Furthermore, since all correlations between the aggression ratings from the different informants were low to moderate, logistic regression models were performed using combinations of total aggression scores from different informants, adjusted by sex and age, to predict ASPD (see Table 4 for combinations tested). In general, AUC values for these models were higher than models with only one informant’s aggression rating included; the highest AUC values were seen in the model with all age 14 ratings and the model with all 5 ratings (AUC: 0.81 and 0.82, respectively) (Table 4, Fig. 1c). In both of these models, only the ORs for the age 14 teacher and self ratings were significant, therefore, an additional model was run with only these two ratings included. In this model, the ORs for the teacher and self ratings were significant and robust (OR 1.6 and 1.9, respectively), and the AUC value remained high (AUC = 0.80) (Fig. 1d). Additionally, to confirm that the varying sample sizes of the different informant combination models were not driving the differences seen in AUC values, all combinations tested were
