test when sample sizes become extreme, as in the present case. Results were within expected limits and consistent with a limited effect of population stratification on the meta-analysis results44 (Supplementary Table 8). Then, we compared the sign of SNP effect size estimates between EUR-stratified results and within-sibling GWAS results from the UK Biobank, finding sign concordance estimates of 63.4–80% across phenotypes, all of which were significantly higher than would be expected if our results were driven entirely by population stratification or cryptic relatedness and were consistent in magnitude with other large-scale association studies49. Finally, given reduced power in the within-sibling GWAS, we additionally compared the sign of SNP effect size estimates between EUR-stratified 23andMe summary statistics (the largest participating cohort) and EUR-stratified summary statistics with all cohorts except 23andMe, finding sign concordance estimates of 94.3–100%. See the Supplementary Note for further details on the methods and full results, including the list of excluded variants and loci.
