Geography Reference
In-Depth Information
9. Some might worry about the stability of this measure over time. To test its robustness, we constructed
a second k measure using data from 1790 to 1990. That measure yielded a qualitatively identical set of
results.
10. Breschi and Lissoni (2002) independently developed an equivalent approach.
11. Though the magnitude of the gap shrinks, our results remain qualitatively robust to shifting the dividing
line between close and far from a path length of three to a length of four. We use three categories rather
than the distance measure itself for three reasons: (1) calculating the precise distance for the longer paths
in these data would increase the time required to compute it by orders of magnitude (i.e. by months); (2)
dummy variables for individual path lengths lead to some small cell sizes and concomitantly unstable
coei cient estimates; and (3) given our interaction with a quadratic, we i nd the results of the categorical
coding far easier to interpret and understand.
12. All patents list the home address of the inventor on the front page of the patent application. To locate
each inventor, we match the inventor's 3-digit zip code to the latitude and longitude of the center of the
area in which the inventor resides based on information from the US Postal Service. We then use spheri-
cal geometry to calculate the distance between the points. The USPTO includes 5-digit zip information,
but we choose to reduce measurement error by using cleaned data. CHI, an information provider, has
called every patent holder to verify the inventor's location; however, it records this information only at
the 3-digit level.
13. We mean-deviate the variables before creating the interaction terms to facilitate interpretation of the
ef ects (Friedrich 1982). For collaborative proximity, we use Unconnected ij as the excluded category.
14. We allow all patents issued between January 1985 and 30 June 1990 to enter the estimation of the activ-
ity control, meaning that the patents used to calculate it vary in the time during which they can receive
citations. Alternatively, we could select a small set of patents from 1985 and base the measures on the
subsequent i ve years of citations; however, this approach would ignore the patent activity just prior to
our sample.
15. This variable made use of the fact that the USPTO assigns patent numbers sequentially. This assignment
pattern generates a correlation between a patent number and the grant date of the patent of 0.98.
16. We also considered as a control variable the time between the issuance dates of the focal and potentially
citing patents in each dyad. Exploratory analysis revealed small ef ect sizes (though typically signii cant),
and inclusion of the time control had no meaningful impact on the coei cients of central interest.
17. Since the high correlation between a term and its square can force estimates to take opposing signs, we
further tested the validity of our non-monotonic ef ect in two ways: (1) in unreported estimates (available
from the i rst author), we re-estimated the models using a log-quadratic specii cation and found quali-
tatively identical results. Since this functional form can capture decreasing returns without a signii cant
coei cient on the quadratic term, it is less sensitive to these problems; (2) we estimated a model with only
the linear term and interactions and then entered the quadratic terms. In all cases, the addition of the
quadratic terms signii cantly improved the model. (For example, in model 4, the addition of the quadratic
k and its interactions has a c 2 = 70.4, signii cant at p < .00001 with i ve degrees of freedom.)
18. These i gures assume that the two inventors are 665 miles from one another (the average distance in our
sample) and work for dif erent organizations.
19.
This pattern seems consistent with the evolution of the software industry, for instance. Early on, knowl-
edge localized to an extreme: understanding of a new piece of code resided in the head of a single devel-
oper or a small group of developers in a university, government, or large corporate computing facility.
Inventors developed local languages for specii c hardware. Over time, programmers developed tech-
niques for reducing the interdependencies in code. Higher-level languages such as Cobol and C allowed
programmers to divorce code from specii c hardware. Meanwhile, software production has dispersed
geographically - beyond Silicon Valley, Route 128, and IBM's Armonk home, to Seattle, Austin, and
even Bangalore.
Bibliography
Alcacer, Juan and Michelle Gittelman (2004), 'How do I know what you know? Patent examiners and the
generation of patent citations', Review of Economics and Statistics , 88 , 774-9.
Allen, Thomas J. (1977), Managing the Flow of Technology: Technology Transfer and the Dissemination of
Technological Information Within the R&D Organization , Cambridge, MA: MIT Press.
Argote, Linda (1999), Organizational Learning: Creating, Retaining and Transferring Knowledge , Boston:
Kluwer.
Arrow, Kenneth J. (1962), 'Economic welfare and the allocation of resources for invention', in Richard Nelson
(ed.), The Rate and Direction of Inventive Activity , Princeton, NJ: Princeton University Press, pp. 609-24.
Arrow, Kenneth J. (1974), The Limits of Organization , New York: Norton.
Search WWH ::




Custom Search