To address missing data, we used multiple imputation (Little and Rubin, 2002). Missing data for key variables ranged from 0.05% for baseline AA attendance to 7.8% for drinking at months 13-15. Since missing data patterns were non-monotone (i.e., many were intermittently missing), the Markov Chain Monte Carlo (MCMC) method for multiple imputation was used (Gilks et al., 1996). We performed ten imputations using MI and MIANALYZE in SAS 9.2, and reported statistics are averaged across imputations (Barnard and Rubin, 1999).
