Bayesian Probability Boxes in Risk Assessment

Refined risk assessments should increase realism compared with the first tier deterministic risk assessment. This may involve using probabilistic methods which account separately for uncertainty and variability. Analysts use cumulative distribution functions to represent variability, and bounds around these to illustrate uncertainty. In probability bounds analysis, parametric probability boxes (p-boxes) are usually formed using intervals for each parameter. In this paper a Bayesian framework is adopted, which takes account of dependencies between parameters. Bayesian p-boxes use imprecision represented by bounds to summarise the uncertainty surrounding the risk distribution parameters.