The Axiomatic Justification of the Transferable Belief Model

Belief functions have recently been advocated as an alternative to probability functions for representing quantified belief. This new normative model has several merits, but these merits are not sufficient to justify its use. Some ‘axiomatic’ justification is also needed. Indeed the examination of the requirements that underlie the normative models of subjective behaviors provides usually the best if not the only tool to compare them. We present such a set of axioms. In order to show that belief functions are appropriate for representing quantified beliefs, we present and analyze the requirements that should be satisfied when conditioning is introduced and when the domain on which beliefs are assessed changes. The deduced model corresponds to the transferable belief model, i.e. a model for quantified beliefs based on belief functions and independent of any underlying probability model.