Consistency without Inference: Instrumental Variables in Practical Application

I use Monte Carlo simulations, the jackknife and multiple forms of bootstrap to study a comprehensive sample 1309 instrumental variables regressions in 30 papers published journals American Economic Association. simulations based upon show that non-iid error processes highly leveraged regressions, both prominent features work, adversely affect size power IV tests, while increasing bias mean squared relative OLS. Weak instrument pre-tests F-statistics are found be largely uninformative bias. In has little as, despite producing substantively different estimates, it rarely rejects OLS point estimate or null is unbiased, statistical significance excluded instruments exaggerated.