Expert judgement models in quantitative risk assessment

Expert judgement is a valuable source of information in risk management. Especially, risk-based decision making relies significantly on quantitative risk assessment, which requires numerical data describing the initiator event frequencies and conditional probabilities in the risk model. This data is seldom found in databases and has to be elicited from qualified experts. In this report, we discuss some modelling approaches to expert judgement in risk modelling. A classical and a Bayesian expert model is presented and applied to real case expert judgement data. The cornerstone in the models is the log-normal distribution, which is argued to be a satisfactory choice for modelling degree-of-belief type probability distributions with respect to the unknown parameters in a risk model. Expert judgements are qualified according to bias, dispersion, and dependency, which are treated differently in the classical and Bayesian approaches. The differences are pointed out and related to the application task. Differences in the results obtained from the different approaches, as applied to real case expert judgement data, are discussed. Also, the role of a degree-of-belief type probability in risk decision making is discussed.