Assessing Mediation in a Multilevel Structural Equation Modeling Framework for the 2-1-1 and 1-1-1 Design

• In Event: Marriott Roundtable Session 9

Abstract

Researchers have used multilevel structural equation modeling (ML-SEM) to model mediation in a hierarchical context. No one has thoroughly examined and compared the performance of an ML-SEM latent model which incorporates the measurement errors of the indicators with an ML-SEM composite model which treats the latent variable as a composite score. A simulation study was conducted to compare the recovery of the mediated effect using the ML-SEM latent model versus the ML-SEM composite model under various conditions for a 2-1-1 and 1-1-1 design. Type I error rates and power for the mediated effect was larger for the composite model when the number of clusters was small. The mediated effect was unacceptably underestimated for the composite model.

Authors

• Man Chen, University of Wisconsin - Madison

• Tiffany Ann Whittaker, The University of Texas at Austin