PhD STUDENTSHIP - OPEN UNIVERSITY
Applications are invited for a three-year full-time PhD studentship at
the Open University Department of Statistics, Milton Keynes, starting 1
October 2005. The projects available are as follows; details are given
at the end of this message.
1. Bayesian statistical hierarchical modelling of environmental
pollution processes. Supervisor: Dr Alvaro Faria.
2. Fitting four-parameter distributions to regression data. Supervisor:
Professor Chris Jones
3. Exploratory multivariate mixture modelling. Supervisor: Professor
Frank Critchley.
Applicants should have at least a 2(i) Honours degree or a recognised
postgraduate qualification containing a substantial element of
statistics. The successful applicant will be based at the main Open
University campus in Milton Keynes. Although the Open University differs
from other universities in that we don't have undergraduate students on
site, we are in other ways similar to other universities. In particular,
we have a very strong and active Statistics Department. For more
information about the Department, see our website at
http://statistics.open.ac.uk/.
Full tuition fees and research expenses will be paid, and a maintenance
grant starting at 12,000GBP will be payable. The studentship is funded
by the Open University and there are no nationality or residency
restrictions.
For further information on the studentship (but NOT to obtain or return
an application form) please contact Paddy Farrington at:
[log in to unmask]
The prospectus and application form can be dowloaded from
http://www.open.ac.uk/research-degrees/ .
Alternatively, email [log in to unmask], or phone +44
(0)1908 653844, or write to:
The Secretary
Department of Statistics
Faculty of Mathematics and Computing
The Open University
Walton Hall
Milton Keynes MK7 6AA.
Where the application form asks for details of your research topic,
simply state which (one or several) of the projects you are most
interested in, and why. The completed application form should be marked
with the reference code RDFL and returned by 18 March 2005 to the
Department of Statistics (by email to [log in to unmask], or
by post to the address above).
Equal Opportunity is University Policy.
PROJECT SUMMARIES
1. Bayesian statistical hierarchical modelling of environmental
pollution processes. Supervisor: Dr Alvaro Faria. This project addresses
the task of determining the joint probability distribution to assess the
ground contamination levels and related uncertainties caused by
polluting discharges in the environment. The distribution is updated in
time and space as new data in the form of measurements and expert
judgements become available to give real-time estimates of deposition
levels. The task is important as there are many situations where
decision making, such as evacuation and decontamination, relies on the
fast and accurate assessment of ground contamination levels after an
accidental release of hazardous substances. A Bayesian hierarchical
analytical model will be developed and used to produce fast estimates of
the ground contamination levels and their confidence limits in space and
time. The Bayesian element will allow the incorporation of both
objective measurements from instruments as well as subjective
information from experts. Expert intervention can also be carried out to
incorporate the effects of events such as decontamination. The
hierarchical element of this model unlike traditional environmental
statistical models allows the incorporation of other physical-biological
and ecological models to explicitly handle three main sources of
uncertainty: (i) ground contamination readings, (ii) spatial
interpolations and (iii) the underlying model itself. Further to those,
a Markov chain Monte Carlo simulation approach will be implemented for
validation purposes.
2. Fitting four-parameter distributions to regression data. Supervisor:
Professor Chris Jones. The purpose of this project is to explore the
practical utility of using four parameter
(location-scale-skewness-kurtosis) families of distributions as models
for continuous univariate responses in regression contexts. In contrast
with two parameter (typically normal) response distributions, questions
of robustness to both skewness and heavy tails in the response
distribution are automatically taken care of. This approach provides a
natural framework for quantile regression too. Models can be fitted
using a likelihood approach, either fully parametrically or
semi-parametrically. Model selection questions arise. The supervisor has
introduced two particular four parameter distribution families recently
that form natural particularisations of this work: a skew t family
(Jones & Faddy, 2003, JRSSB) and what he calls sinh-arcsinh
transformations of normality. The project is methodological, with scope
for a mix of theory, simulations and/or practical applications.
3. Exploratory multivariate mixture modelling. Supervisor: Professor
Frank Critchley [joint work with Ana Pires and Conceicao Amado
(Lisbon)]. This is an opportunity to take part in a small international
team engaged in developing novel techniques to visually explore
multivariate data for subpopulation structure. Recently, the team has
been developing Principal Axis Analysis as a rotation of standardised
principal components to optimally detect subgroup structure, rotation
being based on preferred directions in the spherised data. As such, it
is particularly well-suited to detecting mixtures of elliptically
contoured distributions. This new methodology is proving both to be of
value in itself and shows promise as a first step towards addressing
wider problems in robust-diagnostic multivariate mixture modelling.
Ability in, and enthusiasm for, statistical computing would be an
advantage.
|