Consistent Underestimation in the Intuitive Summation of Monetary Amounts

Many economic decisions rely on a fast, intuitive system of numerical cognition. When trying to judge the sum of a sequence of numbers, this system produces non-random errors: on average, it underestimates. This underestimation bias is thought to be caused by a compressive scaling of numbers when they are encoded internally. We present two preregistered, incentive-compatible experiments that tested the economic relevance of the underestimation bias. We varied the economic frame of sequences to be summed and deployed both judgment and forced-choice elicitation methods. Experiment 1 (n = 104) showed significant underestimation in the judgment task, with an overall mean bias of approximately -6%. Experiment 2 (n = 501) recorded persistent underestimation bias in both judgment and forced-choice tasks. Clear learning effects imply the effect size likely marks a lower bound. These findings have implications for modelling the cognitive foundations of economic preferences. They also provide insight into how firms' pricing structures can exacerbate biases, causing economic loss.