Therefore, given the number of SNPs used, we can correct the estimate of the variance explained by the SNPs for incomplete LD with causal variants, if causal variants have the same allelic frequency spectrum as genotyped SNPs. Using the same linear model as above, but corrected for this incomplete LD (c = 0), we estimated the proportion of variance explained by causal variants to be 0.54 (s.e. = 0.10, Table 1). This estimate assumes that the LD between SNPs and causal variants is as strong as that between the genotyped SNPs. However, if the causal polymorphisms tend to have lower MAF than the SNPs that have been assayed, as expected from neutral and selection theories of quantitative genetic variation6,22, we expect the LD between SNPs and causal variants to be reduced. When we used SNPs with a MAF < 0.1 as proxies for causal variants we found c = 6.2 × 10−6. Using this value of c to correct for incomplete LD, we estimated the proportion of variance in height explained by causal variants to be 0.84 (s.e. = 0.16;
