IID: independently and indistinguishably distributed

The inability of the Bayesian model to accommodate Ellsberg-type behavior is well-known. This paper focuses on another limitation of the Bayesian model, specific to a dynamic setting, namely the inability to permit a distinction between experiments that are identical and those that are only indistinguishable. It is shown that such a distinction is afforded by recursive multiple-priors utility. Two related technical contributions are the proof of a strong LLN for recursive multiple-priors utility and the extension to sets of priors of the notion of regularity of a probability measure. (C) 2003 Elsevier Science (USA). All rights reserved.