Re: st: MLE of Heckman model with endogeneity

[Nina Parfinenko <nparfin@gmail.com>]

Hi Stas,

Thank you for your answer. This is a snippet of my data:

id employed(Y2) logwage(Y1) experience(X,Z) part-time-work(X)
non-labor-income(Z)
100002 Employed 259.1516 21 1 10.38887
100002 Employed 222.4555 21.91667 1 10.7015
100002 Employed 252.914 22.88889 1 11.11234
100002 Employed 274.1406 23.80556 1 10.79897
100004 Employed 241.0749 11 1 10.58269
100004 Employed 222.7311 12.02778 1 10.95263
100004 Employed 382.0812 12.58333 1 10.72274
100004 Employed 244.6271 13.55556 1 11.40245
100005 Employed 173.5379 0.9444444 1 8.768297
100005 Employed 251.9188 2.083333 1 9.007585
100005 Employed 229.8504 3.027778 1 0
100005 Not employed . 5.000381 0 10.37486
100005 Not employed . 5.000381 0 9.950747
100005 Not employed . 5.000381 0 10.39137
100005 Not employed . 5.000381 0 10.3661
100005 Not employed . 5.000381 0 10.11485

This is panel data of females, with 25 variables in both X & Z, 5 only
in X and 9 only in Z. Out of 39 variables, 31 are dummies. Variables
include age, experience, tenure, marital status, non-labor income,
education, children, enlish-speaking background, occupation, health
and ethnicity.

Thanks for the pointer to -cmp-, it seems to cover both -heckman- and
-ivprobit-, so maybe it can be used for their hybrid. On the other
hand, I suspect my ML function could be non-concave.
Best,
Nina

On Wed, Feb 12, 2014 at 3:29 AM, Stas Kolenikov <skolenik@gmail.com> wrote:
> Ninochka,
>
> can you give a snippet of your data, like the first ten observations
> with a single x, single z, and all the relevant patterns of what is
> observed?
>
> I think David Roodman's -cmp- should be able to handle it, see
> http://www.stata-journal.com/article.html?article=st0224.
>
>
> -- Stas Kolenikov, PhD, PStat (ASA, SSC)
> -- Principal Survey Scientist, Abt SRBI
> -- Opinions stated in this email are mine only, and do not reflect the
> position of my employer
> -- http://stas.kolenikov.name
>
>
>
> On Tue, Feb 11, 2014 at 8:17 AM, Nina Parfinenko <nparfin@gmail.com> wrote:
>> Dear Statalist,
>>
>> I need to estimate a two-equation sample selection model for wage (y1)
>> and hours (y2). y2 is not observed, only whether y2>0 or y2=0. Wage y1
>> is observed only if y2>0. Also y1 is endogenous in the y2 equation.
>> More precisely, the two equations are
>>
>> y1=betaX+e1
>> y2*=alpha*y1+gammaZ+e2
>>
>> and (e1, e2) are jointly normal.
>>
>> This model could be estimated with ivprobit if y1 were always
>> observed. On the other hand, it would be a candidate for heckman
>> command if not for the endogeneity of y1 in the y2 equation. I seems
>> that Stata ml commands do not take endogenous variables either.
>>
>> Is it possible to estimate the above model with maximum likelihood
>> (the 2 equations simultaneously) in Stata using built-in commands? I
>> have already estimated the system with a two-step procedure and now
>> need to estimate it jointly and simulatenously with Maximum
>> Likelihood.
>> Many thanks,
>> Nina
>> *
>> *   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/