Discrete exponential family models for the study of ethnic residential segregation

Previous studies of ethnic residential segregation within urban environments have suggested several factors as potential determinants of the phenomenon. Physical characteristics of the urban environment, demographic and socioeconomic characteristics, preferences for neighborhood composition, and discrimination have all been identified as playing a major role in determining the spatial distribution of ethnic groups within modern cities. However, few studies have attempted to explicitly identify the manner in which residential patterns emerge from the interaction of these factors in field settings. Recently, important strides have been made in this direction by researchers using agent-based and cellular automata models; unfortunately, these efforts have been hampered by a lack of inferential tools to connect theoretical models with extant data.;Using a recently developed statistical framework based on discrete exponential family models that bridges this "inferential gap," this project addresses the following questions: (1) What are, in general, the conditions under which residential systems become segregated? (2) Given information on the current configuration of a residential system, how can we plausibly identify the mechanisms that could have led to that particular configuration? (3) How do metropolitan areas across the United States differ in terms of the mechanisms that drive residential settlement?;I first explore the circumstances in which settlement processes lead to segregation through simulations of both artificial and real urban residential systems with two racial/ethnic groups, focusing in particular on in-group preference (homophily) and out-group avoidance (xenophobia). I then turn to a sample of metropolitan areas and seek to identify the types of processes that could have led to their present residential system configuration, as captured by the 2000 U.S. census. My research shows that homophily and xenophobia shape residential systems in distinct ways: the former is associated with the clustering of households of the same ethnicity, the latter with the separation in space of households of different ethnicities. These mechanisms have the biggest segregating effect on residential systems when acting together. However, spatial residential patterns differ depending on whether one or both ethnic groups are homophilous, and the influence of homophily can be countered by even slight tendencies toward integration. The influence of xenophobia, on the other hand, is more pervasive, and can only be mitigated by equal or higher levels of in-group avoidance. These results suggest that reducing xenophobia and encouraging integration is a more effective policy lever than reducing homophily.;My analysis of Census 2000 data suggests that metropolitan areas in the United States that have only two major racial/ethnic groups show remarkable similarity in terms of xeno-phobia, the tendency of the groups to separate spatially based on ethnicity. On the other hand, they show much wider variation in the levels of ethnic homophily manifested by the minority and majority groups, which are linked mainly to the proportion minority in the metropolitan area: the bigger the minority group, the weaker the minority homophily effect and the stronger the majority homophily effect. I discuss the contribution of the project to our understanding of residential settlement and segregation and suggest some directions for future work.