Bifactor Exploratory Structural Equation Models Versus Traditional Approaches in Predicting External Criteria

Abstract

To examine relationships between the global construct of multidimensional data and external criteria, bifactor exploratory structural equation modeling (B-ESEM) and traditional methods (e.g., unidimensional confirmatory factor analyses, CFA; parceled CFA; and bifactor models) can be used. We compared their performance in a Monte Carlo simulation study. (a) B-ESEM performed the best, followed by parceled CFA and bifactor models, and the unidimensional CFA was the worst. The relative biases of predicting the external criteria with the global construct increased with larger cross-loadings and greater model complexity and decreased with an increasing percentage of uncontaminated correlations (PUC). (b) With few large cross-loadings, relative biases of B-ESEM structural coefficients were negligible. In contrast, PUC determined our choice of unidimensional CFA, parceled CFA, or bifactor models. (c) With many large cross-loadings and high general factor loadings, B-ESEM was preferred. (d) No model was acceptable with many large cross-loadings and low general factor loadings.