Re: st: moptimize gf evaluator

[jpitblado@stata.com (Jeff Pitblado, StataCorp LP)]

Klaus Pforr <kpforr@googlemail.com> had a follow-up question related to
-moptimize()- and the -gf- evaluator types:

> Thank you for extensive and helpful answer. I'm still waiting for my 
> copy of the new edition of "ML with stata".
> 
> And it is correct, that by adding the derivatives of the panel-level log 
> likelihood with respect to the beta coefficients, I enable my evaluator 
> to correctly handle the robust, cluster and svy options, isn't it? I 
> understand the 2006 edition of "ML with stata" and the help files, that 
> I need the observation level scores for that.
> 
> I want to implement a modified conditional logit estimator, and I want 
> it to handle the cluster option (robust and weights are not that 
> important, but they seem to processable, once you specify the scores, 
> which you need for the cluster option anyway). Therefore it seems 
> necessary to implement the evaluator as a gf2 type and not as d2 type 
> evaluator. I need the gradient and the hessian to gain speed and 
> precision, as the likelihood has a computationally complex denominator.

Yes, the -gf- evaluator types allow for robust, cluster, and -svy- options.

In my previous post, change the line

	moptimize_result_display(M)

to

	moptimize_result_display(M, "robust")

to get the robust standard errors for the fitted model.

Panels must be nested within clusters for -moptimize()- to compute clustered
variances.


There is currently a bug in how -moptimize()- performs the conformability
check of the scores with the data/groups.  This bug causes
-moptimize_result_display(M, "robust")- to exit with a conformability error
before it can report the coefficient table.

We'll have this fixed in a future ado-file update; however, in the mean time,
just change the call

	moptimize_init_by(M, panel)

to

	moptimize_init_by(M, "i")

to get -moptimize_result_display()- to report the robust standard errors in
the coefficient table.

--Jeff
jpitblado@stata.com
*
*   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/