[FieldTrip] Cluster-based permutation test with different sample sizes by channel

[Schoffelen, J.M. (Jan Mathijs)]

Dear Soren,

As long as you use a test-statistic that is not too sensitive for the sample size (such as the T-statistic), and the imbalance across regions in number of contributing samples is not too large, I’d say you’re fine. Also, if you use a non-parametric cluster-threshold statistical sensitivity is not too compromised

Just to be sure,  you don’t need to repeat the test for each of the regions separately (which requires a multiple comparison correction for the number of tests), but you could run the statistics function with an empty neighbour definition, preventing clustering across space.

Best wishes,
Jan-Mathijs


> On 23 May 2019, at 19:29, Soren Emmanuel Wainio-Theberge <swain083 at uottawa.ca> wrote:
> 
> Hello all, 
> 
> My name is Soren Wainio-Theberge, and I'm working in the Mind, Brain Imaging and Neuroethics unit in Ottawa, Canada. I had a question about using cluster-based permutation tests with my data. I'm working with a large ECoG dataset, and I've grouped all the electrodes into regions according to an atlas. I wanted to do a cluster-based permutation test on the region-based results (since I have reason to suspect region-specific effects, but don't want to go as far as specifying a priori ROIs). However, since this is ECoG data each subject has only a subset of the full number of regions represented, depending on their electrode placement. As such, each region will have a different sample size. At the moment I'm only clustering across channels, not across time or time-frequency.
> 
> My fix was to permute each channel separately, rather than all together, which seems to work code-wise. My question is whether this a statistically sound thing to do - is the cluster test sensitive to differences in sample size for each region?
> 
> Best and thanks very much, 
> Soren Wainio-Theberge
> _______________________________________________
> fieldtrip mailing list
> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
> https://doi.org/10.1371/journal.pcbi.1002202


_______________________________________________
fieldtrip mailing list
https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
https://doi.org/10.1371/journal.pcbi.1002202