# Re: st: Re: Re: selection bias with bivariate probit

 From Partha Deb To statalist@hsphsun2.harvard.edu Subject Re: st: Re: Re: selection bias with bivariate probit Date Wed, 20 May 2009 21:09:59 -0400

Unless I've understood the specification incorrectly, there are at least 2 ways to do it in Stata.

1. Ignore the binary nature of both outcome and treatment and use linear IV (-ivregress-, -ivreg2-).

2. Assume bivariate normality (which -cmp- does as well) and use the bivariate probit model (-biprobit- ).

cheers,

Partha

Martin Weiss wrote:
<>

Sam privately asked for more details on my post. I think that the user-written package -cmp- could be of use to him, so I recommend he install it by typing -ssc install cmp- within Stata (while being connected to the internet) and subsequently take a look at the help file by typing - help cmp- . The file is quite comprehensive, with clickable examples for most constellations of data -cmp- is designed to handle. I hope that Sam succeeds and that other listers can give additional guidance to him...

HTH
Martin
_______________________
----- Original Message ----- From: "Martin Weiss" <martin.weiss1@gmx.de>
To: <statalist@hsphsun2.harvard.edu>
Sent: Wednesday, May 20, 2009 11:18 PM
Subject: st: Re: selection bias with bivariate probit

<>

Try -ssc d cmp-

HTH
Martin
_______________________
----- Original Message ----- From: "Sam Lee" <sunghanl@uic.edu>
To: <statalist@hsphsun2.harvard.edu>
Sent: Wednesday, May 20, 2009 11:15 PM
Subject: st: selection bias with bivariate probit

Hello all,

I'm looking for a procedure to control selection bias, but a bit special case that I don't know whether STATA supports this. Outcome regression: dichotomous dependent variable (high earnings or not) y = beta * x + gamma * z (college or not) Selection regression: dichotomous choice dependent variable (college degree or not) z = alpha * w

Classic Heckman procedure deals with the situation where a selection model with a dummy dependent variable (work or not) and an outcome model with a continuous dependent variable (wage) with truncated (only observed if chosen). My challenges to use a Heckman procedure are two folds: (1) a dummy variable (high wage or not instead of continuous wage) in an outcome model (2) outcome observations not truncated (we observe earnings for both college degree and non-college degree) - so this is more of a treatment model instead of a selection model. STATA has "heckprob" dealing with the first problem and "treatreg" dealing with the second problem. But so far I couldn't find any stata function or ado file dealing with both extensions.

As far as I know, we can include both mills ratio of selected and not-selected in the outcome model from the selection model. Then I'll have a consistent coefficient estimate for gamma. But I don't know how to correct std dev of gamma in the outcome probit model (I guess since it is difficult to deal with covariance matrix, I prefer a standard procedure supporting this correction).

Your help will be greatly appreciated.

Sincerely,

Sam Lee

Sam Lee

Department of Accounting

University of Illinois (RM# 2302)
601 S. Morgan St, Chicago Il 60607

Ph:312.413.2131 Fax:312.996.4520

*
*   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/

*
*   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/

*
*   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/

--
Partha Deb
Professor of Economics
Hunter College
ph:  (212) 772-5435
fax: (212) 772-5398
http://urban.hunter.cuny.edu/~deb/

Emancipate yourselves from mental slavery
None but ourselves can free our minds.
- Bob Marley

*
*   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/