Fig. 2.

Illustration of the PLS analysis pipeline. The cross-covariance matrix between imaging data and behavioural data is estimated. This matrix is then inputted to singular value decomposition, yielding singular values for each LV, as well as imaging, behavioural, and subject weights. Then, loadings onto each LV were calculated as the correlation between subject weights and the original imaging and behavioural data, respectively. PLS; partial least squares. LV; latent variable.