Dost -
There are at least two effects here when you include the reaction times as
a regressor.
First, the effect that you were predicting, signal which was apportioned
to the group effect is now apportioned to the readction times. Note, that
if there is shared variance between your group EV and your reaction times,
i.e. if your reaction times differ on average between the two groups, then
the way in which this shared variance is apportioned can be influenced by
orthogonalisation. If you want to remove any possible influence of
reaction times on your group effect, you should orthogonalise the group EV
with respect to the reaction times - all the shared variance will then be
apportioned to the reaction times.
The second effect of including reaction times can explain your "new
activation". There will be some portion of the reaction time EV which is
_not_ predicted by group membership. When the model is fit, this portion
will effectively "soak up" variance from the residuals. The estimated
residual variance is reduced and hence so is the standard error on the
group level parameter estimate -> higher z-scores.
Hope this is useful
T
-------------------------------------------------------------------------------
Tim Behrens
Centre for Functional MRI of the Brain
The John Radcliffe Hospital
Headley Way Oxford OX3 9DU
Oxford University
Work 01865 222782
Mobile 07980 884537
-------------------------------------------------------------------------------
On Tue, 16 Nov 2004, Dost Ongur wrote:
> Dear all,
> We have been analyzing a dataset where there was a significant difference
> between the reaction times of the two groups in our task (15 in each
> group). In an effort to probe the effects of this difference, I added a
> new EV with the demeaned reactions times for each subject.
>
> Some interesting between-group blood flow differences disappeared,
> indicating that these were related to the reaction time (e.g. the longer
> the reaction time, the more blood flow to this area). Some between-group
> differences remained the same, indicating these were not related to the
> reaction times.
>
> Interestingly, there is now a new area of activation in the between-group
> contrast following the addition of the reaction time covariate. We are
> trying to understand what this might mean. Does this mean that activity in
> this area was negatively correlated with reaction times (the faster the
> performance, the more the blood flow)? Or is there some more complicated
> relationship?
>
> Thanks,
> Dost
>
|