So I got into a discussion recently about the interpretation of "significant clusters" when using NBS toolbox.

I have always considered this advice from Fieldtrip when interpreting them: https://www.fieldtriptoolbox.org/faq/how... (which has functions for non-parametric cluster statistics but for clusters in time and space using MEG instead of clusters in connectivity matrices - but I think the principle is the same regarding how the permutations are performed) - this says we cannot say too much about the actual size and extent of a significant cluster. With this interpretation, all that can be said from the statistical test is that there is a significant difference between the connectivity matrices of two groups after rejecting the null hypothesis which is "there is no difference between the connectivity matrices between two groups".

However, a lot of articles using NBS don't seem to take this advice and interpret their "significant cluster" in quite a lot of detail (e.g. interpreting the size as significant and the exact cluster of significant edges in quite some detail). Is there any reason I am missing why the fieldtrip advice does not equally apply for NBS as well?

Thanks in advance!

Hi William, 

this is a bit of an open ended question. Yes - care is needed when interpreting a cluster/component and particular caution is needed when drawing conclusions about specific edges within a cluster. Depending on how the NBS is implemented (i.e. TFCE vs standard vs NBS-predict) the method can offer weak or strong control of the FWE. General assumptions of permutation testing (i.e. exchangeability assumption) can also play a role. It is often possible to draw some conclusions about the size, extent and location of a cluster, but yes caution is definitely needed. 

best, Andrew

Originally posted by willhedley: