Geography Reference
In-Depth Information
conversations with the US Patent Oi ce indicate that applicants often intentionally
obfuscate their descriptions to diminish the value of the knowledge revealed (Stern,
2001).
Second, the use of citations as an indicator of knowledge l ows has been cast into
doubt recently by the work of Alcacer and Gittelman (2004), who i nd that examiners
add 40 percent of the citations found on US patents. On the one hand, this i nding is
comforting as it suggests that examiners actively work to prevent applicants from exclud-
ing citations to relevant prior art for strategic reasons, such as those mentioned above.
It is nonetheless potentially problematic for our study to the extent that examiners most
frequently insert socially proximate citations to patents of intermediate interdependence.
The few studies that analyze the characteristics of examiner-added citations, however,
show no evidence of such a bias (Alcacer and Gittelman, 2004; Sampat, 2004). Indeed,
self
-citations - which almost certainly rel ect true knowledge l ows - as frequently come
from examiners as from inventors. This suggests to us that, on balance, examiner inter-
vention
improves
the quality of patent data for our purposes and cannot account for our
results. Consistent with this conclusion, Duguet and MacGarvie (2005) i nd that i rms'
patent citation patterns match their survey responses regarding technology acquisition
and dispersion. At worst, if examiners add citations that do not rel ect true knowledge
l ows and do so in an unbiased way, this should only add noise, increasing the dii culty
of i nding statistical support for our hypothesis.
Third, patents admittedly of er imperfect measures of invention. Inventors may limit
their patent applications to a subset of their discoveries, and one must ask whether
this selection process biases our results. Inventors most likely seek legal protection
when a patent raises a meaningful barrier to imitation (e.g. when inventing around the
patent proves dii cult), when the invention will not quickly become obsolete, and when
few alternative 'natural' defenses protect the knowledge (Levin et al., 1987). Of these
conditions, the last seems most germane to our study. It implies that our sample may
under-represent inventions that involve highly tacit, causally ambiguous and complex
knowledge. Empirical research, however, suggests that this selection bias may not exist:
Cohen et al. (2000), for example, i nd that i rms in industries with complex products dis-
proportionately choose to patent.
Finally, we recognize that patents represent but one embodiment of knowledge.
Though we have no reasons to expect a priori
that they should dif er from other pieces of
knowledge, they may. Despite this potential limitation on the scope of the applicability
of our results, patents of er an excellent i rst test bed for our ideas for the reasons noted
above.
Case- control design
Our unit of analysis is a patent dyad, one patent issued in May or June of 1990 and one
issued later that may or may not cite the i rst. Hence our approach conceptually follows
that of other studies of the likelihood of tie formation - in this case, the likelihood that
a future patent builds on the knowledge embodied in one of our focal patents. These
studies have typically estimated tie formation on the entire matrix of possible rela-
tions (e.g. Gulati, 1995; Podolny, 1994). This approach has two disadvantages. With
large numbers of nodes, in this case patents, it can generate enormous, sparse matrices,
increasing the dii culty of estimation and variable construction. In our situation, this
