A Performance Comparison of Methods for Instrumental Variables Regression When Treatment Is Binary

• Activity Number: 474
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Jessica Pruszynski*+ and Purushottam Laud and Rodney Sparapani and Robert McCulloch
• Companies: Medical College of Wisconsin and Medical College of Wisconsin and Medical College of Wisconsin and The University of Texas
• Keywords: Bayesian methods ; instrumental variables ; binary data ; comparative effectiveness research ; Markov chain Monte Carlo

Abstract

Instrumental variable methods are used to estimate causal relationships when performing a randomized experiment is not possible. Traditional instrumental variable methods in econometrics do not account for the presence of a binary or categorical variable. In this paper, we consider an instrumental variables model in which the treatment is a binary variable. We compare the performance of a two stage least squares moment based approach to a model-based Bayesian approach under two scenarios. In the first scenario, we ignore the binary treatment and proceed with the analysis as if the treatment were continuous. In the second scenario, we apply the correct model with the knowledge that the treatment is binary. Under both conditions, we compare, using repeated data simulation, the moment-based and Bayesian approaches using such quantities as power, bias, mean squared error, and confidence/credible interval length and coverage.