A partial least squares approach for within-subject correlations in neuroscience

Partial least squares correlation (PLSC) is a multivariate technique used in neuroscience to capture patterns of relationships between neural data and covariates of interest. Conventionally, PLSC captures between-subjects correlations (e.g., between average neural activation and average task performance). However, within-subject correlations (e.g., trial-by-trial relationships between neural activation and subjective ratings or stimulus attributes) are often of interest as well. Thus, we introduce an extension of PLSC to capture within-subject correlations. We validate this approach against related parametric methods and demonstrate that it detects known effects in simulated data and reproduces published findings in open fMRI and EEG datasets. This method is freely available as part of the pyplsc and mne-plsc Python packages.