Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st:
From
[email protected] (Jeff Pitblado, StataCorp LP)
To
[email protected]
Subject
Re: st:
Date
Mon, 24 Jan 2011 14:23:39 -0600
Christian Gregory <[email protected]> has a couple of questions about
output in the iteration log from -ml- when the -showtolerance- option is
specified:
> I'm using ml to estimate a model.
>
> 1. I'm using <showtolerance> option: the output reports "ptol" and
> "vtol." Do these refer to difference in likelihood or parameter
> estimates? I cannot find reference to them in any of the manuals.
>
> 2. In computing the scaled gradient (g()*H*g()'), how does Stata compute
> the Hessian? I'm using an lf1 type estimator: does it compute the
> Hessian numerically from the gradients that I have written? Or does it
> compute them numerically from numeric gradients that it calculates?
The 'ptol' identifier refers to the relative difference compared to the
coefficient tolerance. The coefficient tolerance is specifed via -ml-'s
-tolerance()- option.
The 'vtol' identifier refers to the relative difference compared to the value
tolerance (also known as the log-likelihood tolerance for -ml-). The
log-likelihood tolerance is secified via -ml-'s -ltolerance()- option.
For an -lf1- evaluator, -ml- computes the Hessian matrix by taking numeric
derivatives of the gradient resulting from the scores returned by the -lf1-
evaluator.
--Jeff
[email protected]
*
* 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/