Based on the best-fitting model, class membership was utilized in a two-step process. First, an individual was assigned to the latent class with the highest probability, and discrete class memberships were compared using post hoc validators such as gender, age at onset of regular drinking (from the first interview in which the individual began drinking regularly), and DSM-IV AD. Interpretation of the classes was based on the validators. Gender and DSM-IV AD differences between classes were assessed with a chi-squared test; number of DSM-IV criteria, age and age at onset of regular drinking were tested using an analysis of variance (ANOVA), with gender and birth cohort as covariates. Tukey-adjusted t-tests were employed to test for pairwise differences when there was a significant (p<0.05) main effect of latent class assignment.
