February 2012 Event

Our February 2012 event was held on Wednesday 29th from 4-6pm. The event was held in the Poldhu room within the Kay building (No 24. on the campus map).

The event comprised a series of 15-minute presentations followed by refreshments and an opportunity to meet other participants.

Copies of the slides can be obtained by following the appropriate links below.




Martin Pitts

Using quantitative data in (Roman) archaeology

Applying statistical methodologies to fragmentary and necessarily incomplete archaeological data poses a range of challenges. In this presentation I briefly outline how the application of basic as well as multivariate methods can shed new light on such data, from pottery to human remains, with particular reference to addressing interdisciplinary themes such as identity, imperialism and health inequalities.

Slides are available as pdf.


Bruce Ingleby

Problems with modelling errors in short-range weather forecasts

Weather forecasts need the combination of observational information with short range forecasts (data assimilation) and as part of this we need a model of the errors in the short range forecasts.  In setting up this model we use training data from two sets of forecasts and treat the temperature and humidity differences between them as a proxy for forecast error.  We can predict part of the humidity error from the temperature error but the nature of this relationship changes with the relative humidity. Here, we're particularly interested in this relationship for low relative humidity levels where the regression models that we use appear to break down. In this talk we will demonstrate the difficulties we've encountered and invite audience members to suggest alternative approaches that might help overcome them.

Slides are available as pdf.


Martin Hoyle

Cost-effectiveness and meta-analysis of clinical evidence of drugs and medical devices for use on the NHS

PenTAG, the Peninsula Technology Assessment Group, is one of nine university-based research groups contracted to produce high quality reviews of the clinical and cost-effectiveness of drugs and medical devices for NICE, the National Institute for Health and Clinical Excellence, in the UK.  NICE's decisions on whether to recommend treatments for use on the NHS are based heavily on the results of these reviews.  We have worked on many diseases, including Alzheimer's disease, chronic myeloid leukaemia, colorectal cancer and renal cell carcinoma.  Cost-effectiveness is usually assessed by Markov cohort models, but sometimes discrete event simulations.  Statistical techniques include survival analysis (very common), regression and meta-analysis using WinBUGs.  The methodology for conducting cost-effectiveness analyses is constantly evolving.

Slides are available as pdf.


Tim Naylor


Astronomers tend to have "cost limited" datasets, and so it is important that they make the bestuse of the data they have got, as a result of which they tend to have a good grounding in statistics.However, there is a sharp divide between the cosmologists, who now almost exclusively use fully Bayesian analysis, and the stellar astrophysicists who tend to have more empirical methodsfor data exploration.  In this short talk I will give a few examples of the sorts of analysis which arebeing undertaken, to give a flavour of where astronomy could make contributions to the broaderstatistics field.

Slides are available as pdf.


Peter Wingfield Digby

Occupational Injury Statistics – an example of official statistics at home and abroad.

Peter Wingfield-Digby, an independent statistical consultant, has spent all his working life in official statistics. He will briefly review the development of official statistics in this country, and concentrate on how one topic – statistics on occupational injuries – has been dealt with both here and overseas.

Slides are available as pdf‌.

Please contact us at events@exista.org if you have any questions regarding this event.