Missing Data Strategies for Multilevel Models (303271)

Keywords: multilevel, missing data, monte carlo simulation

The current poster proposes to examine common methods of missing data handling strategies in multilevel (i.e., clustered) data sets (Raudenbush & Bryk, 2002). It will compare traditional strategies of missing data handling (i.e., listwise/pairwise deletion) against newer imputation (expectation maximization and multiple imputation) and maximum-likelihood based (full-information maximum-likelihood) methods (Enders, 2010). A small-scale Monte Carlo simulation will be conducted in SAS 9.4 using data sets with 30 clusters with 10 observations per cluster. The data set will feature a single individual-level outcome variable regressed on a single cluster-level predictor variable, a single individual-level group-mean-centered predictor variable, and their cross-level interaction. Both predictor variables will have a medium effect-size (R2 = .06) and there will be a moderate amount of clustering (ICC = .30). Missingness on the individual-level predictor will be varied (1%, 5%, 15%, 25% missing values) in a missing at random (MAR; Rubin, 1976) pattern. Methods will be evaluated in light of their impact on parameter estimates and standard-error-based (i.e., Type I/II error) outcomes.