Dear Andrew,
> I'm analysing some PET data, and am interested in looking at regions where
> activity correlates with RT, and was wondering what was the best way to
> proceed.
> We have 6 subjects, 12 scans per subject
> When I choose the multi-subject covariates only design, there appear to be
> two ways to proceed. One is to not have covariate * subject interactions,
> to mean centre the RTs, and then to make the following contrasts: 1 to look
> at regions where activity correlates with increasing RT and -1 the converse.
> The second is to select covariate by subject interactions, and then to have
> the following contrasts: 1 1 1 1 1 1 and -1 -1 -1 -1 -1 -1.
> Could someone help explain what the differences are between these two
> analyses?
If you don't model the subject by covariate interaction, you assume that
there is a common slope of the covariate over all subjects. This saves
you some degrees of freedom, but essentially you're assuming that the
slope of these covariates is the same for each subject, if there is some
component in the observations, which can be explained by your covariate.
If you do model subject by covariate interaction, you don't make this
assumption of the same slope for each subject, but allow for fitting a
different slope for each subject. Note that this model can also be used
to generate subject specific contrast-images, which you can use as input
to a second level analysis.
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]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|