A0329
Title: Tests for heterogeneous treatment effect
Authors: Fangzhou Yu - University of New South Wales (Australia) [presenting]
Abstract: Two hypothesis tests are developed for heterogeneous treatment effects. The method is focused on the null hypothesis that the conditional treatment effects are zero for all covariate values and the null hypothesis that the conditional treatment effects are constant for all covariate values. The tests are applied to the treatment effects identified under the unconfoundedness assumption and the local effects identified by a binary instrumental variable. The test statistics are based on the Wald statistic of the best linear projection coefficients of the treatment effects on the covariates with coefficients estimated by regressing the augmented inverse propensity-weighted outcome on the covariates. First parametric tests assuming parametric forms of the potential outcomes and the propensity score are derived, and then the parametric assumptions are relaxed by allowing for nonparametric/high-dimensional models to derive semiparametric tests where Double/Debiased Machine Learning estimates the projection coefficients. The finite sample performance of the tests is demonstrated using simulated experimental and survey datasets. The use of the tests in two applications is illustrated regarding the effect of being the only child on the mental health of the only children and the effect of 401(k) participation on the net financial assets of the participants.