Hi,

It doesn't make sense to have 8 EVs and only 8 subjects!  You will not get sensible results from such an analysis.  You need more subjects (or less covariates, but even then 8 subjects is a very small number).

All the best,
        Mark

On 19 Jul 2013, at 18:26, Ricky Savjani <[log in to unmask]> wrote:

> Hi FSL experts,
>
> I am trying to run a third-level analysis using the "Single-Group Average with Additional Covariate" as explained on the FSL GLM wiki on my 4D COPE image. My Design matrix looks like this:
>

> <image.png>

>
> where EV1 is to represent the group mean activation and EV2 - EV8 are 7 covariates that I have already demeaned.
>
> The contrast matrix looks like this:
>

> <image.png>

>
> where I am to get brain areas that linearly correlate each behavioral measure with BOLD activation.
>
> Here is a copy of my randomise command:
>
> randomise -i cope1_merged.nii.gz -o "covariates_" -d covariate2.mat -t covariate2.con -m avg152T1_gray_thr_bin.nii.gz -n 5000 -T
>
> The problem is that the resultant tfce_corrp images have so much significance. Also, the range of the tstat images are huge (millions!). I know something must not be right here. Also, i tried the other option of not demeaning my data and then using "-D" option in randomise. This resulted in values of 0.00 for every tfce_corrp image.
>
> I would very much appreciate any help with this!
>
> Thanks,
>
> -Ricky