All twin modeling was conducted with the statistical software package OpenMx (2011), which uses maximum likelihood estimation to evaluate the extent to which variance in each phenotype can be attributed to A, C, and E. Twin models were fitted to raw categorical data. Modeling was conducted with the AA subsample, the EA subsample, and the full sample. Three bivariate triangular (also known as Cholesky) decomposition analyses were fitted to assess the degree of overlap in genetic and environmental influences between age at first drink and problem use as well as heritabilities for each phenotype. A series of sub-models were tested to assess the statistical significance of pathways representing additive genetic, shared environmental, and non-shared environmental influences, both within and across phenotypes, to derive the best-fitting bivariate models. The sub-models were tested by calculating the difference between the −2 log likelihood fit of the full model and the nested sub-model, which is distributed as chi-square for the given degrees of freedom.
