noted that employees often hoard knowledge
frequency of relations reduced the probability of (e.g., Davenport & Prusak, 1998), the mechanisms
seeking knowledge across subsidiaries but did not contributing to such reduced motivation to provide
affect search and transfer costs. The intersubsidiary knowledge have not been elucidated. Our findings
network variable capturing network size increased suggest that hoarding may in part be a result of
the probability of seeking knowledge, whereas in- perceived competition between users and potential
tersubsidiary relation strength and degree of per- providers of knowledge and that such competition
ceived competition increased search costs but did is likely to vary across pairs of providers and users.
not affect transfer costs. And the variable denoting The findings suggest that employees are not neces-
the degree of perceived competition in a team’s sarily predisposed to hoarding knowledge; rather,
transfer network increased transfer costs, whereas their tendency to do so can be explained by inter-
relation strength in a transfer network reduced unit patterns of competition in an organization.
transfer costs when transfers involved highly tacit Furthermore, our results indicate that it matters
knowledge. These effects would have been masked who perceives competition: for search and transfer
had we not distinguished between the three phases costs, it is the perceptions of the direct contacts in
of knowledge sharing and decomposed teams’ net- a focal actor’s network and not the perception of
works into three distinct subsets. the focal actor that matter. This pattern makes in-
Before we discuss the implications of these find- tuitive sense: If direct contacts perceive the search-
ings, it is worth acknowledging some limitations of ing actor to be a competitor, they are likely to be
our study. Because we studied only one company, less cooperative and are more likely to hoard their
we cannot claim that these results would necessar- knowledge or even refuse to cooperate at all, irre-
ily hold for all organizations. In particular, the spective of the perceptions of the focal actor.
company we studied relied quite extensively on One implication of our findings is that research
informal relations among employees to accomplish on knowledge sharing needs to address the possi-
work, implying that relational properties played an bly negative aspects of competition in interunit
especially important role in knowledge sharing relations (cf. Tsai, 2002). This implication opens up
here. Other mechanisms for knowledge sharing, in- several avenues for new research. For example,
cluding reliance on headquarters’ coordination and while we treated the extent of perceived competi-
transfer prices between subsidiaries, played rela- tion as an exogenous variable, it may be endoge-
tively small roles in this company, yet they may be nous to the process of knowledge sharing. Two
important mechanisms in different settings. units that share knowledge may, over time, begin to
The company we studied was also characterized develop similar products and services for similar
by a fairly cooperative culture in which new prod- markets because they increasingly rely on the same
uct developers frequently helped one another. knowledge base and ideas. This convergence, in
Teams in other companies that actively encourage turn, may make the relationship more competitive,
competition may experience outright rejection of triggering a process of avoiding competitive rela-
requests for knowledge and strong resistance to tions and forming new interunit relations to ex-
transfers, which will raise search and transfer costs change knowledge (cf. Galunic & Eisenhardt, 2001).
over what the teams in our setting incurred. This Knowledge sharing and the formation of, and pos-
bias may be in a conservative direction, however: sibly dissolution of, interunit relations thus co-
even though we studied an organization with a evolve in such a competitive landscape (Koza &
fairly cooperative culture, we still found strong ef- Lewin, 1999). Although our data were cross-sec-
fects for the competition variables, which may have tional and hence we could not disentangle these
an even more pronounced effect in organizations intertemporal processes, in subsequent studies re-
with highly competitive cultures. searchers could use longitudinal data to analyze
Although these limitations suggest that some when knowledge sharing increases perceived com-
2005 Hansen, Mors, and Løvås 791

petition in interunit relations and how perceived tion strength on knowledge sharing that would not
competition affects the formation and dissolution emerge from studying the transfer process only
of such relations. (measured by the number of engineering months
spent on search). These findings also validate and
refine results reported by Hansen (1999): while that
Multiple Networks, Multiple Phases of
study advanced the argument that weak interunit
Knowledge Sharing
ties facilitate search but undermine the transfer of
The most important implication of our findings tacit knowledge, it had no direct measures of search
is that research on knowledge sharing needs to and transfer outcomes but only an analysis of
fully incorporate the phase level of knowledge project task outcomes in the form of completion
sharing in an organization and the subset level of time. Results reported here show that weak ties
social networks in order to advance a robust theory (i.e., low average relation strength) benefited search
of knowledge sharing. We are not the first to con- by reducing search costs, while strong ties helped
ceive of knowledge sharing as consisting of multi- transfer by reducing transfer costs (that is, engineer
ple phases, yet our approach here represents a months spent on transfer), verifying Hansen’s
broader view than other perspectives. For example, (1999) earlier argument by measuring search and
although Szulanski (1996) considered various steps transfer costs separately.
