On Sun, 27 Dec 2009 02:33:13 +1100, Ian Harding <[log in to unmask]>
wrote:
<snip>
>I've tried serial orthogonalization of the design matrix, followed by
>respecification of the model using the modified motion parameters (such to
>avoid orthogonalization of my effects of interest). However, albeit
> significantly reduced, this does not remove all colinearity. There are
If you serially orthogonalize the columns, starting with the effects of interest,
and only then the effects of no interest, and then replace the modified
(orthogonalized) columns of interest with the originals, then you should have
what you want.
Alternatively, you could take the matrix formed of columns of interest, form
the "residual forming" matrix, and replace the columns of no interest by their
residuals with respect to that matrix, it should also give you what you want.
>similar issues in orthogonalizing the mp's to each effect of interest in a
>serial manner (ie. orthogonalizing to condition 1, then taking those
>parameters and orthogonalizing to condition 2), or in combining all
>task-regressors such to create an "All Stimuli" column, orthogonalizing the
>mp's, and then inputing those mp's back into the original matrix. Soooo....
>I'm stumped.
>
>Any thoughts?
>
>Happy Holidays!
>
>Ian
>
>---
>Ian Harding, PhD Student
>Melbourne Neuropsychiatry Centre
>Department of Psychiatry
>University of Melbourne
>
|