! Yuanyuan Liu Prof. Rust ECON 615: PartII October 30, 2012 Reduced-Form Estimation vs. Structural Modeling in Labor Economics

Liu 1

This analysis is to compare and contrast two labor papers presented at the IZA Young Professionals program at Georgetown University last week. They are Explaining Charter School Effectiveness by Angrist, Pathak and Walters and Matching, Sorting and Wages by Lise, Meghir and Robin. The first one uses a reduced-form approach to study the treatment effect of charter school attendance while the second one develops a structural search-matching model to study optimal labor market policies. As will be discussed below, these two papers ask very different questions; and thus utilize very different empirical approaches, both well-suited for the purposes of each paper.

In the Explaining Charter School Effectiveness paper, the authors ask the following questions: Do charter schools generate achievement gains? What are the causes of charter school treatment effect heterogeneity? They conduct a semi-parametric investigation of heterogeneous potential outcomes, and estimation of LATE gives mixed results for urban and non-urban charter schools. This charter school treatment effect heterogeneity is referred by the authors as “urban charter advantage”: urban charter schools boost student achievement while charter schools in other setting do not. Then they seek to isolate sources of charter effect heterogeneity through both student-level and school-

!

Liu 2

level explanatory variables. On the student-level, student demographics explain some of the urban charter advantage because urban charters are most effective for non-whites and low-baseline achievers; in other words, there is a big gap between urban non-treated counterfactuals and nonurban ones, and going to charter schools bridges this gap because urban charter schools push the scores of their students from a typically low level up to a level much closer to the achievement seen among non-urban charter students. On the school level, they find that over-subscribed charter schools with high-quality lottery records seem more effective than non-lottery schools, and they also find strong evidence supporting the effectiveness of No Excuse. The reduced form approach used in this paper is well-tailored for the purpose of the paper-investigating the effectiveness of one specific policy which was implemented through lottery and has a clear treatment effect interpretation. The charter school admission lotteries provide a random experiment for the study, and the authors can use the sample of lotteried applicants to investigate potential outcomes of compliers and counterfactuals. The choice of a reduced form treatment model is natural having a random sample (to some extent), and also intuitive based on the nature of the question, that is, the policy at question has clear impacts that can be quantified (student achievement), and outcomes on the treated are observed while potential outcomes on the counterfactual non-treated can be found through matching. In addition, it has the advantage of being able to identify and decompose causal effects. However, the choice of reduced form approach also inevitably suffers from sample selection bias because of the nature of the sample. Specifically, the sample used in the paper are students that won charter school enrollment lotteries, which means that they would all potentially go to the over-subscribed (and thus better) charter schools; as a result, the estimation of treatment ef-

!

Liu 3

fect on the treated will be boosted in the sense that it actually measures the treatment effect of a subsample, or the “better” charter schools. Moreover, as the authors have also point out, schools that keep better lottery records seem more effective than non-lottery schools, thus one might reasonably argue that better school management may also partly contribute to the positive treatment effect of urban charter schools. As for the drastically different treatment effect of urban and nonurban charter schools, one might also argue that, since Y0i-urban