> Dear FSL experts, I am working on a data-set which show some very
> interesting patterns but also something I fail to grasp:
> I have 4 conditions, stimulus duration is identical for all 4
> conditions and stimulus frequency is almost identical. At single
> subject level A, B, C en D are estimated and A is contrasted with B
> and C is contrasted with D. All stimuli are presented in the same
> run. At a grand-average level
> A>B differes is a substantial (and expected number of) areas (B>A
> shows little effect).
> C>D shows no differences (also not D>C).
> So far everything is as expected. Now the strange thing (for me)
> C>D has similar copes values (about 20%-40% higher) in terms of
> range but varcopes values that are about 10 times higher. An
> explanation would be that the variance is larger for C>D than A>B
> but as far as I can see this should not be possible since these are
> estimated in the same runs.
it is nor completely clear what you analysis you mean here. Is this
all a 1st or 2nd level analysis?
It is not impossible to have vastly different efficiency (reflected in
the varcopes) for different contrasts at the 1st level. The most
common reason would be different numbers of events, but even when
having the same numbers it can be different due to the exact details
of the timing. If for example an event of type C is always followed by
an event of type D very shortly afterwards the D>C contrast would be
very poorly estimated.
I suggest you have a look at the efficiency of your design (misc tab
in feat gui) and see what you get.
Good luck Jesper