margherita Comola <[email protected]> :
You could:
1. Ignore endog of educ and treat selection into formal work vs. other
categories as a probit, so use -heckman-
2. Ignore selection, find an excluded instrument for educ, e.g. some
policy change that induces some people to have more educ than
otherwise identical people, and use -ivregress-
3. Use -cmp- (from SSC) to model the various equations as a system.
The second option is probably best IMHO, in the sense of giving
consistent estimates under various error distributions, but it
requires a very large dataset and good excluded instruments for
reasonable inference.
On Fri, Sep 11, 2009 at 8:01 AM, margherita Comola
<[email protected]> wrote:
> Dear Statalisters,
>
> I need to estimate a selection model with the following features:
>
> -the selection equation is a multinomial logit (3 alternatives). As
> usual, the outcome variable is available for one of the alternatives
> only.
> -one of the regressors is endogenous both to the selection AND to the
> outcome equation.
>
> In particular, the setting is the following: I am estimating a wage
> equation in a setting where wage is available for formal workers only
> (not for informal or non-participants). The selection equation is
> multinomial: non-participants, formal workers, informal workers.
> Education (that can be coded as a dummy or as a continuous variable)
> is endogenous to both equations.
>
> Any idea?
*
* 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/
