To follow up on Steve's reply, SPM has always reported and FSL's randomise will soon report uncorrected cluster P-values. I would use them very cautiously, though. They only apply to the case where you can uniquely identify a cluster before you have analyzed the data, as they don't account for searching over the brain for clusters. This is in fact worse and more slippery than with uncorrected voxel-wise P-values, as you always know where voxel x,y,z is before you analyze your data, but a, say, DLPFC cluster may not exist, or maybe very difficult to uniquely identify a priori.
Another concern, in the permutation/randomise setting, is that the uncorrected cluster P-values crtically depend on a stationarity assumption, i.e. homogeneous smoothness. If the smoothness varies dramatically, as it does in VBM or other structural data, the uncorrected P's shouldn't be trusted (clusters from very smooth regions will tend to be significant, even if there is no true effect). The FWE-corrected cluster P-values, on the other hand, don't depend on any stationarity assumptions and good for use with any type of data.
-Tom
On Wed, Jun 25, 2008 at 8:47 PM, Daniel Peterson <[log in to unmask]> wrote:
Hi All,
I was wondering if anyone knew a way to get the uncorrected p-values from the suprathreshold
cluster stats in randomise. Likewise for TFCE and mass clustering.
Thanks,
-Dan
-- ____________________________________________ Thomas Nichols, PhD Director, Modelling & Genetics GlaxoSmithKline Clinical Imaging Centre
Senior Research Fellow
Oxford University FMRIB Centre