Dear Cutter,
Well. See the error message and the mentioned m file:
% In a first pass, voxels over which non-sphericity will be estimated are
% selected using an 'effects of interest' F-contrast (can be specified in
% SPM.xVi.Fcontrast) and critical threshold taken from SPM defaults
% stats.<modality>.UFp.
In your case, this F contrasts results in no sig. voxels at that threshold, so no voxels can be selected, so you obtain an error message. Simple as that. This doesn't mean something has gone wrong, maybe there are just no differences for this particular model (maybe this results from an inefficient experimental design, inefficient single-subject designs), which can well be different if you deselect some of the subjects or combine them in a different way.
Basically there are two options, you could switch to "equal variance" or you could modify the default threshold. The first one is probably preferable.
However, as you try to run a mixed-design ANOVA with Flexible factorial, in which the group effect is going to be invalid due to Flexible relying on a pooled error representing the within-subject error, which is inappropriate for the group effect, I would recommend to turn to a different approach. See recent https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;b5ca289f.1603 for suggestions
- the two-sample t-test approach
- the approach described by Guillaume, basically the same
- GLM Flex
Finally, dunno whether this has to mean anything as SPM doesn't plot every single path, but looking at the paths of the con images on the design_matrix.png it always reads "con_0002". So I would double-check whether you really have selected the correct, individual con images.
Best
Helmut
|