Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: RE: control function with probit in first stage


From   "Schaffer, Mark E" <[email protected]>
To   <[email protected]>
Subject   st: RE: RE: control function with probit in first stage
Date   Fri, 13 Apr 2007 22:28:02 +0100

Deepak, 

> -----Original Message-----
> From: [email protected] 
> [mailto:[email protected]] On Behalf Of 
> Deepak Hegde
> Sent: 13 April 2007 19:40
> To: [email protected]
> Subject: st: RE: control function with probit in first stage 
> 
> Wooldridge (2002 p612) has a section on this.  Imbens (p3 of 
> the below linked file) has a short introduction too: 
> http://are.berkeley.edu/courses/ARE213/spring2006/lect18_06apr
> 10.pdf, as does http://www.stata.com/meeting/5nasug/wiv.pdf (p4) 

I'm a bit confused here.  The 2nd and 3rd examples are for the control
function approach applied in the linear model, when including the
residuals from the first stage is an alternative way of obtaining 2SLS
in the second stage coefficients.  But in these cases, the first stage
residuals are from a simple linear regression.  The probit doesn't have
residuals as such, though you can calculate "generalized residuals" -
see Scott Merryman's posting from some years ago (though my notes say
there might be a typo in it):

http://www.stata.com/statalist/archive/2003-12/msg00650.html

I am pretty sure that the inverse Mills ratio approach used in
-treatreg, twostep-  is in fact the correct control function approach.
If you work out the generalized residual for the probit following
Scott's posting, you should get exactly the lambda that -treatreg- uses
in the 2nd part of the -twostep- estimation.

Hope this is useful.

Cheers,
Mark

> The reason I am using the control function approach is 
> because I am explicitly interested in the coefficient/std 
> errors on the residual (from the first-step) in the 
> second-stage regression. 
> 
> Thanks,
> Deepak
> 
> 
> 
> 
> 
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of 
> Schaffer, Mark E
> Sent: Friday, April 13, 2007 10:49 AM
> To: [email protected]
> Subject: st: RE: RE: RE: control function with probit in first stage 
> 
> Deepak,
> 
> > -----Original Message-----
> > From: [email protected]
> > [mailto:[email protected]] On Behalf Of Deepak 
> > Hegde
> > Sent: Friday, April 13, 2007 6:20 PM
> > To: [email protected]
> > Subject: st: RE: RE: control function with probit in first stage
> > 
> > Hi
> > 
> > Mark, Thanks.
> > 
> > I am trying to explicitly use the *residuals* from my first step 
> > (probit model used to predict treatment) as a regressor in 
> my primary 
> > model of interest.  I think -treatreg- uses the *predicted 
> values* of 
> > the treatment variable from probit instead.
> 
> It uses the inverse Mills ratio, if I'm not mistaken.  Also 
> if I'm not mistaken, this is the standard two-step procedure. 
>  Do you have a specific reason not to use this, or perhaps a 
> reference for the procedure you want to follow?
> 
> Cheers,
> Mark
> 
> > 
> > So, ideas on how I can use the first stage residuals in the second 
> > stage will be particularly appreciated..
> > 
> > Many Thanks,
> > Deepak
> > 
> > 
> > -----Original Message-----
> > From: [email protected]
> > [mailto:[email protected]] On Behalf Of 
> Schaffer, 
> > Mark E
> > Sent: Friday, April 13, 2007 1:22 AM
> > To: [email protected]
> > Subject: st: RE: control function with probit in first stage
> > 
> > Deepak,
> > 
> > > -----Original Message-----
> > > From: [email protected]
> > > [mailto:[email protected]] On Behalf Of Deepak 
> > > Hegde
> > > Sent: 13 April 2007 05:36
> > > To: [email protected]
> > > Subject: st: control function with probit in first stage
> > > 
> > > Hi all,
> > > 
> > > I am trying to correct for a model with a potential
> > endogenous binary
> > > explanatory variable by using the control function approach.
> > >  
> > > Since my endogenous variable is binary, my first stage is 
> a probit, 
> > > and the second stage is OLS (with the estimated errors from the 
> > > first-stage as one of the explanatory variables).
> > > 
> > > Can someone please point me to an easy way of 
> implementing this in 
> > > Stata to get correct standard errors and estimates?
> > > Code/Routines will be particularly helpful, but I'll also
> > be thankful
> > > for illustrations, etc.
> > 
> > Why not use Stata's built-in -treatreg-?
> > 
> > --Mark
> > 
> > > Thanks a lot! 
> > > --- Deepak
> > >  
> > 
> > *
> > *   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/
> 

*
*   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/



© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index