Predictions using Latino effect sizes that were not adjusted for genetic ancestry (LATunadj, EUR+LATunadj, EUR+LATunadj+ANC, as compared to LAT, EUR+LAT, EUR+LAT+ANC) were much less accurate (S2 Table), as in previous work (C.-Y. Chen, Han, Hunter, Kraft, & Price, 2015); this is consistent with the fact that LATunadj predictions were dominated by genetic ancestry (adjusted R2 = 0.37; S3 Table). We also observed a modest correlation (adjusted R2 = 0.025) between the EUR prediction and genetic ancestry (S3 Table), again reflecting small genetic effects of ancestry on phenotype that can arise from random genetic drift between populations at causal markers. The relative performance of the different prediction methods was similar in simulations in which phenotypes explicitly contained an ancestry term, representing environmentally-driven stratification (S4 Table).
