Missing item scores in multilevel cross-classified educational data: An empirical Bayes Multiple Imputation approach

Previous research has shown that the multiple imputation (MI) approach faces challenges in giving plausible estimates of missing item scores in multilevel item-response data. The direct likelihood (DL) analysis approach on the other hand has been shown to give reliable estimates for statistics of interest in such clustered data. But MI can still have some practical advantages over the DL analysis. Through a simulation study, we propose an empirical Bayes MI approach that considers additional challenges resulting from a cross-classified multilevel data structure with substantial amounts of missing binary item scores. Results based on the proposed imputation model are compared to those obtained from the DL analysis.status: publishe