I think it follows from basic theory of
order statistics that the further you
go out in the tails, the more wobbly things
are. And that's probably intuitive in any
case. The fact that you are dealing with
conditional quantiles wouldn't seem to
alter that.
If you're in search of some rule that things
will be OK if sample size is so much, that would
seem a poor way to summarize, given a more likely
continuous dependence of error on sample size.
Better to get an idea of quite
how unreliable your results are by bootstrapping?
Nick
n.j.cox@durham.ac.uk
Jeannette Wicks-Lim
> I am trying to estimate something like the following:
> qreg lnwage pwl labcon pwllabcon
>
> where pwl=1 when prevailing wage law is in effect and 0 when it is not
> labcon=1 if worker is a construction laborer, 0 if not
> pwllabcon = pwl*labcon, so that it equals 1 if pwl is in
> effect and workers
> is construction laborer, 0 otherwise
> sample is restricted to laborers (across industries)
>
> because I am looking at one state, I have only 30
> observations on workers
> for whom pwlabcon=1. I assume that this sample size is too
> small to estimate
> anything but the median regression; but what guidelines are
> there for the
> necessary cell size if you want to estimate qregs for the
> 10th or 90th
> quantile?
*
* 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/