Everyone is welcome to attend the following two talks:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Friday 12th May 2000, 2:15pm
Room SM1.3, School of Mathematics, University of Bristol
Statistical analysis of performance indicators in UK higher
education
David Draper (University of Bath)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Friday 9th June 2000, 2pm
Room TBA, University of Bath
Space-time prediction for on-line disease surveillance
Peter Diggle (Medical Statistics Unit, Lancaster University)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
For further details, see our Web site
http://www.stats.bris.ac.uk/~guy/Avon/
or contact Nicky Welton ([log in to unmask], Tel:
(0117) 965 6261 ext. 3227).
Location maps for Bristol and Bath Universities can be
obtained from http://www.bris.ac.uk/directions.html
and http://www.bath.ac.uk/opendays/maps.htm respectively.
Abstracts:
~~~~~~~~~~
*** Statistical analysis of performance indicators in
*** UK higher education
David Draper (University of Bath)
In recent years the Higher Education Funding Council for
England (HEFCE) and the corresponding bodies in Scotland,
Wales, and Northern Ireland have become increasingly
interested in monitoring the quality with which universities
carry out their public mandate. Last December HEFCE
published results on a number of performance indicators for
1996-97, including drop-out rates after the first year of
undergraduate study, comparing observed (O) and expected (E)
outcomes for all 165 government-funded higher education
institutions in the UK (this publicly available report makes
interesting reading).
Interpreting this effort in the language of causal
inference, with a binary outcome such as dropout at the
student level, (a) the process of students choosing and
attending universities gives rise to an observational study
(rather than a controlled experiment), meaning that
measuring and controlling for potential confounding factors
(PCFs) is crucial; (b) the supposedly causal factor at the
university level is the underlying - and *unobserved* --
quality of the university; and (c) there are many
student-level PCFs (including age, A-level qualifications,
and subject of study). HEFCE's method of computing expected
outcomes given the PCFs turns out to be a form of indirect
standardization, which is almost equivalent to a version of
hierarchical model-based inference in which the university
effects are treated as fixed.
A PhD student, Mark Gittoes, and I have shown that when the
difference D = ( O - E ) is correctly calibrated the
numbers of unusually under- and over-achieving universities
in 1996-97, as far as dropout rates were concerned, were
both surprisingly high. In this talk I will (1) describe a
method for calibrating D; (2) relate this method to
regression reformulations of the problem, both hierarchical
and non-hierarchical; and (3) present league-table-style
results for all UK universities and discuss their policy
implications. Time permitting, I will also (4) compare
fixed-effects and random-effects formulations of the
model-based version of the HEFCE approach on their ability
to correctly identify "good" and "bad" universities, and
(5) explore the effects of unobserved student-level PCFs on
the results.
*** Space-time prediction for on-line disease
*** surveillance
Peter Diggle, Anders Brix and Julia Kelsall
(Medical Statistics Unit, Lancaster University)
The work in this talk was motivated by the following
problem in environmental epidemiology, which is the subject
of a three-year collaborative research project between
Lancaster University and the Southampton Public Health
Laboratory Service. Individual cases of acute
gastro-infection within a health authority region are
indexed by the residential location of the patient and the
date of reporting of symptoms. These data are held
centrally and updated daily. In addition, the approximate
spatial distribution of the population at risk is known
from census data. The objective is to develop an automatic
surveillance system in order to identify and quantify
anomalous patterns of incidence which could indicate
spatially localised changes in disease risk. Identified
anomalies would then be subject to further investigations
of various kinds, including refinement of the diagnosis of
individual cases through pathological analysis, with a view
to detecting common causes of any genuine outbreaks.
Our general approach to this problem is to build a
space-time stochastic model for disease incidence and to
use the model to construct a predictive distribution for
the current value of the risk in the neighbourhood of an
arbitrary location. Our basic modelling assumption is that
the point process of disease incidence is a log-Gaussian
Cox process, i.e. a Poisson process whose space-time
intensity function is l(x,t) = exp{Z(x,t)} where Z(x,t) is
a Gaussian process.
In the talk, I will describe the exploratory analysis of
historical data using kernel smoothing methods, leading to
a detailed formulation of the point process model. I will
then discuss associated methods for estimating model
parameters and for identifying changes in the underlying
space-time intensity.
The full development and implementation of an on-line
surveillance system will involve many statistical modelling
issues concerned with both spatial and non-spatial aspects
of the problem. For example, one of many such issues is the
need to adjust for different reporting rates amongst
general practitioners and other health care providers
which, if undetected, could induce spurious local peaks in
incidence.
Acknowledgement: This work is supported financially by the
UK Department of Health.
----------------------------------------
Dr. Nicky Welton,
Office 2P10, CSM,
University of the West of England,
Bristol. BS16 1QY.
Email: [log in to unmask]
Tel: (0117) 965 6261 ext 3227
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|