On specifying heterogeneity in knowledge production functions

Within the Geography of Innovation literature, the Knowledge Production Function approach has become a reference framework to investigate the presence of localized knowledge spillovers and spatial econometric tools have been applied to study interregional spillovers. A linear specification for the KPF is assumed linking patents to R&D expenditure. This approach however suffers of different drawbacks. First patent applications are count data in nature. Patents per inhabitants may produce an unrealistic picture of the spatial distribution of innovative activities. Secondly, spatial heterogeneity is not usually observed, producing both omitted variables bias and spatial correlation in the error structure. Third, a positive R&D-patents linkage may arise as a spurious correlation if market size is not observed, causing R&D to be endogenous. This paper uses a regional cross section model to study the spatial distribution of high tech patents across 232 European regions in the period 2005/2006 to address these issues. Two main processes drive technological change in the model: research activities and knowledge generated outside firms and in a second moment embedded through either formal or informal acquisition. Among the different knowledge sources we particularly focus on the role of firms working in Knowledge Intensive Business Services and on that of universities. In developing the empirical model we take into account that a) patents are count data; b) the exclusion of market size will cause biased and inconsistent model parameters estimates; c) estimates of interregional spillovers may be biased by the omission of heterogeneity in the model specification. Empirical results indicate that, as expected, a count data distribution best fits the data, producing less spatially autocorrelated residuals. Regional innovative activity is explained by both investments in research and localization of KIBS, but only the first generates positive interregional externalities. Scientific universities do not directly affect the production of new knowledge. However, different knowledge production processes characterize regions with and without scientific universities, with R&D driving innovation in the sooner and KIBS in the latter. Finally, most of what are assumed to be interregional spillovers reveal to be, at a more careful inquiry, effect due to unaccounted spatial heterogeneity in regional innovation.