Hi Allison,
>
> So now I'm really confused. I don't completely understand the
> format of
> SPM.Vi, so it's hard to verify what's being done. So, if someone
> couldexplain
>
>
>
> 1) What is the proper way to non-sphericity correct a PET
> study with 2
> or more (unequal) groups, 2 scans each
>
> 2) The proper way to non-sphericity correct a PET study with
> 2 or more
> (unequal) groups, 1 scan each
As I understand your design you have two groups of subjects, two scans per subject. I'm assuming these two scans are before and after some intervention.
In a study like this there are three potentially intresting effects, main effect of group (is there a difference between groups?), main effect of intervention (does it have an effect?) and the interaction (is the intervention more effecitve in one group than in the other?).
If you chuck all the scans into a PET-design you can answer all these questions, BUT you will have strong correlations between the two scans in each subject. SPM allows for the estimation of those correlations, thereby avoiding the problem of inflated degrees of freedom that might otherwise cause you to draw unwarranted conclusions.
What we should remember though is that the resulting variance-covariance matrix that is used to pre-whiten your data is an _estimate_, i.e. not the truth. My take on it is that whenever it is possible to design your second level analyses such that you can avoid variance-component estimation, you should. It is really mainly when you want to use F-tests spanning several basis-functions per condition at the second level that it is necessary.
In your case you can quite easily avoid the problem by performing two second level analyses.
In one you use as your input the mean across the two scans in each subject as your input-images and set up a two-sample t-test to look for differences between groups.
In the second you use the difference (subtraction image) between the two scans as your input-image, set up a two-sample t-test and the [1 1] contrast will give you the intevention effect and the [1 -1] and [-1 1] contrasts will give you the one-tailed interactions.
You can of course still model unequal variances between groups, but to be honest I suspect it would make b**er all difference.
>
> 3) The format of SPM.Vi.Vi
I'll pass on this one. Sorry.
Good luck Jesper
|