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: not concave log pseudolik in gnbreg
From
"JVerkuilen (Gmail)" <[email protected]>
To
[email protected]
Subject
Re: st: not concave log pseudolik in gnbreg
Date
Thu, 10 Jan 2013 19:07:55 -0500
On Thu, Jan 10, 2013 at 5:20 PM, [email protected]
<[email protected]> wrote:
> I am running a model in -gnbreg-
>
> The log takes many iterations
Never a good sign.
and the final part reads as follows:
>
> Iteration 1320:log pseudolikelihood = -399.5054
> Iteration 1321:log pseudolikelihood = -398.9085
> Iteration 1322:log pseudolikelihood = -398.79423
> Iteration 1323:log pseudolikelihood = -398.77882
> Iteration 1324:log pseudolikelihood = -398.74555 (not concave)
> Iteration 1325:log pseudolikelihood = -398.74486
> Iteration 1326:log pseudolikelihood = -398.74102
> Iteration 1327:log pseudolikelihood = -398.74077 (not concave)
> Iteration 1328:log pseudolikelihood = -398.74077
>
Not 100% sure but I'd consider switching to -nbreg- first just to see
if the model fits that way without complicating the overdispersion
parameter, which I suspect is the culprit. Also output more from the
last iterations, such as the final Hessian or the covariance matrix of
the parameters (estat vce). That may indicate if the problem is
isolated to one parameter.
And try putting difficult on and force it to iterate longer, say to
1500 iterations.
I've found that the NB model is a bit fragile. It's often dominated by
one or two outliers or some other misspecification issue, so you might
want to look for that.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/