A drawback of these multivariate methods is that their power depends on the specific configuration of phenotypic correlations and on the location of the GV effect (e.g., on the latent dimension, or specific to one of many correlated phenotypes). For instance, when the ideal model (Figure 1a) holds, MANOVA is decidedly less powerful than univariate analyses based on sum scores. MANOVA, however, easily outperforms the sum score approach when the GV affects only one of multiple strongly correlated variables (e.g., Figure 1d–1f) [4]–[5], [14].
