For more details see the Bayes linear methods home page. Michael Goldstein and I have written a book, Bayes linear statistics, for Wiley, which appeared in April 2007.
This research is a collaboration between statisticians and an environmental chemist, and largely concerns whether it is possible to use the chemical properties of pesticides to predict a pesticide's potential to pollute the environment. This is a very rich area for future research as there are major shortcomings with current practice, and considerable though addressable difficulties with data sources and statistical modelling of the underlying relationships.
This concerns a study carried out by Jenny Argyle for her PhD, supervised by Allan Seheult and myself. The basic question of interest is whether it is possible to detect at an early stage a failure to thrive in infants by examining the individual child's growth curve and relating the curve to national standards and the child's growth history. The main results are a prediction method for child growth, and the finding that the correlation structure between successive measurements is Markov.
This research, which arose initially from internal consultancy, is
collaborative with John Carpenter and Justine Schneider of the Centre for
Applied Social Studies,
One current project concerns the use of Bayesian belief networks to model
relationships for social policy and to support decision making. This began as
methodology used to model variables affecting good practice in the support of
people with some form of disability, for example learning disability or mental
health problems. In particular, the aim was to place such people in worthwhile
and satisfying employment. This research was funded by the Department of Health
and lead to methodology which could be implemented in the form of a computer
tool. This tool is now being developed to commercial prototype, part funded by
the charity SCOPE and part-funded by the
This work concerns a major research study in collaboration with British
Telecommunications Plc and statisticians and computer scientists at the
A prototype tool has been constructed to implement the research. Funding for this has been provided jointly by British Telecommunications Plc and Codeworks Ltd, http://www.codeworks.net/.
This work concerns a study in which statisticians (myself, Frank Coolen, Malcolm Farrow) are working with experienced physicians in Accident and Emergency departments in the main North Eastern hospitals to improve our understanding of sub-arachnoid haemorrhage (SAH). This is a rare but often fatal disease involving the bursting of blood vessels at the periphery of the brain. It is hypothesized that ``warning leaks'' (minor haemorrhages) can precede major (usually fatal) haemorrhages. Properly diagnosed warning leaks lead to treatment which can prevent a further major haemorrhage. However, there is a suspicion that such warning leaks, for which the major symptom is severe headache, are often misdiagnosed for a variety of reasons. The work is thus aimed at exploring whether it is possible to identify particular combinations of patient signs and symptoms that indicate SAH with high probability. The approach we take is to exploit the Bayesian graphical model structures established by David Speigelhalter, Stefan Lauritzen, and others.
This is work aimed at developing methods to support the decision making of commodity traders, in particular by understanding better the uncertainties involved. This work has been funded by the EPSRC as a Knowledge Transfer Partnership (number 38) with Energy Scitech Ltd..
This relates to PhD work carried out by Jonathan Cumming., as an industrial CASE studentship, funded by EPSRC and sponsored by the Smith Institute. The collaborating company was Etech Ltd. The background is that physicians and surgeons collect vast amounts of data on patients, often taking the same sets of measurements before and after (sometimes repeatedly) an intervention - for example, an operation. The question is whether anything useful can be gleaned from the data. In particular, do the measurements indicate that the operation was effective; do the measurements imply that one operation type was more effective than an alternative, and so forth. The kind of support offered is two-fold. First, better ways of examining large amounts of data in parallel, using graphics. Secondly, making sense of the data by structuring it appropriately and then modelling and exploiting relationships between the variables concerned.
(0191 or +44 191) 334 3121
(0191 or +44 191) 334 3051