Making (Good) Matches: Building and Piloting an Online Platform for Experimental School Choice Research

In this project, I examine the role of matching as it is applied to the problem of U.S. public school choice. Across the country, families submit ordered preferences to their districts, which function as centralized clearinghouses for the two-sided market over student demand and available seats. Although most districts leverage student-optimal, strategyproof mechanisms such as the deferred acceptance algorithm, many families still submit untruthful preferences in an attempt to “game” the system and obtain a better outcome. My project fits into the broader game theoretic literature on truth-telling behavior in the school choice problem. Using modern web technologies such as NextJS, FastAPI, and MongoDB, I developed MatchEd, a platform for configuring and running online, crowd-sourced experiments related to the school choice problem. The platform offers researchers visual and programmatic interfaces via the web and its REST API respectively, and it generates links that can be inputted into online crowd-sourced marketplaces such as Amazon Mturk.

To test the platform, I conducted a pilot study to examine participants’ learning dynamics in repeated school choice games. I analyze the differences in truth-telling between a control group that plays a one-shot game and an experimental group that plays five simulated practice rounds before the actual game. The results support that the dominant strategy is “learnable,” even over a small number of games. Given the results of the pilot study, I argue that further research is needed to develop practical methods of countering sub-optimal behavior among participants in strategyproof school choice matching mechanisms.