The Influence of Dennis Lindley's Work on Applied Statistical Practice

• Activity Number: 58
• Type: Invited
• Date/Time: Sunday, August 9, 2015 : 4:00 PM to 5:50 PM
• Sponsor: Memorial
• Author(s): David Draper*
• Companies: UC Santa Cruz/eBay Research Labs
• Keywords: Bayesian ; Conditional exchangeability ; Utility ; Inference ; Prediction ; Decision-making

Abstract

Dennis Lindley (1923-2013), winner of the Royal Statistical Society Guy Medals in Silver and Gold (the latter representing a lifetime achievement award: there have only been 36 recipients from 1892-2014), was a towering figure in Bayesian theory and methodology for more than 60 years, from the 1950s through the 2000s. In this talk I will focus on an aspect of Dennis's work that is not as frequently discussed, namely that his methodological and theoretical contributions have had a broad and deep influence on applied statistical practice. The capstone of this thread of Dennis's influence is his emphasis on the central role of conditional exchangeability judgments in applied statistical modeling: the only way to successfully reason outward inferentially from sample to population or predictively from observed data to future data is to make sound judgments about the level of conditioning at which observed and unobserved units are similar. Dennis was also prescient about the importance of practical Bayesian decision theory, based on real-world-relevant utility functions (rather than artificial constructs such as squared-error loss), for meaningful Bayesian progress in the 21st century.