On Tue, Oct 13, 2009 at 1:05 PM, James Shaw <[email protected]> wrote:
> Thanks for the prompt response. I suspected that the estimated
> weights might factor into generating the scores, though I was not sure
> how to implement them. So, should I interpret the "scale" in your
> code to be the derivative of the objective function with respect to
> the parameters of interest?
I've been thinking aloud looking at -rreg.ado- code. You would need to
go over it line by line, figure out what each variable and scalar
means, and map them into the definition of the "score".
> I generally agree with your concerns about the jackknife, though it
> seems to yield estimates that are similar to those provided by the
> bootstrap in this case. I am not a huge fan of M estimation. The
> only benefit it provides over quantile regression is efficiency, and
> the resulting estimates lack the interpretability of quantile
> regression estimates. I am in need of a robust estimator for
> longitudinal data that will not entail resampling for variance
> estimation since I am resampling at another stage in my code.
> Unfortunately, my options are limited.
I see. Well as I said, to get the scores and analytic standard errors,
you'd need to hack the -rreg.ado- code, which is not that difficult.
(Everybody does something like that from time to time, believe me :)).
Just as with every hacking exercise, copy the official file into
myrreg.ado and work with myrreg.ado.
--
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/
