Bookmark and Share

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]

st: RE: "Alternate Solutions Exist" in Quantile Regressions (qreg)


From   "Nick Cox" <n.j.cox@durham.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: "Alternate Solutions Exist" in Quantile Regressions (qreg)
Date   Tue, 16 Mar 2010 19:20:39 -0000

There's a much simpler option: try a much simpler model. It seems that
your model has about 20 predictors and that may be too complicated for
your data to bear. 

In terms of your question: One extra predictor which is Gaussian noise
may be enough. How much noise is difficult for anyone not in sight of
your data to guess at but starting small and working up would seem
easiest. Clearly the numbers to compare with the values of your -finw-. 

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

Jen Zhen

when running
- qreg finw  ry5-ry13  d1995-d2003 d2005-d2007 ,  wlsiter(50) -
we get the note that "alternate solutions exist" in between iterations.

In an earlier list post
(http://www.stata.com/statalist/archive/2008-10/msg00519.html), it was
suggested to check whether the different alternate solutions differ
significantly, but Stata by default seems to give us only one. Do you
know a way to view the others, or other approaches to tackle this
issue?
Another suggestion made in that thread was to perturb the data and see
which of the answers remains, but we're not sure how to exactly
implement this perturbation. Should we just take all variables
currently used in the regression, add to them some small random error,
and then rerun the regression with those new variables? How large or
small should the error sensibly be?

*
*   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   |   Site index