To determine which time-frequency components to include in our Cholesky model, we conducted PCA on the 15 measures (5 TF-PCA components, 3 electrodes) plus EXT, separately for each gender. The raw data were centered about the variable means and scaled by their respective standard deviations, equivalent to PCA on the correlation matrix. Although most of the variation in the data was due to the 15 time-frequency measures, the second (unrotated) component in both genders included a loading for the EXT composite in addition to certain TF measures, while the third component loaded almost entirely on the EXT composite. The pattern was clear: the TF measures and EXT each loaded on separate components, and one component included loadings for both, thus capturing covariance between the two domains. TF-PC1 (with peak energy centered around 2.5 Hz and 330 ms; see Figure 2) and TF-PC4 (with peak energy also centered around 2.5 Hz, but occurring slightly later in time, around 400 ms; see Figure 2) both loaded with EXT, with very similar loading patterns and magnitude of loadings for the two genders, thus
