Conventional Psychometric Practices Preordain Support for Unidimensionality of Diagnostic-Criteria Sets: Substance Use Disorder as a Case Example

Psychometric methods are inseparable from contemporary psychiatric classification efforts but were not developed with psychopathology in mind. Here, we consider how conventional psychometric approaches are ill-equipped to characterize the structure of mental disorders and lead researchers to adopt unidimensional conceptualizations of them. Using the substance-use-disorder criteria set from the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders as a case example, we examine the validity of a unidimensional factor model in a large epidemiologic sample ( N = 104,944) across 11 classes of psychoactive drugs. Traditional fit indices, factor loadings, and eigenvalues appeared to justify a unidimensional model, but evidence for unidimensionality faltered when using updated model-fit indices and model-based fit benchmarks and manipulating the extent of skew by restricting analyses to symptomatic individuals. Exploratory factor analyses further underscored evidence of multidimensionality in substance use disorder. We situate our findings in the historic literature that documents the hazards of underfactored models and offer suggestions for future classification efforts.