Geography Reference
In-Depth Information
Table 16
.
7
Geographic matching: % frequency for socially connected patents at the
MSA level (citing and controls)
Geodesic distance
Citing (no. of triples)
Control (no. of triples)
1-5
29.4
29.5
6-10
13.5
9.6
11-20
7.0
7.0
>20
28.8
4.8
Unconnected
9.7
8.2
All
12.0
9.0
Table 16
.
8
Geographical matching at MSA level: % frequency and test of proportions*
No. of obs
Citing:
% match
Control:
% match
z
- statistic
(
P
>
z
)
All inventors (JTH experiment)
10988
12.01
9.05
7.16***
Only connected citing-cited
4779
14.69
10.40
6.33***
patents, and related controls
Only unconnected citing-cited
6209
9.95
8.00
3.80***
patents, and related controls
Notes:
* One-tail z-test on dif erence between co-location percentages: 2.81 (0.0234)***
Statistically signii cant at 1%***, 5%**, 10%*. One tail
z
- test
P
(
Z
>
z
).
than 20 degrees of separation. Overall, these results suggest that, once we control for
social ties, there is no reason to expect an association between geographical proximity
and citations.
Analysis: the JTH experiment
As a i rst step in our analysis, in the 'all inventors' row of Table 16.8 we replicate the JTH
exercise on our data. The second and third columns of the table report the percentage of,
respectively, citing and control patents that are co-located with the originating ones, at
the MSA level, while the fourth column reports the
z
-statistic for the dif erence between
the two and (in brackets) the result of a 1-tail test on the hypothesis that
p
c
>
p
nc
(where
p
c
is the co-location probability of citing patents, and
p
nc
the co-location probability of
controls).
16
The geographic matching rates we found for both the citing and the control patents
were almost double those reported by JTH (at the city level). To the extent that locali-
zation ef ects are expected to fade with time, these dif erences may be explained by the
shorter period of observation we used (9 years maximum, from 1991 to 1999, in our
sample, against a maximum of 14 years in the original JTH exercise).
As for dif erences between citing and control patents, however, our results are very
close to the original JTH ones. The proportion of citing patents co-located with originat-
ing ones is signii cantly greater than the proportion of control patents.
