Eliciting Model Structures for Multivariate Probabilistic Risk Analysis

Frontiers in Applied Mathematics and Statistics

Mark Burgman¹  Hannah Layman¹  Simon French² 

[1] London, United Kingdom;Manchester, United Kingdom;

关键词: parameters;    model structures;    uncertainty;    elicitation;    cognitive bias;   

DOI  :  10.3389/fams.2021.668037

来源: Frontiers

【 摘 要 】

Notionally objective probabilistic risk models, built around ideas of cause and effect, are used to predict impacts and evaluate trade-offs. In this paper, we focus on the use of expert judgement to fill gaps left by insufficient data and understanding. Psychological and contextual phenomena such as anchoring, availability bias, confirmation bias and overconfidence are pervasive and have powerful effects on individual judgements. Research across a range of fields has found that groups have access to more diverse information and ways of thinking about problems, and routinely outperform credentialled individuals on judgement and prediction tasks. In structured group elicitation, individuals make initial independent judgements, opinions are respected, participants consider the judgements made by others, and they may have the opportunity to reconsider and revise their initial estimates. Estimates may be aggregated using behavioural, mathematical or combined approaches. In contrast, mathematical modelers have been slower to accept that the host of psychological frailties and contextual biases that afflict judgements about parameters and events may also influence model assumptions and structures. Few, if any, quantitative risk analyses embrace sources of uncertainty comprehensively. However, several recent innovations aim to anticipate behavioural and social biases in model construction and to mitigate their effects. In this paper, we outline approaches to eliciting and combining alternative ideas of cause and effect. We discuss the translation of ideas into equations and assumptions, assessing the potential for psychological and social factors to affect the construction of models. We outline the strengths and weaknesses of recent advances in structured, group-based model construction that may accommodate a variety of understandings about cause and effect.

【 授权许可】

CC BY