Finally, latent growth modeling (LGM) using structural equation modeling procedures was used to evaluate the effects of treatment and possible mediators of treatment on PDA outcomes over the 27 months of follow-up. A bootstrapping resampling procedure was used to estimate model test statistic p values and parameter standard errors under the nonnormal data conditions in this dataset (Arbuckle, 2006). A total of 153 cases with complete data were available for these analyses. Those who were eliminated from these analyses on the basis of missing data did not differ from those included in the LGM on age, sex, ethnicity, education, marital status, baseline drinking, alcohol dependence, size of social network, AA involvement, treatment attendance, or treatment assignment.
