Has anyone run into a situation where higher level (ie group) stats in
a ANOVA result in a very speckled image? In essence, there is
virtually no smoothness in the image resulting in many single voxels
being suprathreshold when their neighbors aren't even close. Oddly,
this only comes up when we correct the compute zfstat images to take
into account a random factor in the design.
In brief, we have 2x2x2 design where one factor is Group and obviously
this is between subject. The other two factors are within-subject,
leading to a mixed design equivalent to the "4. FFR" (fixed fixed
random) situation at the bottom of the FEAT-Higher level analysis
page. So we setup the analysis as shown in the design matrix above
except that EV8 was replaced by subject-specific EVs to account for
the within-subject repeated measurements. The analysis runs and the
results for the within-subject fixed factors look ok. But when we
apply the 4. FFR correction (eg fslmaths fstat1 -div fstat5 fstata;
ftoz -zout zfstata fstata 1 1) the resulting zfstata image is
speckled. In fact, it looks like the mean structural caught the
measles -- kind of pretty, but not what we were looking for.
Clearly we've made some kind of fundamental mistake, so any help
spotting and correcting it would be much appreciated.
Thanks.
joe
Joseph Devlin, Ph. D.
Cognitive, Perceptual & Brain Sciences, UCL
Gower Street
London, WC1E 6BT
email: [log in to unmask]
|