st: RE: -probit- does not converge, but does -dprobit-

[Nick Cox <n.j.cox@durham.ac.uk>]

The ML engine was changed in Stata 11. A side-effect is that some models that converged before do not converge now. There are various things that you can do. 

1. Reconsider your model. Perhaps it's good news: Stata has found out a fragile model. 

2. Reconsider precisely how it is fitted, e.g. by tuning the optimisation options. 

3. Send the output and the dataset to tech support. 

4. Provide more details for people to comment. 

Nick 
n.j.cox@durham.ac.uk 

Duha Altindag

I am using version 11.1. I have two data sets: old.dta and new.dta.
They are almost identical, but new.dta has more observations. That is,
old.dta is a subset of new.dta.

Sometime ago, with an earlier version of Stata I had used old.dta and
run -probit- and -dprobit-. That time, I could get marginal effects
using both -probit- and -dprobit-.

Now, I use new.dta and run -probit- : it doesn't converge.
Then I use run -dprobit- : it does converge and gives me marginal effects.

Then I experiment with old.dta and Stata 11.1. I get the same thing:
-probit- does not converge, but -dprobit- does.

Previously, someone complained about a similar issue and he was
advised to contact Stata Support.
http://www.stata.com/statalist/archive/2010-08/msg01331.html


Is there an update on the issue? Would you advise me to contact the
Support as well? Or does someone have a quick advice?

I know -dprobit- is not an official command anymore. But does it mean
its results cannot be trusted? Should I trust -probit- or -dprobit-?

*
*   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/