Dear FSL people,
I use FSL on a windows XP.
I have 24 subjects and each subject did 3 sessions. The sessions are the
same experiment (basically, I had each subject do the experiment 3 times to
get better power/enough trials).
I have succesfully performed a first level analysis for each session
separately (so 72 separate analyses).
I would now like to make the leap to higher level analyses. As far as I
know, I should first do a second level analysis for each subject over the 3
sessions (to get the mean effect for each subject) and then finally a third
level analysis over the 24 subjects (to get the mean effect for all
I run into trouble when trying to do the second level analysis. According
to the manual, 'putting all the subjects into one second level analysis
should make the estimation of cross-session variance more robust'. However,
In my case this leads to a second level analysis with 72 'inputs' (the rows
when specifying the GLM) and 24 EV's. Unfortunately, FSL does not seem to
be able to cope with this amount (my FSL basically freezes when I try to
specify so many EV's). I suppose I don't quite understand why putting all
the subjects into 1 second level analysis (as opposed to for example doing
24 separate second level analyses) would be better, since you're not taking
between subject variances into account yet.
So basically, here's my question: Would I loose statistical power when I
decided to instead split this second level analysis up? My idea is to first
do the first 12 subjects (so 36 'inputs' and 12 EV's) and then the second
12 subjects. Is there a problem with doing the second level analysis this
way, seeing as I should still wind up with 24 copes for the third level
analysis... To be honest, I don't see any other way for me to analyze my