Re: st: selection equation Heckman

[Jeffrey Wooldridge <jmwooldridge60@gmail.com>]

question so I'll give you my reasoning. In the vast majority of
applications of Heckman's method the outcome equation is of interest,
with the selection equation serving to model the selection process. I
view the selection equation as essentially a reduced form, much as the
reduced form for linear instrumental variables estimation. If you put
extra variables in the outcome equation that are not also in the
selection equation then you are imposing restrictions on the reduced
form. For robustness reasons this is rarely down in IV estimation and
I discourage it in the Heckman model for the same reason. Now, if you
are sure the extra variables in the outcome equation do not belong in
the selection equation then there is no problem. But how can we ever
be sure of that? If imposing the exclusion restrictions in the
selection equation changes the estimates by a lot -- more precisely,
in a statistically significant way -- then there is little
justification for not including all exogenous variables in the
selection equation.

Of course one can test the exclusion restrictions in the selection
equation, but one may still make the wrong decision -- as always
happens in pre-testing situations.

I hope this helps.

Cheers, Jeff W.

On Mon, Oct 15, 2012 at 9:43 AM, Nick Cox <njcoxstata@gmail.com> wrote:
> I don't think that thread is relevant, unfortunately.
>
> Nick
>
> On Mon, Oct 15, 2012 at 2:39 PM, Lars Folkestad
> <lfolkestad@health.sdu.dk> wrote:
>> This probably does not help and your probably asking about something else but there was The longest discussion about heckman in the beginning of The month. Starting with:
>>
>> http://www.stata.com/statalist/archive/2012-10/msg00187.html
>>
>> Mvh
>> Lars Folkestad
>>
>>
>> Den 15/10/2012 kl. 15.31 skrev "Jan Wynen" <Jan.Wynen@kuleuven.be>:
>>
>>> Dear all,
>>>
>>> I am currently trying to estimate a Heckman model, whereby I have a selection equation which includes socio- demographic variables and a unique selection variable. The outcome equation also includes these socio- demographic variables (minus the selection variable) but also includes extra variables which are only visible if the selection equation equals one. According to Wooldridge (Econometric Analysis of Cross Section and Panel Data, 2nd ed. Cambridge: MIT Press.2010, Chapter 19, pp.803-806) the variables in the outcome equation should be a strict- subset of the ones in the selection equation. However, according to Verbeek (A guide to modern econometrics,  1st ed. Wiley.2000, Chapter 7, pp. 243) the inclusion of extra variables in the outcome equation is no problem, if they have a zero coefficient in the selection equation.
>>>
>>> Is estimating a model as indicated above impossible using the Heckman procedure (will it lead to a problem with endogeneity)? Or is the inclusion of extra variables (not appearing in the selection equation) in the outcome equation OK, and if so does anybody have more references to literature saying this practice is OK? I have found an earlier post on this, however the references were wrong.
>>>
>>> I have already posted this question two weeks ago, my apologies if this is regarded as being impatient.
>>>
>>> Best and many thanks,
>>>
>>> Jan
>>>
>>>
>>>
>>>
>>>
>>> *
>>> *   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/
>>
>> *
>> *   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/
>
> *
> *   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/

*
*   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/