For each outcome, we first fit a best-fitting baseline model including fixed effects for sex, age, and polynomials of age up to the cubic term (for analyses of count data outcomes) or squared term (for analyses of binary outcomes). The best-fitting baseline model included a sexby-age interaction for all continuous phenotypes, but not for the binary alcohol dependence or abuse phenotype. Preliminary model fitting did not support the inclusion of an interaction term between sex and ALDH2 genotype, nor any role for elder sibling sex, for any alcohol use outcomes, so these terms were excluded from the baseline model and subsequent model comparisons. To the baseline model, we fit a series of models adding terms first for the effect of ALDH2 genotype, then for the change in ALDH2 effect over time (that is, an ALDH2 by age interaction). Next, separately for the effects of parental alcohol use and misuse (ie: the parental combined alcohol problem index), elder sibling alcohol use and misuse (ie: the elder sibling combined alcohol problem index), and peer deviance, we fit a successive series of incrementing
