An Evaluation of the Use of Multilevel Structural Equation Modeling for Mediation Analyses

• In Event: Advanced Topics in Structural Equation Modeling

Abstract

The primary goal of this study was to assess how well use of multilevel structural equation modeling versus conventional multilevel modeling recovered the true indirect effect value when both the mediating, M, and outcome, Y, variables entail multiple indicators of their respective latent factors. Data were generated to fit an upper-level (2->2->1) mediation model with X and M measured at level two and Y at level one. In the simulation study, the following factors were manipulated, including: the true intra-class correlation values, the within- and between-cluster sample sizes, the degree of measurement error in the mediating variable’s indicators, and the magnitude of the true mediated effect. Relative parameter and SE bias and confidence interval coverage (using PRODCLIN) results are discussed.

Authors

• Xin Li, The University of Texas - Austin

• Susan Natasha Beretvas, The University of Texas - Austin