Dear Andre,
what I found out in the meantime is that the subject factor itself
should cover any covariates in a repeated-measures design that stay the
same across measurements (e.g, sex of subjects). You can try this out in
MATLAB or any statistics software (I only did it with the regress()
function in MATLAB): First create the model as you want it to be in SPM
(with subject factors and covariates) and afterwards drop your
covariates. The estimates for groups, time points, etc. should stay the
same, only those for the subject factors and covariates change. If you
are not explicitly interested in these, you can simply ignore covariates
of such a structure. Broadly speaking, the subject factors cover
characteristics of your repeated data that stay the same (you can
imagine it to be something like the mean over time points for each
subject). If your covariates should now also account for fixed
characteristics, the respective variance is shared between your
covariates and subject factors, without influencing other estimates.
Subject factors will cover subject-specific variance, which was not
already accounted for by your covariates.
I didn't add this to my original mail because it does not solve the main
problem of not being able to set up a proper contrast anymore. Moreover,
this solution just covers covariates of the structure stated above. I
hope this helps and brings us closer to a general solution.
Best,
Manfred
Am 20.09.2017 um 22:17 schrieb Andre Machado:
> I have set up a second level 2x2 flexible factorial with 19 people in one group and 20 in the other, and a within subjects factor with two levels for condition. In general, I have followed the instructions contained in Glascher and Gitelman's guide, and included subjects as a factor (and modeled for it). Everything works fine except the contrast manager.
>
> My situation was very similar to the one encountered in this email
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1707&L=spm&P=R43211&1=spm&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4
> And any answers to it could help me out.
>
> I had this batch running with no problems, until I decided to model the subject factor. One of problems that came up, is that If I add covariates (age, sex, blood pressure), I stop being able to make the contrast for group effect work. Simply adding 0's for them when setting up the group contrast is not working.
>
> Anyone can help?
|