Hi,
while I don't know if this is the right thing to do in general concerning TBSS, I can help you with the error message. Presumably, you have split your cohort into two groups in the Glm gui. When you do that, Glm asks you to provide separate design matrix entries for each group (and zeroes for the other group), i.e. something along the lines of:
1 0
1 0
1 0
0 1
0 1
0 1
In your 4D voxel confound file, there is only one entry for the entire cohort, obviously. That's why Glm complains. Actually, randomise does not care about this, so you can ignore the error message. However, you should perhaps demean your 4D image using fslmaths...
You *could* theoretically create 4D files for every group, with zero-images for the rest of the cohort, and provide randomise with two 4D images (this actually works). Search the list or email me on how to do this. However, the difference between this and the standard approach is the same as splitting/demeaning any other regressor of no interest across groups, so this should be done deliberately, and not only because some gui tells you so.
Cheers,
Cornelius
On Tue, Nov 27, 2012 at 2:38 AM, Maren Strenziok
<[log in to unmask]> wrote:
Hi,
I processed FA maps from 2 groups (trained/non-trained) and 2 time points (pre-training/post-training) with tbss, did an interaction analysis with randomise, and found areas of significant decrease in FA over time in one group. I then extracted FA values from locations of peak intensity in the tfce-corrected maps and found that there were baseline differences between the two groups. I would like to control for these pre-training differences. I understand that I can use the --vxf and --vxl options in randomise to add the baseline maps as voxelwise regressors. When I set up the design matrix by setting the Number of additional, voxel-dependent EVs to 1 and loading a 4D file that contains all subjects' pre-training FA maps at the bottom (EV3 (vox), I get the following error message:
Problem with processing the model: Warning- design matrix uses different groups (for different variances), but these do not contain "separable" EVs for the different groups (it is necessary that, for each EV, only one of these groups as non-zero values).
Is using the additional, voxel-dependent EV the right way to correct for baseline differences in FA and, if so, why do I get this error message?
Thanks, Maren
Cornelius Werner
cornelius.werner<at>gmail.com