Dear all,
I'm having a trouble using a 2SLS estimation with two instrumental variables and two endogenous variables. The problem I have is with the F-statistics and the different weak instruments tests or underidentification tests. I have read that nowadays the literature has converged to using the Montiel Olea-Pflueger effective first-stage F-statistic to test wether instruments are weak or not in the special case of one instrumental variable and one endogenous variable, but, however, this is not my case. Also is important to say that my errors are clustered, so I cannot use Cragg-Donald F-statistic or Stock-Yogo critical values.
When I run the command "ivreg2" it shows me several tests, including the underidentification test or the weak instruments test using Kleibergen-Paap rk LM statistic and Kleibergen-Paap Wald F statistic respectively. My question is: which of them is important to report in this case? Should I report Anderson-Rubin test? Which conditions should I keep in mind when reporting one statistic or another? As far as I have seen the literature linked to my research topic only shows Kleibergen-Paap Wald F statistic, but generally they don't say why.
Thank you for your attention.
I'm having a trouble using a 2SLS estimation with two instrumental variables and two endogenous variables. The problem I have is with the F-statistics and the different weak instruments tests or underidentification tests. I have read that nowadays the literature has converged to using the Montiel Olea-Pflueger effective first-stage F-statistic to test wether instruments are weak or not in the special case of one instrumental variable and one endogenous variable, but, however, this is not my case. Also is important to say that my errors are clustered, so I cannot use Cragg-Donald F-statistic or Stock-Yogo critical values.
When I run the command "ivreg2" it shows me several tests, including the underidentification test or the weak instruments test using Kleibergen-Paap rk LM statistic and Kleibergen-Paap Wald F statistic respectively. My question is: which of them is important to report in this case? Should I report Anderson-Rubin test? Which conditions should I keep in mind when reporting one statistic or another? As far as I have seen the literature linked to my research topic only shows Kleibergen-Paap Wald F statistic, but generally they don't say why.
Thank you for your attention.
Comment