Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Re: hessian condition number


From   "Clive Nicholas" <Clive.Nicholas@newcastle.ac.uk>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Re: hessian condition number
Date   Tue, 30 Mar 2004 03:43:29 +0100 (BST)

Marcello Pagano wrote:

> The log is a measure of how many digits of accuracy are lost in the
> calculations with floating point arithmetic.  Been around a long time.

Kit Baum wrote:

>> The condition number of a p.d. matrix is related to the ratio of its
>> largest to smallest eigenvalues. They are the roots of the
>> characteristic polynomial of the matrix. "findit condition number". I
>> have never heard of taking the log10 of the condition number;
>> well-conditioned regressor matrices are usually condition<20 (with
>> condition being the positive sqrt of the ratio). See Belsley et al. as
>> referenced under [r] regression diagnostics.

Thanks for your help and contributions. As I had posted previously, Dow
and Endersby's recent article in Electoral Studies, in which they compare
the merits of multinomial logit and multinomial probit (at least as far as
political science is concerned), proposed reporting the log10 of the
hessian as a standard diagnostic to either model. I'm saying neither that
they are right nor that they are wrong: I simply wanted to see how you did
it in Stata. Now I know (I think).

Still, if you're a Stata fundamentalist and you don't have access to
LIMDEP, you won't worry too much as to the merits of MNL and MNP, since -
for now - you can only fit MNL models. Unless somebody at College Station
has future plans afoot...

CLIVE NICHOLAS        |t: 0(44)191 222 5969
Politics              |e: clive.nicholas@ncl.ac.uk
Newcastle University  |http://www.ncl.ac.uk/geps
*
*   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/



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