[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Re: IV tests in Stata

From   Kit Baum <>
Subject   st: Re: IV tests in Stata
Date   Sat, 6 Dec 2008 07:30:17 -0500

< >
I think there are two things being confused here. A test of the hypothesis that the excluded instruments are orthogonal to the error process, commonly known as a test of overidentification or overidentifying restrictions, can only be done when your IV specification is overidentified (when you have more excluded instruments than included endogenous variables). You cannot challenge the instruments in an exactly identified specification.

On the other hand that orthogonality hypothesis is only one of three which should be tested for a given instrument. The other two are:

1) meaningful correlation between instrument(s) and endogenous variables: the various 'weak instruments' tests

2) the exclusion restriction: the instrument does not itself belong in the equation. Now if it does, it might be either exogenous or endogenous. Naturally, if it is excluded, it will be correlated with the error process, and will likely trigger rejection of the first hypothesis. That can readily be tested by including it in the equation. If you only have one instrument, you will have to find something else to identify the equation, but you do want to ensure that the hypothesis that the instrument has only an indirect effect on y (through the endogenous variable(s)) rather than a direct effect on y is reasonable for your instrument(s).

PS> Stata, unlike SAS, TSP, RATS, MATLAB, SPSS etc. is not an acronym.

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

On Dec 6, 2008, at 02:33 , Carlos wrote:

Do all the parametric tests of sensitivity of IVs to the violation of
the exclusion restriction that are available in STATA require more
than one IV (=overidentification tests)?  In other words, what if I
have only one IV?  Is ti possible to run such a sensitivity test in

*   For searches and help try:

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