Multidimensional Ellsberg

The classical Ellsberg experiment presents individuals with a choice problem in which the probability of winning a prize is unknown (ambiguous). In this paper we study how individuals make choices between gambles in which the ambiguity is in di↵erent dimensions: the winning probability, the amount of the prize and the payment date, and many combinations. While the decision-theoretic models accommodate a rich variety of behaviors, we present experimental evidence that points at one systematic behavioral pattern: (i) no ambiguity is preferred to ambiguity on any single dimension and to ambiguity on multiple dimensions, and (ii) “correlated” ambiguity on multiple dimensions is preferred to ambiguity on any single dimension. JEL Classification: C91, D11, D81.