Figure 1g illustrates the power results of 12 selected simulation scenarios for the sum score procedure, MANOVA and TATES, given a GV explaining .5% of the phenotypic variance. As expected [4]–[6], the sum score procedure has excellent power to detect the GV, if the phenotypic data are generated according to a 1-common factor model, and the GV effect is on this factor (Figure 1g: A1–A3). However, if either the location of the GV effect or the data-generating process is different, the power of the univariate sum score procedure drops to levels often <.10 (Figure 1g: B1,C1,D1,E1–3,F1–3). In 9 out of the 12 scenarios we considered, the power of TATES was 2.5 to 9 times higher than the power of the sum score procedure. As expected [4]–[5], [14],the power of MANOVA is especially high if the GV effect is specific to only one of many highly correlated phenotypes (Figure 1g: D1,E1). The power of MANOVA drops if the phenotypic correlations are lower, or if multiple phenotypes are subject to the GV effect. In contrast, TATES is only slightly less powerful than
