Dear David,
> Thanks for the help. I'm afraid, however, that I don't understand
> entirely. Given four conditions and five subjects (no covariates, no
> nuisance variables), the contrast vector would be [c1 c2 c3 c4 s1 s2 s3 s4
> s5], where c=condition and s=subject).
>
> I am specifically interested in a contrast image that would show, for
> example, condition 2 minus condition 1 (i.e., [-1 1 0 0] for the first
> subject (and on through all 5; other contrasts of interest are [0 -1 1 0]
> and [0 -1 0 1]).
>
> 1. I can't work out how your example does that.
Oh I see - I had assumed you simply wanted to compare the two groups.
In fact you want a group by condition interaction (i.e. the group
differences in specific activations). The solution to your problem is
even simpler - In your case there is no distinction between a fixed
and mixed (random) effects analysis. You can test for the interaction
with a contrast at the first level like
[-1 1 0 0 1 -1 0 0 ...
assuming the first for columns of the design matrix model the 4
conditions for group 1 and then for group 2. The reason that you do
not need take contrasts to the second level is that there are no
repeated measures (i.e. there is no within-subject variability, only
between subject variability for any given condition or difference in
condition effects).
with very best wishes - Karl
|