Dear Otto,
> I have a question regarding the power of SPM results. Specifically I have
> FDG PET studies obtained in 17 adult controls and 7 children older than 6
> year of age. I performed a routine SPM analysis at the 0.05 level and no
> significant differences can be detected. The reviewer however argues that
> this is a null finding of group differences between small groups and thus
> I cannot conclude that subtle differences do not exist.
> How can I determine what power my results have?
For voxel-wise power estimates, one can apply the non-central t or F
distributions - see for example Van Horn et al, NeuroImage 7, 97-107. I once
wrote some software to do voxelwise power calculations on SPM96 analyses that
may or may not be useful to you:
ftp://ftp.mrc-cbu.cam.ac.uk/pub/imaging/Power
However, I'm not sure whether the voxel-wise approach is answering the correct
question. For example, an obvious question might be how much power you have to
detect a change in a given brain region, and this question is more complex to
answer within SPM, as even for a small region you are likely to have more than
a single measurement's worth of data. You could of course reduce the problem
by using regions of interest.
Good luck,
Matthew
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|