Post-Selection Estimation

• Activity Number: 609
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Amit Meir* and Yoav Benjamini
• Companies: University of Washington and Tel Aviv University
• Keywords: Selection Bias ; Selective inference ; Conditional estimation ; Data Carving ; Data Splitting

Abstract

The common practice of using the same data to select the inferential questions of interest and to answer them, referred to as selective inference, has been receiving more attention recently in view of the replicability crisis in science. In previous work we explored the selective inference problem in the context of correlating regions of activity in the imaged brain with personal behavioral characteristics, and offered maximum likelihood estimator of the conditional distribution. Here we propose a more general framework for conditional estimation, as well as two post selection estimators: one based on the recently proposed carving approach, the second on a penalized quasi-likelihood approach. We demonstrate the favorable properties of the estimators via analysis and simulations.