Hi Mirko,
> I would really like to get help. I have two groups and subjects had two
> scans before and after treatment.
With this kind of repeated-measures data, you probably want to model
subject effects in the design matrix, together with a time effect, a
time-group interaction (probably the treatment effect you are
interested in) and your covariate(s). So your design matrix would have
many more than 4 regressors/columns, at least:
(number of subjects) + 2 + (number of covariates)
SPM5's "flexible factorial" model should allow you to set this kind of
design up; either with subject as a factor, or, perhaps simpler,
without it but with dependence between levels of the time factor. See
the following thread, but note that your two-time case won't have the
complications with the dependencies.
http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0701&L=SPM&P=R47197
In the latter case, I think two columns for group, one for time-effect
and one for time-group interaction, plus the covariate(s) will be needed.
I'm afraid I'm not sure how to set either design up in SPM2.
As an alternative, which is probably simpler, and which I think is
equivalent, you could create time2-time1 difference images for each
subject, e.g. with:
http://www.cs.ucl.ac.uk/staff/gridgway/vbm/make_diffs.m
and then just enter these in a two-group t-test with your covariates.
I think this only allows you to test the time-group interaction (and
any interactions between time and your covariates, and the three-way
time-group-covariates interactions), but that's probably what you
need, I guess...
> In one group subjects were treated
> with an active drug while in another were treated with placebo. All
> subject have also behavioral variable measure before and after treatment
> which are used as covariates. I have problem n evaluating interactions
> with covariates
It's not quite clear which interactions you are interested in, but I
think you'd probably need these entered explicitly as product terms in
the design matrix. E.g. separate grp1*cov1 grp2*cov1 regressors.
> because the SPM permits me to enter only four numbers
> for contrasts (e.g. 1 -1 -1 1), which in my interpretation could not
> include covariates interactions.
Just to clarify this, t-contrasts must have the same number of
elements as there are regressors/columns in the design matrix. For
some simple interactions, e.g. time-group, it is often possible to get
interactions with a kind of difference-of-differences contrast, but as
far as I understand it (probably not far enough) in general,
interactions are products of regressors.
I hope this is of help, please post to the list again if you want more
details, or if I have misunderstood your data/aims.
Ged.
|