In addition to our EUR GWAS meta-analysis and our trans-ancestry meta-analysis (METAFE), we re-analysed five down-sampled GWASs as shown in Table 2. These down-sampled GWASs include various iterations of previous efforts of the GIANT consortium and have a sample size varying between around 130,000 and 2.5 million (EUR participants from 23andMe). To ensure sufficient genomic coverage of HM3 SNPs we imputed GWAS summary statistics from Lango Allen et al.19, Wood et al.20 and Yengo et al.3. with ImpG-Summary (v.1.0.1)64 using haplotypes from 1KGP as a LD reference. GWAS summary statistics from Lango Allen et al. only contain P values (P), height-increasing alleles and per-SNP sample sizes (N). Therefore, we first calculated Z-scores (Z) from P values assuming that Z-scores are normally distributed, then derived SNP effects (β) and corresponding standard errors (s.e.) using linear regression theory as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta =Z/\sqrt{2{\rm{MAF}}\times (1-{\rm{MAF}})\times \left(N+{Z}^{2}\right)}$$\end{document}β=Z/2MAF×(1−MAF)×N+Z2 and SE = β/Z. Imputed GWAS summary statistics from these three studies are made publicly available on the GIANT consortium website (see ‘URLs’ section). We next performed a COJO analysis
