Hi everyone,
I have a question about cluster-extent versus TFCE thresholding in FSL. Specifically, I ran an analysis with TFCE and FWE correction and reported my results in a manuscript. We were asked to look “more” at a different region which was not significant with our TFCE-FWE analysis, and we found it to be active when using a standard hard cluster-forming thresholding approach (i.e., Z>3.1, p<0.05)
I’m surprised by this since I had understood TFCE to be more sensitive than a standard “hard-threshold” approach.
Does anyone know why this might have happened or what it might mean?
thank you
martin
That p<0.05 is actually FWE cluster corrected, so even worse than FDR! And, just for completeness, TFCE is also p<0.05 FWE corrected.
I think the main difference between the the two approaches is that in one I am setting the cluster determining threshold (the Z>3.1) myself, so I’m saying I’m snubbing clusters smaller that that, whereas TFCE uses a mathemagical approach to side-step having to set, a priori, a hard cluster threshold size, and I only need to set the desired familywise alpha.
Anyways, again, what is “befuddling” me is how come do I get a significant cluster with what should be the less sensitive off the two tools (i.e., the cluster extent approach)?
martin