the variation of the relationship at causal loci owing to imperfect LD. However, the heritability estimate recovered when we adjusted relationship estimates using equation [9] (Online Methods; c = 0). In case (ii), even if we included the causal variants in the analysis, we still underestimated heritability, because the causal variants have lower frequency than the SNPs, on average, and have less LD with the SNPs than the SNPs have with other SNPs. Similarly, when we adjusted the relationship estimates with equation [9] (c = 6.2 × 10−6), we obtained unbiased estimates of h2. These results are consistent with the inference we draw from the empirical data. The results show that the estimate of variance caused by causal variants is unbiased regardless of the number of SNPs used, provided the method proposed here is employed.
