Bi-directional probabilistic assessment

Abstract

Conditioning has been shown to be one approach for improving subjective probabilistic assessments of a target uncertainty [5]. In practice, conditioning on only one distinction is often employed in an attempt to improve the quality of the assessment while minimizing the level of effort for constructing a distribution for the target uncertainty. Once the actual assessment begins it is not uncommon that the expert realizes that much of his knowledge and beliefs are represented best in the reverse conditioning order. By assessing both conditional orderings (bi-directional assessment) and subsequently “balancing” the two perspectives while reducing assessment errors, a better representation of the expert's knowledge and beliefs is constructed than by a conventional, uni-directional or direct assessment. After the expert has attempted to reduce the assessment errors, total reconciliation of the distributions can be achieved through nonlinear programming.