Figure 6 shows the final structural model containing the mediating variables that emerged in the preceding analyses. The model was constructed using all the variables described thus far. Initially, a saturated model was constructed in which all possible paths were included. (Paths pointing backward in time were not allowed.) Modification indices were consulted to adjust the model to optimize fit with the data. The resulting model was an excellent fit to the data (model chi-square = 122.51; df= 106; p = .13; CFI=1.00, RMSEA=.036). As shown in this model, direct paths from the treatment contrast term to the PDA slope and intercept latent variables were not significant. The effect of treatment appears to have been mediated through effects on AA attendance at posttreatment, Number of Abstinent Friends in the social network (measured at posttreatment), coping skills at 15 months, and self-efficacy at posttreatment and at 15 months. In addition, the mediating variables were influenced to some extent by baseline drinking. Finally, self-efficacy at 15 months, which appeared as a strong predictor of both PDA slope, was itself a function of posttreatment drinking, posttreatment self-efficacy, treatment condition (being higher in NS), and Number of Abstinent Friends.
