The LGM was conducted in two steps. First, the growth model was estimated without the covariates (i.e., unconditional model). Various change patterns were explored; this article reports the model that best fit the data in terms of Bayesian information criteria. Second, the covariates were incorporated into the growth model to examine their effects on the change rate in the numbers of drinks consumed (i.e., conditional model). In these ZIP models, the coefficients for the count part reflected effects on the level of HED, whereas the coefficients for the inflation part reflected effects on the probability of not engaging in HED. Coefficients for the intercepts indicated effects of covariates at the initial time point (age 12 years), whereas coefficients for the slopes indicated effects on changes as age increased yearly. In the inflated part, a positive effect of a covariate on the slope implies higher levels of the covariate would lead to a higher probability of remaining in the zero category (i.e., a higher probability of abstaining from HED). Whereas, a negative effect implies higher levels of a covariate would lead
