Evaluation of aggregation of audit evidence under uncertainty: An empirical study of belief functions

An audit is a process of gathering and evaluating evidence with the purpose of providing a report on a company's financial statements. The level of support that evidence provides for an audit objective must be evaluated. Each evaluation is combined with other evaluations to form an opinion on account balances and the financial statements. Recently, evidential networks using Dempster-Shafer belief functions have been developed to assist in the evaluation and aggregation of evidence. Belief functions are an uncertainty framework for expressing the support that evidence provides for an audit objective or other variable.;Two experiments are conducted in this dissertation. The first looks at whether decision-makers who are reading scenarios about everyday events can correctly match the type of uncertainty in the scenario with the uncertainty framework that best describes that type of uncertainty. Forty-four college students evaluated eight scenarios that contained uncertainty that could be modeled using either subjective probabilities or belief functions. For two of the eight scenarios, the decision-makers correctly matched the scenario and the related uncertainty framework.;In the second experiment, forty-nine auditors evaluated five pieces of audit evidence for audit objectives related to each of four accounts. The auditors were asked to evaluate the support provided by each piece of evidence individually and also asked to provide an evaluation of the combined support of all five pieces of evidence for the account. For each account, one of the five pieces of evidence was varied along two dimensions of task complexity, simplicity and clarity. Theory suggests that combining evidence with a higher degree of task complexity should become more difficult and should lead to greater inconsistency between auditors' evaluations of the strength of the evidence. It was hypothesized that the evidential network would combine the evidence more consistently when using the same inputs as the auditors. The results indicate that there is no statistically significant difference between the consistency of the evidential network and the auditors.;One very interesting result of this research indicates that auditors use belief functions rather than subjective probabilities to express the support that the evidence provides for the audit objectives.