Ricardo Ovaldia
>
> --- Nick Cox <[email protected]> in reply to my
> post asking if -qreg mpg foreign- followed by
> -test foreign- tests the null that the MEDIANS of the
> two groups (foreign and domestic)are the same? wrote:
>
> > 1. Yes, in this circumstance it bears this
> > interpretation given the relationship between the
> > median and minimising sums of absolute deviations.
> >
> > 2. But it's not a test that I would want to run
> > with. The double exponential model, while a
> > numerically tractable and appropriate model to work
> > with when being moderately pessimistic about the
> > tails of an error distribution, is not often a very
> > realistic model when comparing two data
> > distributions.
>
> Now, if I include another variable in the model such
> as -price-, in the case of linear regression I can
> interpret the results from -test foreign- as a ttest
> of foreign adjusted for -price-. Does this
> interpretation also applies to -qreg- except with
> respect to medians? Or is there a better way to
> compare adjusted medians?
I stuck my neck out previously, with no one either
supporting or disagreeing. I'll stick my neck
out a bit further, but no more than that.
The analogy now seems to be under very severe strain.
In the case of regression and means, lots of things
fall out neatly as a consequence of the algebra,
and that of course is one motivation for choosing
the method in the first place.
I don't think there is really very much that
you can say algebraically about conditional medians
in such circumstances, so at best you would be
arguing on a rather tenuous analogy.
Nick
[email protected]
*
* 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/
