Proceedings of the Annual Meeting of the Cognitive Science Society

Research in decision-making has recently begun to empha-size predictive accuracy as the dominant principle for design-ing and evaluating choice models. This emphasis has led tothe development of increasingly more precise models of hu-mans’ risk preferences, as measured in certain experimentalparadigms built upon certainty equivalence testing. In thispaper, we argue that the level of precision attained by recentchoice models is unexpected, because human preferences areirreducibly noisy. We support this argument by conducting ex-periments to measure intra-observer consistency in choice be-havior in two common risk preference paradigms: decisionsfrom description and experience. We find that while currentchoice models of decisions from experience align fairly wellwith the upper limits of choice consistency seen in our experi-mental data, choice models for decisions from description aresignificantly more consistent with humans’ choices than thehumans themselves are consistent with their own choices. Wediscuss some theoretical and practical implications of our re-sults.