If the relationship at causal loci is predicted without error by the observed SNPs, β should equal one. When we applied this approach in our data, we found that for any of the MAF threshold θ, var(Ajk) is proportional to N whereas cov(Gjk, Ajk) is constant, irrespective of N (Supplementary Fig. 5). Consequently, we established an empirical linear relationship between β and the number of SNPs,