Ulric B. and Evelyn L. Bray Social Sciences Seminar

Abstract: People often make simplifying assumptions to understand complicated systems. This paper connects this observation with the human tendency toward overprecision: excessive certainty in the accuracy of one's beliefs. Intuitively, simplified models ignore some variation-inducing possibilities, which causes people to disregard this variation in their predictions, which causes them to be over-precise. We first develop a theoretical model of people who properly recognize uncertainty conditional on the assumptions they make, but do not fully appreciate how their assumptions affect their beliefs. This simple model leads to a variety of predictions, such as concrete connections between model uncertainty, overprecision, and disagreement. We explore these predictions in an experimental setting where we can cleanly vary within-model and across-model uncertainty. Consistent with the theory, subjects respond approximately correctly to increases in within-model uncertainty, but mostly ignore across-model uncertainty, leading to our predicted connections. Finally, we analyze observational data from the Survey of Professional Forecasters, finding that forecasters are overprecise, and validating the connection between overprecision and disagreement.