We obtained un-imputed genotypes from Psychiatric Genomics Consortium (PGC) controls from seven European cohorts genotyped on the same array (Supplementary Table 2). To simulate population stratification on a continental scale, we assigned case/control status based on cohort membership, then computed association statistics for each pair of cohorts (note that in this simulation setup the expected mean χ2-statistic is 1+bNFST, where b is the correlation between phenotype and ancestry and N is sample size, ref 12). To simulate population stratification on a national scale, we computed the top three principal components within each cohort, then computed association statistics using each of these principal components as phenotypes. Quantile-quantile (QQ) plots from simulations with population stratification and polygenicity show indistinguishable patterns of inflation (Fig. 1a,b), but the average LD Score regression intercept was approximately equal to λGC in simulations with population stratification (see Supplementary Table 3a for simulations with continental-scale stratification and Supplementary Table 4a for simulations with national-scale stratification), and near 1 in simulations with polygenicity (Supplementary Figures 1–5). Furthermore the qualitative appearance of the pattern of inflation as a function of
