Hello everyone,
First off, thank you to the developers for the great toolbox. I’m using it to run a searchlight RSA analysis comparing neural similarity to behavioral ratings and wanted to check that my script is doing what I think it’s doing. From reading the existing posts here, I understand that using the rsa_beta output measure can be a bit tricky.
I have four conditions across three runs. I created behavioral RDMs by calculating pairwise distances between conditions. Here are relevant snippets of my script, based on the decoding_searchlight_crossnobis_filled example:
cfg.decoding.software = ‘similarity’;
cfg.decoding.method = ‘classification’;
cfg.decoding.train.classification.model_parameters = ‘euc’;
cfg.results.output = {‘rsa_beta’,‘other_meandist’};
cfg.scale.method = ‘cov’; % we scale by noise covariance
cfg.scale.estimation = ‘separate’;
cfg.scale.shrinkage = ‘lw2’; % Ledoit-Wolf shrinkage retaining variances
cfg.files.components.matrix = {target_dsm}; % pairwise distance matrix based on behavioral ratings
cfg.files.components.index = ones(length(target_dsm))';
The rsa_beta output looks to be on very different scales for each participant and I see from the documentation that they’re not normalized. Will this be an issue for conducting the prevalence inference analysis at the group level?
Thank you in advance for any insights!
Best,
Heather