st: xtivreg2: Endogenous variables

["Fitzgerald, James" <J.Fitzgerald2@ucc.ie>]

Hello Statalist users

I am implementing xtivreg2 (Schaeffer, 2010) in STATA 11.2.

I am using lags of my included variables to act as instruments for any of the variables found to be endogenous based on difference in Hansen tests i.e. orthog() option is used to test the orthogonality of each individual included variable (I first test the orthogonality of the excluded instruments and eliminate those that do not satisfy orthogonality conditions, such that the Hansen test of all excluded instruments yields a p-value >0.1, and thus a rejection of the null in the difference in Hansen tests indicates that it is the included variable that is endogenous).

My question is, if any of the included variables are found to be endogenous and the KP wald stat indicates only very weak instruments exist, must I drop the variable from the model, or simply conclude that the coefficient is uninterpretable and leave the endogenous variable in the model? I am reluctant to drop the variables as I am comparing coefficient signs and significance levels generated from a number of different samples, and I am worried that I cannot properly compare results across samples if the models are differently specified.

What is the effect of leaving an endogenous variable in the model?

Thanks

James
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/