The School of Engineering, Computer Science and Mathematics website has been designed in line with modern Internet technologies (XHTML, CSS, DOM) but your browser does not appear to properly support some of the required features.

So that you are not at a disadvantage, you will still see all of the information content but in a less organised form. However, you will have a better browsing experience if you upgrade to a modern browser.

Past Statistics/OR Seminars by Reverse Date


Statistics and Operational Research Seminars (Historical)

RSS Local Group Meeting
Bayesian spatial partition modelling in epidemiological case-control studies
Carmen Fernández (Lancaster University)
10 Mar 2005Exeter University Laver, 321 Thurs 2pmStatistics & Operational Research
The talk considers developments in the modelling of case-control data where each sample individual is associated with a geographical location. A logistic regression framework is used and residual spatial variation is flexibly accommodated via the use of Voronoi tessellations with unknown numbers and locations of tiles. Modifications for matched case-control studies are also discussed.


Postgraduate Seminar
Some investigations in discriminant analysis with mixed variables
Nor Idayu Mahat (University of Exeter)
24 Feb 2005Exeter University Laver, 321 Thurs. 2pmStatistics & Operational Research
The location model has been developed for the treatment of mixed categorical and continuous variables in discriminant analysis. In its recent development, Asparoukhov and Krzanowski (2000) suggested estimating the parameters of the location model by using non-parametric smoothing procedures. This approach overcomes some deficiencies in Maximum Likelihood Estimation and Linear Model Estimation. However, the choice of smoothing parameters by maximising the leave-one-out pseudo-likelihood function suggested in this approach depends on distributional assumptions This talk will describe how the smoothing parameters can instead be chosen by optimising either the error rate or the Brier score, neither of which make distributional assumptions. Some investigations on other possible smoothing procedures also will be discussed.


Postgraduate Seminar
Spatial Survival Analysis and the FMD epidemic in Devon
Trevelyan McKinley (University of Exeter)
2 Dec 2004Exeter University Laver, 321 Thurs 2pmStatistics & Operational Research
The magnitude of the potential economic impacts of epidemic animal disease events have been highlighted in recent years through outbreaks such as the foot-and-mouth disease epidemic in the United Kingdom during 2001. This talk reports some initial work from a project in conjunction with the Veterinary Laboratory Agency, Weybridge, which is looking at the feasibility of using survival modelling to develop dynamic space-time predictions of survivor and hazard functions for individual farm premises as an animal disease epidemic progresses. Survival analysis used in a spatial context is a potentially useful approach to quantifying the risk of infection of susceptible premises within future time periods given the characteristics of these premises and their geographical location relative to potential sources of infection. Results from such analyses could provide powerful insights into the patterns of infection, such as regional differences in the dynamics of the epidemic, and can assist in optimising various aspects of the operational response activities, such as targeting of at-risk farms. We consider various possible model formulations and apply a range of these to data from Devon for the 2001 UK FMD epidemic.


RSS Local Group Meeting:
Competing Risks: a brief introduction
Martin Crowder (Imperial College)
18 Nov 2004Exeter University Laver, 321 Thurs 2pmStatistics & Operational Research
The origins of Competing Risks date back to Bernoulli's attempt in 1760 to disentangle the risk of dying from smallpox from other risks. Much subsequent work has been demographic and actuarial in nature and although obviously of potential relevance in Reliability and Survival Analysis, applications to those fields are quite recent. The talk will cover some of the basic ideas and application of the subject.


Statistics Seminar
Limited and full information estimation and goodness-of-fit testing in 2^n contingency tables: A unified framework
Albert Maydeu-Olivares (Faculty of Psychology, University of Barcelona)
12 Nov 2004Exeter University Laver B74 Friday 4pmStatistics & Operational Research
High-dimensional contingency tables tend to be sparse and standard goodness-of-fit statistics such as chi-square cannot be used without pooling categories. As an improvement on arbitrary pooling, for goodness-of-fit of large 2^n contingency tables, we propose classes of quadratic form statistics based on the residuals of margins or multivariate moments up to order r. These classes of test statistics are asymptotically chi-square. Further, the marginal residuals are useful for diagnosing lack of fit of parametric models. We show that when r is small (r = 2,3) the proposed statistics have better small sample properties and are asymptotically more powerful than chi-square for some useful multivariate binary models. Related to these test statistics is a class of limited information estimators based on low-dimensional margins. We show that these estimators have high efficiency for one commonly used latent trait model for binary data (the two parameter logistic IRT model)


