We identified 36 loci with multiple associations for height. We have shown by examples of pairs of multiple associated SNPs that marginal SNP effects will be underestimated (overestimated) if their trait increasing alleles are negatively (positively) correlated, consistent with the findings from a GWAS of gene expressions25. However, this is not necessarily always the case when there are more than two associated SNPs in LD with each other, and the generality of these results depends on the actual LD correlations of all segregating causal variants at a locus. If one of the associated SNPs at each locus is causative, then there must be multiple causal variants in that region, because the joint effects have already taken the LD into account, such that, conditional on the causal variant in the model, the effects of any of its proxies would not be statistically significant. However, it is unlikely that the SNPs themselves are causative, because the ~2.5 million SNPs in the HapMap 2 panel of Utah residents of Northern and Western European ancestry (CEU)26 represent only a fraction of all the polymorphisms
