Abstract:
|
Policymakers are tasked with making decisions under uncertainty. In practice, the frequentist framework for policy evaluation focuses on testing the hypothesis that program impacts are equal to zero. Stakeholders often view the resulting 'thumbs up-thumbs down' inference as restrictive. By contrast, the Bayesian evaluation framework provides intuitive, informative, probabilistic inference such as "There is a 70% chance that the intervention improved the outcome of interest by at least 5%." Furthermore, the conventional approach to policy evaluation often tests many hypotheses separately (by outcomes, by time periods, by geographic regions). By contrast, a Bayesian model can piece together disparate data sources to obtain a more precise impact estimate, reducing the likelihood that important but modest-sized effects go unrecognized for lack of statistical power. In this talk - using data from the evaluation of an initiative of the Affordable Care Act as a motivating example - we will summarize the ways in which Bayesian methods can provide a flexible and powerful tool for policy evaluation, and we will discuss the challenge and tradeoffs of the Bayesian approach.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.