I'm looking at fMRI data with multiple sessions (or "runs"). Some of the
trial types don't occur in every single session for a given subject,
because they're behavior-dependent. In my case, these are "incorrect"
trials, and occasionally a subject will not make an error in a given
session.
Is it crucial that the sum of contrast weights within a given session be
zero, or does it suffice that the sum across all sessions vanish?
The disadvantage of insisting that sums vanish across every session is
that the weights will vary across sessions. Not that I think that
necessarily matters for the validity of the model, though it does mean I
have to be more careful showing others how to choose the weights. On the
other hand, perhaps there's something wrong with allowing nonzero sums
within a session. But given that SPM already takes into account the
effect of run, and furthermore does grand mean scaling, I can't see any
_likely_ scenarios where it would matter much, though obviously one can
think of possibilities.
|