Hello Steve,
Thanks for your reply. I do understand that demeaning is no guarantee for removing collinearity, but that was not really my question. Rather: shouldn't the coefficient time series estimated with raw and demeaned regressors be perfectly correlated?
Actually, I just looked again at the fsl_glm help file and I may have found the reason for the apparent inconsistency: '-demean' removes the mean not only from the regressors but *also* from the EPI time series, and that, I believe, *is* going to change the estimated time series beyond a simple constant shift.
thanks again
best
giuseppe
|