Measuring the Effects of Segregation in the Presence of Social Spillovers: A Nonparametric Approach

Bryan S. Graham, Guido W. Imbens, Geert Ridder2010 · DOI

Introduction

Social scientists have long debated the impact of segregation—whether by race, gender, or socioeconomic status—on individual and collective outcomes. While the existence of "peer effects" is widely accepted, measuring the actual impact of policy-driven reallocations remains a significant econometric challenge. In their paper, "Measuring the Effects of Segregation in the Presence of Social Spillovers," Bryan S. Graham and Geert Ridder introduce a nonparametric framework designed to evaluate how moving individuals between social groups affects average outcomes and inequality.

The Challenge of Associational Redistribution

Most existing research on peer effects asks a simple question: "What happens if we add one more 'high-type' individual to a group?" However, as economist Steven Durlauf has noted, policymakers rarely have the luxury of simply adding resources. Instead, they face a problem of associational redistribution: to add a high-achieving student to one classroom, they must remove them from another.

Traditional parametric models often assume a linear relationship between group composition and outcomes. This assumption is risky; if the real-world relationship is non-linear—for example, if the benefits of peer quality diminish after a certain threshold—linear models will provide misleading policy advice. Graham and Ridder’s nonparametric approach avoids these assumptions, allowing the data to reveal the true shape of social spillovers.

A New Nonparametric Toolkit

The authors propose three primary metrics (estimands) to quantify the effects of reallocation:

  1. Average Spillover Effect (ASE): A global measure of the strength of peer effects within a population.
  2. Local Segregation Outcome Effect (LSOE): This measures how a marginal increase in segregation from the current status quo affects the overall average outcome. It balances "complementarity" (whether high types benefit more from each other) against "curvature" (diminishing returns).
  3. Local Segregation Inequality Effect (LSIE): This quantifies how changes in segregation affect the achievement gap between high and low types.

Beyond these metrics, the paper addresses the Social Planner’s Problem. If an administrator could perfectly assign individuals to groups to maximize total welfare, what would the optimal distribution look like? The authors demonstrate that the optimal solution often involves a mix of highly integrated and highly segregated groups, depending on the "concave envelope" of the production function and the total availability of "high-type" individuals.

Case Study: Gender Segregation in Project STAR

To demonstrate their framework, the authors analyze data from Project STAR, a large-scale randomized experiment in Tennessee. They focus on how the gender composition of kindergarten classrooms (the proportion of girls) affects mathematics achievement. Because students and teachers were randomly assigned, the data provides a clean environment to identify causal peer effects without the bias of self-sorting.

Key Findings:

  • Positive Spillovers: Both boys and girls perform significantly better in math when placed in classrooms with a higher proportion of girls.
  • The Neutrality of Segregation on Averages: Interestingly, the LSOE for gender segregation was found to be near zero. This means that while individual classrooms change, the total average math score across the school system would remain unchanged by a small increase in gender segregation. The gains for girls in female-heavy classrooms are almost perfectly offset by the losses for boys in male-heavy classrooms.
  • Widening the Gap: While average scores remain stable, the LSIE indicates that segregation would significantly increase the achievement gap. In this context, girls (the higher-performing group) would pull further ahead, while boys would fall further behind.
  • Optimization Potential: The authors’ simulation suggests that a social planner could increase average math scores by approximately 0.04 standard deviations through optimal reallocation compared to the random assignment status quo.

Conclusion

Graham and Ridder’s work provides a rigorous bridge between theoretical models of social interaction and practical policy evaluation. By focusing on the net effects of reallocation rather than isolated peer effects, their framework offers a more realistic tool for evaluating school assignment policies, housing vouchers, and other interventions aimed at managing social diversity. Their findings in Project STAR serve as a cautionary tale: policies that seem neutral on average may still have profound implications for social inequality.

Summary updated

Abstract

In this paper we nonparametrically analyze the effects of reallocating individuals across social groups in the presence of social spillovers. Individuals are either 'high' or 'low' types. Own outcomes may vary with the fraction of high types in one's social group. We characterize the average outcome and inequality effects of small increases in segregation by type. We also provide a measure of average spillover strength. We generalize the setup used by We relate our reallocation estimands to this theory. For each estimand we provide conditions for nonparametric identification, propose estimators, and characterize their large sample properties. We also consider the social planner's problem. We illustrate our approach by studying the effects of sex segregation in classrooms on mathematics achievement.
BibTeX
@misc{graham2010measuring,
  author = {Bryan S. Graham and Guido W. Imbens and Geert Ridder},
  title = {{Measuring the Effects of Segregation in the Presence of Social Spillovers: A Nonparametric Approach}},
  year = {2010},
  doi = {10.3386/w16499},
}
Figure