how would an appropriate second level analysis be designed,
if 10 subjects were scanned 3 times (using the same paradigm),
and - for proof of concept - we want to pool all 30 runs to the test
the null hypothesis "no activation in a region X"?
It seems that in a one-sample t-test against zero (as one
one would use for a normal random effects analysis) is not correct here due to
the different variances between the subjects.
thanks very much for any hints,
Max Planck Institute of Psychiatry
NMR Research Group
Mail: [log in to unmask]