What's after searchlight.scores_?

Hello everyone,

I would like to know what people usually do after getting the searchlight.scores_
before going for further analysis such as RSA.

I saved the searchlight_img = new_img_like() image from the searchlight.scores_ as in the tutorial,
so now I have 1 image for each participant, in which the “t-value” now should indicate the classification accuracy, if I understand right.
I wanted to visualize the overall searchlight results, so I used mean_img(searchlight_imgs) to average all the searchlight_img I have, and then I found the accuracy peak drops a lot in this averaged image (.38 in individual to .27 in average, chance level = .25), so I wonder maybe this is not the right way.


  1. Am I right that the mean_img() function averages images regarding the whole-brain voxels so the cluster information will be averaged out?
  2. How can I check which clusters get high scores in general after individual searchlight?
  3. I do notice the example for comparing to massively univariate analysis. Should I do this for individual results, or for the group results (if so, which way is more proper to combine all the individual results)?
  4. Another question: if I have group argument for the cross-validation in decoding, should I do the same group for permutation test?

Many thanks and have a good day.

Author posted SecondLevelModel/non_parametric_inference/permuted_ols after searchlight.scores_ instead.