Re: st: heckman selection model with endogenous covariates

[Austin Nichols <austinnichols@gmail.com>]

James <James.Rao@agr.uni-goettingen.de> :
The selection model estimated via MLE (or a hurdle model using
equations like you have given below; since you don't specify errors or
their distributions, or true functional forms, the equations do not
actually dictate an estimator) requires very strong distributional
assumptions; if you are willing instead to assume that
  E(labor|X) = exp(Xb)v
with v a mean one error that may also be zero (e.g. gamma distributed
or some mixture), then you have a GLM model that looks like -poisson-
and can be estimated with that command.  I.e. it looks like
 ln(labor) = Xb + e
when labor>0 but allows labor=0 as well. If you assume that a column
of X is endogenous, but you have an instrument, you can use a GMM
version of that model; see the help file for -ivpois- on SSC or the
new -gmm- command in Stata 11.
http://fmwww.bc.edu/repec/bocode/i/ivpois.html
http://www.stata.com/help.cgi?gmm

On Thu, Apr 8, 2010 at 5:13 AM, Rao, James
<James.Rao@agr.uni-goettingen.de> wrote:
> Dear users,
>
> Am interested in estimating a lbor demand equation as follows:
>
> ln_labor=alpha + b1_X + b2_treatment
>
> Since some respondents do not use hired labor, i specify the probability of
> using hired labor as:
>
> prob(labor)=b3_Z + b4_treatment
>
> and then estimate the two equations jointly following heckman selection
> approach.
>
> My concern is that the treatment variable in  both equations is potentially
> endogenous. Anyone with an idea of how I can address this potential
> endogeneity within the framework of heckman selection models?
>

*
*   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/