Localized Knowledge Spillovers: Evidence from the Agglomeration of American R&D Labs and Patent Data

This working paper supersedes WP 12-22, WP 11-42, and WP 10-33. We employ a unique data set to examine the spatial clustering of private R&D labs, and, using patent citations data, we provide evidence of localized knowledge spillovers within these clusters. Jaffe, Trajtenberg, and Henderson (1993, hereafter JTH) provide an aggregate measure of the importance of knowledge spillovers at either the state or metropolitan area level. However, much information is lost regarding differences in the localization of knowledge spillovers in specific geographic areas. In this article, we show that such differences can be quite substantial. Instead of using fixed spatial boundaries, we develop a new procedure — the multiscale core-cluster approach — for identifying the location and size of specific R&D clusters. This approach allows us to better capture the geographic extent of knowledge spillovers. We examine the evidence for knowledge spillovers within R&D clusters in two regions: the Northeast Corridor and California. In the former, we find that citations are from three to six times more likely to come from the same cluster as earlier patents than in comparable control samples. Our results are even stronger for labs located in California: Citations are roughly 10 to 12 times more likely to come from the same cluster. Our tests reveal evidence of the attenuation of localization effects as distance increases: The localization of knowledge spillovers is strongest at small spatial scales (5 miles or less) and diminishes rapidly with distance. At the smallest spatial scales, our localization statistics are generally much larger than JTH report for the metropolitan areas included in their tests.