Re: st: problems convering maximum likelihood in Stata 11

[Maarten buis <maartenbuis@yahoo.co.uk>]

--- On Thu, 11/3/10, Jon Schwabish <jschwabi@yahoo.com> wrote:
> In running a simple logit model
> (albeit on a very big data set) in Stata 11, my likelihood
> function won't converge; in running the same model on the
> same data set in Stata 10, the function converges in about 5
> iterations. 
> 
> Is there something inherently different in Stata 11 about
> the likelihood process? I have tried adding all sorts of
> options in Stata 11 (nrtolerance, difficult, etc) all to
> little effect.

Are you using analytical second derivatives (i.e. using the
d2 method)? What changed between Stata 10 and 11 is that in
11 Stata by default assumes that your evaluator program 
returns the Hessian matrix rather than the negative Hessian
matrix in Stata < 11. Adding the -negh- option should 
solve that problem for Stata 11. If you want your program 
to work in Stata 10 and 11, you should instead add the 
-version 10- command at the top of the program that calls
-ml- (i.e. not the evaluator program, but the "main"
program.)  

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------


      

*
*   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/