Causal Inference When the Intervention Is Prioritization

• Activity Number: 670
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Benjamin Ogorek*+ and Shannon Terry
• Companies: Nationwide Insurance and Nationwide Insurance
• Keywords: Permutation ; Causal Inference ; Unequal probability sampling ; cycle decomposition ; Horvitz-Thompson estimator ; Prioritization

Abstract

A permutation of an ordered set is a valid cause according to the Rubin Causal Model (RCM). Intuitively, such a permutation could represent a prioritization of actions to influence an outcome of interest. From the standpoint of the RCM, many permutations may be applied, and each defines a potential outcome. In this research, we develop finite-population causal inference for estimating the causal effect of a prioritization. As causal effects are relative, population summaries from both a control and treatment permutation are compared. The control prioritization need not be a random - it may arise from an established paradigm. A modified Fisher-Yates shuffle, based on the cycle decomposition of the treatment permutation, comprises the probabilistic selection mechanism for the treatment group. Those units not selected for the treatment group either align with the control permutation, or define an additional permutation distinct from both the control and treatment permutations. First-order selection probabilities are studied, and approximate inference based on the Horvitz-Thompson and Hanson-Hurwitz estimators is presented. A simulation demonstrates the utility of this approach.