RSS Local Group Meeting:
Measurement error, power and sample size in gene-environment interaction studies
Jian'an Luan (University of Cambridge)
11 Jun 2003Laver None NoneStatistics & Operational Research
See here for abstract


RSS Local Group Meeting:
Machine Learning Techniques for Bioinformatics
Colin Campbell (University of Bristol)
29 May 2003Laver None NoneStatistics & Operational Research
See here for abstract


Operational Research Seminar:
A cash flow criterion for controlling a base stock inventory system
Roger Hill (University of Exeter)
22 May 2003Laver None NoneStatistics & Operational Research
It is common practice to control certain inventory systems using a 'base stock' policy and the standard approach is then to associate time-weighted costs with holding stock and with having unsatisfied demand and to determine the base stock which minimises the long run average total cost per unit time. This approach ignores the impact of the control policy on the timing of the cash flows associated with payments made to suppliers and revenues received from customers. The approach made here is to concentrate on cash flows and determine the control policy which optimises an appropriate cash flow measure. The impact of the control system is measured in two ways. Firstly, if a customer order is met immediately from stock then we determine for how long that unit has been held in stock and compute the corresponding compound loss of interest resulting from having paid the supplier for it early. Secondly, if a customer order is not met from stock then we determine the delay in meeting that demand and the consequent loss of interest which could have been earned on the customer payment if it had not been delayed. The objective is to minimise the total expected cash flow impact per unit time. A solution procedure is given and a comparison is made between this approach and other ways of controlling this system.


Statistics Seminar:
Bootstrapping SIR for Dimension Assessment in a General Regression Problem
Santiago Velilla (Universidad Carlos III de Madrid)
8 May 2003Laver None NoneStatistics & Operational Research
A bootstrap method is constructed for assessing the dimension of a general regression problem. A resampled version of the matrix used in the SIR method of Li (1991) is obtained, and the bootstrap distributions of the statistics of interest characterized. The proposed methodology incorporates both formal and graphical inference procedures and can be considered as an alternative to the permutation test of Cook and Yin (2001).


Statistics Seminar:
Nonparametric classification exploiting separation of populations
Adolfo Hernandez (University of Exeter)
13 Mar 2003Laver None NoneStatistics & Operational Research
Kernel discriminant analyisis is greatly affected by the well-known phenomenon often referred to as the 'curse of dimensionality'. This causes bad behaviour of the rules because of the amount of data required as the dimension of the problem increases. In this seminar two dimension reduction methods are proposed, based on the concept of separation of populations. The basic idea is firstly to obtain a dimension reduction subspace through the maximization of certain functionals which can be seen as indexes of separation, and, secondly, to evaluate a reduced kernel discriminant rule in that subspace. The good behaviour of these methods is justified both theoretically and also through application to data sets where comparisons with other methods proposed in the literature can be established.


RSS Local Group Meeting:
Markov Chain Monte Carlo exact inference for binomial and multinomial logistic regression models
John MacDonald (University of Southampton)
27 Feb 2003Laver None NoneStatistics & Operational Research
See here for abstract


Statistics Seminar:
Spatial modelling of childhood malaria in the Gambia
Rana Moyeed (University of Plymouth)
20 Feb 2003Laver None NoneStatistics & Operational Research
A spatial generalized linear mixed model is developed to describe the variation in malarial prevalence amongst a sample of village resident children in the Gambia. The response from each child is a binary indicator of the presence of malarial parasites in a blood-sample. The model includes terms for the effects of child-level covariates, village-level covariates and separate components for residual spatial and non-spatial extra-binomial variation. The results show that the extra-binomial variation is spatially structured, suggesting an environmental effect rather than variation in familial susceptibility. The method of inference was Bayesian using vague priors and a Markov chain Monte Carlo implementation.


