Analysing the independence hypothesis in models for rare errors: an application to auditing

Abstract

Populations where items are error free are found in many areas of applied statistics, such as auditing and actuarial science. Decisions are then made by inferring the total error in a population. This parameter is usually modelled by two parametric structures under the assumption of prior independence. This paper explores the usefulness of robust Bayesian techniques in the setting of an applied problem. The results reveal a dramatic lack of robustness with regard to the independence hypothesis