Self-organizing maps and the US urban spatial structure

In this paper we consider urban spatial structure in US cities using a multidimensional approach. We select six key variables (commuting costs, density, employment dispersion and concentration, land-use mix, polycentricity, and size) from the urban literature and define measures to quantify them. We then apply these measures to 359 metropolitan areas from the 2000 US Census. The adopted methodological strategy combines two novel techniques for the social sciences to explore the existence of relevant patterns in such multidimensional datasets. Geodesic self-organizing maps (SOM) are used to visualize the whole set of information in a meaningful way, while the recently developed clustering algorithm of the max-<i>p</i> is applied to draw boundaries within the SOM and analyze which cities fall into each of them. <br> <b>Keywords:</b> urban spatial structure, self-organizing maps, US metropolitan areas