Multiple priors and comparative ignorance

Fox and Tversky (1995) observe comparative ignorance: people are more ambiguity averse when they evaluate ambiguous and clear prospects jointly rather than in isolation. To accommodate such patterns, I relax the multiple priors model and allow subjective sets of priors to depend on partitions that are generated by feasible prospects. All sets of priors are derived uniquely on the corresponding partitions. This model violates transitivity, but preserves all other axioms of Gilboa and Schmeidler (1989). A parsimonious special case identifies a threshold partition π such that all π-measurable prospects are ranked via a set of priors P, but all other comparisons rely on a distinct set P⁎. Another refinement derives partition-dependent probabilistic beliefs without imposing ambiguity aversion.