3
$\begingroup$

I would like to know how to correct a p value of a stiatistic when the same test was run over multiple regions of interest. The modality is MEG. So, I have 4 Regions Of Interests, and I ran the same cluster-based permutation test on the averaged timecourse of each ROI. I got more than 10 clusters, each with p < 0.05, but how can I get the corrected p value considering there was more than one region?

Any help will be much appreciated.

$\endgroup$
1
$\begingroup$

You have one p-values for each ROI timecourse (for the largest cluster) indicating whether your conditions are exchangeable. If you want to correct for the 4 ROIs you could use any adequate multiple comparison correction, from classical Bonferroni (which might be too conservative as the tests are likely not completly independent) to Benjamini-Hochberg style FDR correction (but applied to the cluster permutation test p-values).

This leaves us with the question which p-values count, one per ROI (the largest cluster, i.e. smallest p-value) or the p-values for all supra cluster-threshold clusters. I would think the former is more appropriate, as you already corrected for the multiple clusters.

That is, for the Bonferroni correction you would multiply your p-values by 4, (or identically, divide your critical alpha by 4).

This solution is incomplete, as I do not know how to get corrected p-values for the second largest cluster with FDR or Holm-Bonferroni.

$\endgroup$
0
$\begingroup$

I fully endorse Benedikt's answer. Further, you have to specify the question you are asking.

Let's say you want to know whether activity in a face recognition task differs between conditions in FFA, PPA, STS, and FEF and that the location of the difference is what you want to know, then Bonferroni is the correction of choice.

In case you ask the question whether the activity in the brain differs somewhere, but you do not care where specifically, then false discovery rate correction (FDR) is the method of choice.

The former allows statements like it is FFA and not STS, the latter allows statements like the conditions lead to differential activity in a collection of areas, whose identity is not too certain.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.