Han and Jeanette,
You may be interested in this blog posting that Tom Nichols recently
brought to my attention, that provides a recipe for running FEAT without
all the individual 1st level directory stuff:
http://blogs.warwick.ac.uk/nichols/tag/fsl/
The approach outlined in that blog entry assumes you have varcopes. Tom
indicated that you can drop the option --vc=4dvarcope and replace the --
runmode=flame1 with --runmode=ols
cheers,
-MH
On Tue, 2011-11-08 at 16:44 -0600, Jeanette Mumford wrote:
> Ah, I see. I've never successfully tricked feat into running a group
> analysis this way. Since the flame1 algorithm requires the lower
> level variances, this could be where your approach didn't quite work.
> One option is to try running the group analysis with the OLS option,
> as this will not use the lower level varcopes.
>
> Cheers,
> Jeanette
>
> On Tue, Nov 8, 2011 at 3:45 PM, Han Jiang <[log in to unmask]>
> wrote:
> Sorry, I wasn't being very clear.
>
> I generated subject-level maps using a different kind of
> analysis (not provided in FSL), one which provides voxel-wise
> amplitude information within a particular frequency band.
> First I tried using randomise to get group results using those
> maps, and those were somewhat successful. Parallel to that, I
> wanted to generate group level results with Feat (parametric
> testing), so I mimicked individual Feat directories by
> creating reg directories, etc., and just called those subject-
> level maps "cope1.nii.gz" in each manually created .feat/stats
> directory, etc.
>
> My question is regarding differences between results from Feat
> and randomise, that is, why t values would be so much higher
> with group level randomise compared to group level Feat.
>
|