Chi-square tests and t-tests were conducted for bivariate analysis. We conducted generalized estimating equation (GEE) logistic regression models for repeated measures interacting race by treatment to analyze whether the relationship between formal substance use treatment services and abstinence was different for white and black alcohol dependent drinkers, over seven years. GEE models estimate a single overall average group trajectory for the outcome variable and have the advantage of providing unbiased trajectory parameter estimates even when the variance structure is not correctly specified (Diggle et al., 1994; Neuhaus, 1992). This technique allows the inclusion of respondents even if they are missing one or more interviews, and baseline variables can be added for control. Time invariant covariates included in the model were gender, age, marital status, baseline education, baseline income, baseline employment status, baseline DSM-IV alcohol dependence, baseline ASI measures (alcohol, drug, psychiatric, medical, social and legal) and race. Additionally, substance use treatment utilization prior to or at baseline was included in the model assessing substance use treatment utilization. Similarly, AA utilization prior to baseline was included in the model assessing AA
