Individual-level data from each of the seven interview waves that comprise the prospective study were combined, creating one dataset in which subjects’ alcohol and posttraumatic stress history, and AUD and/or PTSD diagnoses were coded positive if a participant reported experiencing them at any interview. The dataset also included information on DSM-IV qualifying PTEs during the interview wave in which they were first reported as well as demographics at each subject’s most recent interview. After running descriptive statistics on the sample’s characteristics and bivariate tests between the alcohol and posttraumatic stress variables of interest, the 11 symptoms used to generate a diagnosis of DSM-5 AUD and the three clusters of symptoms (re-experiencing, numbing, and hyperarousal) used to determine DSM-IV PTSD were fitted to a series of LCA models with two through six class solutions. LCA is a data reduction technique that aims to empirically categorize typologies from a set of observed, discrete variables (McCutcheon, 1987). A multinomial logistic regression, adjusted for age, sex, race, household income, current drinking status, and familial clustering, examined associations of the latent classes with specific traumas
