Abstract:
|
Propensity score methods are an important tool to help reduce confounding in non-experimental study and produce accurate causal effect estimates. Most propensity score methods assume that covariates are measured without error. However, covariates are often measured with error. Recent work has shown that ignoring such error could lead to bias in treatment effect estimates. In this talk, we consider non-experimental settings where an important covariate is measured differently in the treatment and control groups, leading to differential measurement error in the two groups. We propose a flexible Bayesian approach for assessing sensitivity to differential measurement error when using propensity score methods, and evaluate its performance using simulation studies. We consider three scenarios: systematic (i.e., a location shift), heteroscedastic (i.e., different variances), and both systematic and heteroscedastic measurement error. We also explore various prior choices (i.e., point mass or weakly-informative prior) on the sensitivity parameter related to the differential measurement error. Since this type of data usually includes little information on the true values of mismeasured cova
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.