SEMINAR 19 December 2011: Morten I. Lau, Durham Business School, Durham University

Abstract

Do individuals behave differently when they have to infer the probability of some uncertain event compared to when they are directly told the probability of an uncertain event? The answer to this question goes to the heart of the application of the Bayesian approach to decision-making under risk and uncertainty. Assume that enough information exists to define a prior probability of some event, and that the sample data allows an individual to infer a unique posterior probability by the correct application of Bayes Rule. The traditional approach would then treat that probability exactly the same as it would treat a probability that had been given directly to the individual. However, there exists a large literature that suggests that individuals do not apply Bayes Rule correctly when making decisions that depend on them correctly pooling prior information and sample data. We replicate and extend a classic experimental study of this issue, due to Griffin and Tversky [1992], making sure that we employ methods that addressed many of the sneaking suspicions that economists have when they come across experiments run by psychologists. We find evidence that exposes the structural nature of violations and the subjective expected utility model over alternatives, and the manner in which these violations might be explained.