Comparing the Performance of Multivariate Multilevel Modeling to Traditional Analyses With Complete and Incomplete Data

• In Event: Multilevel Modeling: Estimation, Model Selection, and Sample Size

Abstract

A multivariate multilevel model (MVMM) extends standard multilevel modeling by including multiple dependent variables and thus could be used in place of traditional multivariate analyses. In the context of a two group study with two dependent variables, a simulation study was conducted to compare the performance of MVMM to traditional MANOVA and a series of Bonferonni-adjusted analyses. The results showed that across various effect and sample sizes, response correlations, and missingness levels, MVMM has greater power than traditional analyses. While the Type I error rate for the overall multivariate null hypothesis can be elevated with MVMM, especially with small sample size, the Type I error rate for the test of a specific dependent variable is accurate.

Authors

• Ryoungsun Park, Wayne State University

• Keenan A. Pituch, The University of Texas - Austin

• Jiseon Kim, University of Washington - Seattle

• Hyewon Chung, Chungnam National University

• Barbara G. Dodd, The University of Texas - Austin