Dear Richard (and SPM'ers)
Thanks very much for your helpful response to our question about using
masking to to perform a conjunction analysis.
I think we understand conceptually what we want to do, and wondered if
anyone could give us practical advice on how to carry out the analysis.
We have two conditions, A and B, and have modelled our baseline (N)
explicitly. We have carried out a second-level (random-effects) analysis
with SPM99 on both the A-N and B-N contrasts, without any problems. Now
we are interested in performing a conjunction analysis to answer the
question about what A-N and B-N share, using masking (inclusive).
We have no problem performing this analysis at the single-subject level,
but what we expected to be able to do was to go to the second level and
average each of the contrasts created by the A-N and B-N conjunction
analysis. However, we have been unable to find anything that seems like
the correct file to use in such an analysis. SPM99 doesn't seem to allow
us to do the conjunction analysis on the grouped data, because both
the A-N and B-N must be appear in the same dialog, and, in the case of our
random-effects analysis, these two contrasts are in different files.
I know we must be doing something wrong (and, probably something fairly
obvious) -- does anyone out there have some practical advice for
performing this analysis?
Best wishes,
Christy Marshuetz
> >I have what seems to be a simple problem. I have 2 conditions in my study
> >(A and B). I want to see what activiations A and B share for each subject
> >and I would like to average this for the 12 subjects.
>
> If you have a baseline condition, N, then 'activations A and B share', i.e.
> voxels which appear in both of the contrasts A vs N and B vs N, are given
> by the conjunction of these two contrasts. If the baseline is modelled
> implicitly then you could do the conjunction of the main effect of A (1 0)
> and the main effect of B (0 1). However, if there is no baseline
> condition, then you can only compare A with B, in which case there are no
> 'activations' which 'A and B share', since there are no voxels which appear
> in both A vs B and B vs A. I have assumed that you are talking about an
> epoch-based design. If you have used an event-related design, however,
> then there is effectively a 'baseline' anyway, and you can do the main
> effect contrasts specified above.
>
> If you want to 'average' for 12 subjects, you need to choose between a
> fixed-effects analysis and a random-effects analysis,
>
> Best wishes,
>
> Richard.
>
> from: Dr Richard Perry,
> Clinical Research Fellow, Wellcome Department of Cognitive Neurology,
> Darwin Building, University College London, Gower Street, London WC1E 6BT.
> Tel: 0171 504 2187; e mail: [log in to unmask]
> Pager: 04325 253 566.
>
>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|