On Thu, 17 Mar 2005 16:57:58 -0500, Daniel Weissman <[log in to unmask]>
wrote:
>Dear Stephen,
>
>I quite agree that following Euclidean normalization (and subsequent
>mean-centering) the betas for two different parametric regressors might
not
>be directly comparable. As you point out, the two regressors are not
scaled
>by the same factor when using a Euclidean transformation. So, it sounds
as
>if you agree with my concerns regarding the use of Euclidean normalization
>for parametric regressors, at least for cases where an experimenter wants
to
>explicitly compare the betas for two different parametric regressors. Is
>that right?
Yes, in the example you gave, one would have to be careful.
>If so, what would be the best transformation to use if one wanted to
>directly compare the betas for two parametric regressors?
If the values of the parameters are directly comparable as in the example
you gave, you'd want to either not scale the parameters, or scale them
identically.
>As to my question of whether there is any reason to prefer a Euclidean
>normalization or a Z-transformation of the parametric values, I don't
quite
>understand your statement that the z-transformation is not linear since,
in
>fact, the z-transformation is linear.
Maybe I misunderstood what you meant by z-transformation. What's the
formula?
Best,
S
>Thanks very much for the response!
>Daniel
>
>Daniel Weissman
>Center for Cognitive Neuroscience
>Duke University
>Durham, NC 27705
|