As shown in Fig. 4a, height phenotypes for worldwide populations (y-axis) are positively correlated with GIANT-based18 polygenic scores for height (x-axis), but not with UK-Biobank-based polygenic scores (4b) or East Asian GWAS based polygenic scores (4c). Polygenic scores constructed using only genome-wide significant variants from GIANT (top left) were positively correlated with height phenotypes (r = .67, p = .002), as were scores constructed using larger numbers of GIANT-based variants (e.g., all variants, top right, r = .59, p = .008). Results in 4b and 4c demonstrate that correlations (or lack of correlations) between height and polygenic scores for height are dependent on discovery GWAS. There are numerous reasons why polygenic scores differ between studies. However, recent findings suggest that correction for population stratification may not have been adequate in GIANT30,31, and therefore the positive correlations observed in 4a could be partially due to uncorrected population stratification. The dependence of correlation estimates on discovery GWAS is further illustrated in 4c, in which the point estimate for correlation between height and East Asian GWAS based polygenic scores for height is negative
