The best-fitting trivariate model is shown with unstandardized path coefficients in Fig. 1. The proportion of total variance attributable to genetic and environmental sources is reported with 95% confidence limits in Table 2. The final model was derived by conducting a series of tests, in two stages. First, pathways whose confidence intervals in the full model included 0 (the A pathway for alcohol and the C pathways for all three phenotypes) were dropped one at a time and the resulting sub-models were compared to the full model. Of these, only the removal of the C pathway for alcohol significantly affected model fit. In the second step, starting with the pathway with the lowest Δχ2 value (C component for cannabis; Δχ2(1) =−3.85), the remaining three pathways were dropped in combination and tested against the reduced model from the previous step. No deterioration in model fit was observed when the C component for cigarettes was dropped in combination with the C component for cannabis (Δχ2(2) =−7.41), but dropping the A pathway for alcohol resulted in a poorer fit to the data (Δχ2(2)
