were in the same direction, power of these four methods decreased with increasing residual correlation between the traits, which also corroborates previous findings [7], [12]. PCHAT and TATES were relatively independent of the underlying (genetic) correlations of the traits. For PCHAT, this can be explained by the fact that it constructs the optimal linear combination of traits from an heritability point of view, thereby essentially removing the influence of residual correlation on power [4]. For TATES, it was described that the power was not influenced by opposite effects of the QTL on the traits, because of its reliance on p-value information [15]. UV-MA did however severely suffer from a negative genetic correlation between the traits; indeed, in this scenario it performed equal or worse than standard UV analysis. These findings are not unexpected; a negative genetic correlation between the traits is disastrous for the power of a MA, because the direction of effect is taken into account. An alternative meta-analysis approach is Fisher's method [27]. As it combines univariate p-values into one test statistic, similar to TATES, it does not suffer from a differential sign of effect. For scenarios with a negative genetic correlation, Fisher's method performed better than UV-MA
