Essays on Matching and Market Design.

Abstract

Using a combination of experimental, theoretical, computational and empirical methods, my dissertation studies matching and market design with applications to education policy including school choice and college admissions. I tackle three problems: the effect of standardized tests on matching mechanisms (Chapter 2); experimental evidence of matching in a large market (Chapter 3); and quasi-experimental evidence of the theoretical properties of matching mechanisms (Chapter 4).

In Chapter 2, I investigate the matching of college admissions, where students' admission priorities and colleges' preferences over students are misaligned, due the imperfect measure of student aptitudes by standardized entrance tests. I show that in this case any matching mechanism that is stable with regard to priority is not stable with regard to preference. The resulting instability leads to market unraveling. However, a manipulable mechanism, combined with limited information about priorities, may succeed in mending this market failure. A laboratory experiment confirms this theoretical prediction.

In Chapter 3, we study the role market size plays in school choice. We evaluate the performance of the Boston and the Deferred Acceptance (DA) mechanism in laboratory with different market sizes. The results show that increasing the market size from 4 to 40 students per match increases participant truth-telling under the DA but decreases it under the Boston mechanism, leading to a decrease in efficiency but no change in the large stability advantage of the DA over the Boston mechanism. Furthermore, increasing the scale to 4,000 students per match has no effect on either individual behavior or mechanism performance. Our results indicate that "large market" in practice is smaller than in theory.

In Chapter 4, we evaluate the Immediate Acceptance (Boston) mechanism and the parallel mechanism in college admissions, both in the laboratory and with naturally-occurring data. Through both channels, we find that the more emphasis a mechanism put on the first choice, the more likely students rely on their rankings in the test to manipulate their reported preferences, which confirms the theoretical predictions. Although in the laboratory, the parallel mechanism proves to be more stable, we do not observe economically significant difference in stability in the field.