Hi Volkmar and Marko,

Thanks a lot for your help with this. I have figured out in the meanwhile which subjects were responsible for the estimation failure. Please see attached the mask.img for these two. While subjects 1 has a small black hole, the second should be perfect. As pointed out by Marko, the tiny black should also be not the problem.

Based on the fact that everything went well for my first contrast of interest (consistent across subjects), it is likely that the failure is due to the nature of the second contrast, which can be different across subjects and scans.

Maybe it's worth to describe this second contrast a bit more. I have a longitudinal design with two visits and two scans per visit. Each scan has 12 regressors, so 48 in the first level design. The second contrast of interest depends on subjective ratings (assessed after each scan, 4 times in total) and can either be -1/4 1/4 -1/4 1/4 or 1/4 -1/4 1/4 -1/4. For example, for one of the problematic subject's the contrast looks as follows:

for visit A: -1/4 1/4 -1/4 1/4 0 0 0 0 0 0 0 0 1/4 -1/4 1/4 -1/4 0 0 0 0 0 0 0 0
for visit B: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1/4 -1/4 1/4 -1/4 0 0 0 0 0 0 0 0 1/4 -1/4 1/4 -1/4 (see attached)

This second contrast (for visit B) crashed the estimation of the second level batch. Do you know why?

I have also another second level batch which did not run. The contrast images for this batch represented the average across visit A and B, so e.g.

-1/8 1/8 -1/8 1/8 0 0 0 0 0 0 0 0 1/8 -1/8 1/8 -1/8 0 0 0 0 0 0 0 0 1/8 -1/8 1/8 -1/8 0 0 0 0 0 0 0 0 1/8 -1/8 1/8 -1/8 0 0 0 0 0 0 0 0 

For this analysis, 4 subjects produced the error message.

Any suggestions are very welcome, thank you very much.
André

________________________________________
Von: SPM (Statistical Parametric Mapping) [[log in to unmask]]" im Auftrag von "Volkmar Glauche [[log in to unmask]]
Gesendet: Donnerstag, 1. Oktober 2015 15:10
An: [log in to unmask]
Betreff: Re: [SPM] AW: [SPM] estimation of second level batch

Hi André,

since the error message says "there are no significant voxels" it is most likely that at least some voxels survived the masking. However, your model does not seem to explain sufficient variance in the inmask voxels. You should check the mask image whether it includes the brain regions you would expect. If not, you need to go back to the individual subjects and check the mask images from 1st level. Since some of your contrasts from 1st level seem to give results at group level, this is quite unlikely. Another reason could be that there are very inconsistent or small effects for some of the contrasts. Again, you will have to look at the 1st level results for the contrast that fails at  group level.

Best,

Volkmar