- On the Determinants and Implications of School Choice: Semi-Structural Simulations for Chile
School choice is one of the most widely debated policies aimed at increasing student welfare in different countries. Proponents argue that school choice may create incentives for schools to increase productivity, offer a product closer to student demands, and expand the choice set for poor students. Opponents, in contrast, argue that school choice may increase segregation, decrease school quality for poor students by moving good peers to other schools, and produce competition in irrelevant school attributes if parents do not care about education outcomes. Most researchers use reduced-form methods to study these claims. For instance, some papers analyze the effect of interschool competition on test scores and other measures, finding mixed evidence.1 Other papers use a variety of methods to study the process of choice by parents. This paper uses semi-structural estimates of parents’ preferences from an earlier work to study the effects of school choice on both student welfare and socioeconomic segregation.2 To the best of our knowledge, this is the first time school choice has been evaluated using this kind of approach, in which preferences are explicitly taken into account. [End Page 197]
We evaluate these effects in the context of the Chilean quasi-voucher system. The Chilean system, which has been in operation since 1981, has four key characteristics: private and public schools receive government subsidies proportional to school enrollment, based on a (mostly) flat per-student subsidy; students are free to apply to any school that receives government subsidies; voucher schools are free to choose students among the pool of applicants and may charge top-ups; and school entry is relatively easy. These conditions make the Chilean experience probably the most massive school choice program in the world.3
We use data on school choice for a sample of students living in the Metropolitan Area of Santiago, which has the broadest school choice in the country owing to the high entry of voucher schools and the existence of relatively low transportation costs.4 We consider data for 2002 to capture a period in which the choice system was mostly consolidated: the bulk of school entry has taken place, and information on test scores is available from the mid-1990s onward. We use semi-structural estimates from our earlier work, which follows the literature on horizontal differentiation in the attribute space developed by Berry, Levinsohn, and Pakes, among others.5 We model school choice as a discrete process in which parents choose schools based on attributes such as characteristics of peers (mean and standard deviation of income and mother’s education at the school level), indicators of the development of cognitive abilities (test scores), indicators of the development of noncognitive skills (discipline and the teaching of religious values), proxies for transportation costs (distance from school to the center of the municipality in which the student lives and a dummy for whether the school is close to a subway station), the top-up charged by the school, a dummy indicating whether the school has an extended day, and a dummy for whether the school is single-sex or coeducational.6 We further allow the choice to depend on an unobserved [End Page 198] (for us) school effect, which is common to all students, and interactions between the set of observed school attributes and student characteristics such as household income, mother’s education and age, gender, preferences for the teaching of values, and a proxy for parents’ expectations of student potential. We thus allow for considerable heterogeneity in preferences, which is supported by the data.
The paper also presents reduced-form estimates in which we relate the decision to attend school outside the home municipality as a function of differences in the average quality (both the mean and the standard deviation) of the destination and the home municipality and some socioeconomic controls. We see these results as a benchmark for semi-structural estimates. Interestingly, the estimated effects are quite similar to structural estimates, and we find that students react more strongly to differences in quality in urban markets with more competition than...