Bayesian Inference for Regression Discontinuity Designs

• Activity Number: 503
• Type: Invited
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Alessandra Mattei*+ and Fan Li and Fabrizia Mealli
• Companies: University of Florence and Duke University and University of Florence
• Keywords: Bayesian ; Causal Effects ; Instrumental variables ; Multivariate Outcomes ; Regression Discontinuity

Abstract

Regression discontinuity designs (RDD) identify causal effects of interventions by exploiting treatment assignment mechanisms that are discontinuous functions of observed covariates. In this paper we consider a non-standard fuzzy RDD setup, where the treatment is determined by both a covariate and an application status. In this setting, causal estimands and estimation strategies are different from those in the standard instrumental variable approach to fuzzy RDDs. A Bayesian approach is developed for drawing inferences on the causal effect of interest and multivariate outcomes are utilized to sharpen the analysis. Model checking procedures via posterior predictive checks are also discussed to evaluate the fit of the proposed Bayesian models to the observed data. The method is applied to evaluate the effects of Italian university grant on student dropout and academic performances.