a covariate (Supplementary Note) suggests that it was not an important heritability determinant in NTR. However, the important effect of sampling variation has not been fully explored. First, we assessed the gamma fit by artificially adding sampling error to the “true” distribution, showing that it fits our estimated h2 (Figure 3a). A similar approach quantifies the impact of sample size (Supplementary Note), again using the gamma model obtained from NTR, but inflating the sampling variation to reflect the smaller MuTHER sample size. The resulting estimated h2 distribution is similar to that reported in MuTHER (Figure 3b). We suggest that, despite other differences between the studies, much of the apparent differences may be attributable to sample size effects. Analysis of the recent Brisbane Systems Genetics twin study47 suggested a similar conclusion (Supplementary Figure 4a, Supplementary Note). Although we conclude the underlying heritability in all of these studies may be comparable, this is a distributional statement, and larger sample sizes are desirable in terms of accuracy. Supplementary Figure 4b shows accuracy prediction as a function of sample size – even with the NTR sample size we predict that the rank correlation between true vs. estimated heritability is only slightly greater than 0.5.
