Dear Christopher,
>| I want to reduce that SPM{Z} to an index of change over time, and compare
>| that index across subjects in treatment. One approach would be to sum the
>| size of all clusters which reach significance for the bivariate comparison
>| {k, Z}. Some in my group are concerned about the stability of a measure
>| such as "k" on an SPM output in such a contrast.
>|
>| Would you accept this as a reasonable measure?
I assume here that, as you have such a spatially heterogenous population,
you wish to derive some kind of 'summary statistic' which will allow you to
make inferences about the distributed pattern of differences which you see.
After a wee bit of thought, I can't really see a problem with using a
simple measure (although I'm prepared to be proved wrong...), like all
voxels which survive your cluster threshold, to compare your groups. One
could simply take these values and evaluate them, off-line, using a
two-sample t-test as before, comparing patient's 'k' to your group 'k'.
This analysis asks the question: 'where do I see ANY changes in my patients
vs. controls'. If you're comfortable with this, then go ahead.
I would advise that you have as many normals as possible in your control
group, though, as I think that inference based on this kind of summary
measure will suffer from a lack of sensitivity.
I am unsure if I'm being blasphemous and taking God's name in vain by
uttering these sacred words, but: 'hope this helps'
Good luck!
Best
Dave
==============================================================================
David J. McGonigle. Tel: 44 171 833 7486
Wellcome Trust PhD Neuroscience Programme. Fax: 44 171 813 1420 (work)
Wellcome Dept. of Cognitive Neurology. 44 181 675 7039 (home)
12 Queen Square. email: [log in to unmask]
London WC1N 3BG. http://www.fil.ion.ucl.ac.uk
==============================================================================
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|