A0174
Title: A general proposal for model-free difference-in-differences
Authors: Daniel Henderson - University of Alabama (United States) [presenting]
Stefan Sperlich - University of Geneva (Switzerland)
Abstract: A general framework is proposed for model-free difference-in-differences analysis with confounders. Following the natural steps in practice, we start by searching for the preferred data setup, namely the simultaneous selection of confounders and potential data (outcome) transformations. We then offer a test for the credibility of identification assumptions. The treatment effects themselves are estimated in two steps: first, the heterogeneous effects stratified along with the confounders, then second, the average treatment effect(s) for the population(s) of interest. We suggest bootstrap procedures to calculate the standard errors of these estimates and significance tests. We study the asymptotic statistics and the finite sample behavior (via simulations) of our tests and estimators. We address practical issues that arise such as bandwidth selection, incorporating sample weights and dealing with discrete data in both the outcome variable and set of confounders.