I don't think the concept of "contribution" can be formalized well.
Fields' decomposition is based on a set of axioms typical for
inequality literature, and even at that, not everybody likes it that
much. There are generalizations of those decompostions
(Shorrocks-Owen-Shapley decompositions), but again those are more
appropriate for distribution analysis. The only case when regression
decompositions would be pretty unambiguous would be when your
regressors are orthogonal (which you often have in the appropriately
design analysis of variance exercises). Otherwise, you should think of
what you are conditioning on (other regressors). With the panel data
and autocorrelations, you also have a fraction of variance due to the
panel structure, and the concept of explained variance gets shaky
since there is no clear variance to look it, there are also lags from
the prior times. So... go ahead and formalize the concept of the
contribution, and then the measure should crystallize out -- that's
been the way it worked out in the distribution literature.
On 9/11/06, Olga Gorbachev Melloni <[email protected]> wrote:
Hi,
I am looking for routine that computes contribution of each explanatory
variable to the total Y (dependent var).
I know there is a routine gfields, but it seems to be developed for
inequality literature, is there anything in it that stops me from using it
on generic regression? and if so, do other routines exist?
I am trying to assess how important each regressor is in my model and
whether their specific contributions change with time. I have a panel data
set and i was running a newey2.
--
Stas Kolenikov
http://stas.kolenikov.name
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
