Hi Matthew,
generally, I'd think that one should keep as much data as possible.
if you have multiple conditions per session, so that even if you have an
error-only condition, there might still be other events that would be
useful to you. I'd expect that a good approach would model error trials
explicitly as a condition. If there is no correct trial left for a
specific condition, of course, you shouldn't model the correct trials of
this condition.
if you have only one condition per session, i..e., there are no correct
trials at all in that session, I wouldn't bother with modelling the
session. The reason is that you presumably have already enough degrees
of freedom for a fixed-effects within-subject analysis. And at the
second level, you wouldn't model the missing session anyway (because you
don't have observed effects for this session).
Stefan
> Dear SPMers,
>
> I have recently collected event-related data on an executive function task
> over 6 sessions (or runs). The paradigm is such that there are many error
> trials. Occasionally a participant has no correct trials for a given
> condition in a session, hence SPM multi-session analyses are confounded.
>
> I do not know what to do except simply exclude those sessions where this
> occurs (it is rare), but I want to avoid losing SNR. I thought that
> proportionally scaling the data volume-wise first, then modelling as a
> single session might be OK. Can anyone advise me?
>
> Thankyou,
>
> Matthew.
>
>
>
--
Dr. Stefan Kiebel
Wellcome Dept of Imaging Neuroscience
Institute of Neurology, UCL
12 Queen Square
London WC1N 3BG
Phone: (+44) 20 7833 7478
Fax: (+44) 20 7813 1420
|