To identify an optimal LCA measurement model, we specified models with 2 to 8 classes and evaluated parameter estimates and model fit. Parameter estimates included item response probabilities and latent class prevalence estimates. Item response probabilities reflected the probability of meeting criteria for each psychiatric disorder conditional upon membership in a latent class. Latent class prevalence estimates corresponded to the proportion of respondents that would be members a latent class accounting for measurement error. Gender and AUD severity were covariates in the LCA due to their association with internalizing and externalizing psychiatric comorbidity (Bucholz et al. 2000; Kessler et al. 2005; Dawson et al. 2010). Hence, latent class prevalence estimates were adjusted for gender and AUD severity.
