Re: st: Relative Importance of predictors in regression

[David Hoaglin <dchoaglin@gmail.com>]

Jorge,

Thanks for that link.  Harring's remark that the proportion of
variance accounted for by the overlap can be negative indicates that
the "shared variance" is an artificial construct.  If the "shared
variance" were real, it would be possible to give a definition for it
and then estimate it.  If one had a negative shared variance and chose
to express in linear units by taking its square root, the result would
be an imaginary number --- perhaps a fitting outcome.

One could, I suppose, define some measure of shared variance by using
the principal components for the predictors, but that would not
necessarily be complementary to the unique contributions of the
individual predictors (which, themselves, are not complementary).

David Hoaglin

On Wed, Nov 6, 2013 at 9:53 AM, Jorge Eduardo Pérez Pérez
<jorge_perez@brown.edu> wrote:
> You are right about the MS. In my calculations I used the partial SS
> themselves. These should add up to the R squared.
>
> I don't really have much on the way of interpretation of the shared
> variance, in fact I found that it is hard to interpret it here:
>
> http://www.education.umd.edu/EDMS/fac/Harring/Past-Classes/EDMS651/Notes/MRA-MS.pdf
>
> --------------------------------------------
> Jorge Eduardo Pérez Pérez
> Graduate Student
> Department of Economics
> Brown University

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/