Naïve Sampling and Format Dependence in Subjective Probability Calibration

Previous data suggest that probability judgments can often be fairly realistic, but production of probability intervals promotes extreme overconfidence bias (Juslin & Persson 2002; Juslin, Wennerholm, & Olsson, 1999; Klayman, Soll, González-Vallejo, & Barlas, 1999). We present a novel explanation of this format-dependence effect in terms of a naïve sampling model. The model assumes that people process distributional information in an unbiased manner, but they are naïve in the sense that they uncritically take sample properties as estimates of population properties. A Monte Carlo simulation demonstrates that the model predicts format dependence, with extreme overconfidence bias for interval production that practically disappears for probability assessments for the same intervals. In the last section this novel prediction is empirically verified by data from human participants.