Hi Philip,
>I have set up an EV for the task and have configured a separate file for
>the EV plus the demeaned RT (as a covariate) for the corresponding trial. I
>have also configured a contrast across these two cases (i.e., with and
>without the RT covariate). The reasoning (which is quite possibly wrong) is
>that any contingent ROIs identified by this contrast will reflect areas
>that are particularly correlated with the behavioural response.
The problem with this reasoning is that activity correlated to the
behavioral response is likely to be correlated to both the unmodulated
regressor as well as the RT modulated regressor. In fact, there are many
linear combinations of the two that could be plausible decision related
activity. Furthermore, there is a range of linear combinations that are
very difficult to interpret (e.g. no significant unmodulated response, but
significant RT modulated response).
If you have a neuronal model for the decision related activity (e.g. neurons
start firing at the onset of a stimulus and end firing at the response) then
you should probably model them as such i.e. make the shape of your
regressors match the neuronal activity that you are looking for.
cheers,
jack
|