Frank Eaton <FEaton@NFOCFgroup.com> asks about standard errors after quantile
regression with -qreg-,
> My attempts to run qreg yield plausible parameter estimates, but the
> SEs are all very small, such as 3.92e-14. Does anyone have an idea
> what might cause this problem? Is it something in my data?
My best guess is yes, it is Frank's data.
Things I might check if I were Frank:
1) Does my dependent variable have much range, or does it only take on a
few values?
2) Am I asking for a quantile in the tail when there is insufficient
data to characterize that tail?
Either of these can cause problems with the estimation of the density of true
residuals because they essentially violate the assumption of continuity.
Frank might also consider using -bsqreg- because it makes fewer assumptions in
estimating the standard errors. Though if (1) is true, I would still be
suspicious of any results because the location of the quantile is being
determined almost exclusively by the estimates and they are working against a
dependent variable with few observed quantiles.
With the limited information at hand, these are just wild conjectures.
Frank, if this isn't helpful and you cannot see anything in your data, I
suggest you contact Stata technical support <tech-support@stata.com> and
provide as much information as you can. They will probably need your data,
because I suspect that is where the problem lies.
In response to the question about SEs produced by -qreg-, they are based on an
asymptotic sandwich/Huber type estimator suggested by Koennker and Basset
(1982) and operationalized by Rogers (1993). As noted above, bootstrapped
standard errors can also be obtained. All of the references can be found in
[R] qreg.
-- Vince
vwiggins@stata.com
*
* 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/