Prior to the GWAS analyses, we tested for sex and age effects in our sample in a linear regression model with binary adult antisocial behavior as the dependent variable. We conducted genome-wide association analyses in three study designs using imputation dosage genotypes: 1) combined studies, logistic regression on case-control status with sex, age and study as covariates 2) combined studies, linear regression on symptom count, same covariates as 1), 3) repeated analyses 1 and 2 for the two studies separately with age and sex as covariates. This allowed us to determine consistency among the associations across the studies. Given our family based sample, Merlin offline [26] was used since it accounts for family relationships including MZ twins. Minx (as implemented in Merlin) was used to perform association analyses on the X-chromosome. Ancestry principal components were not significantly associated with the phenotypes and were not included as covariates.
