INSEAD Working Paper 2011/44/ST revised version of 2010/03/ST
Geographic localization of knowledge spillovers is a long-held tenet of economic geography. However, empirical research has examined this phenomenon by considering only one geographic unit (country, state or metropolitan area) at a time, and has not accounted for spatial distance in such analyses. We accomplish both using a choice-based sampling framework to estimate the likelihood of knowledge flow, as represented by a citation between random patents. In addition to a robust country effect, we find a puzzling persistence of state-level localization that cannot be explained merely as an outcome of spatial proximity. This effect is found to be more than just a manifestation of greater mobility or closer networks within states, suggesting a role for state-level institutions. As a demonstration that state-level policy could influence knowledge flow patterns, we find that a natural experiment wherein Michigan inadvertently started enforcing non-compete agreements indeed led to a decrease in localized spillovers in the state.