> I'm curious about the validity or otherwise of using basic ANOVA or ANCOVA
> models to explore single-patient versus group differences. In particular,
> using combined contrasts, conjunctions etc to analyse the G,W and say CSF
> tissue classes together. The "groups" in the model are represented by
> selections of corresponding tissue classes of a particular subject and
> control subject groups and so on.
>
> So the question is...would that approach necessitate a more complex
> multivariate model or is a basic model with contrasts adequate? Does
> anybody have any experience with this?
Ideally, you would use a multivariate method to do a voxel-wise simultaneous
analysis of the different tissue classes. Unfortunately, such tests are not
done by SPM - although GRF theory has now been worked out for some
multivariate models:
http://www.math.mcgill.ca/~keith/hbm2004/hbm2004.htm
The closest that SPM would currently come to performing multivariate testing
is to use a non-sphericity correction. This makes a number of assumptions
(i.e. that the form of the covariance is the same for all voxels), which are
unlikely to be valid for VBM data. Depending where you look in the images,
the covariance between e.g. grey and matter will vary by quite a lot.
As for conjunctions and stuff, I will let someone else try to answer that.
Best regards,
-John
|