My main area of research lies in the field of extreme value theory. In particular, over the last few years my research has focussed on the estimation of environmental extremes in the presence of serial correlation, with applications to wind speed, sea-surge and rainfall extremes. The JRSSB (read) paper by Davison and Smith (1990) resulted in a shift away from the more traditional analysis of extremes, using block maxima, towards more efficient threshold-based procedures. However, the most widely used pragmatic solution to strong short-term serial correlation - runs declustering - can be overly wasteful of data, often resulting in very large estimates of variability attached to key quantities of interest (e.g. return levels) and impractically wide associated confidence intervals. Fawcett and Walshaw (2006), Fawcett and Walshaw (2007), Fawcett and Walshaw (2008) and Fawcett and Walshaw (2012) consider various approaches for making better use of our data on extremes, including inference within the Bayesian paradigm; a review of these methods, with a practical guide for estimating return levels, is given in Fawcett and Walshaw's (2013) review paper. I am currently collaborating with Dave Walshaw and Simone Padoan, investigating the use of spatial models for extreme rainfall within regions of the UK.
Since 2008 I have collaborated with colleagues in the Transport Operations Research Group to investigate the role of fully Bayesian methods in assessing the effectiveness of road safety schemes, with a particular focus on mobile safety cameras in the Northumbria police force area. Such safety remedial measures are often implemented after a run of abnormally high casualty counts, and so any simple before-after study is bound to be subject to the effects of regression-to-the-mean (RTM). The current "gold standard" is to use an empirical Bayes procedure to separate real treatment effects from those of RTM. However, work in Thorpe and Fawcett (2012) and Fawcett and Thorpe (2013) highlights the restrictions of such an approach, relative to a fully Bayesian analysis; in particular, over-optimistic estimates of variability attached to predictions of casualty frequency, and the restriction to possibly unrealistic model structures. This work with Neil Thorpe is ongoing, in partnership with the North East Traffic & Accident Data Unit (TADU) and PTV Group in Karlsruhe, Germany. See the recent discussion of our work in the magazine Local Transport Today.
In December 2014 Neil Thorpe and I were awarded a University Strategic Development Grant for Impact Generating Activity, to develop software tools for identifying road traffic accident hotspots. Click here for more information.
Although most of my methodological research has it's roots in real-world, applied problems (see above), my collaborative work with colleagues in the Transport Operations Research Group and the Food and Human Nutrition Research Centre has led to publications and reports in the non-stats literature. Recent work includes investigating data linkage beteen police and hospital records of road casualty accidents (Thorpe and Fawcett, 2012); assessing the effectiveness of a new type of pig feed on grower-finisher Landrace X pigs; and investigating biomarkers of wholegrain food intake (Haldar et al., 2010).
I am also interested in researching innovative teaching methods. In 2008, I was involved in a project to assess the effectiveness of computer-based assessments in a large statistics service course (Fawcett et al., 2008), and I am currently the lead investigator on a University-funded teaching innovation project looking at case-based learning and teaching methods in Statistics service courses. I also recently took part in a University College London/HE Academy-funded project examining the role of the "teaching researcher".
Recent conference talks/posters
A Novel Approach to Collision Hotspot Identification Accounting for Regression-To-the-Mean and Trend, Transport Practitioners' Meeting, London, July 2015.
Hierarchical Models for Environmental Extremes, 3rd Meeting of the Nordic-Baltic Biometric Society, Turku, Finland, June 2011 (invited).
Bayesian Inference for Clustered Extremes, 20th Annual Meeting of the International Environmetrics Society, University of Bologna Paolo Fortunati, Italy, July 2009 (invited).
A Hierarchical Model for Extreme Wind Speeds, 9th ISBA World Meeting, Hamilton Island, Australia, July 2008.
Hierarchical Models for Extreme Wind Speeds, 19th Annual Meeting of the International Environmetrics Society, University of British Columbia Okanagen, Kelowna, Canada, June 2008 (invited).
Short course: Modelling Environmental Extremes, University of British Columbia Okanagen, Kelowna, Canada, June 2008.
Estimating Casualty Reductions from Road Safety Measures, Transport Operations Research Group, School of Civil Engineering and Geosciences, Newcastle University, March 2008 (invited).
Improved Estimation for Temporally Clustered Extremes, 18th Annual Meeting of the International Environmetrics Society, Mikulov, Czech Republic, August 2007 (invited).
A hierarchical Model for Extreme Wind Speeds, 1st Joint Meeing of the IMS/IBSA, San Juan, Puerto Rico, July 2003.
I have given many talks to the Applied Probability and Statistics (ASP) group in the School of Maths & Stats over the last few years, as well as the (now defunct) Bayesian Statistics group.My two most recent are available here: June 2013 and March 2012.