Thanks!
This was on a computer that I'm not managing, made them update the binaries and now it works! Thanks RIck!
I have a related question, I will try to make it clear - It might by difficult, haha.
Task: Run 3dMVM on a huge number of datasets. Getting the stats: No problem, it does them slice by slice. Problem: using the -resid option. It keeps all images (4D) in memory.
The solution: Use 3dZcutup and split the data into 9 stacks of about 10 slices each. Run 9 instances of 3dMVM and glue the stats and residuals together afterwards.
New Problem:
This works for all stacks apart from the most buttom and top ones (slices0-9) and slices 82-91) - Then I get the classic 3dMVM error where it suggests that the model might be wrong or the table might be wrongly formatted. This is run via a script and it works for all other stacks. My guess is that there are a lot of 0 voxels and very little brain at the top 10 and bottom 10 slcies.
If I increase the bottom and top stacks (e.g. merge stack0,1,2), then the stats work again - But I reach the breaking point of memory, it can't make the residuals.
My question:
If I my end goal is to the the stats + residuals for the bottom 10 slices. Can I make a "fake" dataset, containing 20 slices: first 10 are bottom slices that I'm interested in and the next 10 could be from the middle of the brain, just to trick MVM to think that the model/stats are correct?
Or does this screw up the residuals? Do they depend on the adjesent slices? If the residuals are independent of adjecent slices, then I can probably do this.
Thanks!