Re: st: questions about quantile regresion

[Constantine Daskalakis <constantine.daskalakis@mail.tju.edu>]

At 08:13 PM 10/20/02, Shige Song wrote:
> Dear All,
>
> I am trying to using quantile regression to explore age pattern at first 
> marriage. I have two questions when I read through the Stata Manual. First 
> of all, Stata's quantile regression facilities do not handle probability 
> (sampling) weight. "Qreg" only handle "fweight" and "aweight" while 
> "sqreg", "bsqreg", and "iqreg" do not allow weighting of any kind. My 
> sample contains two parts with one part being over-sampled (1:4), so I 
> really need to use some sort of weighting scheme to adjust this, any 
> suggestions?
>
> My second question is about censoring. Based on my understanding, there 
> are two kinds of censoring here. First, not everybody get married 
> eventually. Second, for the very young cohort (e.g. those who are at their 
> 20s at the time of interview),  large proportion of them do not have an 
> age at first marriage because they have not been married (although most of 
> them eventually will). Is it possible to handle this using quantile 
> regression? Can Stata do it?
>
> Thanks in advance!
>
>
> Shige Song
> Department of Sociology, UCLA
>
> *
> *   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/
First, why use -qreg-? Age (time from birth) to first marriage seems to be 
a typical time-to-event problem. So, you are looking to survival analyses 
(Kaplan-Meier, Cox, etc). Look at the -st- set of commands in Stata.

Now, as to the weighting, I'm not sure that the survival commands allow 
weighting either (have never tried it). Certainly, any inferences for the 
original population will be biased if you don't use the weights. But I 
don't see why the COMPARISON across groups (say, men vs. women etc) would 
be a problem without using the weights. For example, in experimental 
design, we often fix the distribution of predictors to something artificial 
(eg, clinical trials fix 1:1 ratio between treatment and control) and we 
don't use any weighting.

Best,
cd



________________________________________________________________

Constantine Daskalakis, ScD
Assistant Professor,
Biostatistics Section, Thomas Jefferson University,
125 S. 9th St. #402, Philadelphia, PA 19107
   Tel: 215-955-5695
   Fax: 215-503-3804
   Email: constantine.daskalakis@mail.tju.edu
   Webpage: http://www.kcc.tju.edu/Science/SharedFacilities/Biostatistics

*
*   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/