The majority of the full sample self-identified as White (90.4%), but we selected individuals for the current analysis on the basis of the first 10 genetic principal components computed with EIGENSTRAT (Price et al., 2006). Since EIGENSTRAT solutions are sensitive to close relatives, one member of each close relative pair was excluded. Full details are provided in Miller et al. (in press). To identify individuals as White for the present analysis we first found the centroid in the 10-dimensional principal component space for subjects thought to be white, and then computed the distance of every individual from this centroid. A small number closest to the centroid were used to compute the centroid and variance-covariance matrix to compute Mahalanobis distance to the centroid. A hyperellipsoid of constant Mahalanobis distance to the centroid was constructed such that points inside were those closest to the centroid of the white group. The hyperellipsoid was continuously expanded, updating the centroid and variance-covariance matrix every time another subject was added to the white group inside the hyperellipsoid. Expansion continued until further expansion started to bring in
