Ntzani et al. [22] examined differences in effect sizes from a set of 108 associations discovered by GWAS and for which data for various ancestries was available. Because of the sophistication of their approach, they had to focus on 12 diseases and 4 anthropometric traits, as well as on a relatively short (∼30) list of GWAS that either use samples with different ancestries in the replication stage or compare their own results with previous papers using different ancestries [22]. In contrast, we took the simpler approach of studying replicability in the studies with largest sample size, so we could gather attempts from multiple GWAS on the same diseases and were able to construct a larger database. Ntzani et al. [22] found overall consistency in effect direction across ancestries (∼82%, similar to ours of 85%), but with modest correlations in effect sizes, (rho≈0.33) that would seem contradictory with the large correlation in odds ratios we report here. Nevertheless, an almost identical correlation in OR would have been observed if limiting the study to the 22 SNPs that are shared between Ntzani
