Dear Marco,
> I have a few questions concerning a multi-subjects fMRI experiment.
> For each subject we acquired 4 separate scan-sessions according to the
> following experimental design:
> session 1,2: Ce Cr Ce Cr Ce Cr Ce Cr Ae Ar Ae Ar Ae Ar Ae Ar
> session 3,4: Ce Cr Ce Cr Ce Cr Ce Cr Be Br Be Br Be Br Be Br
> where:
> e = encoding
> r = retrieval
> C = baseline
> A = condition1
> B = condition2
>
> Up to now, I have been able to analyze contrasts between conditions
> belonging to sessions of the same type (e.g. Ae-Ce; Ar-Cr for sessions 1
> and 2) both using a fixed-effect and a 2nd-level random effect model.
>
> In addition, however, I would like to compare conditions belonging to
> sessions of a different type - though I know it would have been much
> better to include all conditions in one session. E.g. compare Ae-Be. I
> have tried to do this on a random-effect basis, by computing the simple
> main effects for each subject (Ae-Ce and Be-Ce) and the entering the
> corresponding con*.images into a paired t-test (Ae-Ce vs Be-Ce). My
> questions here are: what am I actually looking by this approach? At
> interaction effects [of the type (Ae-Ce) - (Be-Ce)]?
Yes indeed. You could construe your design as a 3 factor design:
Factor 1: C vs. Active (A or B) (2 levels)
Factor 2: e vs. r (2 levels)
Factor 3: Condition 1 vs. Condition 2 (2 levels)
Your effect is a 2 way interaction Factor 1 x Factor 3, under e.
> Do I have to mask this contrast by another contrast (e.g. main effect)
> and how can I specify the masking contrast?
You do not have to but if you wanted t; use a 2nd-level model with
(Ae-Ce) in one column and (Be-Ce) in another (plus the constant term).
Then mask [1 -1 0] with [1 1 1]. The latter is the main effect of
Factor 1.
> Is there any way to look at direct comparisons (Ae-Be)?
Yes; just do a two sample t test on contrasts testing for Ae and Be
separately. These will be the same as the beta???.img.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|