- MNE version: 1.4.2
- operating system: Windows 10
Dear collegues,
We are computing permutation statistics for mne_connectivity by using the function:
conpy.cluster_permutation_test (more or less equal to mne.stats.permutation_cluster_test)
We have 2 experimental conditions with 15 subjects.
Can you provide some insights about permutation testing regarding the following:
- Is the code that we use only permute subjects or there is something more that is change for one permutation?
- If we have 15 subjects can we achieve 1000 permutations with them? In general, what is the largest number of permutations we can reach with 15 subjects ?
- What is the minimum change in the dataset, which happen during shuffling? We assume that by just changing one observation from the first condition with one observation from the second condition is not correct to compute permutation, as this will actually mean practically no change and thus - increase of type 1 error. Thus it should be some minimum amount of shuffling for one permutation, shoundn’t it?
- What can happen if having 15 subjects is too little to obtain 1000 permutations? What mne or conpy permutation testing shuffle instead? How does it achieve the given number?
Thank you in advance for your ideas and insights!