Glabus, Michael wrote:
> Using the full factorial model (PET toolbox), one is presented with the
> option to apply ANCOVA in two separate places:
>
> Under Factor:ANCOVA and later in the prescription options under Global
> normalisation:Normalisation:ANCOVA.
>
> Can anyone clarify which of these is the most appropriate for VBM?
> I have a feeling that the latter was the option that used to be relevant
> in previous incarnations of the software.
>
> (This question has been asked by someone else in a previous message, but
> no reply)
Hi Michael,
Still no reply as far as I can see... it seems this is a tough one...
I'll have a guess:
Judging from the help under Global normalisation:Normalisation, I
think the idea is that the options under the Factor:ANCOVA divide the
global covariate up into separate covariates for each level of the
factor, whereas the global option uses a single covariate.
So, as a very simple example (assuming my guess is correct!), with a
two-sample t-test, with (arbitrarily) two scans in group 1, three in
group 2, and some random global covariates, the global option would
give design matrix:
[1 0 27
1 0 42
0 1 34
0 1 28
0 1 33]
while the factor option would give design matrix:
[1 0 27 0
1 0 42 0
0 1 0 34
0 1 0 28
0 1 0 33]
I would guess that only the first (global) of these was available in
earlier versions of SPM, so you might just want to stick to that for
simplicity.
In my simple example, the second (factor) would allow groups 1 and 2
to have a different dependence on the globals, e.g. in VBM maybe
patients' local GM has less dependence on their global GM than for
controls...
So would this be better? Well, I think that's one of those difficult
statistical design questions (like should you adjust for global GM at
all?) that you'll probably have to decide for yourself -- or
investigate both, then report the prettiest SPMs ;-)
Hope this helps (and isn't totally wrong!)
Ged.
|