Geography Reference
In-Depth Information
15.3).
14
We then weighted these parameters according to the patent's class assignments
(equation 15.4), where
p
ik
indicates the proportion of patent
k
's subclass memberships
that fall in class
i
.
Citations
j
(
before 7/90
)
Count of patents
j
in subclass
i
m
i
5
j
[
i
Average citations in patent class
i
(15.3)
;
Technology mean control patent
k
M
k
5
p
ik
m
i
(15.4)
;
The models also include controls for several other factors.
Same class
is a dummy
variable denoting whether the two patents in each dyad belong to the same primary
technological class.
Recent technology
is the mean of the patent numbers of the focal pat-
ent's prior art (higher numbers indicating more recent technology).
15
The models include
counts of two types of backward patent citation. First, they include a tally of the number
of citations to patent
prior art
. Second, the models include a control for the number of
non-patent prior art
citations (e.g. references to published articles).
Number of classes
is a
count of the number of major classes and
number of subclasses
is a count of the number
of subclasses to which the focal patent is assigned. Descriptive statistics appear in Table
15.1.
16
5. Results
The results appear in Table 15.2. Model 1 estimates the ef ects of the control variables
alone, and model 2 introduces interdependence,
k
.
Model 3 provides the i rst test of our core hypothesis by interacting interdepend-
ence with collaboration-based indicators of social proximity. The results provide three
pieces of support for the hypothesis. First, the positive sign on
k
x
Close collabora-
tion
coupled with the negative sign on
k
2
x
Close collaboration
indicates that the gap
in citation probability between close and unconnected inventors rises and then falls,
peaking when the source knowledge displays moderate interdependence. (Recall that
Unconnected
is the excluded category, so the coei cients related to
Close collaboration
capture dif erences between close and unconnected inventors.) Second, by subtracting
the coei cients for
Far collaboration
from the coei cients for
Close collaboration
, we see
that the largest gap between close and far inventors also appears for moderate
k
. Third,
the coei cient estimates suggest that the greatest gap between far and unconnected
inventors arises for moderate
k
(though with much smaller magnitude; see below). In
sum, our primary measure for social proximity provides strong support for our core
hypothesis.
17
Model 4 adds interactions of interdependence with geographic and organizational
proximity. Both proxies for social proximity display the expected inverted-U relation-
ship, though only the results for geographic proximity show strong statistical signii cance.
Coei cients for the collaboration-based measures retain their signs and signii cance, as
do most of the coei cients for the control variables.
Based on model 4, Figure 15.2 traces out, as a function of interdependence, how
many times more likely a citation is for collaboratively close pairs of inventors than for
unconnected pairs, for close pairs than for far pairs, and for far pairs than for uncon-
nected pairs. (We set all other variables to their mean values for the purpose of creating
this chart.) The i gure shows vividly that the maximal dif erence in citation probabilities
