Dear Nici and Filip,
I am by no means an expert with regard to RFX, thus my advice should be
taken with a grain of salt.
From my perspective, your idea is not quite compatible with the general
intention of RFX analyses in neuroimaging, i.e. trying to make an inference
with regard to the population(s) from which the subjects were sampled. By
using one contrast image for each individual run and therefore several
contrast images for each subject on the second level of the analysis, you
are no longer analysing variance *between subjects*, but variance *between
sessions within _and_ across subjects*. In other words, you are
contaminating your between-subject variance (differences between contrast
images of different subjects) with within-subject variance (differences
between contrast images that belong to the same subject). Including the
within-subject variance into your analysis no longer guards you against the
possibility that one subject skews the results just by virtue of her/his
exceptionally large inter-session variability. Consequently, you are no
longer able to make any inference about the general population from which
subjects were drawn. Instead, your inference is only valid with regard to
the particular group of subjects studied - but this is not what you
intended, I suppose.
Hope this makes sense - any corrections are most welcome.
Best wishes,
Klaas
|