Figure 2.

Examining the bias in the eigenvalue estimates of pleiotropic covariance. (A) We estimated the per-generation mutational variance in the trait combination defined by the eigenvector m₁ both as the eigenvalue of m₁ (from multivariate analysis using model 2; y-axis) and from the univariate analysis of index traits defined by m₁ (the univariate mutational variance from model 1; x-axis). Estimates for all 171 m₁ are plotted, with points shaded in black falling more than two standard deviations from the mean difference between the two estimates per trait. The dotted line indicates the 1:1 relationship between estimates, and the solid line indicates the linear regression through the observed points (excluding the seven, shaded, outliers). (B) The per-generation mutational variance in m₁ in the observed data (eigenvalue of m₁, y-axis) and in shuffled data (after projection, x-axis). The solid line indicates equality (1:1) of variance in shuffled and observed data; six m₁ fall below this line and are shown in solid black. The 11 m₁ for which there was no variance in the shuffled data (variance less than zero) are shaded.