Extended Data Fig. 2: Calibration of FMR jackknife standard errors.

Simulations were performed under a normal mixture model with small-, medium- and large-effect SNPs (similar to Fig. 1d), at sample size N=460k, N=145k or N=50k. For different effect-size thresholds, I calculated the standard error of the proportion of random-effect heritability explained by SNPs with effect sizes less than that threshold. Bar plots show root-mean-squared jackknife standard errors (blue) and empirical standard errors (orange) based on 25 replicates. At large sample size (N=460k), standard errors were sometimes underestimated, probably due to the nonnegativity constraints in the regression. Caution is needed when making comparisons between the genetic architecture of different traits, as underestimated standard errors could lead to false-positive differences.