A nonparametric test for industrial specialization

Abstract

Urban economists hypothesize that industrial diversity matters for urban growth and development, but metrics for empirically testing this relationship are limited to simple concentration metrics (e.g. location quotient) or summary diversity indices (e.g. Gini, Herfindahl). As shown by recent advances in how we measure localization and specialization, these measures of industrial diversity may be subject to bias under small samples or the Modifiable Areal Unit Problem. Furthermore, empirically examining industrial diversity requires statistically testing for patterns of industry mix that deviate from random firm location. Extending recent work by [S. Billings & E. Johnson. 2012. A Nonparametric Test for Industrial Specialization. Journal of Urban Economics. 71(3):312-331.], we develop a nonparametric microdata based test for industrial co-specialization. Our test employs establishment densities for specific pairs of industries, a population counterfactual, and a new correction for multiple hypothesis testing to determine the statistical significance of co-specialization across both places and industries. The results of these pairwise tests are then mapped out as networks of proximate industries unique to each place within our study area. We use pairs and triads of industries to highlight specific four digit industries that may drive co-specialization and a larger network of industrial diversification. Results give us new understanding of the relationship between industrial co-specialization and urbanization, with manufacturing industries tending to be more co-specialized in less dense areas than business services, while business services show more connected and transitive spatial networks. Finally, we discuss the role that intransitivities in industry triads may play in the econometric identification of co-specialization and underlying place specific agglomerative forces.