I strongly suggest you go through the SPM manual, and the lecture slides available on the FIL website to get a full understanding of these issues.
All things equal, p=.005 is *less* stringent than p=.001. So you should see more voxels.
k=20 means it will only include clusters with >= 20 voxels.
You should always use some form of multiple comparisons correction, either FWE or FDR.
________________________________________
From: SPM (Statistical Parametric Mapping) [[log in to unmask]] on behalf of Melissa Levesque, Ms. [[log in to unmask]]
Sent: Thursday, February 20, 2014 12:33 PM
To: [log in to unmask]
Subject: [SPM] thresholds of significance
Dear spm experts,
I'm looking at results of second level analyses I've done and I'm confused with the thresholding. I started to do the analyses with the default values of p=0.001 and extent thresholds=0, then redid them with p=0.005 and extent thresholds=20 since that is what we had written in our protocol. I don't really understand what's going on, because at p=0.005 (compared with p=0.001), some results get more significant and others get less significant.
Can anyone explain this to me?
We have a higher k at p=0.005. I'm thinking the issue might be that since more voxels are included in the analysis at p=0.005, some of these might be less significant and that explains why are overall cluster level significance decreases. Is that it?
Thank you for your help,
Melissa
|