A Bayesian Approach to Multilevel Structural Equation Modeling With Continuous and Dichotomous Outcomes

• In Event: Roundtable Session 14
  In Roundtable Session: Issues in Multilevel Latent Variable Modeling

Abstract

Determining the appropriate method of estimation is an important step in the application of multilevel structural equation modeling (SEM). This paper compares the impact of prior distributions in the context of multilevel SEM with continuous and dichotomous factor indicators under various conditions of cluster size, number of clusters, and intraclass correlation (ICC). The use of informative and weakly informative priors in the Bayesian estimation framework improved the accuracy and efficiency of model estimates under conditions where frequentist estimation performed poorly (i.e., a small number of clusters, small cluster sizes, and low ICCs). Overall, the results of this study suggest that Bayesian estimation of multilevel SEM may be an attractive alternative to frequentist estimation when priors are selected carefully.

Authors

• James Patrick Clifton, University of California - Merced

• Sarah Depaoli, University of California - Merced