Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: Quantile Regression
Steve Samuels <email@example.com>
Re: st: Quantile Regression
Tue, 2 Oct 2012 19:34:30 -0400
Oops. In the first paragaph, read "or 3) -sqreg- rejects replicates..."
> On Oct 2, 2012, at 4:18 PM, Robert Davidson wrote:
> Hello Statalist,
> I estimated a simultaneous quantile regression using sqreg and thought
> I was getting bootstrapped standard errors because it says it is
> bootstrapping. However, I get different standard errors then when I
> estimate the same quantiles separately using bsqreg. Does someone
> know the reason for this? It matters to me because my sample has
> several million observations and employs fixed effects and takes many
> hours to run. I would like to use sqreg because I can set my computer
> to run overnight and have it complete in the morning (or maybe mid
> afternoon) but if I use bsqreg I have to check it every couple of
> hours and re run the code with the new quantile.
Without details (see FAQ 3.3 first sentence), we can only guess. This
could happen if 1) you did not set the same random seed before each
-sqreg- and -bsqreg- command; 2) the number of bootstrap replicates
differed between -sqreg- and -bsqreg- runs; or 3) -sqreg-
rejects replicates in which convergence failed for any quantile.
By the way, the manual states that -sqreg- is faster than -bsqreg-.
> On a related note, is there a way to compute sampling weights for
> quantile regressions? I believe it only accepts frequency weights and
> analytic weights as listed options. This would dramatically reduce
> the run time for some of my programs and not use up all the memory and
> then have them crash.
I'm not sure what this question means. I don't see how sampling weights
would save time in any case, even if the quantile regression commands
did accept them as an option.
You could get away with utilizing analytic weights in place of sampling
weights, as long as the standard errors were correct. Somebody (Jeff
Pitblado?) has said that analytic weights + robust standard errors =
probability weights. But the qreg BS standard errors won't be robust if
the data came from multistage sample surveys. Standard errors for such
data should be based on variation between first-stage clusters (PSUs).
Unfortunately, the -vce- option in the -qreg- commands does not accept
I've never had the luxury of having so many observations to analyze. I
imagine that almost every simple model can be rejected, so that model
building and validation are real challenges.
* For searches and help try: