Missing Data Imputation versus Full Information Maximum Likelihood with Second-Level Dependencies

Missing data in the presence of upper level dependencies in multilevel models have never been thoroughly examined. Whereas first-level subjects are independent over time, the second-level subjects might exhibit nonzero covariances over time. This study compares 2 missing data techniques in the presence of a second-level dependency: multiple imputation (MI) and full information maximum likelihood (FIML), which were compared in an SAS simulation study. The data was generated with varying levels of missing data, dependencies at the second level, and different sample sizes at both the first and second levels. Results show FIML is superior to MI as it correctly estimates standard errors.