Eliciting engineering judgments in human reliability assessment

The data scarcity is one of the main problems for human reliability analyses. This theme has also emerged from the introduction of sophisticated techniques into the area, such as the formalism of Bayesian belief networks that permit the utilization of multidisciplinary sources of information. Such flexibility makes possible more accurate models which in many cases may require data accessible only qualitatively. In these cases, an alternative way is to elicit such information through protocols directed to experts. Among the methods cited in the literature, the one proposed by (G.C. Nadler et al., 2001) promotes (1) the self-knowledge of the expert about his numeric beliefs and (2) indicators that ensure mathematically the quality of both the questionnaire for elicitation and the expert opinions. However, the high number of questions may render such method inappropriate for quantifying a large number of parameters. Furthermore, the quality indicators proposed by (G.C. Nadler et al., 2001) do not guarantee reliable results when the questionnaire is reduced. Therefore, this paper proposes a quality indicator that, when combined with the known ones, allows for an adequate quantification of qualitative knowledge about human reliability problems with a reduced number of questions. It is also presented cost reduction results obtained from the application of the proposed approach in two real problems