Dear Marjolein,
> I tried to compare a group of patients (P) and a group of controls (C). All
> subjects had two water-PET scans, 1 activation (a) and 1 rest (r) scan. I
> did a multi-group analysis with the following contrast: Pa Pr Ca Cr: 1 -1 -1
> 1
>
> What is in simple words the meaning of the regions that show up in the SPM
> (patients more activated than controls, or controls more activated than
> patients, or activation vs rest, or rest vs activation) ?
>
> I tried to figure this out by writing the contrast in different ways:
> (Pa-Pr)-(Ca-Cr) can also be written as (Pa-Ca)-(Pr-Cr) and as
> (Pa+Cr)-(Ca+Pr).
> This doesn't make sense to me, so I became even more confused.
All are equally valid interpretations. The first (Pa-Pr)-(Ca-Cr) is
the most usual, namely greater activation in the Patient group. This
corresponds to a condition x group interaction.
> Then I realised that I really don't know what a multi-group analysis in SPM
> actually is, what SPM actually does. Can you please clare up this matter for
> me??
There is nothing special about the name 'multi-group analysis'. It is
simply a linear model in which you have modeled the effects of
condition within two groups with appropriate dummy variables. The name
'multi-group' is just there to make it easy for you to assign the scans
to each row of the design matrix using SPMs GUI.
> Because I only have one scan per condition I can't perform a random effects
> analysis, but I really want to compare the differences in the patientgroup
> with the differences in the controlgroup. Is this comparison possible in SPM
> in a fixed effects analysis? Or do I have to perform t-tests and then
> compare the results simply 'on sight'?
Your first analysis is doing exactly what you want. Because there is
only one condition effect (activation minus rest) from each subject
there is no distinction between a fixed- and random-effects analysis.
This distinction only emerges when there are repeated measures over
units (e.g. more than one scan per condition in each subject).
I hope this helps - Karl
|