Closed jwdink closed 9 years ago
OK, I gotcha. It looks like the M&O paper is an example of the more general bootstrapping technique, and then for the specific version used for time-series, they're actually going off of the "clustering of adjacent time-samples" technique from http://www.ai.mit.edu/events/talks/fMRI/papers/permutation_tests2.pdf
I'm having trouble finding anything on the "sum t" test.
Also, I think you were right that the cluster-based technique not being as sophisticated as I thought it was
This paper refers to both sum-t and Maris & Oostenveld:
Brainstorming note to self: make selection of t-threshold less arbitrary by finding values where derivative for f(t) = num_clusters is closer to zero
We should call these "cluster analyses" -- you should also read about the "sum t" and see whether this p-value idea is better/worse than the sum-t test.