GLM-based MVPA and its extension to 'cross-decoding'
We¹ previously proposed an approach to multivariate pattern analysis of fMRI data which is not based on classifiers, but on the general linear model (GLM), bringing it into a common framework with univariate analyses. A possibility opened by classifiers, cross-decoding, i.e. to train on one pair of conditions and to test on another pair, was only tentatively covered by that proposal. We now follow up on that, and propose cross-MANOVA as a a comprehensive approach, covering both the previous cross-validated MANOVA using a single contrast and cross-analyses using two contrasts.
¹ Allefeld & Haynes, Searchlight-based multi-voxel pattern analysis of fMRI by cross-validated MANOVA, NeuroImage 2014