Improvement and Evaluation of Multiple Imputation by Heckman's One-Step MLE for Binary MNAR Outcomes and Various Types of MAR Covariates

Missing data is inevitable in clinical epidemiology. It becomes one of the major challenges in the analyses and can potentially undermine the validity of results and conclusions. Although methods for handling missing data with mechanisms of missing completely at random (MCAR) or missing at random (MAR) have been widely researched, methods adapted for the missing not at random (MNAR) mechanism are less studied. Galimard et al. (2018) have derived a method to use multiple imputation by Heckman's One-Step ML Estimation for binary MNAR outcome and continuous MAR covariates (MIHEml). This dissertation focuses on updating MIHEml in terms of methodology and programming. Both the simulation and the following empirical analysis prove that a more upgraded and generalized MIHEml algorithm manages to improve the estimation accuracy and prediction accuracy, comparing with complete case analysis to handle a binary MNAR outcome and various types of MAR covariates in the same process. Thus, this study reassures the effectiveness and applicability of the updated MIHEml.