Comparing Bias Reduction Approaches for Estimating Average Treatment Effect in Multiple Quasi-Experiment Contexts

• In Event: 42.034 - Estimation and Model Advances for Causal Inference and Treatment Effects

Abstract

The purpose of this paper is to provide guidance in choice of analytic bias reduction methods for educational studies in which the goal is to estimate a treatment effect in the presence of selection bias into treatment. In addition, issues of dimensionality, collinearity, omitted confounders, missing outcomes, and non-independence may be factors influencing analytic choice. The present paper investigates these issues by systematically comparing performance of common bias reduction techniques, including multiple regression/analysis of covariance (direct control of group differences on covariates), propensity score (PS) matching methods (1:1 or 1:n), and inverse probability weighting on the PS (IPW). Preliminary simulation results showed that MR may be preferred for situations without dimensionality or collinearity issues; otherwise IPW may be preferred

Authors

• Elizabeth A. Sanders, University of Washington

• Elizabeth A. Dietrich, University of Washington - Seattle