Hi, You can see how the error in the p-value relates to the number of permutations on the wiki: http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Randomise/Theory#Monte_Carlo_Permutation_Tests All the best, Mark On 27 Feb 2014, at 08:13, Alain Imaging <[log in to unmask]<mailto:[log in to unmask]>> wrote: Thanks a lot Mark. Sometimes the simplest answer is the one I'm not able to see, apparently. I will go for a smaller number of permutation (any hint for the minimum number that still will give me some "proxy" results ?). Best Alain > Date: Thu, 27 Feb 2014 06:34:55 +0000 > From: [log in to unmask]<mailto:[log in to unmask]> > Subject: Re: [FSL] fastest way to check vertex analysis results > To: [log in to unmask]<mailto:[log in to unmask]> > > Hi, > > You can run randomise with a smaller number of samples, which will speed it up. The results will be less accurate (the p-values will vary) but it will let you see if you have anything near the threshold or not. > > All the best, > Mark > > > On 24 Feb 2014, at 11:00, Alain Imaging <[log in to unmask]<mailto:[log in to unmask]>> wrote: > > > Dear all, > > > > I would like to have an hint on a rather practical question. > > > > I'm conducting exploratory analysis between several behavioral measures of the same construct and shape using vertex analysis. > > What I would like to know is if there is a quick method to check if it could be valuable perform a (time consuming) randomise for a certain behavioral measure. > > I can only think to running randomise with inference off (but then, to which distribution compare the t ?) or performing the old shape analysis using --surfaceout and performing the fdr correction. > > Are there any other methods less time consuming than randomise ? And if not, which of the two methods that I outlined is the most suitable and will give me the best idea of the results of randomise ? > > > > Thanks in advance > > > > Best > > > > Alain