Three Essays on Semiparametric Econometric Evaluation Methods

This dissertation introduces three novel approaches for the econometric evaluation of heterogeneous treatment effects. The proposed methods consider the effects of a binary treatment on different characteristics of the outcome distribution.Section 1 proposes an estimation method for various average treatment effects using weighted integrals of estimates of marginal treatment effects. Marginal treatment effects measure the effect of a treatment variable on the outcome for individuals who are indifferent between participating or not. By integration of suitable weighted marginal treatment effects, several average treatment effects can be calculated. The general assumptions of the method enable the use of various nonparametric estimators for the weights and the marginal treatment effects.Section 2 describes a method to estimate quantile treatment effects of a binary treatment variable on censored durations. The effects of interest are differences between quantiles of the counterfactual outcomes in both treatment states. Identification is based on the conditional independence assumption. Estimators for the whole population and for the subgroup of participants are proposed, consistency, asymptotic normality, and consistency of the variance estimators are shown. A simple transformation of the effects is given, which enables an easy interpretation of the results. A test procedure for several hypotheses on the whole quantile treatment effect process is described. Pros and cons of the method compared to other approaches are discussed.Section 3 describes methods to test for distributional treatment effects under the conditional independence assumption. The differences between latent outcome distributions are judged by testing hypotheses of distributional equality and stochastic dominance. Furthermore, semiparametric efficient versions of the test statistics are given. The latter test statistics are double robust, i.e., they are consistent under misspecification of either the outcome equation or the propensity score. Consistent bootstrap procedures for deriving critical values of all tests are proposed.