Table II presents the power of the three trend tests of association at a 5% significance level for a high-risk allele frequency of 20% and an allelic odds ratio of 1.5, illustrating in more detail the effect of varying levels population structure. Power is estimated over 5,000 replicates of data, for a range of FST between the case-control population and external cohorts. Also presented are mean maximum likelihood estimates of the allelic odds ratio, together with 5- and 95-percentiles across the 5,000 replicates of data. Clearly, T_F is most powerful for large FST, but it is anticonservative, and thus is not a valid test of association. T_Fmds is, in principle, more powerful than T_CC, although the difference between the tests decreases and approaches zero, as FST becomes very large. This is a result of the fact that the external samples are so genetically different from the case-control source population that they add very little information to the disease association after adjustment for the axes of genetic variation.
