Re: st: How to combine clad with clusters and weights?

[Nick Cox <njcoxstata@gmail.com>]

-clad- is from STB-58 (2000):

STB-58  sg153 . . . . . . . Censored least absolute deviations estimator: CLAD
        (help clad if installed)  D. Jolliffe, B. Krushelnytskyy, & A. Semykina
        11/00   pp.13--16; STB Reprints Vol 10, pp.240--244
        censored least absolute value deviations (CLAD) estimator
        with bootstrap estimates of its sampling variance (a
        generalization of qreg that is robust to heteroscedasticity)

Please remember that you are asked to state _where_ user-written
programs you refer to come from.

With -clad- as with almost every other program what the help says is
the full story. There is no support for any kind of weights, etc. If
it's not in the help it's not supported.

I can't provide more positive help. It's my impression that lots of
what is regarded as standard apparatus that is tied in a bundle with a
variance-covariance-least squares flavour does not carry over easily,
if at all, to anything based on a different norm, least absolute
deviations. Also, -clad- is -- as a program published in 2000 and not
updated (e.g.) to match Stata's -svy- support -- in effect a dead end.

If -qreg- or one of its relatives doesn't provide what you seek you
are faced with a difficult programming problem.

Indeed, is there literature talking about what you seek or are you
supposing that it should exist by analogy with other modelling
approaches?

Nick

On Tue, Feb 12, 2013 at 12:00 PM, Jossi Steen-Knudsen <jsk@rff.dk> wrote:

> Does anyone know whether the user-written command for censored quantile regressions -clad- can be used in combination with sampling weights (pweights)?
>
> Can I somehow just multiply the weights if no option is available?
>
> Also, the data are in clusters, so is there any way to tell -clad- that standard errors should take account of clustering?
>
> For other regressions than -clad- I use the svyset-setup, which in my case is defined as:
>
> svyset vid [pweight=weightlong], fpc(one) || _n, strata(strataID) fpc(ssrlong) singleunit(scaled)
>
> In other words, it is the above framework I somehow would like to implement in a censored quantile regression.
>
> Any help on weights or clusters would be highly appreciated.
*
*   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/