Hi Fatima,
> I am attaching my design matrix. I can set up the following t contrast
> zeros(1,24) 0 0 1 -1 (thus testing within subject factor stim1>stim2)
>
> but I cannot set up
>
> zeros(1,24) 1 -1 0 0 (between subject factor grp1>grp2)
>
> the answer always is "invalid contrast"
I think the problem is basically that the subject factor makes the
group factor irrelevant; any differences between the groups can be
modelled by the different subject regressors. This is because the
subjects are nested within the groups; if subject was crossed with
another factor like stimulus, then I think that would be estimable.
I think what you are probably really interested in is the interaction
between group and stimulus. While you can't estimate a group (main)
effect if you model all the subject effects, you should still be able
to estimate the "extra" stimulus of the second group wrt the first.
Hopefully, if you select the interaction of your group and stim
factors, and drop the main effect of group, you should get a design
matrix which makes sense, and a contrast which is estimable. If not,
it might be easier to use a paired t-test, and specify a covariate
with ones where stim=2 and group=2 and zeros elsewhere. A contrast
with a one over this column would then test the interaction of
interest. I hope -- insert my usual disclaimer about not being a
statistician...
Best,
Ged
|