David J. Laws and Anthony O'Hagan
University of Sheffield
Publication details: Applied Statistics 49, 577-590, 2000.
In this paper we describe a Bayesian model for a scenario in which the population of errors contains many zeroes and there is a known covariate. This kind of structure typically occurs in auditing, and we use auditing as the driving application of the method. Our model is based on a categorisation of the error population together with a Bayesian non--parametric method of modelling errors within some of the categories. Inference is through simulation or an approximating method based on moments. We conclude with an example based on a data set provided by the UK's National Audit Office.
Keywords: Auditing; Bayesian inference; Dirichlet process; Log--normal distribu tion; Nonparametric; Rare errors; Simulation.