JeanLucAnton
commented
3 years ago Mixed Effects analysis ? For each subject, we generally work with an accuracy map, obtained by calculating the average of the N accuracy maps obtained from the N cross-validations. It might be interesting to use these N maps to bring them to the second level, the idea being to take into account the variance between subjects but also within subjects : a subject with accuracy maps obtained by cross-validation that are too variable should not "weigh" as much as a "good" subject.
Sylvain proposes to test mixed-effect analyses : how to launch this type of analysis? With which tool?
Sylvain: Here is a possible pipeline for running a group analysis on accuracy maps obtained with a "searchlight decoding" : 1. subject by subject, run the searchlight decoding (in the native space) 2. get an accuracy map per subjet (nota : it is the mean of the accuracy maps obtained during the intra-subject cross-validation) 3. subtract the theoritical chance value (0.5 a two categories classification, 0.33 for three categories ...) to the accuracy maps (for example with imcalc dans SPM) 4. warp these maps to a standard space (for ex : MNI) with your prefered method 5. use these maps as entry of a RFX group analysis in SPM. It is a good approach to use a non-parametrical permutation test (like SnPM)