Nichols, thank you for your suggestions. Following your tips, I've read the
chapter of Wooldridge (2002). However, I have some questions about the
procedure.
First, I have more than one policy variables which are suspected to be
endogenous. Let's say that I have three policy variables P1, P2 and P3. And,
these policy variables are not continuous variables but dummy variables.
Then, is it just ok to conduct the procedure to test the endogeneity of
these variables as follows?
.logit P1 X Z
.predict vhat1, resid
.logit P2 X Z
.predict vhat2, resid
.logit P3 X Z
.predict vhat3, resid
.poisson NF P1 P2 P3 X vhat1 vhat2 vhat3
.test vhat1 vhat2 vhat3
And, I'm also suspecting that there is serious heteroskedasticity since the
data is cross-country data. Is it ok to combine <robust> command of STATA
with the procedure to address the heteroskedasticity problem?
-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Austin Nichols
Sent: Tuesday, May 16, 2006 2:25 PM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: Endogeneity Tests for Count Data
Assuming you have an instrument Z, you can regress government policy
var P on Z and X, save the error, and include the estimated error in
the "second stage" poisson regression.
. reg P X Z
. predict vhat, resid
. poisson NF P X vhat
. test vhat
A test that the coef on vhat is zero is a test of endogeneity of P.
See Wooldridge's EACSPD
http://www.stata.com/bookstore/cspd.html
and the references therein, and the method of moments approach
outlined there (pp.664-665 to start).
On 5/16/06, Jun S Kang <mediaecons@gmail.com> wrote:
> Hello. I'm conducting a statistical test where dependent variable is count
> data about the number of firms in a specific industry across countries as
> follows:
>
> # of firms in country i = f (X, government policy variables)
>
> Some countries do not have any firms in the industry. The highest number of firms in a country is 59. I have a potential endogeneity problem since there are some possibilities that the policy variables are endogenously determined. However, I could not find how to test for the possible endogeneity in the context of count data. Is it possible to treat the count dependent variable as continuous data and conduct usual Wu-Hausman tests for endogeneity using the IVREG2 procedure with endog option? Or do I need other than the IVREG2 procedure?