Multiscale Segregation: Multilevel Modeling of Dissimilarity—Challenging the Stylized Fact That Segregation Is Greater the Finer the Spatial Scale

Abstract

A very large literature has explored the intensity of urban residential segregation using the index of dissimilarity. Several recent studies have undertaken such analyses at multiple spatial scales, invariably reaching the conclusion that the finer grained the spatial scale, the greater the segregation. Such findings, however, overstate the intensity of segregation at finer spatial scales because they fail to take into account an argument made by Duncan et al. some sixty years ago that indexes derived from fine-scale analyses must necessarily incorporate those from coarser scales, with the consequence that finer scale segregation is invariably overestimated. Moreover, most studies ignore stochastic variation that results in upward bias in the estimates of segregation. This article demonstrates the importance of a recently developed multilevel modeling procedure that identifies the “true” intensity of segregation at every level in a spatial hierarchy net of its intensity at other levels and also net of stochastic variation This is illustrated by both a simulated data set and an empirical study of an English city, with the latter raising important substantive issues regarding the interpretation of segregation patterns and the processes underlying them. Key Words: dissimilarity, multilevel modeling, scale, segregation.