We used GCTA to decompose the phenotypic variance in each measure into variance components due to the additive effects of all genotyped SNPs and residual effects. GCTA consists of two steps in which the genetic similarity between all pairs of individuals is obtained via a pairwise genetic relationship matrix (GRM), followed by construction of a mixed effects model using genetic similarity as a random effect to predict each phenotype. In our identification of subjects of European descent from SAGE, we used GCTA to systematically remove one of any pair of individuals who were more related than second cousins in order to control for cryptic relatedness, which could artificially inflate SNP-heritability estimates. The GRM used in all analyses comprised the 2596 unrelated individuals. Univariate and bivariate models were fitted to the phenotypic data while controlling for age, gender, study origin (to account for mean differences/batch effects between the different samples within SAGE), and the first five ancestral principal components to account for stratification effects within individuals of European descent (11).
