RSA group analysis

Hi all,

I’d like to perform RSA-Searchlight (representational similarity analysis) on fMRI data of a group of subject, each having their own stimuli-RDM, using Python, currently via rsatoolbox.

I can easily get single-subject searchlight similarity maps. A bit less easily, but I can also calculate (subject-specific) p-value maps based on permutation tests.

Next I want is the group level analysis, which here I would like to ask for help.
Things I can think about:
1- Simply average the similarities
2- Like (1) but after r2z conversion (should be negligible since RSA correlation value are usually small)
3- convert p-value maps to z-maps and run 1smp-ttest

I know it should be obvious but somehow the RSA guides I’ve read concentrate in estimating the single-subject similarities.

Thank you