Hi Thomas,

Probably this is best run all at second-level, and with the standard
(mixed-effects) framework, so that you are asking the question - "are the
fitted effects at second-level greater than the session variability?". So
you would put the 3*6 first-level analyses into the second-level design.

Your covariate of no interest looks ok. You probably also want to take out
the mean effect with an EV of all 1s. Then your main question should be
orthogonalised with respect to this, i.e., demeaned.

Regards, Steve.



On Wed, 7 Jan 2004, Thomas Mierdorf wrote:

> Hi,
>
> I am wondering if it might be possible to use a
> three level mixed-effects GLM to analyse my
> single-subject repeated-measure fMRI data which
> looks like this:
>
> First-level:
> - blockdesign (106 scans)
> - three conditions repeated three times each
>    + 7 confounds (incl. headmovement estimates)
>
> Second-level:
> - the measurement was repeated twice resulting in
> three sessions
> - a covariate of no interest at this level could
> be "adaption" (1 0 -1)
>
> Third-level:
> - the three sessions were repeated six times
> seperated by a week each
> - after the first week we introduced a treatment
> which stopped right before the fifth week
> - a covariate of no interest at this level could be
> again "adaption" (5 3 1 -1 -3 -5)
> - I want to test for positive or negative BOLD
> responses which change with treatment, i.e. either
> look at differences between all of the weeks or
> use a covariate which resembles our hypothesis:
> 1-1-2-3-4-2
>
> Does this make sense while all data is coming from
> a single subject? How do I calculate the effective
> degrees of freedom (which are hopefully not 6 weeks
> minus 2 covariates = 4 as in "random effects"
> analysis)?
>
> Unfortunately, the control condition turned out to
> change with treatment too, therefore I am not able
> to use the contrast "(A or B) > C" at the first-level,
> but only "(A or B) > 0" or "(A or B) < 0". Apart
> from introducing the problem of intra-subject
> variability of activation (which would have been
> controlled by the control condition), does this
> complicate anything?
>
> I would be grateful for any advice.
>
> All the best,
> Thomas
>
>
>  -------------------------------
>  Thomas Mierdorf, Dipl. Psych.
>
>  Institute of Experimental Psychology
>  Heinrich-Heine-University, D-40225 Duesseldorf, Germany
>
>  voice: ++492118112010;  fax: ++492118114522
>  email: [log in to unmask]
>

 Stephen M. Smith  DPhil
 Associate Director, FMRIB and Analysis Research Coordinator

 Oxford University Centre for Functional MRI of the Brain
 John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
 +44 (0) 1865 222726  (fax 222717)

 [log in to unmask]  http://www.fmrib.ox.ac.uk/~steve