the potential to distort the results of what is being measured. In this study, the MIMIC models involve only one latent factor representing the unidimensional trait of alcohol use disorder severity. Without covariates, MIMIC models are equivalent to IRT models, as both of these models can be represented mathematically as factor analytic models. The link between IRT models and confirmatory factor analysis for dichotomous variables has been discussed elsewhere (Glockner-Rist and Hoijtink, 2003; MacIntosh and Hashim, 2003). Similar to factor analysis, IRT modeling establishes the relationship between some latent traits and their manifest indicators. By applying two-parameter logistic IRT models to the 11 AUD symptom criteria, we were able to assess the probabilities of symptom item endorsements across different values of the latent factor and to evaluate the psychometric properties of the symptom criteria. Hays and colleagues (2000) gave a detailed overview of IRT and a discussion of the related methodological issues and practical challenges. Briefly, IRT postulates that a person’s performance on a test item is determined by the underlying trait of that person and by the characteristics of the test item, which includes two key parameters, item threshold and item discrimination. Item threshold is the level at which
