Quantifying Prior Information via Kullback-Leibler Divergence: A New Perspective on Effective Sample Size
This presentation will feature Min Lin, postdoc at the Health and Environment Modeling Laboratory at Ohio State.
About
The Biostatistics seminar series invites researchers from across the nation to discuss methodological research and its implications for a variety of health issues.
Abstract
The effective sample size (ESS) tells us how much information a prior contributes in Bayesian analysis. Yet, existing definitions often fail when the prior and likelihood disagree. In this talk, I introduce a new way to measure ESS using the reverse Kullback-Leibler divergence between an informative prior and a power posterior. This formulation provides a clear optimization-based interpretation and ensures desirable convexity properties. I’ll show how this framework generalizes classical ESS definitions, discuss its theoretical guarantees, and illustrate its behavior through simple examples and applied case studies. Overall, this approach offers an intuitive way to understand "how much data" a prior effectively adds.
Contact
Andy Ni