Hi Andreas

You should get equivalent results setting up your model as either a
paired-samples t-test with 2 covariates, or as a flexible factorial
design (2 x 2) with columns for subject effects.  In either case, what
you want to see in your design matrix is the same as you have now (4
columns indicating scan 1, scan 2, covariate 1, covariate 2), but with
an additional column for every subject to control for subject-specific
variance.  So, if you have 10 subjects, you would have 14 columns in
your design matrix (4 columns for treatment effects, 10 columns for
subject effects).

The discussion of within-subjects ANOVAs covered in the Henson & Penny
guide might be of some help as well:

http://www.mrc-cbu.cam.ac.uk/people/rik.henson/personal/HensonPenny_ANOVA_03.pdf

Figure 5 is what a 1x4 within subject ANOVA design would look like.
In your case you would have 2 conditions + covariates instead of 4
conditions.

(Also note you may get into trouble if the covariates you use for pre
and post scans are very similar; this implies a consistent effect of
subject, which you are modeling out by including the subject columns
in your design matrix...)

Good luck!

Jonathan


On Thu, Mar 12, 2009 at 2:53 PM, Andreas Hahn
<[log in to unmask]> wrote:
> Dear SPM experts,
>
> I have a question regarding the regression analysis in SPM5...
>
> My data consists of
> - 1 patient group,
> - each suject underwent 2 scans (e.g. pre & post scans),
> - for each of the scan and subject I have a covariate (e.g. pre & post error
> rate).
>
> I first calculated the regression (scans & error rate) SEPARATELY for each
> scan (pre & post), which is quite simple...
> Now I want to know, if the regression for one (e.g. pre) scan is stronger
> than for the other (e.g. post) scan.
>
> Jonathan Peelle recently offered me a great explanation how to do this if
> the two scans are independent using an ANOVA setting (two groups, two
> covariates, contrast manager setting to 0 0 1 -1).
> However, now I want to account for the dependency of the two scans (like
> within a paired T-test).
> I first thought simply to change the "Factors -> Independence" setting to
> "No" but then the covariate within my design matrix changes and I do not
> understand why!
>
> Any suggestions would be greatly appreciated :)
> Thanks a lot,
> Andreas.
>