Re: st: Missing confidence intervals for median after using -bootstrap- or -bpmedian-

["JVerkuilen (Gmail)" <jvverkuilen@gmail.com>]

On Mon, Nov 12, 2012 at 7:49 PM, Nick Cox <njcoxstata@gmail.com> wrote:
> I am not a statistician; in fact many, perhaps most, people on this
> list wouldn't call themselves statisticians.

I only call myself a statistician outside the company of
statisticians! ;) That's because it's much easier not to have to go
through a lot of rigamarole of explanation of what I do. Now that
statistics is trendy, I don't get the "ohhh, I hated that class! I'm
terrible at statistics!" like I did for many years. It is mildly
pleasant.



It seems likely therefore that the
> confidence interval for anything will be narrow at worst. However, it
> seems likely also from your results that you have lots of ties.

The ties issue is my guess. -tabulate- the variable. If it's highly
concentrated on only a few values nonparametric estimates of the
median's sampling distribution may simply become degenerate. The dirty
little secret that many users of non-parametric statistics like to
avoid is that continuity of response is pretty crucial for classical
nonparametrics to have reasonable performance. Quantile regression, in
particular, hates ties.

There have been some attempts at "ordinalizing" methods like quantile
regression using jittering and other tie-breaking techniques. An
example would be TORQUE by Grace Hong for ordinal regression (I can't
find a clean cite to this work or I'd provide one). They aren't in
Stata yet to my knowledge, though that may be changing.
*
*   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/