A Variation on Ellsberg∗

Ellsberg’s experiment involved a gamble with no ambiguity (N) and a gamble where the prize that could be won is objectively known, but the winning probability depends on the (ambiguous) urn’s composition (P). We extend this by including a gamble where the winning probability is objectively known, but the prize depends on the urn’s composition (C), and also gambles where both the probability and the prize depend on the urn’s composition, and can either be correlated positively (D) or negatively (M). Among transitive subjects who prefer N to P, 40% prefer D to N, 74% prefer D to P, 97% prefer D to M, and the modal ranking (about 39%) satisfies D<N<P,C. We show that this behavior is compatible with the Max-Min Expected Utility model if every prior in the set of priors has a high enough variance, a property that we call ‘skeptical