Thank you Anderson! I never looked into the equation of it (e.g., the ones on Wikipedia).Just realize it does work that way.
Kai
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]]
On Behalf Of Anderson M. Winkler
Sent: 2017年2月22日
4:35
To: [log in to unmask]
Subject: Re: [FSL] Unusual PALM results with cfdrp and fdrp
Hi Kai Wang,
This can happen. FDR depends not only on the number of tests, but also how strong is the evidence for the null is, that is, it depends on the p-values of the additional tests that are included when the extra contrasts
are accounted for.
For example, correcting for C1 only may not yield anything on the map for C1, but when correction includes both C1 and C2, if C2 has lots of signal spread throughout, this may make it easier for tests on C1 to become
significant.
It's different than FWER-correction, which doesn't depend on the strength of the "other" tests. FDR does.
All the best,
Anderson
On 22 February 2017 at 01:29, Kai Wang <[log in to unmask]> wrote:
Dear experts,
I was looking at PALM results for a co-variate analysis when I found that the result with FDR-corrected across contrasts (palm_vox_tstat_cfdrp_c3.nii.gz; fslstats -R value, 0-0.997741) was more significant that that with FDR-corrected within contrast (palm_vox_tstat_cfdrp_c3.nii.gz; fslstats -R value, 0-0.564641). This is confusing to me, since cfdrp should be more strict than fdrp which does not correct for multiple tests within the modality.
My PALM options were: -n 5000 -noniiclass -ise -corrcon -eb XXX.csv -ee -T -C 3.290527 -save1-p -fdr
Thank you
Kai Wang
Postdoctoral Researcher
Institute of Cognitive Science, University of Colorado Boulder
1777 Exposition Dr., Boulder, CO, 80301
[log in to unmask]