To assess robustness to normality of the marker effects, the simulations were repeated with the effects drawn from Laplace distributions and then rescaled to give the same explained variance and correlation as before. Instead of π0 = 0.95 and P<0.1, simulations with π0 = 0 and P<1 were performed to verify that this situation does not assume normality. The results in Table S4, Table S5 and Table S6 confirm this to be the case, whereas when π0 = 0.99 and P<0.001 the analytic expressions tend to underestimate the power and accuracy. This is due to the heavier tails of the Laplace distribution compared to the normal, and quantitatively different results would be seen for different generating models. Again, bias and under-coverage is seen when there are 1000 markers with effects.
