Statalist The Stata Listserver

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

Re: st: Problem with ML estimation

Subject   Re: st: Problem with ML estimation
Date   Fri, 18 Aug 2006 15:37:30 -0400

Thanks for your comments. You are right that I might be encountering machine precison problems. Probably, I will have to change the likelihood function to ensure that the maximand is a finite number.


----- Original Message -----
From: Stas Kolenikov <>
Date: Friday, August 18, 2006 2:44 pm
Subject: Re: st: Problem with ML estimation

> Oops, that was right, you had four dependent variables in your -ml-
> statement, that's what I was thinking about. I still believe that you
> have identification problems with your program and your data. As you
> most likely know, +\infty for Stata is a missing value, and if the
> maximum is achieved at infinity, then the likelihood is flat in the
> neighborhood of it, and that is what Stata complains about: it cannot
> compute the derivative with adequate precision, as the subtraction of
> the numerator and/or denominator in the numerical derivative is
> machine zero. So even though the model might be technically
> identified, it is empirically underidentified (i.e., for a given
> combination of parameters and the data), and it is not identified in
> computer arithmetics precision.
> On 8/18/06, DEEPANKAR BASU <> wrote:
> > I am not sure I understand your comment. I am trying to estimate 
> two parameters
> > (both constants) and not four. These constants are those 
> contained in `theta3' and
> > `theta5'. Since I have three distinct observations, the model 
> should work. Or am
> > I missing something?
> >
> > I can ofcourse analytically see that the result of the estimation 
> should give me
> > `theta3' as +\infty (or a very large number). Would it be a 
> problem for STATA if the
> > result is +\infty?
> -- 
> Stas Kolenikov
> *
> *   For searches and help try:
> *
> *
> *

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index