A hierarchical Bayes model for multilocation auditing

The paper provides a Bayesian analysis of a practical problem in auditing in which substantial prior information needs to be combined with limited sample data. The specific context of the paper is a multilocation audit in which auditors take a two-stage sample of transactions from different sites within an organization. The hierarchical model that we propose has the flexibility to deal with this sort of stratified population in which some strata are not sampled and in which strata are not exchangeable. A careful, thorough elicitation of auditors' prior information is another key feature of our approach. Inference from our model is by direct Monte Carlo simulation. The example data set is given in the context of an audit carried out by the UK's National Audit Office who funded this research.