Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Re: collinearity in IV

From   Kit Baum <[email protected]>
To   [email protected]
Subject   st: Re: collinearity in IV
Date   Thu, 5 May 2005 07:36:19 -0400

Binzhen was puzzled about what ivreg2 did here. The instrument set he has specified is

(iota T X W V TW TV)

where as I understand it TW = T*W, TV = T*V. He should be able to see what is happening in the estimation by using the -first- option.

In an earlier message he suggested that he wanted to take the instrument generated by regressing Z on the above list and multiply it by T (the time dummy) and run a regression of Y on Z*, TZ*, T and X where Z* is an instrument for Z, TZ* = T * Z*. But that OLS regression (which could be run) will not be the instrumental variables estimator. The instrumental variables estimator regresses each endogenous variable (i.e., potentially correlated with error term) on the set of instruments above, and uses both the instruments and the original data in deriving the IV results and their consistent covariance matrix. You cannot pick and choose which RHS variables are to be instrumented by which variables; that is not IV from the standpoint of econometric theory.

I suspect that examination of the first stage regression results will shed some light on the problems that Binzhen reports regarding the covariance matrix of the moment conditions. He might want to use ivreg2 without the gmm option (as I suspect he is doing).

Kit Baum, Boston College Economics

On May 5, 2005, at 2:33 AM, Binzhen wrote:

I am runing the following model:
Here,  TZ=T*Z is the interaction term, T is the time dummy (=1 after
the policy change), and Z are the continuous endogenous treatments, X
are exogenous variables.

The command I use is: ivreg2 Y T X (TZ Z=W V TW TV)

I got an error message for the first stage regression:

Error: covariance matrix of moment conditions not of full rank;
       test stats not reported and standard errors above may be
*   For searches and help try:

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