Common pitfalls during model specification in psychophysiological interaction analysis

Abstract

Psychophysiological interaction (PPI) analysis is a widely used regression method in functional neuroimaging for capturing task-dependent changes in connectivity from a seed region. The present work identifies, and provides corrections for, common methodological pitfalls in PPI analysis that compromise model validity. Firstly, if the seed time series is extracted with prewhitening, the temporal structure of the signal is altered and subsequent deconvolution of prewhitened data becomes suboptimal. Furthermore, prewhitening again during model fitting results in double prewhitening of the seed regressor. Secondly, a failure to mean-centre the task regressor when calculating the interaction term can also lead to model misspecification and potentially spurious inferences. By using simulations and empirical language fMRI data from the Australian Epilepsy Project, we demonstrate the adverse effects of these issues, and how they are resolved when corrected. A systematic review of current practices revealed widespread model misspecification, and underreporting of methods, in published PPI studies. We provide clearer reporting guidelines, and advocate for appropriate methods for handling of prewhitening and mean-centring to ensure the validity of PPI analyses.