Figure 3 presents the power of the three trend tests of association at a 5% significance level, this time for external cohorts which are divergent from the case-control source population (FST = 0.01), otherwise with the same simulation parameters as before. This time, there is some loss of power for T_Fmds compared to T_F. However, T_F is an anticonservative test of association in this setting (false-positive error rate is 8.3%), and thus is inappropriate. Furthermore, despite the reduction in power of T_Fmds, it remains more powerful than T_CC. The same conclusions are reached, irrespective of high-risk allele frequency (see supplementary Fig. 2 for a high-risk allele frequency of 5%).
