I'm experiencing the 'There are no significant voxels' error when I am running a 2nd level test. I would greatly appreciate it if someone could explain why this error occurs, and why SPM can't compute the test in this situation.
Let me provide some specifics:
I am running SPM5 r1782 on Matlab 7.8.0 (R2009a) on RedHat Enterprise Linux 5 on a Dell server with 8 64bit Intel CPUs .
I have run a first level GLM for each of 26 subjects on event-related fMRI data.
For each subject, I have calculated the same 18 contrasts.
For each of these 18 contrasts, I have run various 2nd level tests:
* One-sample t-tests with all 26 subjects for all 18 contrasts.
* One-sample t-tests with group A (12 of 26 Ss) for all 18 contrasts.
* One-sample t-tests with group B (other 14 of 26 Ss) for all 18 contrasts.
* Two-sample t-tests for group A versus group B for _17_ of the contrasts.
** For a single contrast, the two-sample t-test fails with the 'There are no significant voxels' error.
I have found that when I remove a particular subject (who is in group A) from the failed two-sample t-test, it runs successfully. However, when I look at the 1st level con_#### file for this subject, I see nothing obviously amiss, and this con_#### file works fine for the one-sample t-test with all subjects, and the one-sample t-test for group A.
When I look in the spm_spm code at where the problem is occurring, I see that the code has determined that "ReML hyperparameters are needed for xVi.V" (line 736) and then it finds voxels that satisfy the inequality "sum((Hsqr*beta).^2,1)/trMV > UF*ResSS/trRV" (line 742). The problem is that no voxels satisfy this inequality, and so the error is triggered on line 849.
What is this inequality testing for?
Why does at least one voxel need to pass this inequality for the t-test to be calculated?
What does it mean about the data if no voxels pass this test?
Why doesn't SPM just compute the test and let me decide what threshold I want to use to determine significance?
Is there a particular type of problem I should be looking for in the data for the subject that appears to be causing this error?
Note that I tried commenting out the error check in spm_spm, and rerunning it, but this just led to an array indexing error further down in the code, presumably because it was assuming a non-zero number of voxels met the inequality.
thanks for your help,
Psychological and Brain Sciences