 |
 |
Laura, Please see the "Counting Permutations" section of the randomise help page http://www.fmrib.ox.ac.uk/fsl/randomise/index.html You'll find that a two-sample t with n1=3, n2=3 will yield 20 permutations. The test is still exact but will not be very powerful, as the smallest attainable P-value (corrected or uncorrected) is 1/20=0.05. -Tom On Tue, Sep 30, 2008 at 5:19 PM, Gwenaëlle DOUAUD <[log in to unmask]>wrote: > Hi Laura, > > it is perfectly plausible that you don't get any significant result, > especially after correction for multiple comparisons. Look at your > uncorrected p-values or t-values maps to have an idea of the pattern of > possible abnormalities (see randomise manual). > The number of permutations is based on the number of subjects you've > entered in your model. With only 6 subjects, you won't need 5000 > permutations. > > Cheers, > Gwenaelle > > > --- En date de : Mar 30.9.08, Laura D <[log in to unmask]> a écrit > : > > > De: Laura D <[log in to unmask]> > > Objet: [FSL] no significant clusters! > > À: [log in to unmask] > > Date: Mardi 30 Septembre 2008, 16h42 > > Hi all-- > > > > I'm just learning to use FSL, and this forum's been > > very helpful in > > answering all my questions when things go wrong or the > > instructions are > > incomplete. However, I've run into something there > > aren't any answers for so > > far: I'm getting no significant clusters, even when my > > 1-p is set all the > > way to .4! > > > > All files have been generated properly, everything's > > been smooth. Admittedly > > I've only used 3 controls and 3 patients, as I wanted > > things to run quickly. > > Still, there ought to be something at p = .6! Here're > > the two randomise > > commands I ran (I tried s=2,3,4, all with nothing), per the > > "Running FSL-VBM > > (Detail)" displaying results step: > > > > randomise -i GM_mod_merg_s3 -m GM_mask -o fslvbm -d > > design.mat -t design.con > > -T -n 5000 –V > > > > fslview $FSLDIR/data/standard/MNI152_T1_2mm > > fslvbm_tfce_corrp_tstat1 -l > > Red-Yellow -b 0.4,1 > > > > One thing I also found curious was that apparently only 20 > > permutations were > > run with the first command, although I specify 5000. Is > > that normal? > > > > Thanks! > > > > > -- ____________________________________________ Thomas Nichols, PhD Director, Modelling & Genetics GlaxoSmithKline Clinical Imaging Centre Senior Research Fellow Oxford University FMRIB Centre