David Spiegelhalter - Research Interests

Research Interests

Spiegelhalter's research interests are in

• Bayesian approach to clinical trials, expert systems and complex modelling and epidemiology.
• Graphical models of conditional independence. He wrote several papers in the 1980s that showed how probability could be incorporated into expert systems, a problem that seemed intractable at the time. Spiegelhalter showed that while frequentist probability did not lend itself to expert systems, Bayesian probability most certainly did.
• Statistical software. In the 1990s Spiegelhalter led the Medical Research Council team that developed WinBUGS ("Bayesian analysis Using Gibbs Sampling"), a statistical-modeling system allowing hierarchical prior distributions. WinBUGS and its successor OpenBUGS specifies graphical models using acyclic directed graphs whose nodes are random variables, which are updated using Gibbs sampling (an updating method for Markov chain Monte Carlo (MCMC) simulation). Earlier Bayesian software had required that the probability distribution for the observed data be an exponential family and that the prior be its conjugate distribution. Allowing flexible choices of prior distributions simplified hierarchical modeling and helped to promote multilevel models, which became widely used in epidemiology and education.
• General issues in clinical trials, including cluster randomisation, meta-analysis and ethical monitoring.
• Monitoring and comparing clinical and public-health outcomes and their associated publication as performance indicators.
• Public understanding of risk, including promoting concepts such as the micromort (a one in a million chance of death). Media reporting of statistics, risk and probability and the wider conception of uncertainty as going beyond what is measured to model uncertainty, the unknown and the unmeasurable.