Bayesian and Non-Bayesian Approaches in Econometrics and the Social Sciences

概要

論文の詳細を見る
After briefly discussing general objectives of all the sciences, namely learning from our data in order to explain the past, predict the future and make good decisions, I compare Bayesian and non-Bayesian approaches with respect to their capacity to realize these objectives. Specifically, alternative learning, estimation, testing, prediction and decision-making procedures are compared and evaluated. It is pointed out that many scientists learn without use of a formal learning model while Bayesian scientists learn using a formal learning model, Bayes' Theorem. Then I review traditional and my recent derivations of Bayes' Theorem and generalizations of it using an information theory optimization approach. These optimal learning models are shown to be 100% efficient in the sense that they have the property that input information = output information. Some comparisons of such learning models to psychological learning models of Einhorn and Hogarth and others are made. Then applications of the new models to solve practical statistical problems are presented along with procedures for comparing old and new Bayesian solutions. Finally, remarks on the future of Bayesian analysis are presented.
日本行動計量学会の論文
2005-08-24