Re: st: importance of independent variables

[Jeph Herrin <stata@spandrel.net>]

This doesn't exactly answer your specific questions, but one useful 
approach for assigning "importance" to independent variables is that of 
"random forests" (RF).

This is an algorithm which over many iterations selects a random subset 
of Xs and does a random split sample validated regression of the Ys 
against the Xs to produce a score. This generates a great big round 
robin tournament of Xs against each other, and results in a final 
"importance score" for each X, independent of the other Xs.

Unfortunately, I don't think anyone has implemented any version of this 
in Stata. Though I have considered doing so when I have a few spare 
days, so far I have only used it in R, which has RF algorithms available 
for linear, logit, and survival models (with multiple imputation for 
missing values of covariates that are not Xs).

hope this helps.

cheers,
Jeph

On 12/6/2012 4:17 AM, A. Berâ wrote:
> Dear Stata Users,
>
> I would like to compare the importance of independent variables (Xs)
> in explaining the dependent variable (y).
>
> What is the best way of doing this and how can this be done in Stata?
>
> For example, how can I partition the variance of y into components
> explained by individual Xs? Is there any user-written program that you
> know of for this purpose?
>
> Any help is appreciated. Thanks.
>
> --
> abdullah bera
> *
> *   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/
*
*   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/