Dear Statalisters, I am working with a panel dataset and I am estimating an IV-2SLS model with duo endogenous regressors and their interaction with a third exogenous variable (so I have 4 endogenous variables in total). Since I cannot find good instruments for my endogenous variables I use the ‘ivreg2h’ command in STATA to perform instrumental variables estimation using heteroskedasticity-based instruments (I end up with 32 excluded instruments).
My question is about the identification tests on which I have to base and finally report. Since I use the ‘cluster’ option and I receive robust standard errors clustered at the ‘id’ level I report the Hansen’s J test for which I fail to reject the null hypothesis (p>0.05) of the validity of the over-identifying restrictions. But, I am a bit confused about the weak or under-identification tests. First, with respect to the weak identification test the critical values of Kleibergen-Paap rk Wald F statistic are not available in order to know if I reject or fail to reject the null hypothesis. Then, with respect to under-identification test the Kleibergen-Paap rk LM statistic suggests that I fail to reject the null hypothesis that the model is underidentified.
Looking for help in the forum I found some previous posts that they are quite useful:
http://www.stata.com/statalist/archi.../msg00136.html
http://www.stata.com/statalist/archi.../msg00483.html
However, when I use the ‘xtivreg’ command, with all the generated instruments from ‘ivreg2h’, following by 'xtoverid, noi' which provides the Anderson-Rubin statistic (Wald test) I reject the null hypothesis that coefficients of the endogenous regressors in the structural equation are jointly equal to zero. Therefore, I am wondering which of the statistics above are finally valid to be reported and if it would be enough to report the Anderson-Rubin statistic for weak identification, the Hansen’s J test for over-identification and an additional Davidson-MacKinnon test of exogeneity.
Any response or comments would be much appreciated.
Thank you in advance
My question is about the identification tests on which I have to base and finally report. Since I use the ‘cluster’ option and I receive robust standard errors clustered at the ‘id’ level I report the Hansen’s J test for which I fail to reject the null hypothesis (p>0.05) of the validity of the over-identifying restrictions. But, I am a bit confused about the weak or under-identification tests. First, with respect to the weak identification test the critical values of Kleibergen-Paap rk Wald F statistic are not available in order to know if I reject or fail to reject the null hypothesis. Then, with respect to under-identification test the Kleibergen-Paap rk LM statistic suggests that I fail to reject the null hypothesis that the model is underidentified.
Looking for help in the forum I found some previous posts that they are quite useful:
http://www.stata.com/statalist/archi.../msg00136.html
http://www.stata.com/statalist/archi.../msg00483.html
However, when I use the ‘xtivreg’ command, with all the generated instruments from ‘ivreg2h’, following by 'xtoverid, noi' which provides the Anderson-Rubin statistic (Wald test) I reject the null hypothesis that coefficients of the endogenous regressors in the structural equation are jointly equal to zero. Therefore, I am wondering which of the statistics above are finally valid to be reported and if it would be enough to report the Anderson-Rubin statistic for weak identification, the Hansen’s J test for over-identification and an additional Davidson-MacKinnon test of exogeneity.
Any response or comments would be much appreciated.
Thank you in advance
