First, the answer to your original question, as far as I'm aware, is no. It
would be nice if Stata had many more sample selection models available, but
alas only the standard Heckman and the basic bivariate probit with sample
selection have been coded. (I'm actually trying to modify heckprob to
handle an ordered probit as the regression outcome, but working out the ml
portion of the code is a bit of a struggle.)
Second, I don't think we can judge whether you have a correctly specified
model from your description, so it seems fair to assume that you have it
right. Something worth answering, however, is whether or not the error
terms in the two probits are actually correlated. If not, then you've got
two independent selection equations rather than a bivariate probit. In any
case, you could try to estimate the model in a manner analogous to a classic
Heckman model where the truncated bivariate normal distribution is used to
estimate the hazard term included in the second stage. The standard errors
would also have to be corrected, of course.
Note that LIMDEP can estimate this model.
-----Original Message-----
From: [email protected]
[mailto:[email protected]]On Behalf Of
[email protected]
Sent: Thursday, September 26, 2002 1:06 PM
To: [email protected]
Cc: [email protected]
Subject: RE: st: sample selection with bivariate probit
Erik, thanks for you reply. Here the parameters of interest are the
parameters of the earnings equation. The context is international migration.
z=1 refers to migrants, z=0 refers to non-migrants. What I need is sample
selectivity corrected returns to characteristics in the lnw equation. In the
data I observe lnw for a given individual either for z=1 or z=0, but not for
both (previous message is misleading - sorry ). I.e. for those who migrate I
only observe lnw|z=1 and for non-migrants I only observe lnw|z=0. I want to
predict the earnings of a migrant had that individual not migrated ( and
vice versa for non-migrants). Given returns to characteristics of the same
individual can be different in the host country and in the source country I
want to estimate the model separately for z=1 and z=0, and use the estimated
coefficients for my prediction.
>-----Original Message-----
>From: Erik �. S�rensen [mailto:[email protected]]
>Sent: Thursday, September 26, 2002 1:32 PM
>To: [email protected]
>Subject: Re: st: sample selection with bivariate probit
>
>
>On torsdag, sep 26, 2002, at 13:07 America/Montreal,
>[email protected] wrote:
>
>> Dear Users,
>>
>> I am estimating a log earnings equation controlling for
>selection. The
>> selection involves two stages that leads to a bivariate probit with
>> partial
>> observability.
>>
>> lnw=x3b3+e3 observed for z=1 and z=0
>>
>> where z=y1 x y2; z=1 iff y1=y2=1; z=0 if y1=0 or y2=0
>
>I think you need to explain more carefully what is observed, what is
>not observed and what is the parameters of interest. If you get to
>observe lnw=x3b3+e3 no matter what is the value of z, why do you want
>to estimate it seperately for the values of z? Why would this make a
>difference?
>
>Erik
>--
>Erik �. S�rensen, <http://www.geocities.com/erik_oiolf/>
>
*
* 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/
*
* 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/
*
* 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/
