Methodological considerations: Electrophysiology of multisensory interactions in humans

Abstract

A fundamental question in human studies of multisensory integration is to understand how the brain processes involved in bimodal event perception differ from those underlying unisensory perception. The so-called additive model has been widely used to identify the neural operations specifically related to bimodal processing, but also much criticized for the multiple biases it can generate in the estimation of cross-modal interactions. We discuss here (i) what these biases are and how to avoid or minimize them, with particular emphasis on electromagnetic (EEG/MEG) techniques; (ii) why the use of the additive model is not only suitable but also necessary to identify cross-modal interactions from EEG/MEG signals; and (iii) why violations of the additive model lead to different conclusions when used in human EEG/MEG data, fMRI data, or at the single-cell level in animal studies. Finally, we review statistical methods that can be applied to test this model on experimental data at a group or at an individual subject level, emphasizing on the use of randomization tests.