Klaus Keller (KK) on 10/8/2004

231 words) We report results of a series of experiments on decision-making in the presence of irreducibly imprecise probabilities. Subjects faced decisions among actions where the payoffs depend on the probability of drawing balls from an urn whose composition was not fully known. Consistent with the vagueness avoidance hypothesis, Decision Makers (DMs) displayed systematic preferences for riskless actions even at a high premium. This tendency grew with increased vagueness, characterized by the width of the interval of plausible probabilities. We compared two decision aids that portray these imprecise probabilities in different ways: One aid calculates the expected value of alternative actions contingent on probability distributions provided by the DMs, and the other graphically compares the expected value of actions over the entire range of plausible probabilities. Access to either decision aid significantly reduced vagueness avoidance, but the type of aid used influenced systematically the way DMs approached the problem. We compared the DMs‟ choices to predictions of decision models for decision under ignorance and under risk. We found support for the conservative maxi-min criterion, but a subjective expected utility model with probabilities inferred from the partial information available also performed well, especially for low levels of vagueness and in the presence of decision aids. These findings suggest some initial implications for the debate over how to best characterize imprecise probabilistic information for policy-makers involved with decisions involving irreducible uncertainties, such as climate change.