Experimentally Validating Welfare Evaluation of School Vouchers

We leverage a unique two-stage experiment that randomized access to private school vouchers in rural India to estimate willingness-to-pay for greater school choice, a key input for assessing welfare impacts of alternative voucher designs. We find that a voucher targeted to households with limited assets, a proxy for ability-to-pay constraints, yields a marginal value of public funds (MVPF) greater than 3. We obtain this result using a multistep research design that isolates the information contributed by experimental data. We begin by estimating several models of school choice on control data alone, reserving data from treatment markets for model selection via out-of-sample validation. All of the models underpredict take-up of the randomized voucher offer. Evidence from treatment markets points to why: the program induced household search and led private schools to use voucher surplus to attract students. We incorporate these mechanisms into a unified model, estimated with both treatment and control data, that successfully explains the observed choice patterns. Our experience highlights that credible policy analysis requires modeling choice frictions and supply responses and using experimental data in estimation.