Testing for sample selection in the Milwaukee school choice experiment

We examine data collected from the Milwaukee Parental Choice Program to det ermine whether students with unobserved characteristics correlated with ach ievement are more likely to apply to the voucher program. We first estimate probit models for application to the choice program, and then math and rea ding achievement models. By comparing students randomly rejected from the p rogram with students who did not apply, we can test whether applicants and non-applicants have unobservable characteristics that are correlated both w ith the choice of school sector and with the included explanatory variables . We do not find strong evidence that students who apply to participate in the Milwaukee choice program have unmeasured characteristics, such as motiv ation, which systematically differ from non-applicants [JEL C12, I21]. (C) 1999 Elsevier Science Ltd. All rights reserved.