Overprecision is a ubiquitous bias in risk and decision analysis.
•
Several debiasing tools have been proposed but they lack a systematic comparison.
•
We compare three tools for debiasing overprecision and two elicitation protocols.
•
Auto stretching the tails with the fixed value protocol was more effective.
•
We provide recommendations on how to implement this debiasing strategy.
Abstract
Formal expert elicitation is a widely used method for quantifying uncertain variables in decision and risk analysis. When estimating uncertain variables, experts and laypeople exhibit overprecision, meaning that the ranges of their estimates are too narrow. Overprecision, a form of overconfidence, is pervasive and hard to correct, thus posing a challenge to expert elicitation. Following the increasing interest toward improving judgments in Behavioral Operational Research (OR), and the limited evidence about the effectiveness of debiasing tools, the aim of our research is to test the effectiveness of commonly employed practices for debiasing overprecision. We conducted two experiments, testing a set of debiasing techniques when eliciting points of a cumulative distribution functions for general knowledge questions. The debiasing procedures included hypothetical bets, counterfactual argumentation, and automatic stretching to increase the ranges of subjects’ initial estimates. We find that two debiasing strategies that require further reasoning after initial estimates (hypothetical bets and counterfactuals) were not very effective for reducing overprecision, while the use of multipliers that increase the initial range of distributions, coupled with a re-elicitation of the distribution with the new range, provided more positive results. We provide some recommendations for expert elicitation in OR practice, based on our findings, and suggest avenues for further research into debiasing overprecision.
