Statalist The Stata Listserver

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

st: Re: goodness of instruments

From   Kit Baum <>
To   "alessia matano" <>
Subject   st: Re: goodness of instruments
Date   Thu, 24 May 2007 07:30:50 -0400

Dear Alessia

Your comment #1 is a FAQ. Instrumental variables does not work by specifying which instruments apply to which X variables. You have a set of X and a superset Z, including the exogenous elements of X. Please see Ch.8 in my book, or Baum-Schaffer-Stillman Stata Journal 2003, available in working paper form from the URL below. Some collinearity among the Zs does not invalidate their use; you're just projecting Xs on Zs, and the goodness of fit of that projection is not affected by intercorrelations among the Zs as long as they are not perfect.

You should use xtivreg2 (findit xtivreg2) in order to have access to all of the diagnostic techniques available in ivreg2. Using xtivreg2, you can use the endog() option to evaluate whether specific regressors must be treated as endogenous, and employ the test of overidentifying restrictions (Sargan-Hansen) to consider the quality of the instruments, as well as tests for weak instruments.

Kit Baum, Boston College Economics and DIW Berlin
An Introduction to Modern Econometrics Using Stata:

On May 24, 2007, at 6:54 AM, alessia matano wrote:


I would like to ask you one thing on the goodness of instruments.
I have six variables that I consider endogeneous. I m doing the first
stage with xtivreg. and I would like to ask you:

1. Is that correct to instrument each one with its own instrument and
with the exogeneous ones (correcting then the standard errors)? i.e .
not using in the first stage the instruments related to the other
variables (I am afraid on problems of possible multicollinearity).

2. If I run the first stage regression with only the exogeneous and
then with the exogeneous and the instruments, I can I see if the
instrument is good in that case??

how should I compare the r squared?

Thanks a lot
p.s. my regression is done in fixed effects, with aweights, because
some of the variables come from a microlevel collapse (my dependent
var overall, but not my independent of interest)
*   For searches and help try:
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index