Some approximations useful to the use of dependent information sources

Abstract

Abstract Expert judgments have been frequently utilized in probabilistic risk assessments. The aggregation methods used have often been very simplistic and have ignored the critical issues of incorporating experts' biases and inter-expert dependence. On the other hand, theoretical methods available address some of these issues, but suffer from the fact that the input requirements are too demanding. This paper provides practical guidance on the use of the versatile Bayesian expert judgment aggregation model. Guidance is provided by conducting a sensitivity analysis of the experts' weights and posterior standard deviation on the parameters of the Bayesian aggregation model. Experts' weights and posterior standard deviation are most sensitive to experts' standard deviations and are not as sensitive to inter-expert dependence. Experts' location biases influence the posterior mean, but not the posterior variance and weights. The results obtained are also applicable to the more commonly used judgment aggregation procedures, such as simple geometric averaging, because the Bayesian aggregation method encompasses these simple techniques. The results are illustrated with the help of an example and suggest that the uncertainty is underestimated if expert biases are ignored.