Evaluation of Four Multiple Imputation Methods for Handling Missing Binary Outcome Data in the Presence of an Interaction between a Dummy and a Continuous Variable

Multiple imputation by chained equations (MICE) is the most common method for imputing missing data. In MICE algorithm, can be performed using a variety of parametric and nonparametric methods. The default setting in implementation models to include variables as linear terms only with no interactions, but omission interaction may lead biased results. It investigated, simulated real datasets, whether recursive partitioning creates appropriate variability between imputations unbiased parameter estimates confidence intervals. We compared four multiple (MI) methods on dataset. MI included predictive mean matching an term model (MICE-interaction), classification regression tree (CART) specifying (MICE-CART), random forest (RF) (MICE-RF), MICE-Stratified method. first selected secondary data devised experimental design that consisted 40 scenarios (2 × 5 4), which differed rate (10%, 20%, 30%, 40%, 50%), mechanism (MAR MCAR), (MICE-Interaction, MICE-CART, MICE-RF, MICE-Stratified). First, we randomly drew 700 observations replacement 300 times, then were created. evaluation was based raw bias (RB) well five other measurements averaged over repetitions. Next, simulation study, generated 1000 times sample size 700. Then, created each dataset once. For all scenarios, same criteria used evaluate performance study. concluded that, when there effect dummy continuous predictor, substantial gains are possible methods, also, MICE-Interaction always more efficient convenient preserve effects than