Dear Paul,
> We're analysing PET group differences (depressed vs control) with
> SPM96. Problem is, when we do the global normalisation, it seems to
> 'compress' the values for the control group into a much smaller range
> so that the column in the design matrix representing these subjects is
> almost invisible. The values given are tiny compared to the depressed
> patients. If we allow the analysis to run its course, it is clear that
> only the depressed subjects contribute to the resultant SPM(Z)s. The
> global CBFs are indeed consistently lower for the controls but we have
> no problems when we analyse the groups separately. I've tried
> subject-specific, group-specific and simple ANCOVAs.
As Mathew indicated this may reflect the fact that the variance in the
global estimators for the control group is much smaller than for the
depressed group. This difference in variance suggests a scaling
normalization will be more appoproriate (than an AnCova normalization).
I would try this.
With very best wishes - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|