Decision makers weight small probabilities differently when sampling them and when seeing them stated. We disentangle to what extent the gap is due to how decision makers receive information (through description or experience), the literature’s prevailing focus, and what information they receive (population probabilities or sample frequencies), our novel explanation. The latter determines statistical confidence, the extent to which one can know that a choice is superior in expectation. Two lab studies, as well as a review of prior work, reveal sample decisions to respond to statistical confidence. More strongly, in fact, than decisions based on population probabilities, leading to higher payoffs in expectation. Our research thus not only offers a more robust method for identifying description-experience gaps. It also reveals how probability weighting in decisions based on samples — the typical format of real-world decisions — may actually come closer to an unbiased ideal than decisions based on fully specified probabilities — the format frequently used in decision science.
