Geography Reference
In-Depth Information
organization,
but did not
report any US inventor.
14
The choice of excluding citations
from foreign companies implies that our study does not investigate the extent of inter-
national localization of patent citations. This choice was mainly dictated by data con-
straints, as the inclusion of citations coming from foreign organizations would have
implied the construction of a worldwide network of inventors. We also removed all
observations in which citing and originating patents were assigned to the same organiza-
tion (i.e. company self-citations).
We additionally excluded those patent pairs (citing-cited) in which the same indi-
vidual was listed in both documents (personal self-citations). Given that our focus was
especially on movers, we followed Agrawal et al. (2006) and excluded these cases, to the
extent that they could not represent any knowledge l ow between two distinct individu-
als. This left us with a total of 1933 originating (i.e. cited) patents, 2868 citing patents and
3517 citing-originating patent pairs.
In order to create a sample of control patents we randomly extracted, for each citing
patent, another patent with the following characteristics: no citation to the originating
patent and same IPC four-digit class and priority year of the citing patent. This pro-
cedure yielded a sample consisting of 3283 control patents for a total of 3517 control-
originating patent pairs (the same control patent can control for more than one citing
patent).
Next, we 'unbundled' individual inventors of the originating patents: from each patent
we extracted as many observations as the number of inventors listed on it, in order to
obtain two 'patent triples' (citing-cited-inventor and control-cited-inventor) for each
inventor. The resulting sample contains 10988 observations.
We then extracted information on the spatial and social distance between inventors
within each triple in the sample.
In the i rst place, we asked whether at least one inventor from the citing (control)
patent was located in the same current MSA of the inventor of the originating patent,
and created a dummy variable accordingly. We also asked the same question regarding
the prior location of the inventor, if the latter was a mover. Thus, for movers, we ended
up with two dummy variables for each triple, which took a value of one in the case of
geographical matching of the citing (control) patent with the mover's current and prior
location, respectively.
In the second place, for each patent triple, we measured the social distance between
the mover and the inventors of the citing (control) patents. Using information derived
from the network of inventors, we classii ed all patent triple into two (mutually exclusive)
groups according to the linkages connecting the individual inventors that have produced
them:
Connected
: These were patent triples in which there was at least one inventor from
the citing (respectively, control) patent that was connected to the inventor of the
originating patent through a i nite path in the co-invention network; we also cal-
culated the social distance within each triple as the geodesic distance between the
mover and the closest among the co-inventors of the citing (control) patent.
15
Unconnected
●
: These are patent triples whose respective teams of inventors were
not connected to each other in the co-invention network (although we could not
exclude the existence of other types of informal social ties).
●
