Common problems in the elicitation and analysis of expert opinion affecting probabilistic safety assessments
Expert opinion is frequently used in probabilistic safety assessment (PSA), particularly in estimating low probability events. In this paper, we discuss some of the common problems encountered in eliciting and analyzing expert opinion data and offer solutions or recommendations. The problems are: that experts are not naturally Bayesian. People fail to update their existing information to account for new information as it becomes available, as would be predicted by the Bayesian philosophy; that experts cannot be fully calibrated. To calibrate experts, the feedback from the known quantities must be immediate, frequent, and specific to the task; that experts are limited in the number of things that they can mentally juggle at a time to 7 {plus minus} 2; that data gatherers and analysts can introduce bias by unintentionally causing an altering of the expert's thinking or answers; that the level of detail the data, or granularity, can affect the analyses; and the conditioning effect poses difficulties in gathering and analyzing of the expert data. The data that the expert gives can be conditioned on a variety of factors that can affect the analysis and the interpretation of the results. 31 refs.
- Research Organization:
- Los Alamos National Laboratory (LANL), Los Alamos, NM (United States)
- Sponsoring Organization:
- USDOD
- DOE Contract Number:
- W-7405-ENG-36
- OSTI ID:
- 6675899
- Report Number(s):
- LA-UR-90-2604; CONF-9009226-1; ON: DE90014975
- Resource Relation:
- Conference: Committee on safety of nuclear installations (CSNI), Santa Fe, NM (USA), 4-6 Sep 1990
- Country of Publication:
- United States
- Language:
- English
Similar Records
Parameterizing Bayesian network Representations of Social-Behavioral Models by Expert Elicitation
Experts in uncertainty: Opinion and subjective probability in science