RSS Local Group Meeting:
The Forward Search and the Analysis of Multivariate Data
Anthony Atkinson (London School of Economics)
6 Feb 2003Laver None NoneStatistics & Operational Research
See here for abstract


Operational Research Seminar:
An Overview of OR in Sport
Chris Potts (University of Southampton)
16 Jan 2003Laver None NoneStatistics & Operational Research
This talk reviews some contributions that operational research (OR) has made in sport. The contributions roughly fall within the areas of planning and strategy, scheduling, ranking and performance measurement, and prediction. In planning and strategy, we discuss the use of dynamic programming for optimising batting strategies in one-day cricket, and describe how pit stop strategies are determined in formula one motor racing. For scheduling sports fixtures, we indicate how OR has been used in county cricket. Different sports use different methods for measuring the performance of players/teams. We comment briefly on the need for more robust systems. Finally, prediction is important for bookmakers and is of interest to sports enthusiasts. The modelling of the results of soccer games is discussed.


Postgraduate Seminar:
Clustering Gene Expression Data
Heather Turner (University of Exeter)
12 Dec 2002Laver None NoneStatistics & Operational Research
Gene expression data, resulting from the relatively new microarray technology, has many features which make it difficult to analyse. Many traditional techniques fail due to the size of the data sets or the lack of conformity to common assumptions, such as normality or independence.Clustering has proven to be a useful method of discovering functional grouping of genes, one of the main objectives of microarray experiments. A number of clustering algorithms have been specifically developed for gene expression data, designed to be more efficient and more flexible than standard algorithms. Tne of these algorithms is the plaid model, a two-way overlapping clustering method proposed by Lazzeroni and Owen (2002). This talk introduces the plaid model and proposes an alternative optimisation algorithm that may improve its efficiency. Possible extensions of the model will also be discussed.


RSS Local Group Meeting:
Predicting Reliability for Orthopaedic Hip Replacements
Simon Wilson (Trinity College Dublin & Carlos III University Madrid)
4 Dec 2002Laver None NoneStatistics & Operational Research
See here for abstract


RSS Local Group Meeting:
Challenges in Bioinformatics for Statisticians
Wally Gilks ('MRC Biostatistics Unit, Cambridge')
21 Nov 2002Laver None NoneStatistics & Operational Research
See here for abstract


Statistics Seminar:
Denoising real data using complex wavelets
Stuart Barber (University of Bristol)
7 Nov 2002Laver None NoneStatistics & Operational Research
Wavelet shrinkage is an effective nonparametric regression technique when the underlying curve has irregular features such as spikes or discontinuities. The basic idea is simple: take the discrete wavelet transform (DWT) of data consisting of a signal corrupted by noise; shrink the wavelet coefficients to remove the noise; and then invert the DWT to form an estimate of the true underlying curve. Various authors have proposed methods of doing this using real-valued wavelets. Complex-valued versions of some wavelets exist, but are rarely used. We propose two shrinkage techniques which use complex wavelets. Simulation results show that both methods give smaller errors than using state of the art shrinkage rules with real-valued wavelets.


Statistics Seminar:
Bayesian rule based classification
Chris Holmes (Imperial College London)
30 Oct 2002Laver None NoneStatistics & Operational Research
We describe a new method for statistical pattern recognition that is based on probabilistic rule sets. The model constructs a set of first-order rules of the form IF A THEN B where the antecedent A relates to conditions on a set of predictor measurements x and the consequence B relates to changes in the odds function of the conditional probability p(y/x) for a category label y. Rule-set models are highly expressive and interpretable and a key feature of the method is the ease by which expert domain knowledge can be incorporated into the classifier system. A Bayesian framework is used which places a prior distribution over the state space of all probabilistic rule sets. Inference proceeds using stochastic simulation via tailored Markov chain Monte Carlo algorithms. The methodology is illustrated using examples taken from the machine learning literature where typically we have tens or hundreds of predictors and hundreds or thousands of observations.


