We utilized a generalized estimating equation (GEE) framework to control for relatedness via a kinship matrix (Chen & Yang, 2010) in the EA and AA family samples separately. Autosomal SNPs with minor allele frequency (MAF) ≥ 3% (estimated using founders in the EA and AA families separately) were included in the discovery analysis to test for association using an additive model of SNP effects. Due to the skewed distribution of externalizing scores, we transformed scores for the analysis (square root, left anchored at 1). Sex, age of last observation, GWAS array, and the first ten principal components computed from GWAS data were included as covariates. Because we focused primarily on the results from the combined-ancestry meta analysis, we used established thresholds for genome wide significance (p < 5 × 10−8). Finally, we used LD score regression (Bulik-Sullivan et al., 2015) on ancestry-specific summary statistics to 1) determine the degree to which inflation in the association test statistics was due to population stratification rather than polygenic signal; and 2) estimate the narrow sense, SNP-based heritability (h2snp).
