Comparing structural models for internalizing pathology: Latent dimensions, classes, or a mix of both?

In recent decades, the conceptualization of internalizing problems has changed from categorical to dimensional and hybrid approaches. However, most studies have analyzed the structure of internalizing problems at the disorder level using categorical or dimensional approaches, with only a few studies examining the structure at the symptom level, or considering a hybrid approach. This study aimed to compare categorical (latent class analysis), dimensional (confirmatory factor analysis), and hybrid models (semi-parametric factor analysis) of internalizing constructs at the symptom level regarding model fit (structural validity) and prediction (concurrent validity) in four samples: community adults (n = 1072; n = 620), students (n = 378), and patients (n = 485). All participants completed the Inventory of Depression and Anxiety Symptoms-II to assess internalizing symptoms. In two samples, participants completed additional measures to test concurrent validity regarding disability, externalizing symptoms, personality traits, impairments in personality functioning, and quality of life. Dimensional models, particularly those allowing for non-normal distributions, outperformed categorical and hybrid models in terms of structural and concurrent validity (median _adjR² for dimensional models =.18–.16). Our results suggest that future studies should prefer dimensional models to better describe internalizing constructs and predict external variables. The consistent application of dimensional models of internalizing pathology would facilitate the integration of empirical findings in clinical science and enable a more valid and fine-grained assessment of individual mental health problems in clinical practice, thereby enhancing the potential to guide effective personalized interventions.