Thursday, April 06, 2023, 04:00pm - 05:30pm
"Testing Validity of an IV Using A Discontinuously Distributed Covariate"
Abstract: Endogeneity of treatment variables is a significant concern for applied economists. The most common way of addressing endgeneity is to try to find an instrumental variable. Testing that a proposed variable is in fact a valid IV is in general not possible without imposing some restrictions on the structural functions in general. This paper demonstrates that independence of an instrumental variable from the unobservables in the model given controls has a testable implication in a nonparametric model. Using this implication, the paper also proposes a testing approach for the validity of an instrumental variable. Unlike existing tests for validity of an IV, our testing approach can also be used for continuous treatments. The proposed testing approach exploits the availability of an observed control with a distribution function that has a discontinuity at a known point, but is continuous in a neighborhood around that point.
4:00pm-5:30pm | In Person, 3rd Floor Library, NJ Hall | Coordinator: Roger Klein