A Comparison of the Three Bayes Point Estimates in Multivariate Multiple Membership Multilevel Modeling

• In Event: Estimation Methods for Multilevel Models

Abstract

This simulation study examines parameter recovery in Bayesian multivariate multiple membership multilevel modeling under a variety of manipulated conditions including: the number of groups (50, 100, and 200), average group sizes (20 and 40), and multiple membership rate (20% and 40%). In addition, three Bayes point estimates were compared to test the accuracy of parameter estimation. Across all conditions, no substantial relative bias was found for fixed effect estimates or student-level variance estimates. However, the posterior mean and median school-level variance-covariance were over-estimated with smaller group conditions (<=50), while the posterior mode estimates were not substantially biased across any of the conditions. Generally, the coverage rates of 95% credible intervals were close to nominal coverage across all conditions.

Authors

• Hyewon Chung, Chungnam National University

• Jiseon Kim

• Hyeonjeong Jeon, National Youth Policy Institute

• Soyoung Park, Chungnam National University