As reported previously[6], repeated measures latent class models with 2 to 15 classes were estimated and a 7-class model was retained as the optimal solution based on the BIC, aBIC, entropy, and substantive interpretation of the latent classes. The 7-class model also provided an optimal solution in the current study when covariate predictors were included in the model. A description of each of the classes was derived by examining the probabilities of abstinence, low risk drinking, or heavy drinking within each class (Table 2). The 7-classes differed significantly in the expected direction on biological measures of %CDT and GGT in COMBINE and UKATT, respectively, such that the abstainers and low risk drinkers (Classes 5–7) had significantly lower %CDT and GGT than the classes with more self-reported drinking (Classes 1–4).
