Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: Implementing Censored Quantile Regression


From   "Nick Cox" <n.j.cox@durham.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: Implementing Censored Quantile Regression
Date   Mon, 11 Aug 2008 17:05:43 +0100

Your timings are a consequence of your data, your model(s) and a program
-cladtemp- which as far as I can tell is not publicly accessible.

In these circumstances only very general advice seems to be possible. 

1. You need to look inside -cladtemp- to see what might be speeded up. 

2. You might try fitting much simpler models with just a few predictors
to get a sense of how far model complexity is driving times to
convergence. In fields close to me many people tend to throw in far too
many predictors because they want the imprimatur of "objectivity" for
their models. 

3. You might try using a subset of the data to get a sense of how
dataset size is crucial. 

On the basis of a glance at the original -clad- my impression is that
this stuff is getting to take a lot of time, regardless. That doesn't
rule out big speed-ups, but I don't know how people can advise
specifically about a program they can't even see. 

Nick 
n.j.cox@durham.ac.uk 

Sachin Chintawar

Thanks a lot for helping me out on the question of estimation of the
Censored quantile regression. A few notes from my experience and
interaction
with Dr. Wilhelm.
1. As I had earlier noted data failed to converge using the -CLAD
procedure.
I used the -CLADTEMP procedure that was a modified version of -CLAD used
by
Dr. Wilhelm. This handles the problem of convergence much better.

2. I have one question. While the -CLADTEMP procedure does handle the
convergence problem (due to the way convergence criteria is defined)
both
these procedures take a long time. The -CLAD procedure with reps(200)
took
around 3 hours to give me the result (suggesting that the convergence
was
not met) the -CLADTEMP took around 8 hours to give me the bootstrapped
results (same number of reps)
The data that I had used had 12360 observations and 19 variables.

I was wondering if there was a more efficient (in the sense of saving
time)
way to run censored regressions.
Scott Merryman

You might try -clad- published in Stata Technical Bulletin by  Dean
Jolliffe, Bohdan Krushelnytskyy and Anastassia Semykina
(http://www.stata.com/products/stb/journals/stb58.pdf).

You might also find the program in the link below useful:

http://www.stata.com/statalist/archive/2006-09/msg00379.html

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index