CFE 2019: Start Registration
View Submission - CMStatistics
Title: Over-identified regression discontinuity design Authors:  Gregorio Caetano - University of Georgia (United States)
Carolina Caetano - University of Georgia (United States) [presenting]
Juan Carlos Escanciano - Universidad Carlos III de Madrid (Spain)
Abstract: A new identification and estimation strategy is proposed for the Regression Discontinuity Design (RDD). Our approach explores the heterogeneity in the ``first stage'' discontinuities for different values of a covariate to generate over-identifying restrictions. This allows us to identify quantities which cannot be identified with the standard RDD method, including the effects of multiple endogenous variables, multiple marginal effects of a multivalued endogenous variable, and heterogeneous effects conditional on covariates. Additionally, when this method is applied in the standard RDD setting (linear model with a single endogenous variable), identification relies on a weaker relevance condition and has robustness advantages to variations in the bandwidth and heterogeneous treatment effects. We propose a simple estimator, which can be readily applied using packaged software, and show its asymptotic properties. Then, we apply our approach to the problem of identifying the effects of different types of insurance coverage on health care utilization.