Hi - please see the statistics literature on multiple-comparison correction - in general when analysing the whole brain you have to correct for the number of independent tests you carried out - so you have to use the 'corrp' and not the 'p' values, unless you are going to do a separate correction for multiple comparisons, e.g. feeding in the 'p' values into FDR.

TFCE is indeed generally recommended, but voxelwise or cluster-based are also valid outputs from randomise.

Cheers.



On 19 Dec 2009, at 22:03, Ajay Kurani wrote:

Dear FSL Users,

I have been reading documentation regarding Randomise and being that this is
my first analysis I had some questions.  I had a comparison of 17 control
subjects to 17 patients using affine and nonlinear mapping.  After looking
at my tstat1 images for different smoothing (s2 s3 etc) I used randomise to
do 5000 permutations of the data set(after selecting s2).  

randomise -i GM_mod_merg_s2 -m GM_mask -o fslvbm -d design.mat -t design.con
-T -n 5000 -V

The output files are tfce_corrp_tstat and tfce_p_tstat.  When I display
using corrp there are no results while I receive results when using p
(really 1-p)

fslview $FSLDIR/data/standard/MNI152_T1_2mm fslvbm_tfce_corrp_tstat1 -l
Red-Yellow -b 0.949,1
fslview $FSLDIR/data/standard/MNI152_T1_2mm fslvbm_tfce_p_tstat1 -l
Red-Yellow -b 0.949,1

I wanted to find out if it is okay to use the p_tstat1 file to show
significance in results or if we should always only use corrp_tstat?  Is the
difference only that corrp ensures there are less false positives?  Because
the ROI are very small areas I am worried that the corrp_tstat might be
sensitive enough to pick up these areas since they are clearly present in
p_tstat.  Any help or insight would be much appreciated as I am still a bit
fuzzy on the difference between these two parameters.  Also, is TFCE always
recommended or would cluster or voxelwise statistics be recommended if I am
not defining any ROI ahead of time. Thanks.

Sincerely,

Ajay Kurani




---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director,  Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
+44 (0) 1865 222726  (fax 222717)
[log in to unmask]    http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------