A Probabilistic Modeling Approach to the Detection of Industrial Agglomerations: Methodological Framework

Dating from the seminal work of Ellison and Glaeser [11] in 1997, a wealth of evidence for the ubiquity of industrial agglomerations has been published. However, most of these results are based on analyses of single (scalar) indices of agglomeration. Hence it is not surprising that industries deemed to be similar by such indices can often exhibit very different patterns of agglomeration - with respect to the number, size, and spatial extent of individual agglomerations. The purpose of this paper is thus to propose a more detailed spatial analysis of agglomeration in terms of multiple-cluster patterns, where each cluster represents a (roughly) convex set of contiguous regions within which the density of establishments is relatively uniform. The key idea is to develop a simple probability model of multiple clusters, called cluster schemes, and then to seek a “best” cluster scheme for each industry by employing a standard model-selection criterion. Our ultimate objective is to provide a richer characterization of spatial agglomeration patterns that will allow more meaningful comparisons of these patterns across industries