Haxby searchlight example for multiple subjects?

There are very nice examples of running searchlight on Haxby subject 2. In particular, concatenating runs and one label per volume should be pretty easy for me to implement for my data. The thing I haven’t been able to find is an example/tutorial of running the analyses on multiple subjects and obtaining a group result.

After an initial read of Chen 2011, the Group analysis is confusing to me due to the multiple iterations of bootstrapping, and it seems (I hope) there might be a better way.


There is not a unique well-principled way to perform group analysis on this type of problem. You may want to study how the classifier generalized across subjects, or do some kind of one-sample test on classification accuracies. For this you only to store the individualsearchlight-based accuracies in an image or array and then do a one-sample test on it.
Does that make sense ?

1 Like