How diverse can measures of segregation be? Results from Monte Carlo simulations of an agent-based model *

Abstract

Cultural diversity is a complex and multi-faceted concept. Commonly used quantitative measures of the spatial distribution of culturally defined groups—such as segregation, isolation or concentration indexes—have been designed to capture just one feature of this distribution. The strengths and weaknesses of such measures under varying demographic, geographic and behavioral conditions can only be comprehensively assessed empirically. This has been rarely done in the case of multigroup cultural diversity. This paper aims to fill this gap and to provide evidence on the empirical properties of various segregation indexes by means of Monte Carlo replications of agent-based modelling simulations under widely varying assumptions. Schelling’s classical segregation model is used as the theoretical engine to generate patterns of spatial clustering. The data inputs include the initial population, the assumed geography, the number and shares of various cultural groups, and their preferences with respect to co-location. Our Monte Carlo replications of agent-based modelling data generating process produces output maps that enable us to assess the sensitivity of the various measures of segregation to parameter assumptions by means of response surface analysis. We find that, as our simulated city becomes more diverse, stable residential location equilibria require the preference for co-location with one’s own group to be not much more than the group share of the smallest demographic minority. When equilibria exist, the values of the various segregation measures are strongly dependent on the composition of the population across cultural groups, the assumed preferences and the assumed geography. Index values are generally non-decreasing in increasing preference for within-group co-location. More diverse populations yield—for given preferences and geography—a greater degree of spatial clustering. The sensitivity of segregation measures to underlying conditions suggests that meaningful analysis of the impact of segregation requires spatial panel data modelling.