George Box and Bayesian Statistics

• Activity Number: 1
• Type: Other
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: ASA
• Author(s): George C. Tiao*+
• Companies: The University of Chicago

Abstract

I have known Professor Box since 1960. He was my esteemed teacher, long time colleague and friend. I'll focus on his contributions to Bayesian statistics for three topics. 1st, Box became interested in Bayes out of his frustration with the classical frequentist approach to inference, in particular with respect to the 'robustness' problem, a term that he coined. He was the first to realize that a robust statistic may not be relevant to inference, leading to the distinction between inference robustness and criterion robustness. 2nd, he came up with a reasonable and easily understood argument via data translated likelihood to arrive at the prior proposed by Harold Jeffreys which has been widely adopted in practice. Finally his original work on estimating random effects in hierarchical models has, since the advent of MCMC, led to an explosion in the use of Bayesian methods in numerous fields of application including marketing, data mining, machine learning, panel data analysis, small area estimation and others.