Evidence-informed decision-making is important in every domain. In a recent project with The National Archives, we built a decision support tool which allowed them to make a business case for investment in increased security of digital archives. Digital archives, and the materials held in them, are rich, complex and fragile. They are under threat from rapidly evolving technology, outdated policies and a skills gap across the archives sector internationally. However, since preserving digital objects is a relatively recent phenomenon in archival terms, a systematic study of the nature and relative importance of the risk structures had not been done. There was no consensus or consistency about relevant data collection, leading to data gaps.
Building on a recently-developed Bayesian paradigm, the Integrating Decision Support System developed here at Warwick, we partnered with The National Archives, and other archives across the world to undertake a decision analysis and encode the results into a tool (DiAGRAM) to evaluate an archive’s risk profile and designed for easy use by archives nation- and world-wide.
Space Domain Awareness is a rapidly growing field due to the increased risks that space assets face as a consequence of the increasing number of space debris orbiting Earth. Evaluating such risks in a principled manner requires distinguishing and propagating different sources of uncertainty for each of the hundreds of thousands of debris. A delicate balance between risks of collision and unnecessary manoeuvres must be found since the latter uses up the limited amount of fuel that satellites carry, hence reducing their lifetime. With satellites and debris being subjected to different perturbations depending on their orbit and with the limited observability that sensors such as telescopes provide, the amount of uncertainty on the location of space objects can be significant, hence requiring advanced statistical methodologies. Space Domain Awareness therefore provides a truly complex system on which modern society is largely reliant, making the associated risk management crucial but challenging.
National security and policing has made increasing use of sophisticated AI and decision analytic tools to detect criminal and adversarial activities and better protect its own personnel. Over the last 15 years Warwick Statistics and AS&RU have been engaging in a number of frontier projects with different clients to support these developments. All these methods have used Bayesian Decision analyses - a branch of statistics for which Warwick Statistics has played a leading role over many years - to formally and effectively integrate expert judgements with streaming but patchy data sources. These technological transfers began with the study of how the MoD could better communicate command and control directives to their officers. We then worked with the Home Office to develop better ways of appraising the effectiveness of government Prevent policies so that resourcing could be more cost effective. Over the last 5 years this work has accelerated. Working on a number of projects with Turing and GCHQ teams within AS&RU have devised tools for monitoring and frustrating terrorist attacks both by individuals and by gangs. Another stream of work with GCHQ has produced fast way of detecting and frustrating computer network attacks designed to steal documents - a crime called exfiltration. Meanwhile work has continued between AS&RU and MoD in a project which uses sophisticated algorithms that merge information from a number of different sources in order to help detect those manufacturing plants most likely to be guilty of producing large quantities of illicit drugs.
Statistical insight and evidence are increasingly required in court proceedings. We have been involved in training in statistics and epidemiology for lawyers and co-authored best practice guides for lawyers. We have also contributed directly to legal decisions by the production of expert reports; these have included the assessment of epidemiological and clinical trial evidence in a number of legal cases, using our research experience to review evidence. We also produce expert witness reports to estimate life expectancy in a broad range of personal injury cases.
Modelling of energy procurement requires the consideration of several factors such as the weather, climate, energy supply and energy demand. Energy supply itself comes from a variety of sources such as from conventional sources, renewables such as solar, hydro, and wind, and from interconnections with other power grids. An important part of energy security involves effective modelling of energy procurement to meet future demand. Such capacity procurement is typically based on cost-benefit analysis with respect to standard risk metrics which are calculated using a limited number of recent historic years. These approaches are clearly unable to account for extremes in net demand (demand minus supply; for the GB setting, demand minus wind energy) especially for atypical extremes for which there is very sparse or no information available in the historic data, e.g. extremes caused by a changing climate.
To begin to deal with this issue, in collaboration with the University of Edinburgh and the Alan Turing Institute, we proposed a proof of concept design for a mixture model with two components - each of which are pair-copula Bayesian networks - where the first component models the typical extremes and is driven by domain knowledge and historical data, whereas the second component models the atypical extremes and would need to be elicited from domain experts.