We analyzed the ten NFBC66 phenotypes with EMMAX using a three-step procedure (see Online Methods). First, we computed a pairwise relatedness matrix from high-density markers, which we used to represent the sample structure. Second, we estimated the contribution of the sample structure to the phenotype using a variance component model, resulting in an estimated covariance matrix of phenotypes that models the effect of genetic relatedness on the phenotypes. Third, we applied a generalized least square (GLS) F-test24, or alternatively a score test25, at each marker to detect associations accounting for the sample structure using the covariance matrix.
