The Reliability Paradox in Multisample Covariance Structure Models

• In Event: Measurement and Modeling Issues in Structural Equation Modeling

Abstract

The reliability paradox within structural equation modeling (SEM) is the phenomenon whereby data-model fit indices (e.g. SRMR, RMSEA) tend to indicate better overall fit when factor reliability is poorer. In short, indicators with lower loadings pass on relatively less structural misfit information to fit indices compared to indicators with higher loadings, thus leading to a more favorable assessment of data-model fit overall. This paper will explore the reliability paradox, which has only been investigated within the context of single-group SEM analyses, as it relates to testing parameters in multisample mean and covariance structure models. In doing so our hope is to continue to raise awareness of the interpretational challenges associated with fit indices for latent variable structural equation models.

Authors

• Daniel McNeish, University of Maryland

• Gregory R. Hancock, University of Maryland