Following the initial aggression models, models including multiple aggression ratings were considered. First, for each informant, low to moderately correlated (r < 0.65) direct and indirect aggression scores were entered into logistic regression models simultaneously to examine whether separating the total aggression rating into subtypes clarified which type of aggression was contributing most to the ASPD prediction. Second, low or moderately correlated total aggression scores from different informants were entered into logistic regression models simultaneously to examine whether additional assessments of aggression improved ASPD prediction.
