The Effects of Cluster Bias in Multilevel Structural Equation Modeling

• In Event: 44.078 - Advanced Issues in Structural Equation Modeling

Abstract

Multilevel structural equation modeling (MSEM) is an extension of traditional SEM when data have a hierarchical structure. Much of the research on multilevel confirmatory factor analysis, the measurement model component of MSEM, assumes that the measurement model is at least weakly invariant – there are equal factor loadings across the clusters. The present study investigates a simple latent regression MSEM model where this assumption in violated in varying degrees. Results indicate that parameter bias increases when there is greater variance in factor loadings across clusters, and that other common models (SEM, HLM) do not estimate the parameter well, have attenuated standard errors, or both. Results are more severe when the DV (rather than the IV) is not invariant.

Authors

• Christopher Runyon, The University of Texas - Austin

• Tiffany Ann Whittaker, The University of Texas - Austin