Tony O'Hagan - Consulting
Consultancy in auditing
My expertise
I have experience in various kinds of audit. In particular, I was commissioned by the
UK National Audit Office to develop Bayesian methods for multi-location audit.
My clients
In addition to the National Audit Office, I have worked for clients auditing health care
provision in New York state, and for the Audit Bureau of Circulation, which audits
newspaper circulation figures.
Typical challenges
Statistical auditing involves taking samples of transactions and looking for errors. Usually,
we do not expect to find any errors. Other times, we might expect some errors, but the error rate
should be low. In order to estimate low error rates accurately, relatively large samples
are needed, but auditing organisations are invariably under pressure to reduce the auditing
cost by reducing sample sizes. To justify small samples, auditors may argue that it
is unreasonable to engage in such a level of effort when the great majority of their audits
will reveal error rates below the level known as 'materiality'. The challenge is to provide
a rigorous defence of such a position. Bayesian statistics offers such a defence based on
the incorporation of prior knowledge, and can lead to substantial reductions in sample size.
I believe I can offer a unique combination of high expertise in Bayesian statistics with
experience and understanding of auditing.
One possible form of prior information is previous audits of the same organisation. An
organisation with a good track record over their last few audits, in which no errors
have been found, should have a 'lighter touch' audit with a smaller sample size, while one
with a poor recent record should be subjected to more scrutiny with a larger sample.
A variety of schemes of this kind can be considered within a Bayesian approach. I can
deliver adaptive, serial auditing solutions that are efficient and responsive to
recent performance of both the organisation itself and others of its kind.
The magnitudes, or 'taints', of errors are often more important than the raw error rate.
However, statistical methods designed to assess the total magnitude of error face
the complication of nonstandard distributions of taints. I have developed
nonparametric Bayesian methods to address this challenge. Again, I think this
is a unique methodology for a familiar but complex question.
I am happy to consult on these or other problems in auditing.
Updated: 15 February 2008
Maintained by: Tony O'Hagan
