An Empirical Comparison of Multiple Imputation Approaches for Treating Missing Data in Observational Studies

• In Event: Poster Session 10
  In Poster Session: Division D Section 2 Poster Session 2

Abstract

This study compares four multiple imputation approaches for treating missing data in the context of propensity score analysis: Imputing covariates only and imputing covariates and the propensity score are crossed with two methods of treating the imputed data (averaging treatment effects and averaging propensity scores) under MCAR, MAR, and MNAR missing data conditions. The performance of these approaches is evaluated in terms of accuracy and precision of parameter estimates, Type I error, and statistical power. Factors include sample size, proportion of missing data, treatment effect magnitude, correlation between covariates, and number of covariates. Imputing covariates only crossed with averaging treatment effects shows the best performance under MAR and MCAR. None of the MI approaches were adequate under MNAR.

Authors

• Patricia Rodriguez de Gil, University of South Florida

• Jeffrey D. Kromrey, University of South Florida

• Seang-hwane Joo, University of South Florida

• Yan Wang, University of South Florida

• Eun Sook Kim, University of South Florida

• Jessica Montgomery, University of South Florida