Geography Reference
In-Depth Information
The descriptive statistics we presented in section 3 suggest that the dif erent composition
of the citing and control sample may be responsible for this result. We know that the citing
patents are much more likely than the control ones to be socially connected to the originat-
ing ones, and that the cross-i rm inventors responsible for the connection are most often
immobile in space. Hence we expect that controlling for personal connections may reduce
the observed dif erences in co-location rates between citing and control patents.
As the citing patent sample contains a higher number of socially connected patents
with geodesic distance equal to or lower than 5 (which we know to be highly co-located
with originating ones), we also expect to observe a similar result when controlling for
social connections.
The 'only connected' line of Table 16.8 shows that by excluding all the unconnected
patent triples from both the citing and the control samples,
17
the co-location percent-
ages of both the citing and control samples increase. On the contrary, when considering
only the unconnected patents (third line) the co-location percentages go down, although
the dif erence between citing and control patents remains signii cant. When considering
connected patents, the dif erence in co-location percentages between citing and control
samples is signii cantly higher than when considering unconnected ones. This result
suggests that professional ties, as measured in our network of inventors, may explain a
sizeable part of the JTH i ndings. However, the persistence of some locational dif erences
between citing and control triples leaves open the possibility that social ties other than
those captured by our network may also af ect inventors' knowledge exchanges, and
result in patent citations.
Analysis: the role of movers
In order to focus on movers, we restricted our sample to those triples wherein the inven-
tor of the originating patent had signed at least one patent before the originating one(s),
and he/she had done so when residing in a dif erent MSA.
18
Since movers are a very small
subset of all inventors (see section 3) this restriction caused a dramatic drop in our sample
size, from over 10,000 observations to just 594, resulting from 287 originating patents,
279 inventors, 477 citing patens and 499 controls. This i gure represents approximately
5.5 per cent of the initial sample, a percentage that is consistent with the one found by
Agrawal et al. (2006) that is approximately 6 per cent.
The distribution of patent triples between connected and unconnected ones does not
change with regard to the original sample (36 per cent connected triples among citing
ones; 26 per cent connected among controls). The path length distribution of connected
patents, however, changes slightly, with more citing triples connected with paths shorter
than six degrees of separation than in the general sample (34 per cent vs. 24 per cent;
distribution for control sample does not change much; see Table 16.9).
The analysis of co-location patterns reveals some interesting results. Citing patents
do not appear to be more likely to come from the same current locations of movers,
no matter whether we consider the connected or the unconnected ones, or all (Table
16.10).
On the contrary, when we consider movers' prior location, we i nd that citing patents
still tend to be more likely to be signed by inventors from those locations (11 per cent
vs. 8 per cent; see Table 16.10); however, when considering social ties, this result holds
(indeed is reinforced) only when patents are socially connected (14 per cent vs. 8.5 per
