The Role of Mathematical Models in Expert Opinion Utilization. G. E. Apostolakis, MIT, Room 24-221, Cambridge, MA 02139-4307
Using expert judgment in risk assessments is necessary when the available evidence is subject to a number of interpretations or is very weak. The literature contains a large number of papers that propose mathematical models for processing expert opinions. For example, in a Bayesian framework, expert judgment is treated as evidence and is processed like any other kind of evidence, i.e., via Bayes’ theorem. Risk assessors have realized, however, that these models do not capture all the relevant features of the problem. For example, modeling the dependence between experts is a very difficult issue and using multivariate normal or lognormal distributions in the likelihood function of Bayes’ theorem involves some very restrictive assumptions regarding expert correlations. The complete rejection of mathematical models in favor of the so-called behavioral aggregation models is not a wise choice, however, because the great value of mathematical models is the discipline they impose on the analyst. A combination of the two approaches seems to be the best practical way of aggregating expert judgments. The results of a number of mathematical models are presented to the stakeholders (who may be the experts themselves) and they form the starting point of an informed deliberation that eventually leads to aggregation. This framework will be discussed at the symposium, as well as a recent project on seismic risk assessment that proposed a process for structuring the elicitation and quantification process.