MANOVA was not conducted on all available data because in standard MANOVA, cases are excluded listwise, resulting in a very low sample size when missingness is MCAR. In theory, fitting MANOVA on the raw data using Full Information Maximum Likelihood (FIML) is possible in software like LISREL, Mx, or Mplus [33], [35]–[36], but this is time consuming in a genome-wide context. Here, we chose to stick to the common practice of MultiPhen [11] and Plink [13], which is mean imputation.
