DISCUSSION
At the beginning of this paper, we raised a question about
the contributiono f network structuret o the knowledge transfer
process. Previous research had suggested that network
structure was an integral part of the transfer process, but the
"network effect," while widely recognized, had not been
examined directly.W e focused on two distinctf eatures of
network structure, cohesion and range. The evidence indicated
that both network features facilitated knowledge transfer.
The findings are important because they clarify and extend
past research.
The findings clarify and extend the role of strong ties in the
transfer process. Strong connections have occupied a privileged
position in the knowledge transfer process, in part
because such connections are assumed to occur within a
dense web of affiliations. We found that strong ties and
social cohesion were correlated but that it was a mistake to
equate their effects. Each feature made a distinct contribution
to the knowledge transfer process. The benefits provided
by a strong tie did not require social cohesion. The evidence
also provided an important boundary condition for
strong connections. Previous research had assumed that
strong ties were even more valuable for the transfer of
knowledge that was tacit or difficult to codify. Hansen (1999),
in his analysis of team performance, found that team performance
increased as a function of the strength of connections
from the team to the broader organization and also as a function
of the kind of knowledge being transferred by the team.
And team performance was even higher when strong ties
were used to transfer tacit knowledge. Based on these findings,
Hansen concluded that a strong tie facilitated the transfer
of tacit knowledge more than it facilitated the transfer of
codified knowledge. He concluded that strong ties should be
used for the transfer of tacit knowledge and weak ties for the
transfer of codified knowledge (Hansen, 2002). We tested
this idea in our analysis. We found some support for the initial
conclusion, that strong ties facilitated the transfer of tacit
knowledge more than they facilitated the transfer of codified
knowledge. But the evidence was weak. Moreover, the
effect disappeared altogether once controls for cohesion and
range were introduced into the model. The contingent effect
of tie strength was actually tapping into the effect for network
structure, providing further support for the need to distinguish
tie strength from network structure in empirical
analysis.
Although we did not find a contingent effect of tie strength
on knowledge transfer, it does not mean that Hansen was
incorrect when he asserted that it is best to match the type of tie to the type of knowledge being transferred. Our results
showed that it is easier to transfer all kinds of knowledge in a
strong tie and more difficult to transfer all kinds of knowledge
in a weak tie. Our results also showed that tacit knowledge
was more difficult to transfer than codified knowledge.
Combined, the two results indicate that it is more efficient to
use strong ties to transfer tacit knowledge and weak ties to
transfer codified knowledge. Given that strong ties require a
greater investment of time, it is inefficient to use strong ties
to transfer codified knowledge. Time spent using a strong tie
to transfer codified knowledge could be spent transferring
tacit knowledge. The greater efficiency here is based on
matching type of tie to type of knowledge. That matching
does not require an interaction between tie strength and type
of knowledge. A significant interaction between tie strength
and type of knowledge would have implied that individuals
exerted significantly more effort during the transfer of tacit
knowledge than they exerted during the transfer of codified
knowledge.
The estimates of the nonlinear effects in model 9 provided
further support for the matching hypothesis. Those estimates
were based on the logs of tie strength, network density, and
network diversity. The model provided the best fit to the
data. Given the nonlinear effect that tie strength had on
knowledge transfer,i t makes sense for an individuatlo allocate
just enough network time and effort to facilitate transfer
and then to allocate the rest of his or her time and effort to
other knowledge transfer relationships. After some point, the
marginalr eturnst o additionalt ime and effort begin to
decline. The individuawl ould be better off allocatingt he additional
time and effort to a different knowledge transfer relationship,
where the time and effort would have more of an
impact.
At the same time, to say that it is more efficient to use weak
ties to transfer codified knowledge, as in the matching
hypothesis, is not to say that network range plays a limited
role during knowledge transfer, as Hansen concluded. The
estimates in table 4 indicated that range eased transfer.
Therefore, by equating weak ties with network range (or
boundary spanning ties), previous research has neglected the
important role that range plays in the transfer process. Our
findings, therefore, clarify existing research by identifying
how strong ties contribute to knowledge transfer but also
when they do not. We extend previous work by identifying
the important role that network range plays during the transfer
process as well, an effect that has been ignored in past
work.
Our work also has important implications for the study of network
structure in general. Previous research has focused on
the benefits of knowledge transfer across a structural hole
(e.g., Stuart and Podolny, 1996) but has not addressed the
problematic nature of such transfers. Presumably, people at
opposite ends of a structural hole do not have much knowledge
in common, which can impede knowledge transfer. The
baseline effects in table 4 illustrated this point. A structural
hole existed between two individuals when those individuals
were not connected through strong third-party ties (the