Operational Research Seminar:
Inventory - too much , too little, or right on?
Geoff Relph (Manchester Business School)
24 Oct 2002Laver None NoneStatistics & Operational Research
Inventory has a major impact on business performance. How do you achieve that elusive balance - customer satisfaction and lower inventory? This talk examines some of the issues involved in better inventory planning. The concept of 'overage inventory' is defined and developed. Case research on inventory management in a manufacturing company is discussed. Options for evaluating and estimating the value of overage and determining simple prioritisation techniques for the desired corrective action needed to reduce the overage are examined. The pragmatic balance between purist academic views of inventory management and the instinctive approach often used in a small business are considered.


RSS Local Group Meeting:
Does the weather God play dice?
David Stephenson (University of Reading)
23 May 2002Laver None NoneStatistics & Operational Research
See here for abstract


RSS Local Group Meeting & AGM:
Something in the air? Multivariate analysis and atmospheric science
Ian Jolliffe (University of Aberdeen)
16 May 2002Laver None NoneStatistics & Operational Research
See here for abstract


Operational Research Seminar:
Base stock inventory policies
Mundappa Pakkala (University of Mangalore)
2 May 2002Laver None NoneStatistics & Operational Research
We consider an inventory model in continuous time. Demand follows a Poisson process and demand during a stockout is backordered. The stock level is controlled by means of a base stock policy in which the balance of physical stock plus stock on order less backorders is maintained at the base stock level. Therefore if a demand occurs then an order for replacement stock is placed immediately. The time for a replacement order to arrive is the lead time - this may be fixed or it may vary. We consider here two variants on this basic policy. First, multi-item demand processes. Second, modelling the cash flows. In each case we discuss the context of the problem, its mathematical formulation and a procedure for finding the optimal solution.


Operational Research Seminar:
Locating ambulances in Riyadh: theoretical developments and practical application
Graham Rand (University of Lancaster)
14 Mar 2002Laver None NoneStatistics & Operational Research
The location of Emegency Medical Services (EMS) is an important problem. Good locations, enabling rapid response, can save lives. Typical OR modelling for these problems tries to improve coverage which is defined as the ability to travel from a service station to a demand point in a pre-specified time. A model was developed to evaluate locations for the Saudi Arabian Red Crescent Society (SARCS), Riyadh City, Saudi Arabia. In this model the usual 0-1 coverage definition (i.e. the demand is covered or not) is replaced by the probability of covering a demand within the target time. Second, once the locations are determined, the minimum number of vehicles at each location that satisfies the required performance levels is determined. Thus, the problem of identifying the optimal locations of a pre-specified number of emergency medical service (EMS) stations is addressed by goal programming. The first goal is to locate these stations so the maximum expected demand can be reached within a pre-specified target time. Then, the second goal is to ensure that any demand arising located within the service area of the station will find at least one vehicle, such as an ambulance, available. Erlang's loss formula is used to identify the arrival rates when it is necessary to add an ambulance in order to maintain the performance level for the availability of ambulances. The use of the model for the Riyadh EMS will be described. This work was undertaken jointly with Othman Alsalloum


RSS Local Group Meeting:
Inference in fMRI experiments using spectral domain methods
Jonathan Marchini (University of Oxford)
28 Feb 2002Laver None NoneStatistics & Operational Research
See here for abstract


Postgraduate Seminar:
Lot sizing policies in an advance ordering environment
Lynette Frick (University of Exeter)
21 Feb 2002Laver None NoneStatistics & Operational Research
In most classic inventory models customer demand is either assumed to be deterministic or stochastic. In some multi-period applications, future demand is only


Statistics Seminar:
Disease mapping of stage-specific cancer incidence data
Leo Knorr-Held (University of Lancaster)
14 Feb 2002Laver None NoneStatistics & Operational Research
We propose two approaches for the spatial analysis of cancer incidence data with additional information on the stage of the disease at time of diagnosis. The two formulations are extensions of commonly used models for multicategorical response data on an ordinal scale. We include spatial and age group effects in both formulations, which we estimate in a nonparametric smooth way. More specifically, we adopt a fully Bayesian approach based on Gaussian pairwise difference priors where additional smoothing parameters are treated as unknown as well. We argue that the proposed methods are useful in monitoring the effectiveness of mass cancer screening and illustrate this through an application to data on cervical cancer in the former German Democratic Republic. The results suggest that there are large spatial differences in the stage-proportions, which indicates spatial variability with respect to the introduction and effectiveness of pap smear screening programs. This is joint work with G Rasser, University of Munich and N Becker, German Cancer Research Center Heidelberg.


