Twin correlations were conducted (separately for AA and EA participants) to determine whether an ACE or ADE model would better fit the data. DZ twin correlations were greater than half of MZ twin correlations in all cases except for problem use in African Americans (rMZ:rDZ was 0.36), thus supporting an ACE model. (A summary of the model fitting steps for the AA subsample, EA subsample, and full sample bivariate models (with fit indices, point estimates and confidence intervals for variance components, and results of tests of change in model fit for each sub-model) is available upon request.) The best fitting models, indicated by the lowest Akaike information criterion (AIC), were ACE for both phenotypes in the EA subsample and ACE for first drink, AE for problem use in the AA subsample. The best fitting model using the full sample, in which pathways for the AA and EA participants were set to be equal, also included A, C, and E variance components for both phenotypes. Additional tests conducted with each of the three best-fitting models to assess significance of the cross-phenotype
