> We are conducting an longitudinal vbm analysis in SPM2 (2 groups:
> patients and controls; 2 scans: baseline scan and one year follow up scan)
> We selected the design: Multigroup: conditions and covariates option in
> order to perform a 2x2 factorial design (1-1-11; -111-1)
> Regarding the global normalisation, should we perform: "proportional
> scaling option" or "no global normalisation option"?
For VBM it's common to use no normalisation, since the images are
(probabilistic) segmentations, unlike direct PET/fMRI measurements
that might have additive/multiplicative confounding global signal changes.
> After that, we understand from the list files that we should chose the
> non-sphericity corretion option, isnīt right?
Yes, since you have longitudinal data, the two levels of your
time/scan factor are dependent. Groups will be independent. I'm not
sure of the mechanics of setting this up in SPM2, but if you can say
something like the time factor has dependent levels / needs
non-sphericity correction but that the group factor has independent
levels and doesn't need correction, then that sounds fine to me.
(actually, a slight complication here, the groups factor might well
have unequal variances in the two levels (e.g. patients more variable)
which would then also be an example of non-sphericity).
An alternative is to model fixed subject effects, i.e. a subject
factor. Like for e.g. a standard paired t-test. Maybe take a look at
this (informative but not very straightforward) thread:
Lastly, another alternative: for two time-points, you might just
prefer to create difference images, and do an unpaired t-test between
the two groups using these. This script might save time in this case:
Hope this helps (sorry it's a bit long!),