Because the final number of samples within the blood and brain groups differed, we performed post-hoc power calculations to compare ability to detect eQTLs (Fig. 1). Based on our previous work in brain (Gibbs et al., 2010), the strength of the association varies substantially for different eQTLs. Therefore, we estimated power over a range of minor allele frequencies and of effect sizes for the eQTLs, using Z as a measure of effect size standard deviations of difference for each minor allele under an additive model. As an example of power in the two datasets at a realistic pair of these parameters, the blood dataset had 98.8% power to detect eQTLs at an effect allele frequency of 0.2 and an additive effect size of Z = 0.5 whereas the brain dataset had 93.9% power to detect the same magnitude of effect. This analysis demonstrates that the difference in power in the two datasets is minimized as the fraction of true eQTL effect sizes rises. For eQTLs with moderate effect sizes (Z > 0.2) we were reasonably powered in both series; therefore, we proceeded to compare the ability to detect eQTLs in both datasets.
