Ulric B. and Evelyn L. Bray Social Sciences Seminar

Abstract: We present a test of the selection on observables assumption that does not require the presence of any excluded instruments or any covariates excluded from the structural function. Our test relies on the presence of a variable, among the set of controls, that has a discontinuous distribution. We develop formal testing procedures for a non-parametric additively separable model for either binary or nite treatment variables. We also outline how our testing strategy can be extended to a non-parametric, nonseparable model. The testing procedures we suggest are easy to implement and should be useful in many empirical situations. Specifically, we demonstrate how our test handles selection concerns which may potentially complicate the study of the impact of participation in the Supplemental Program for Women, Infants, and Children (WIC) on birthweight.

Written with Umair Khalil.

For more information, or if you are interested in attending this online seminar, please contact Letty Diaz by email at [email protected].