To model a polygenic quantitative trait, we assigned per-allele effect sizes drawn from N(0, h2/(2p(1-p))−1/2/M) to varying numbers of causal variants and for varying heritabilities in an approximately unstructured cohort of 1000 Swedes. In all simulation settings, the average LD Score regression intercept was close to one. We note that if there are few causal variants, the LD Score regression estimates are still unbiased, but the standard errors become very large, meaning that this approach is best suited to polygenic traits (Supplementary Figures 3–5).
