We conducted a series of three bi-factor-S-1 models, in which we defined the G factor by not modeling one of three aggression phenotypes: alcohol-related, drug-related, and non-substance-related aggression. Each of these bi-factor-S-1 models exhibited improved fit to the data as compared to the three-factor model (Table 5). In comparing the bi-factor models to each other, drug-related aggression was the worst fitting general factor, non-substance-related aggression was a better fitting general factor, and alcohol-related aggression as the general factor was the best fitting model we tested (Figure 4; Table 5).
