Diversity and Transportation in School Choice

Project Outcomes Statement

Across the United States, many K-12 school districts continue to adopt choice systems that permit students to attend schools outside of traditional zones. One way that school choice is operationalized is through a centralized assignment system that allows students to rank and apply to multiple schools on a single application. The district or governing non-profit then uses this application to assign each student to her highest ranked school contingent on priorities and capacity. Centralized assignment systems have been adopted by many large urban school districts including Boston, Chicago, New York City, Denver, Indianapolis, New Orleans and Washington DC.

In addition to offering more options for families, centralized assignment can potentially alleviate school segregation caused by neighborhood homogeneity. However, these systems present many logistical challenges, including managing increasing transportation costs and deploying complex student sorting algorithms, also known as matching mechanisms.

The purpose of this grant was to study the effects of matching mechanisms on educational outcomes, including diversity. The line of inquiry includes reserve systems that give admission priority to students with certain characteristics. For example, a student may have a better chance of admission to a certain school if she lives in a specific neighborhood, qualifies for Free or Reduced Priced Lunch, or meets academic criteria. Some districts, such as Boston, give preference to students who live close to a school. This policy is considered a compromise that allows for choice while still keeping district transportation costs down.

As a result of this grant, the Principal Investigators published three papers and contributed to the theoretical literature on matching mechanisms. First, in "Explicit vs. Statistical Targeting in Affirmative Action: Theory and Evidence from Chicago's Exam Schools," the authors analyzed Chicago's tier-based preference system that divides neighborhoods into four tiers based on socioeconomic status. An equal amount of seats at each school are reserved for all four tiers and the rest are assigned by merit. One of the goals for this system is to increase diversity at selective or in-demand schools. In addition to assigning tiers, the authors show that it is possible to further target disadvantaged students by changing the order in which seats are assigned. They also show how a the district can exploit the test score distribution across tiers to prioritize disadvantaged students: when disadvantaged students systematically have lower scores than other applicants, the optimal processing order first assigns merit seats and then tier seats.

Another publication, "Efficiency, Justified Envy, and Incentives in Priority-Based Matching," analyzes, compares, and summarizes the properties of three types of mechanisms: Top trading cycles (TTC), deferred acceptance (DA), and serial dictatorship (SD). DA is increasingly popular and used by many districts that implement centralized assignment mechanisms, including Boston Public Schools. TTC was used by New Orleans until 2013, at which time the process switched to DA. SD is not currently used by districts but is considered one possible alternative.

The authors compare the mechanisms across three criteria: strategy-proofness, fairness, and efficiency. DA and TTC mechanisms are strategy-proof, meaning that the best way to ensure a student receives her preferred school is to rank her choices truthfully. This eliminates the gaming aspect observed in previous application processes used by the New York City Department of Education. In terms of fairness, DA completely eliminates justified envy, which occurs when a student prefers a school that accepts a different student with lower priority. Empirical evidence from New Orleans and Boston shows TTC has significantly less justified envy than SD. Finally, TTC and SD are Pareto efficient, meaning that no student could be assigned to a higher priority school without another student being worse-off. There is no mechanism that is both Pareto efficient and without justified envy, but DA dominates other mechanisms without justified envy. These properties can help districts choose which mechanism they should use based on the implied tradeoffs, and the findings have implications for other matching systems outside of education.

The final publication produced by this grant revisits a previous research paper and adds to the literature on forecasting and structural demand models. In a previous publication, the Principal Investigators used structural demand modeling to help simulate the effect of changes to the Boston Public Schools assignment plan. The BPS policy changes restricted the set of schools that each student is allowed to include in their applications based on her residential address. The PIs used several different structural demand models to forecast the outcomes before BPS implemented their plan. In this latest update, the authors validate their ex ante predictions using school choice data after the plan was operationalized. They find that once they control for changes in the environment outside of the models, the demand models are reasonably accurate compared to expectations. Taking changes in the environment into account is shown to be both difficult and important for model accuracy.