Hi Eugene,
Thanks very much for the quick, helpful response! One follow-up re: #2.
>> 2) Regardless of which of these two approaches makes better sense, should
>> I
>> orthogonalize this behavioral EV (or EVs) wrt EVs 1 through 8? I'm
>> thinking
>> yes since it is possible that the number of button presses could vary
>> systematically as a function of condition.
>
> I think this depends on how the regressors overlap. It may be that part of
> the button press responses are influencing the fits of your regressors of
> interest. In this case, you want your new regressors to fully account for
> the button press repsonses. Orthogonalisation of the new regressors would
> mean that the original regressors would still fit to a part of the button
> press response, and the bias caused by the button press on their fits would
> be retained.
OK, going back through the archives, had to remind myself that
orthogonalizing the button press regressors wrt the original
regressors essentially assigns the shared variance to the original
regressors. Is that correct? If so, then perhaps I would want to do
it the other way around and orthogonalize the original regressors wrt
the button press regressors. That would prevent the shared variance
from going into the error term, which is *I think* what would happen
if I were not to orthogonalize at all. Am I on the right track?
Cheers,
Heather
>
> --
> ----------------------------------------------------------------
>
> Eugene Duff, PhD
>
> FMRIB Centre,
> University of Oxford
> John Radcliffe Hospital, Headington OX3 9DU Oxford UK
>
> Ph: +44 (0) 1865 222 739 Fax: +44 (0) 1865 222 717
>
> ----------------------------------------------------------------
>
--
Heather L. Urry
Department of Psychology
Tufts University
490 Boston Avenue
Medford, MA 02155
email: [log in to unmask]
phone: 617-627-3733
fax: 617-627-3181
|