Hi Statalist,

My problem is to evaluate two treatments for bladdercancer. Data is from an observational register. Outcome is dead/not dead. The two treatments are U and V. My instrument (IV) is the fraction of former patients, treated with U, at that very hospital. Thus for each and every patient I have calculated (nU/(nU+nV)) for formerly treated patients. This is the instrumental variable constructed to avoid or mitigate bias due to confounding in the choice of treatment. It could be called the hospital preference. It will take a value 0-1. I will also include (the same) confounders in the two stages, such as gender, age, education and cancer type. Thus, in the first stage I would like to set up the model: Treatment (U/V)= IV+gender+age+edu+ct. In the second stage it would be Death=estimated Treatment (from 1st stage)+gender+age+edu+ct.

I am supposed to use ivreg2 to solve the problem. I am quite unfamiliar with the econometric terminology (endogenous/exogenous/in-/excluded instruments) mostly used in this context. I am unfamiliar with STATA.

Could anyone help me to specify the command?

Kind Regards