Finally, we studied the effect of 10% missingness completely at random (MCAR) or 10% blockwise missingness on the power to detect GVs in three different genotype-phenotype models (see Materials and Methods for details and Tables S19, S20, S21, S22, S23, S24, S25, S26). Power was hardly affected in 1-factor models with the GV effect on the factor. However, if the GV effect was specific to one of the phenotypes (either in factor models or network models), the power of MANOVA usually showed a 5–6% larger drop in power compared to Simes and TATES. Only when the GV effect was specific to a phenotype showing blockwise missingness was the drop in power of Simes and TATES similar to, or even slightly higher (2–3%) than, the power drop observed for MANOVA.
