2.1.1. The Classical Model

The Classical Model (CM) was developed by Cooke [19] and represents an established method for the elicitation of quantitative values. Instead of qualitative approximations, the experts are asked to roughly estimate a value of a given parameter on a continuous scale, e.g., "From a fleet of 100 new aircraft engines how many will fail before 1000 h of operations?" [20]. This approach advocates uncertainty quantification in the form of subjective probability distributions by eliciting from the experts distribution percentiles typically at 5%, 50% and 95% [20,26]. These allow the fitting of a "minimally informative non-parametric distribution" [26] for the estimation.

In order to cope with the empirical control requirement defined by Cooke [19], the method involves the attribution of an individual weight to each expert, called calibration. This is executed by preparing two sets of questions—the so-called seed and target questions, incorporating the same types of elicitation heuristics. The seed evaluations involve quantitative inquiries with known correct answers and are used to "assess formally and auditably" [20] the expert's deviation from the precise solution in a domain, denoted as a calibration score. However, even a perfect calibration does not stand for an informative decision in the form of a narrow probability distribution around the precise value. For this purpose, the information score is introduced which reflects the precision of the answer. Then, the correction factors or weights for each expert represent a combination of the calibration and the information score. The target questions refer to the information of interest. Hence, in order to obtain elicited data with highest possible precision, the experts' assessments are mapped with their corresponding weights.

The CM advocates mathematical aggregation of the varying expert evaluations, e.g., in the form of a Cumulative Distribution Function (see Reference [20] for more detail).
