Generative Models of Segregation: Investigating Model-Generated Patterns of Residential Segregation by Ethnicity and Socioeconomic Status

Abstract

This article considers the potential for using agent models to explore theories of residential segregation in urban areas. Results of generative experiments conducted using an agent-based simulation of segregation dynamics document that varying a small number of model parameters representing constructs from urban-ecological theories of segregation can generate a wide range of qualitatively distinct and substantively interesting segregation patterns. The results suggest how complex, macro-level patterns of residential segregation can arise from a small set of simple micro-level social dynamics operating within particular urban-demographic contexts. The promise and current limitations of agent simulation studies are noted and optimism is expressed regarding the potential for such studies to engage and contribute to the broader research literature on residential segregation.

[An appendix to this article is featured as an online supplement at the publisher's website.]

Keywords:

• agent-based modeling
• race and ethnicity
• segregation

Acknowledgments

The development of the SimSeg program used in this study was supported in part by NIH Grants R43-HD38199 (Simulating Residential Segregation Dynamics: Phase I) and R44-HD038199 (Simulating Residential Segregation Dynamics: Phase II). I also acknowledge the helpful comments of anonymous reviewers and the editors of this special issue.

Notes

¹An extended version of this article includes a more comprehensive listing of relevant studies pertaining to this topic, and also to other topics, that were omitted here to save space.

²See Fossett (2005) for a recent systematic review of this extensive literature.

³Beta distributions generate scores in the bounded range 0–100 and are flexible in terms of the shapes they can assume. The parameters for the beta distribution that yield a mean of 25 and a gini index of intra-group inequality of 35 are α = 2.999 and β = 5.570.

⁴When area stratification is high, distance from the city center explains 80% of variation in housing values. The figure is 0% when area stratification is low.

⁵The gini index for intra-group inequality is fixed at 35 for each group.

⁶Following Fossett (2006), the value of 80% is the median of a distribution of preference goals that is dispersed around this value based on a “logit-normal” distribution. The shape of the resulting preference distribution is similar to those documented in surveys.

⁷Following Fossett (2006), the value of 50% is the median of a distribution of preference goals that is dispersed around this value based on a “logit-normal” distribution. The shape of the resulting preference distribution is similar to those documented in surveys.

⁸Conventional measures of uneven distribution take positive values under random assignment (Winship, 1977). This bias is a potential concern when segregation is measured at small spatial scales as is typical in simulation studies (e.g., Goering, 2006). See the Appendix of the extended version of this paper for a more detailed discussion of the properties of S.

⁹The index of dissimilarity (D) is widely used in conventional segregation studies. But using D in simulation studies is highly problematic; its expected value under random assignment (E[D]) is high when segregation is measured at small spatial scales, particularly when group size is imbalanced (Winship 1977).

¹⁰Contact scores are computed with the focal household excluded. This adjustment eliminates positive bias in V when measuring segregation at small spatial scales.

¹¹I use a refined version _XP_X that excludes the focal household from the contact calculation. This eliminates bias when measuring isolation at small spatial scales.

¹²Possible motivations could include agent error, inaccurate information, compromises in location choices required to satisfy competing preferences, etc.

*Factors: U = Urban structure-area stratification in housing values; S = Minority SES disadvantage; P = Segregation-promoting ethnic preferences; A = Adjacent areas considered for preferences; D = Majority group prejudice and discrimination.

*Factors: U = Urban structure-area stratification in housing values; S = Minority SES disadvantage; P = Segregation-promoting ethnic preferences; A = Adjacent areas considered for preferences; D = Majority group prejudice and discrimination.

¹³Even small, unimportant differences tend to be significant at conventional levels. For example, t tests are statistically significant at least at 0.001 for differences of 0.8 or higher for comparisons on uneven distribution, isolation, and clustering and for differences of 2.5 or higher for comparisons on centrality.