Statistics Seminar:
Cross-validation in additive main effect and multiplicative interaction (AMMI) models
Carlos Tadeu dos Santos Dias (University of Sao Paulo/ESALQ)
7 Feb 2002Laver None NoneStatistics & Operational Research
The additive main effects and multiplicative interaction (AMMI) model has been proposed for the analysis of genotype/environmental data. For plant breeding, the recovery of pattern might be considered to be the principal objective of analysis. However, some problems still remain with the analysis, notably in selecting the number of multiplicative components in the model. Methods based on distributional assumptions do not have sound methodological basis, while existing data-based approaches do not optimise the cross-validation process. This talk will first summarise the AMMI model and outline the available methodology for selecting the number of multiplicative components to include in it. Then two new methods will be described that are based on a full leave-one-out procedure optimising the cross-validation process. Both methods will be illustrated and compared on some unstructured multivariate data. Finally, their application to analysis of GxE interaction will be demonstrated on experimental grain yield data.


RSS Local Group Meeting:
Anticipating catastrophes through extereme value modelling
Stuart Coles (University of Bristol)
24 Jan 2002Laver None NoneStatistics & Operational Research
See here for abstract


Statistics Seminar:
Geostatistical models and applications
Paulo Ribeiro (University of Lancaster)
17 Jan 2002Laver None NoneStatistics & Operational Research
The term 'geostatistics' identifies the part of spatial statistics which is concerned with continuous spatial variation. The term 'model-based geostatistics' was coined by Diggle, Tawn and Moyeed (1998) to mean the application of explicit parametric, stochastic models and formal, likelihood-based, methods of inference to geostatistical problems. Geostatistical methods are currently applied in a wide range of subjects and model-based methods provide further options to tackle challenging pratical problems. Motivated by some pratical applications, this talk discusses model-based geostatistical methods and their computational implementation.


Operational Research Seminar:
A bounding problem in inventory modelling
Roger Hill (University of Exeter)
10 Jan 2002Laver None NoneStatistics & Operational Research
A fundamental inventory model is the stochastic demand, periodic or continuous review, backorder model with linear holding, shortage and ordering costs and a general lead time on replenishment. It is well-established that the optimal control policy for this model is an (s,S) policy and efficient procedures exist for deriving this optimal policy. An important feature of most practical systems is that packaging and handling considerations require that replenishments must be in multiples of some unit of stock transfer q. This talk describes, in outline, the fundamental model and shows how the analysis can be adapted to allow for a general unit of stock transfer q. It finally raises some, as yet unresolved, issues on developing procedures for finding the optimal policy for this modified model.


Analysis of transplant survival rates
Dave Collett (University of Reading)
28 Nov 2001Laver None NoneStatistics & Operational Research


Should small firms be more cautious than large ones? - Dynamic programming models of operations management decisions in small firms.
Lyn Thomas (University of Southampton)
22 Nov 2001Laver None NoneStatistics & Operational Research
Operations management models, like inventory control and production levels have proved very successful in the operations of firms. However they all take as their objective the maximisation of profit or the minimisation of cost. For small firms it could be argued that maximising the probability of survival of the firm is the principal objective. This talk looks at how one can model the operations management decisions under this criterion using dynamic programming and compares the survival probability maximising decisions with the profit maximising ones. It suggests that small firms should be more cautious ( but not too cautious) than large firms.


Postgraduate Seminar: Statistical modelling of performance indicators
Paul Hewson (University of Exeter)
15 Nov 2001Laver None NoneStatistics & Operational Research
Performance Indicators are amongst the most widely published official statistics in the UK. It has been suggested that the UK central government has set over 5,000 targets against these statistics. In contrast to the wealth of numerical data available, less effort has been applied to the statistical analysis of the data. Most work performed to date has been in the educational and health fields, although considerable money has been spent evaluating data envelopment analysis for the Home Office to develop targets for police performance. Using two sets of Performance Indicators, relating to Housing Benefit Administration and Road Safety (reflecting output and outcome indicators), various methods for analysis will be reviewed, particularly approaches based upon generalised linear mixed models. Work in progress to account for the multivariate nature of the data in such models will also be described.


