Reducing Overconfidence in the Interval Judgments of Experts

摘要: Elicitation of expert opinion is important for risk analysis when only limited data are available. Expert often elicited in the form subjective confidence intervals; however, these prone to substantial overconfidence. We investigated influence elicitation question format, particular number steps procedure. In a 3-point procedure, an asked lower limit, upper and best guess, two limits creating interval some assigned level (e.g., 80%). our 4-step experts were also realistic but no was assigned; fourth step rate their anticipated produced. three studies, made predictions rates infectious diseases (Study 1, n = 21 Study 2, 24: epidemiologists public health experts), or marine invertebrate populations 3, 34: ecologists biologists). combined results from studies using meta-analysis, which found average overconfidence 11.9%, 95% CI [3.5, 20.3] (a hit 68.1% 80% intervals)-a decrease compared with previous studies. Studies 2 3 suggest that procedure more likely reduce than (Cohen's d 0.61, [0.04, 1.18]).