of knowledge transfers, including initiation, imple- Our results also showed that the network size has
mentation, ramp-up, and integration of transfers, different impacts at the within-team and intersub-
we go beyond transfer by also including the “front sidiary levels of analysis. Specifically, whereas the
end” of knowledge sharing—the decision to seek density of relations inside a team decreased the
knowledge in the first place, and the search likelihood that it would seek knowledge, the size of
process. a team’s network across subsidiaries increased this
Despite the attention researchers have paid to likelihood. That is, established relations at the
different subsets of networks, such as within-team team level countervailed established relations at
and extrateam subsets, little research compares dif- the subsidiary level. This finding casts new light on
ferent subsets of networks and analyzes their dif- research that has examined the effects of relational
ferent impacts on various phases of knowledge properties on the extent of interunit knowledge
sharing. Our results suggest this scarcity has im- sharing and communication (Gupta & Govindara-
peded further insights into how and why actors jan, 2000; Szulanski, 1996; Zander & Kogut, 1995).
share knowledge across subunits in an organiza- It is not necessarily the absolute number of estab-
tion. To illustrate, consider existing work on the lished interunit relations that explains why organ-
strength of relations and knowledge transfers. Re- ization members share knowledge but, rather, the
searchers have shown that transfers of knowl- relative number of within-team and interunit rela-
edge— especially tacit knowledge— between a pro- tions. For example, even though some teams in our
viding and receiving unit is facilitated to the extent data set had very high numbers of intersubsidiary
that the two parties have a strong established rela- relations (e.g., nine, which is more than one stan-
tion (Hansen, 1999; Szulanski, 1996; Uzzi, 1997). dard deviation above the mean level), a very high
From this research one might conclude that high within-team network density (e.g., 0.60) would off-
average relationship strength will improve knowl- set the positive effect of such a high number of
edge sharing in an organization. Our results indi- intersubsidiary relations on the probability of seek-
cate that this hypothesis is incorrect and that it ing knowledge across subsidiaries.
depends on the particular phase of the knowledge- This emphasis on the relative numbers of within-
sharing process being studied and which subsets of team and interunit relations raises the issue of how
networks are being considered. Specifically, our they relate to each other. Because we did not ana-
results revealed that high average relationship lyze the formation of relations, we cannot predict
strength in within-team networks led to fewer whether the existence of within-team relations
knowledge seeking attempts across subsidiaries crowds out the interunit ones (and vice versa), al-
and, thus, fewer intersubsidiary knowledge-sharing though this may not have been the case in our data
events. High average relation strength in the inter- set, as the correlation between within-team net-
subsidiary network was also linked to higher work density and intersubsidiary network size was
search costs than low average relation strength, low (–.08). Given that the time and energy that can
whereas our results validated the prediction that be devoted to building and maintaining relation-
strong relations in a transfer situation facilitate the ships may be constrained, however, we can specu-
transfer of tacit knowledge. These findings provide late that within-team relations may come at the
a more nuanced perspective on the effects of rela- expense of the development and maintenance of
792 Academy of Management Journal October

interunit relations. If managers see developing and Cummings, J. N. 2004. Work groups, structural diversity,
maintaining relations among members within a and knowledge sharing in a global organization.
unit, such as a team, as a good practice and encour- Management Science, 50: 352–364.
age it (cf. Ghoshal, Korine, & Szulanski, 1994), such Davenport, T., & Prusak, L. 1998. Working knowledge:
a management practice may inadvertently nega- How organizations manage what they know. Bos-
tively affect knowledge sharing across the organi- ton: Harvard Business School Press.
zation to the extent that it leads to fewer interunit Eisenhardt, K. M., & Santos, F. M. 2002. Knowledge-
relations and thus fewer knowledge-sharing events based view: A new theory of strategy? In A. Petti-
(cf. Ghoshal & Bartlett, 1990). grew, H. Thomas, & R. Whittington (Eds.), Hand-
In conclusion, our study shows that the problem book of strategy and management: 139 –164.
