The MIMIC analysis with the covariates gender and age at initiation was used to assess whether these covariates might explain why a two-factor model fit the data better when binge drinking was included as a criterion in our prior report (Beseler et al., 2010). The likelihood ratio test (LRT), Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC) and sample size adjusted BIC were used to assess model fit. We followed the MIMIC analysis with LCA to test whether the model best fit a two, three or four class structure and whether adding binge drinking as a criterion improved the model fit. Estimated model parameters included the prevalence of each latent class and the prior probability that an individual fell into a specific class given a specific item endorsement profile. Class assignment was based on having the highest posterior probability of membership conditional on their individual item profiles. Lastly, we tested a factor mixture model, which allows for a continuous latent factor within each of the classes, indicating that alcohol use severity varies by class due to some common factor within a particular class. These analyses were done in MPlus Version 5 (Muthen & Muthen, 2008).
