Indeterminate probabilities on finite sets (Q1208641)

The representation of beliefs by upper and lower probabilities, something that has received much attention of late, is extended here by supposing that these probabilities are qualified by numerical confidence weights. These weights enable a trade-off to be made among probability judgements, thus allowing the reconciliation of inconsistencies. An axiom system for binary preferences, in which completeness is not assumed and transitivity is weakened, leads to the desired representation, while a generalization of de Finetti's operational method leads to the elicitation of confidence-weighted probabilities. Applications of the model to the reconciliation of inconsistent probability judgements and to Bayesian sensitivity analysis are presented.