# Re: st: quantile regression question

 From Robert Duval <[email protected]> To [email protected] Subject Re: st: quantile regression question Date Mon, 25 Jul 2005 23:20:01 -0400

```Hi Tony (or Mario as it reads on the email account!)

You don't need to split the whole sample. An interaction term is ok
too. You can have categorical variables indicating at which level of
the distribution of ability the individual is... and interact
experience with those dummies. That should give you what you want
without having to split the sample.

I guess my main point is to be careful with the interpretation of
qreg. Again it fits a line over the distribution of the error term of
y after controlling x, not over the distribution of x.

cheers,
robert

On 7/25/05, Robert Duval <[email protected]> wrote:
> rephrasing a little bit my own previous statement...
>
> " If you're not including variable z as a regressor in the conditioning
> set then qreg won't give you what you want"
>
> since even if you were to include z as a regressor qreg won't give you
> what you stated as your goal.
> robert
>
> On 7/25/05, Robert Duval <[email protected]> wrote:
> > quantile regression fits a line at different points of the CONDITIONAL
> > distribution of y given x, i.e.  f(y|x).
> >
> >  At each fixed value of x (i.e. x=x0) f(y|x=x0) gives you the
> > distribution of y after controlling for x (a distribution of the
> > errors in other words). Qreg fits a line at a given (prespecifed)
> > quantile of such distribution.
> >
> > If you're not including variable z as a regressor in the conditioning
> > set then qreg won't give you what you want as stated in your goal (
> > "coefficients of experience in a series of  wage regressions by
> > quantile of the distribuiton of ability").
> >
> > If you're willing to assume exogeneity on your excluded variable z, it
> > seems to me that it would suffice to just split the sample by
> > quantiles of the excluded variable and run an ordinary regression over
> > each sample (a fully interactive model in other words).
> >
> > robert
> >
> >
> > On 7/25/05, Antonio Fanari <[email protected]> wrote:
> > > Hi,
> > > is it possible to run quantile regressions where the quantiles are not
> > > defined over the independent variable but over some other variable?
> > > Say you have wages as a dependent variable and another variable measuring
> > > ability (e.g. an IQ test). Suppose you want to see if the effect of, say,
> > > experience on wages varies across quantiles of ability. One way is to
> > > include interactions of experience*ability in a traditional OLS regression.
> > > However, I was wondering if there is a way to run quantile regressions,
> > > instead. So one would come up with coefficients of experience in a series of
> > > wage regressions by quantile of the distribuiton of ability.
> > > Please forgive me if the question does not make any sense, but I would
> > > appreciate any feedback!
> > > Thanks.
> > > Tony Fanari
> > >
> > > _________________________________________________________________
> > > Don't just search. Find. Check out the new MSN Search!
> > > http://search.msn.click-url.com/go/onm00200636ave/direct/01/
> > >
> > > *
> > > *   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/
> > >
> >
>

*
*   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/
```