Our method does produce biased estimates when model assumptions are violated. Specifically, if the genetic distance we estimated over common variants (0.182; see Online Methods) differs from the distribution over causal variants, our method can be either inflated or deflated. If selection were acting on the causal variants their FSTC could be higher or lower depending on the direction of selection. In the case of positive selection in one of the ancestral groups but not the other, the true value of FSTC will be larger than our genome-wide estimate and so our h2 estimate will be inflated. For example, estimates for white blood cell count were larger than hpub2, due to strong selective pressure at the Duffy locus27,37. However, strong positive selection is believed to be rare in recent human evolution29. If a large proportion of phenotypic variance is due to rare variants then incorrect estimates of FSTC may induce bias. However, previous reports suggest that rare variation explains a small proportion of total heritability25,26.
