The dependent variables were log transformed to improve computational efficiency during the modeling procedure. In the high school model, because of the number of parameters and the fact that there were no expectations of differential prediction of the time-varying predictors with the alcohol slopes across the high school years, the regression of the individual time points on the time-varying predictors were set to be equal across time. Models were tested using Mplus (Muthén and Muthén, 1998–2006).
