See inline responses below.
On Mon, Oct 8, 2012 at 8:30 AM, Stefanie Beck
<[log in to unmask]> wrote:
> Dear SPM Experts,
>
> I am trying to run a 2-sample ttests (2 groups, placebo and drug, each 29
> Subjects) and keep getting the error message 'there are no significant
> voxels' during model estimation. Running a one sample ttest with all those
> subjects in one group works fine.
>
> The 2-sample script does run, if a specific subject is excluded. But most
> strangely, if this subject is kept in the group, but another subject is
> excluded, the script still runs fine. This is very strange, since the
> problem does not seem to be depending on one specific subject.
>
> It is also the case, that the script only crashes for a few contrasts,
> estimating other contrasts defined from the same GLM works fine.
>
>
> There have been a few suggestions on the mailing list, including the
> possibility, that the single subject contrast images (1st level analysis) do
> not overlap. But how could I check this?
Use the checkreg button in SPM. You'll have to do them in batches.
However, since everything works in a one-sample t-test, then this is
unlikely the reason for the error.
>
> Annother suggestion has been made that there is simply not a single
> significant voxel for the between group comparison for this contrast. But if
> this is the case, which threshold does SPM use to define a voxel as
> significant? Is it for example, that all the error message 'no significant
> voxels' occurs if no voxel has a p - vale < 0.05?
The criteria is <0.001. The error means that no voxels were significant.
>
> Do you know of any other reasons the script does not run?
The issue of no significant voxels only effects the hyperparameter
estimates. If you set the variance to be equal, then the models should
estimate without a problem. What I suspect is happening is that you
have a few subjects that are outliers and are causing the group
difference to disappear. You could change spm_defaults.m (ufp field),
but I wouldn't recommend that. I'd just turn off the variance
correction. I'm not sure if there is a better solution at this point
to these two options.
I'd double check that you are entering the right subjects as well.
>
>
> Any help would be highly appreciated!
>
> Thank you,
>
> Stefanie
>
> --
> Dipl. Psych. Stefanie Beck
> Technische Universitaet Dresden,
> Faculty for Mathematic and Natural Science
> Department of Psychology, Professorship for General Psychology
> Zellescher Weg 17, 01062 Dresden, Germany
>
> Office Zellescher Weg 17, BZW, Room 324
> Phone +49 351 463 346 76
> Email [log in to unmask]
|