CMStatistics 2017: Start Registration
View Submission - CMStatistics
B1632
Title: Robust estimation of treatment effects in a latent-variable framework Authors:  Mikhail Zhelonkin - Erasmus University Rotterdam (Netherlands) [presenting]
Abstract: The policy evaluation is one of the central problems in modern economics. Unfortunately, it is often impossible to perform a randomized experiments in order to evaluate the treatment effects. Hence, the data from observational studies has to be used. In this case the sample is typically non-random and one has either to correct for selectivity or to impose (conditional) independence assumption. Since this assumption is often unrealistic, the structural latent variable model is used. The parametric estimators (although, they are straightforward to compute and to interpret) have been criticized for sensitivity to the departures from the distributional assumptions. The alternative semi- and non-parametric estimators have complex identification and are limited to estimation of certain parameter(s) of interest but do not allow for the general evaluation and interpretation of the model. We employ a previous latent-variable framework. We study the robustness properties of the estimators of four principal parameters (average treatment effect, average treatment effect on the treated, local average treatment effect and marginal treatment effect), and propose the robust alternatives.