We study the location and productivity of more than 1,000 research and development (R&D)Â labs located in the Northeast corridor of the U.S. Using a variety of spatial econometricÂ techniques, we find that these labs are substantially more concentrated in space than theÂ underlying distribution of manufacturing activity. Ripleyâ€™s K-function tests over a variety ofÂ spatial scales reveal that the strongest evidence of concentration occurs at two discrete distances:Â one at about one-quarter of a mile and another at about 40 miles. These findings are consistentÂ with empirical research that suggests that some spillovers depreciate very rapidly with distance,Â while others operate at the spatial scale of labor markets. We also find that R&D labs in someÂ industries (e.g., chemicals, including drugs) are substantially more spatially concentrated thanÂ are R&D labs as a whole.
Tests using local K-functions reveal several concentrations of R&D labs (Boston, New York-Northern New Jersey, Philadelphia-Wilmington, and Washington, DC) that appear to representÂ research clusters. We verify this conjecture using significance-maximizing techniques (e.g.,Â SATSCAN) that also address econometric issues related to â€œmultiple testingâ€ and spatialÂ autocorrelation.
We develop a new procedure for identifying clusters â€“ the multiscale core-cluster approach â€” toÂ identify labs that appear to be clustered at a variety of spatial scales. We document that whileÂ locations in these clusters are often related to basic infrastructure, such as access to major roads,Â there is significant variation in the composition of labs across these clusters. Finally, we showÂ that R&D labs located in clusters defined by this approach are, all else equal, substantially moreÂ productive in terms of the patents or citation-weighted patents they receive.