Separate linear latent growth curve models were then used to examine changes in heavy drinking and problems from high school through college. Models included latent intercept and slope variables for either frequency of heavy drinking or problems. Weights from the latent intercept variables to each of the 8 indicators (heavy drinking/problems assessed over 8 equally spaced time points) were constrained to 1. Weights from the latent slope variables to the indicators were given equally spaced values ranging from −7 to 0. To model linear increases in heavy drinking and problems over time, we set the intercept to the final time point. Had we set the intercept to time 1, AO would have predicted prior drinking for all participants who initiated use after the initial assessment. An additional advantage of setting the intercept to the final time point was that analyses in which AO and delay predicted intercepts represented a truly prospective test of the effects of AO and delay on later drinking. The disturbances for the latent intercept and slope were allowed to freely covary. For adjacent time points, the
