Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: variable selection and bootstrap


From   Stas Kolenikov <skolenik@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: variable selection and bootstrap
Date   Sat, 21 Mar 2009 23:51:00 -0500

It looks to me that you are mixing two things up: 1. uncertainty about
the parameter estimates (usually addressed in the form of standard
errors of the estimates), and 2. uncertainty about the model (which is
usually addressed by selecting the model and waving hands about the
well known properties of AIC/BIC, or something like that).

If you have orthogonal factors, you can arguably combine the two by
taking the bootstrap samples, fitting the model, and writing down the
coefficient estimates if the coefficient is in the model or zero if
not. Then you can summarize the bootstrap results and try to make
sense of these. (There are caveats regarding identifiability and lack
of scale invariance in binary dependent variables though, at least
relevant for econometricians.) If your factors are not orthogonal then
different models will have different population parameters to converge
to, and switching between models would lead to all sorts of problems.

In fact, this procedure is pretty close to what Bayesians would do,
except they would sample the parameters rather than the data.

On 3/21/09, aegl <aegl@orange.fr> wrote:
> Hello,
>
>  I need to run a logistic regression including several X variables . I
> therefore plan to use backwards selection to select the best predictors and
> I have read that it is possible to combine this approach with boostrap.
>  I have tried using the following syntax :
>
>  stepwise, pr(.10): logistic Y  X1 X2 X3 X4 X5 X6 X7, vce(bootstrap)
>
>  This gives a result that is different from using an alternative approach:
> first running the backwards selection on the original sample, and then using
> the bootstrap to compute the CIs.
>
>  Does anybody know what I am doing by using this syntax ? Is there another
> way to run backwards selection procedures with bootstrap ?
>
>  Thank you very much for any input.
>
>  Joel Rebloch'
>  Data analyst
>  Grenoble, France
>
>
>
>
>
>  *
>  *   For searches and help try:
>  *   http://www.stata.com/help.cgi?search
>  *   http://www.stata.com/support/statalist/faq
>  *   http://www.ats.ucla.edu/stat/stata/
>


-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index