Addressing the Limitations of Confirmatory Factor Analysis of Posttraumatic Stress Disorder: An Application of Exploratory Structural Equation Modeling

Abstract

To date, confirmatory factor analysis (CFA) has been the predominant method for empirically evaluating the latent factor structure of posttraumatic stress disorder (PTSD). A voluminous CFA literature provided a strong empirical rationale for separating effortful avoidance from emotional numbing symptoms in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5; APA, 2013) PTSD criteria. Nevertheless, CFA studies of PTSD also have generated increasingly complex structural models, which lack a priori theoretical justification and parsimony, yield highly correlated factors, and include multiple factors with only two indicators. It is possible that the CFA requirement of constraining factor cross-loadings to zero has biased the findings of previous PTSD CFAs. Accordingly, the primary aim of the present study was to evaluate DSM-5 PTSD factor structure using exploratory structural equation modeling (ESEM; Asparouhov & Muthén, 2009), which permits cross-loadings and thereby might circumvent some of the limitations of CFA (Prudon, 2015). Participants were trauma-exposed undergraduates (N = 1,139) who completed the PTSD Checklist for DSM-5 (PCL-5; Weathers et al. 2013a). Analyses focused on comparisons between the four-factor DSM-5 model of PTSD and the seven-factor Hybrid model, using both ESEM and CFA. As expected, in both four- and seven-factor ESEM models there were a number of significant cross-loadings, which justified the use of ESEM. For both ESEM and CFA, the seven-factor Hybrid model was superior in fit to the four-factor DSM-5 model, and thus was accepted as the best representation of PTSD. However, compared to its seven-factor CFA counterpart, the seven-factor ESEM model was superior in fit, had lower factor loadings and factor intercorrelations, and exhibited better-differentiated patterns of associations with external correlates. These findings suggest that ESEM is more useful than CFA for investigations of PTSD factor structure. They also suggest that the superior fit of the seven-factor Hybrid model is not attributable solely to CFA misspecifications, and thus support its continued use in PTSD construct validity research.