Dear Kristin,
> I am conducting a PET experiment to identify differences in r-CBF between
> a
> group of patients (18 subjects) and a group of healthy controls (18
> matched
> subjects). Each subject was scanned 4 times during rest, and the meanimage
> of these 4 scans was entered into a categorical comparsion between the two
> groups (Compare Populations 1 scan pr. subject SPM99 (age included as a
> nuisance covariate).
>
> As the patient group consists of subjects with different levels of
> disability I would like to see if there is a correlation between
> disability
> level and changes in r-CBF (and still include age as a nuisance
> covariate).
> The obvious thing would be to use the "Multiple subjects - Covariates
> only"
> model, but this is not possible with only one scan per subject in SPM99.
> In
> SPM96 however, it works very vell, - how come?
>
I don't have access to SPM96 any longer, so I can't answer that one.
In the SPM99 'multiple subj cov only' option each subject uses 1 degree of
freedom (block) so that with 1 scan/subj your model would become
overparameterised, hence the trick to use a single-subj model.
> I have tried to use the
> "Single Subject Covariates Only" model instead and this comes out with
> good
> results (that is, I pretend that all scans are performed on the same
> subject
> during different disabiliity levels).
>
> Is there a problem in variance estimation if you enter different subjects
> as
> a single subject? Intuitively one would think that the variance is much
> greater between 18 subjects, than within 18 scans from the same subject.
>
The between-subject variance is indeed very likely to be much
greater than the within-subject variance . But, as you are interested in the
association between subject-to subject differences in disability and / or
age and subj to subj diffs in mean rCBF, between-subject variance is the
relevant source of variance, similar to 2nd level or RFX analyses.
Good luck
Dick Veltman
|