Abstract
<p>Psychometric techniques are increasingly critical for refining the diagnostic criteria for substance use disorders (SUD), but core assumptions of these techniques are often untested and potentially unmet. We examined the assumption of local independence – that indicators are uncorrelated after conditioning on their latent variable – among SUD criteria across 11 classes of psychoactive drugs using data from the National Survey of Drug Use and Health surveys from 2021 to 2023 (N = 165,249) by applying unique variable analysis, factor analysis, and item response theory analysis to identify psychometric redundancy. Across all drugs, unique variable analysis identified 44 instances of psychometric redundancy (out of the 585 possible instances) in the SUD criteria set, with variation across drugs. Item response theory suggested that redundancy also varied across SUD severity. Gave up activities and role interference were redundant across all drugs, and time spent and larger/longer were redundant for six drugs. Dropping any of these redundant criteria resulted in a gain in SUD information, with information gained being nearly equivalent within redundant pairs. We discuss psychometric and diagnostic implications of criterion redundancy, including but not limited to inflated support for a unidimensional SUD structure, skewing of the SUD definition toward sources of redundancy (e.g., functional impairment), and inflated misclassification rates (particularly false positives). We offer several remedies to psychometric redundancy, including modeling residual covariances, consolidating redundant items into testlets, applying gating criteria where applicable, and using cognitive interviewing to maximally distinguish conceptually unique but psychometrically overlapping criteria.</p>