The principal component analysis implemented in EIGENSTRAT is sensitive to pairs of close relatives, so one member of every such pair was excluded in initial computations, but the genotypes of relatives were then projected onto the components constructed from the set of unrelated subjects (this was accomplished using the “-w poplist” option to smartpca.perl). We identified relative pairs, even a few between-family pairs, by running a PLINK “–genome” analysis to produce a matrix of “Z1” identity-by-descent (IBD) statistics for all pairs of individuals. The Z1 statistic is the estimated probability that two individuals share exactly one allele from a common ancestor, averaged across the autosomal genome. We chose to classify pairs of individuals as unrelated if their Z1 was less than .10, which would be roughly midway between first-cousin once-removed and second-cousin except that the ethnic structure of our sample tends to amplify distant relationships in the PLINK analysis (inflating Z1 somewhat, but inflating Z2 and Pi-Hat much more), especially for members of minority groups. The matrix of Z1 statistics was used to select the largest possible subgroup of unrelated
