Valid T-Ratio Inference for IV

In the single IV model, current practice relies on first-stage F exceeding some threshold (e.g., 10) as a criterion for trusting t-ratio inferences, even though this yields an anti-conservative test. We show that true 5 percent test instead requires greater than 104.7. Maintaining 10 replacing critical value 1.96 with 3.43. re-examine 57 AER papers and find corrected inference causes half of initially presumed statistically significant results to be insignificant. introduce more powerful test, tF procedure, which provides F-dependent adjusted values.