Speaker: Cindy Yu, Professor, Department of Statistics, Iowa State University

Title: Treatment Effect Estimation Using Quantile Regression Imputation Under Non-ignorable Assumption

Abstract: Most of causal inferences (CI) assume missing at random assumption, i.e. the treatment selection probability depends on covariates only. However in real applications, it is often found that the selection probability also depends on potential outcomes, which leads to incorrect treatment effect estimation if this selection bias is not adjusted appropriately. In this work, we propose an iterative estimating-solving (ES) algorithm to impute potential outcomes via semiparametric quantile regression under MNAR assumption. Model identification under MNAR is carefully discussed. We validate the proposed estimator through theories of convergence and large sample properties, and simulation studies. The proposed method is applied to a real data application.