In the additional impulsivity ratings analyses, significant mean differences in impulsivity between ASPD and non-ASPD cases were seen (for males, as well as females for age 14 ratings) (Online Resource 3). Most correlations between informants and aggression and impulsivity scores were low to moderate (r range 0.10–0.54), with teacher age 12 and 14 aggression and impulsivity scores having higher correlations (0.73 and 0.68, respectively). In univariate impulsivity models, all informant impulsivity scores significantly predicted ASPD. However, in impulsivity residual models (i.e., each impulsivity measure uncorrelated with corresponding aggression measure), only self and co-twin age 14 ratings still predicted ASPD. Furthermore, in aggression residual models (i.e., each aggression measure uncorrelated with corresponding impulsivity measure), all of the age 14 aggression ratings (teacher, self, co-twin) still predicted ASPD. Finally, when comparing a multi-variate, multi-rater aggression and impulsivity model to the multi-rater aggression model with all 5 ratings simultaneously in the model, the teacher 14 and self 14 aggression scores were still the only significant predictors of ASPD and AUC values were the same (0.82), though a likelihood ratio test (comparing these two ‘best’ models) indicated the aggression + impulsivity model was significantly better (p = 0.015).
