Table 1 also shows power for the same simulations. There is overall a modest drop in power for our approach, except under strong positive genetic correlation where there is a small increase. The power loss is greatest under strong negative genetic correlation. For individual SNPs, substantial differences can occur between methods. The most extreme cases entail a greater gain in power for the unadjusted analysis than for our approach, although this must be offset against the increased type-1 error. Supplementary Tables 1 and 2 show absolute bias and mean square error. The pattern is similar in that the adjusted analysis has less bias, although this is offset by its increased standard error so that the differences in mean square error are barely discernible.
