Comparison of latent variable and psychological network models in PROMIS data: output metrics and factor structure.

Purpose: Much research is still needed to compare traditional latent variable models such as confirmatory factor analysis (CFA) to emerging psychometric models such as the Gaussian graphical model (GGM). Previous comparisons of GGM centrality indices with factor loadings from CFA have discovered redundancies, and investigations into how well a GGM-based alternative to exploratory factor analysis (i.e., exploratory graph analysis, or EGA) is able to recover the hypothesized factor structure show mixed results. Importantly, such comparisons have not typically been examined in real mental and physical health symptom data, despite such data being an excellent candidate for the GGM. Our goal was to extend previous work by comparing the GGM and CFA using data from Wave 1 of the Patient Reported Outcomes Measurement Information System (PROMIS).

Methods: Models were fit to PROMIS data based on 16 test forms designed to measure 9 mental and physical health domains. Our analyses borrowed a two-stage approach for handling missing data from the structural equation modeling literature.

Results: We found weaker correspondence between centrality indices and factor loadings than found by previous research, but in a similar pattern of correspondence. EGA recommended a factor structure discrepant with PROMIS domains in most cases yet may be taken to provide substantive insight into the dimensionality of PROMIS domains.

Conclusion: In real mental and physical health data, the GGM and EGA may provide complementary information to traditional CFA metrics.