of knowledge sharing can be usefully conceptual- London: Sage.
ized as consisting of three phases that to a large Galunic, D. C., & Eisenhardt K. M. 2001. Architectural
extent are affected by an actor’s different subsets of innovation and modular corporate forms. Academy
social networks in an organization. As the various of Management Journal, 44: 1129 –1149.
effects of the degree of perceived competition, re- Gatignon, H., Tushman, M. L., Smith, W., & Anderson, P.
lation strength, and network size illustrate, disen- 2002. A structural approach to assessing innovation:
tangling interunit knowledge sharing into the three Construct development of innovation locus, type,
distinct phases of deciding to seek knowledge, and characteristics. Management Science, 48:
searching, and transferring can yield significant 1103–1122.
new insights into how network variables govern the Ghoshal, S., & Bartlett, C. A. 1990. The multinational
occurrence and effectiveness of knowledge sharing corporation as an interorganizational network.
in organizations. By adopting a multiple network Academy of Management Review, 15: 603– 625.
perspective on knowledge sharing, researchers can
Ghoshal, S., Korine, H., & Szulanski, G. 1994. Interunit
gain a fine-grained understanding of how different communication in multinational corporations. Man-
subsets of an actor’s relations in an organization agement Science, 40: 96 –110.
contribute to knowledge-sharing behaviors and
Granovetter, M. S. 1973. The strength of weak ties. Amer-
outcomes.
ican Journal of Sociology, 6: 1360 –1380.
Gulati, R. 1995. Does familiarity breed trust? The impli-
cations of repeated ties for contractual choice in
REFERENCES alliances. Academy of Management Journal, 38:
Aiken, L. S., & West, S. G. 1991. Multiple regression: 85–112.
Testing and interpreting interactions. Thousand Gupta, A. K., & Govindarajan, V. 2000. Knowledge flows
Oaks, CA: Sage. within multinational corporations. Strategic Man-
Ancona, D. G., & Caldwell, D. F. 1992. Bridging the agement Journal, 21: 473– 496.
boundary: External activity and performance in or- Hansen, M. T. 1999. The search-transfer problem: The
ganizational teams. Administrative Science Quar- role of weak ties in sharing knowledge across organ-
terly, 37: 634 – 665. ization subunits. Administrative Science Quarterly,
Austin, J. R. 2003. Transactive memory in organizational 44: 82–111.
groups: The effects of content, consensus, specializa-
Hansen, M. T. 2002. Knowledge networks: Explaining
tion, and accuracy on group performance. Journal of
effective knowledge sharing in multiunit companies.
Applied Psychology, 88: 866 – 878.
Organization Science, 13: 232–248.
Borgatti, S. P., & Cross, R. 2003. A relational view of
Hansen, M. T., Podolny, J. M. & Pfeffer, J. 2001. So many
information seeking and learning in social networks.
ties, so little time: A task contingency perspective on
Management Science, 49: 432– 445.
the value of corporate social capital in organizations.
Brewer, M. B. 1979. Ingroup bias in the minimal inter- In K. S. Cook & J. Hagan (Eds.), Research in the
group situation: A cognitive motivational analysis. sociology of organizations, vol. 18: 21–57. Green-
Psychological Bulletin, 86: 307–324. wich, CT: JAI Press.
Burt, R. S. 1992. Structural holes: The social structure Hayes, R. H., & Clark, K. B. 1985. Exploring the sources
of competition. Cambridge, MA: Harvard University of productivity differences at the factory level.
Press. New York: Wiley.
Cohen, J. 1988. Statistical power analysis for the be- Katz, R., & Allen, T. J. 1988. Investigating the not in-
havioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum. vented here (NIH) syndrome: A look at the perfor-
Cohen, W., & Levinthal, D. 1990. Absorptive capacity: A mance, tenure, and communication patterns of 50
new perspective on learning and innovation. Ad- R&D project groups. In M. L. Tushman & W. L. Moore
ministrative Science Quarterly, 35: 128 –152. (Eds.), Readings in the management of innovation
2005 Hansen, Mors, and Løvås 793

(2nd ed.): 293–309. New York: Ballinger/Harper & of intergroup behavior. In S. Worchel & W. G. Austin
Row. (Eds.), Psychology of intergroup relations (2nd ed.):
Koza, M. P., & Lewin, A. Y. 1999. The co-evolution of 7–24. Nelson Hall.
