Dear Scott,
Thanks for pointing the error out. I had used it in a simple regression
framework. As the error term in a count model does not assume multivariate
normality, we cannot use it for a count model. Dear David: sorry for the
confusion.
Thanks,
Prabhu
----- Original Message -----
From: "Scott Merryman" <[email protected]>
To: <[email protected]>
Sent: Friday, April 25, 2003 8:54 AM
Subject: st: Re: Re: heckman and negative binomial?
> ----- Original Message -----
> From: "Vimalanand S. Prabhu" <[email protected]>
> > Dear David,
> >
> > Yes, you can use the heckman two-step framework
> >
> > Create a dummy variable for those people who are sent for treatment. Use
> the
> > heckman command. Calculate the Inverse Mills Ratio, and use it as an
> > independent variable in the final model - which can be nbreg .....
> >
>
> I believe this two-stage approach will only work if the second stage is a
> linear regression model.
>
> One possible alternative is to use -gllamm- According to the Sophia
> Rabe-Hesketh's notes from the 2nd Dutch & German Users' Group meeting (
> http://www.stata.com/support/meeting/2dutch/abstracts.html ) the Heckman
> selection model can be estimated in -gllamm- and gllamm is able to
estimate
> poisson models with: family(poisson) link(log).
>
> Scott
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
