Dear Bruno,
> we thank Karl for the quick answer to our question regarding the
> evaluation of an interaction between a contrast and a performance
> variable in an fMRI study. Now we would like to ask, whether an
> analysis applying a parametric modulation would also be a way to go.
> Again here an outline of one of our fMRI experiments:
>
> a.) 2 conditions ABABABAB, of which we think that only one is modulated
> with the performance variable.
>
> b.) 6 scans per epoch, 9 subjects
>
> We entered all 432 scans (9 subjects x 2 conditions x 4 repetitions x 6
> scans) as one session. Then we specified 1 trial and chose the
> parametric modulation 'other', entering 36 values (each value was
> entered 4 times since only 1 performance variable exists per person but
> the trial was repeated 4 times). We chose a 'linear' expansion and
> selected the F-contrast 'effects of interest' in the results section.
>
> Is this procedure OK? The results look good - but is it really
> appropriate to concatenate the scans of different subjects in one
> session?
This is perfectly fine as an analysis and represents the equivalent
fixed-effect analysis to the random-effect analysis I outlined
previously. Note here that you have to treat all the scans as one
session because your effect of interest is a session x condition
interaction (which would lie in the confound space in a multi-session
design).
If you get good results that great (it may have been that other
non-specific session x condition interactions could have inflated the
error variance too much).
Remember that the sphere of inference from this analysis only applies
to the specific subjects studied, at the time of study, and cannot be
generalized to the population. This is a qualification of the
inference.
With best wishes - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|