INTERNATIONAL KNOWLEDGE FLOWS: RESULTS 6

All of the results discussed so far derive from estimation of the “full” model of Eq. 3. For comparison to our earlier paper, as well as for the light it sheds on the interaction of different effects, it is useful to consider briefly how the results differ in less complete or different models. In particular, our earlier research did not exclude self-citations, and did not include the “technological proximity” effect. These effects are interesting in their own right, and may also be expected to interact in important ways with geographic localization. Table 5 summarizes the results with and without these non-geographic effects. Generally, excluding self-cites significantly reduces the apparent geographic localization, as well as reducing the extent to which that localization “fades.”

That is, the citation intensity from other countries, relative to the domestic citation rate, is lower in columns 1 and 2 than in columns 3 and 4 in the first year, but is higher in columns 1 and 2 than in columns 3 and 4 after 20 years. What this means is that self-cites are highly geographically localized (which should not be a surprise) and generally come at shorter lags (Trajtenberg, Henderson and Jaffe, 1997). Thus including them creates strong localization particularly in early years; excluding them dilutes localization; this weaker initial localization then also fades less.

w6507-17
w6507-18
Inclusion of the technological proximity parameter has an effect similar to the exclusion of self-cites, but much smaller. That is, except for citations to the U.S., Column 3 shows slightly less localization than Column 4 (and Column 1 slightly less than Column 2), whereas both Columns 1 and 2 show dramatically less than either 3 or 4. What this suggests is that citations within the same patent class have a slight tendency to geographic localization, but, not surprisingly, much less so than citations within the same organization. Finally, there does not appear to be much interaction between the self-cite and technological proximity effects. The parameter у is not much different in column 2 from in column 4. What this means is that self-citations exhibit approximately the same tendency toward concentration in the same patent class as non-self-citations.