Understanding the social context of the Schelling segregation model

doi:10.1073/pnas.0708155105

Understanding the social context of the Schelling segregation model

William A. V. Clark^a,^b and
Mark Fossett^c

^aUniversity of California, Los Angeles, CA 90095; and
^cTexas A&M University, College Station, TX 77843

Edited by Susan Hanson, Clark University, Worcester, MA, and approved January 22, 2008 (received for review August 30, 2007)

Abstract

A recent article [Vinkovic D, Kirman A (2006) Proc Natl Acad Sci USA 103:19261–19265] showing that the Schelling model has a physical analogue extends our understanding of the model. However, prior research has already outlined a mathematical basis for the Schelling model and simulations based on it have already enhanced our understanding of the social dynamics that underlie the model, something that the physical analogue does not address. Research in social science has provided a formal basis for the segregative outcomes resulting from the residential selection process and simulations have replicated relevant spatial outcomes under different specifications of the residential dynamics. New and increasingly detailed survey data on preferences demonstrates the embeddedness of the Schelling selection process in the social behaviors of choosing alternative residential compositions. It also demonstrates that, in the multicultural context, seemingly mild preferences for living with similar neighbors carry the potential to be strong determinants for own race selectivity and residential segregation.

Footnotes

^bTo whom correspondence should be addressed. E-mail: wclark{at}geog.ucla.edu
Author contributions: W.A.V.C. and M.F. designed research, performed research, analyzed data, and wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
↵ d SimSeg can implement two- and three-group simulations. The shift from two to three groups is sufficient to reveal many differences in the implications of preferences in the multigroup context.
↵ e This is a conservative setting because specifying groups as equal in size is optimal for maintaining integration in Schelling-style segregation models (22). Higher levels of segregation result when we use more realistic demographic distributions (e.g., 65, 25, and 10).
↵ f Targets for same-group presence are the same for all groups. We ran simulations in which preference targets were dispersed around the mean (as seen in the distributions in Fig. 2) and simulations in which targets were homogeneous. The results were similar either way, indicating that, in the main, preference effects revolve around the central tendency of the preference distribution.
↵ g This specification is used in recent studies (22, 23) and is similar to the 5 × 5 square neighborhoods Schelling used in some simulations (2). Significantly, preference effects are robust over varying specifications of neighborhood size, shape, and type (e.g., bounded areas versus site-centered areas).
↵ h The results shown here are from representative simulation experiments. We performed 500 replications for each design to establish the statistical norm for quantitative results.
↵ i V is a well known measure with many attractive technical properties. We also computed scores for the more widely used dissimilarity index (D). Reporting V is a conservative choice; scores for D are always higher and suggest even more dramatic segregation results.
↵ j We omit results for simulations with preferences set to zero. Not surprisingly, segregation did not emerge in these simulations.
↵ k The shift from small-to-large scale segregation patterns occurs at low levels of agent vision (e.g., 4–16 units). Computational burden increases geometrically as local agent vision increases. Intermediate vision produces large-scale segregation with low computational burden.
↵ l When multiple factors affect segregation, quantitative assessments of their separate effects depend critically on how the analysis is performed. When each is “added” or “removed” from a model system in which no other segregation dynamics are active, the impact can be large. However, when they are “added” or “removed” from a model system in which multiple segregation dynamics are active, their impact is usually small.