network alliances: A longitudinal analysis of an in- Tsai, W. 2001. Knowledge transfer in intraorganizational
ternational professional service network. Organiza- networks: Effects of network position and absorptive
tion Science, 10: 638 – 653. capacity on business unit innovation and perfor-
Levin, D., & Cross, R. 2004. The strength of weak ties you mance. Academy of Management Journal, 44: 996 –
can trust: The mediating role of trust in effective 1004.
knowledge transfer. Management Science, 50: Tsai, W. 2002. Social structure of “coopetition” within a
1477–1490. multiunit organization: Coordination, competition,
Manning, W. G., Duan, N., & Rogers, W. H. 1987. Monte and intraorganizational knowledge sharing. Organi-
Carlo evidence on the choice between sample selec- zation Science, 13(2): 179 –190.
tion and two-part models. Journal of Econometrics, Uzzi, B. 1997. Social structure and competition in inter-
35: 59 – 82. firm networks: The paradox of embeddedness. Ad-
Moulton, B. R. 1986. Random group effects and the pre- ministrative Science Quarterly, 42: 35– 67.
cision of regression estimates. Journal of Economet- Wasserman, S., & Faust, K. 1994. Social network anal-
rics, 32: 385–397. ysis: Methods and applications. New York: Cam-
Nohria, N., & Ghoshal, S. 1997. The differentiated net- bridge University Press.
work: Organizing multinational corporations for Wilder, D. A. 1978. Reduction of intergroup discrimina-
value creation. San Francisco: Jossey-Bass. tion through individuation of the out-group. Journal
Oakes, P. J., Haslam, S. A., Morrison, B., & Grace, D. 1995. of Personality and Social Psychology, 36: 1361–
Becoming an in-group: Reexamining the impact of 1374.
familiarity on perceptions of group homogeneity. Zander, U., & Kogut, B. 1995. Knowledge and the speed
Social Psychology Quarterly, 58: 52– 61. of transfer and imitation of organizational capabili-
Podolny, J. M., & Baron, J. N. 1997. Relationships and ties: An empirical test. Organization Science, 6:
resources: Social networks and mobility in the work- 76 –92.
place. American Sociological Review, 62: 673– 693.
Podsakoff, P. M., & Organ, D. W. 1986. Self-reports in
organizational research: Problems and prospects.
Journal of Management, 12: 531–544. Morten T. Hansen (morten.hansen@insead.edu) is a pro-
Reagans, R., & McEvily, B. 2003. Network structure and fessor at INSEAD, France. He was formerly an associate
knowledge transfer: The effects of cohesion and professor at Harvard Business School and received his
range. Administrative Science Quarterly, 48: 240 – Ph.D. from the Graduate School of Business at Stanford
267. University. His research has been concerned with the
role of knowledge in organizations and, specifically, with
Rogers, W. 1993. Regression standard errors in clustered
the problems of searching for and transferring knowledge
samples. Stata Technical Bulletin, 13: 19 –23.
across dispersed subunits in an organization.
Schulz, M. 2001. The uncertain relevance of newness:
Organizational learning and knowledge flows. Acad- Marie Louise Mors is a postdoctoral fellow at the MIT
emy of Management Journal, 44: 661– 681. Sloan School of Management. She received her Ph.D.
from INSEAD. Her research interests include social net-
Schulz, M. 2003. Pathways of relevance: Exploring in- works and processes of knowledge creation and sharing
flows of knowledge into subunits of multinational in global organizations.
corporations. Organization Science, 14: 440 – 459.
Bjørn Løvås was an assistant professor of strategic and
Sorenson, O. 2000. The effect of population level learn- international management at the London Business
ing on market entry: The American automobile in- School. He received his Ph.D. from INSEAD. His research
dustry. Social Science Research, 29: 307–326. and teaching focused on strategies for innovation and
Szulanski, G. 1996. Exploring internal stickiness: Imped- growth in knowledge and talent intensive industries,
iments to the transfer of best practice within the with emphasis on how firms manage and compete on the
firm. Strategic Management Journal 17(winter spe- informal relationships of their members.
cial issue): 27– 43.
Tajfel, H., & Turner, J. C. 1986. The social identity theory