Dear Rebecca,
> I am trying to analyse a drug study in which subjects were scanned
> on two occasions, once on drug, once on placebo. Blind, randomised,
> counterbalanced. It seems to me that in the analysis of this
> there are really 3 levels of variance - the within session
> variance (modelled at the first level) and the between subject
> variance (modelled at the second level) but also the
> within-subject- but-between-session variance, which intuitively falls
> between the two in terms of contribution to overall variance.
Yes... this within-subject-between-session variance is taken into
account by the session or block effect in the design matrix. So, if this
variance component is an additive component in each session, the
1st-level model should already embody this component.
>
> I have tried incorporating the between session variance in the first
> level analysis to just generate a single contrast image per subject
> which then feeds into a 1-sample t-test at the second level. This
> gives very nice results.
Sounds fine to me...
> I have also tried doing it at the second
> level by making 2 contrast images per subject, one for drug and one
> for placebo then feeding these into a paired t-tests. This gives very
> different (not nice at all) results.
I'm no expert at this, but both analyses should be able to pick up the
effect you're after. In the one sample t-test, you do this by specifying
the contrast [1] and in the 2-sample t-test by e.g. [-1 1]
(placebo/drug). The difference between these models is that the 2-sample
t-test uses more df for parameter estimation, i.e. n+1 dfs, where n is
the number of subjects. Whereas the 1-sample t-test uses only 1 df. The
difference in results might be caused (1) by the difference in the
residual dfs and (2) by different residual components. In your case, I
would stick to the 1-sample t-test, because it answers your question and
gives you maximum df.
Stefan
--
Stefan Kiebel
Functional Imaging Laboratory
Wellcome Dept. of Cognitive Neurology
12 Queen Square
WC1N 3BG London, UK
Tel.: +44-(0)20-7833-7478
FAX : -7813-1420
email: [log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|