RSS Meeting: Modelling spatial-temporal processes for hydrology and climate
Valerie Isham (University College London)
7 Nov 2001Laver None NoneStatistics & Operational Research
A review will be given of some of the spatial-temporal models developed by an interdisciplinary team from University College and Imperial College for use in the context of hydrological design. Approaches using both point-process-based stochastic models and statistical, generalised linear, models (GLMs) will be described. A strength of the former models is their ability to represent high space-time resolution, while the latter more easily enable spatial and temporal nonstationarities to be incorporated. These models are also being used to investigate other climatological processes, such as temperature and wind speed, where there is a particular focus on questions of the influence of long-range effects (e.g., El Nino), and climate change.


RSS Meeting: Bayes 'n' drugs 'n' sporting role
Phil Brown (University of Kent)
25 Oct 2001Laver None NoneStatistics & Operational Research
A joint EU/IOC international project centered on St. Thomas's Hospital London has been looking at detection of growth hormone abuse in sport. We describe approaches to modelling multivariate markers of GH intake through time to discriminate between those that were treated with growth hormone and those on placebo in a double-blind study.


Estimating abundance from data containing many zeros
Alan Welsh (University of Southampton)
18 Oct 2001Laver None NoneStatistics & Operational Research
North East Herald Cay is a small but ecologically significant coral cay in the Coral Sea, about 350 km off the coast of Queensland, Australia. As part of the development of a monitoring program, we consider the problem of estimating the number of nests of different species of seabirds on North East Herald Cay based on surveys of 10mx10m quadrats along transects across the Cay. We consider three approaches based on different plausible models. Our main findings are that an approach based on a conditional negative binomial model which allows for additional zeros in the data works well and that a transform-both-sides regression approach produces badly biased estimates and should not be used. We discuss our experience of collecting the data, applying the methodology to the available data and discuss the implications for monitoring nesting on North East Herald Cay.


Statistics in sport and games
Frank Duckworth ('Editor, RSS News')
10 Oct 2001Laver None NoneStatistics & Operational Research


Special RSS Meeting and AGM at Plymouth University (Robbins Seminar Room 2): A Heretic's View of Placebos and Ethics in Clinical Trials
Stephen Senn ('University College, London')
31 May 2001Laver None NoneStatistics & Operational Research


Postgraduate Seminar: Using State Space models to investigate the effect of vitamin A supplement on diarrhoea
Valeska Andreozzi (University of Exeter)
24 May 2001Laver None NoneStatistics & Operational Research


Postgraduate Seminar: A Score Test for Zero-inflated Negative Binomial Models
Naratip Jansakul (University of Exeter)
24 May 2001Laver None NoneStatistics & Operational Research


The size of orders from customers, characterisation, forecasting and implications
Roy Johnston (Warwick University)
18 Apr 2001Laver None NoneStatistics & Operational Research


Using the Randomisation in Specifying the Mixed Models and ANOVA tables
Chris Brien (University of South Australia)
16 Mar 2001Laver None NoneStatistics & Operational Research


Postgraduate Seminar: Analysis of Multivariate Process Control Data
Julie Badcock (University of Exeter)
8 Mar 2001Laver None NoneStatistics & Operational Research


Postgraduate Seminar:The Evolution of Trees: Application of Genetic Algorithms to Network Optimisation
Evan Thompson (University of Exeter)
8 Mar 2001Laver None NoneStatistics & Operational Research


Use of discrete event simulation in the evaluation of screening for Helicobacter pylori for the prevention of peptic ulcers and gastric cancer.
Ruth Davis (University of Southampton)
1 Mar 2001Laver None NoneStatistics & Operational Research


RSS meeting: Estimating Mixtures of Regressions
Merrilee Hurn (University of Bath)
15 Feb 2001Laver None NoneStatistics & Operational Research


RSS meeting: Independent component analysis: flexible sources and non-stationary mixing
Richard Everson (University of Exeter (Computer Science))
25 Jan 2001Laver None NoneStatistics & Operational Research