Mobile safety cameras: estimating casualty reductions and the demand for secondary healthcare (2013)

Author(s): Fawcett L, Thorpe N

    Abstract: We consider a fully Bayesian analysis of road casualty data at 56 designated mobile safety camera sites in the Northumbria Police Force area in the UK. It is well documented that regression to the mean (RTM) can exaggerate the effectiveness of road safety measures, and, since the 1980s, an empirical Bayes estimation framework has become the standard tool for separating real treatment effects from those of RTM. In this paper, we show that, relative to a fully Bayesian treatment, the empirical Bayes method is over-optimistic when quantifying the variability of estimates of casualty frequency. More realistically, a fully Bayesian analysis allows population-level estimates to contribute to the uncertainty in such estimates of casualty frequency. Implementing a fully Bayesian analysis via Markov chain Monte Carlo also provides a more flexible and complete inferential procedure. We assess the sensitivity of estimates of treatment effectiveness, as well as the expected monetary value of prevention owing to the implementation of the safety cameras, to different model specifications, which include the construction of informative priors for some parameters.

    • Type of Article: Research paper
    • Date: 15-07-2013
    • Journal: Journal of Applied Statistics
    • Volume: 40
    • Issue: 11
    • Pages: 2385-2406
    • Publisher: Routledge
    • Publication type: Article
    • Bibliographic status: Published

    Keywords: Markov chain Monte Carlo, mobile safety cameras, negative binomial distribution, Northumbria Safety Camera Partnership, regression to the mean.


    Dr Lee Fawcett
    Lecturer in Statistics