Bridge Estimation: Estimator Hybridization for Efficiency and Robustness

• Activity Number: 114
• Type: Topic Contributed
• Date/Time: Monday, August 2, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Russell T. Shinohara*+ and Constantine Frangakis
• Companies: Johns Hopkins Bloomberg School of Public Health and Johns Hopkins Bloomberg School of Public Health
• Keywords: Estimation ; Robustness ; Superefficiency

Abstract

The decision between an efficient estimator that may be inconsistent and a nonparametric estimator that is consistent on a larger set of truths often irks researchers. We propose a methodology that combines estimators to yield not only nonparametric consistency but also efficiency properties enjoyed by parametric estimators. We show that these "bridge" estimators are superefficient on the same large set of truths as the original nonparametric estimator. We also demonstrate in small samples that the bridge estimators are more efficient than either of the original estimators in a wide variety of scenarios.