Using Belief Functions to Represent Degrees of Belief

Abstract Belief functions are an alternative or supplement to probabilities for representing the degree to which we believe in various hypotheses, highlighting different aspects of subjects′ uses of evidence. In particular, belief functions have potential for situations in which the belief assessor is interested in judging the degree of support or justification that the evidence affords hypotheses, e.g., in legal situations. Noting that there is a lack of empirical evidence testing the theoretical claims in favor of using belief functions, we constructed an experiment to determine (a) if subjects could be trained in the meanings of belief-function responses; and, (b) once trained, how they use those belief functions in a hypothetical legal setting. We found that subjects could use belief functions, identified limits to belief functions′ descriptive representativeness, and discovered patterns in subjects′ uses of belief functions that inform our understanding of subjects′ uses of evidence.