# RE: st: IV and sample selection

 From "Ngo,PT (pgr)" To Subject RE: st: IV and sample selection Date Fri, 18 Apr 2003 12:16:44 +0100

```Dear Erdal,

Is it that your dependent variable Y is censored, so that your model would be a censored regression model with an endogenous explanatory variable S that is fully observed ?

Wooldridge, in his new book (the econometric analysis of cross-section and panel data) addresses this in section 16.6.2 (p. 530-533) on censored tobit model with an endogenous explanatory variable.  He refers to the two-stage Smith-Blundell procedure in order to test for the endogeneity of S in censored regression model and to obtain average partial effects.

Also, there is a command in Stata which you may want to check: ivtobit.

I hope I have not misundertood your problem,

Thi Minh Ngo
PhD candidate
London School of Economics
DESTIN

------------------------------

Date: Thu, 17 Apr 2003 17:02:28 -0400
From: "Erdal Tekin" <prcete@langate.gsu.edu>
Subject: st: IV and sample selection

Hi,
I am trying to estimate two equations specified as the following:

S = Xa + Zb +e1
Y = cS + Xd + e2

S is an endogenous variable.  Normally, these two equations can be estimated using the ivreg command and identification is achieved by the exclusions restrictions (Z's).

Here is the problem,  S, Z, and X are  observed for everybody in the sample, but Y is observed for only if P=1 and P is another endogenous variable.  Thus, I need to estimate an equation for P as well.

Is it possible to estimate these three equations together in STATA which will take care of both the selection and the endogeneity issues?

How about I do a selection correction for Y equation first and construct a predicted Y for eveybody in the sample, then estimate S and Y equations using the ivreg?  Would this work fine?