How Near-Misses Influence Decision Making Under Risk: A Missed Opportunity for Learning
研究组织为何难以从险失事件中学习，发现人们常将险失视为成功，从而降低感知风险并做出更冒险的后续决策，错失了学习机会。
Although organizations appear to learn from obvious failures, we argue that it is harder for them to learn from “near-misses”—events in which chance played a role in averting failure. In this paper, we formalize the concept of near-misses and hypothesize that organizations and managers fail to learn from near-misses because they evaluate such events as successes and thus feel safer about the situation. We distinguish perceived (“felt”) risk from calculated statistical risk and propose that lower levels of perceived risk encourage people with near-miss information to make riskier subsequent decisions compared to people without near-miss information. In our first study, we confirm the tendency to evaluate near-misses as successes by having participants rate a project manager whose decisions result in either (a) mission success, (b) near-miss, or (c) failure. Participants (both students and NASA employees and contractors) give similar ratings to managers whose decisions produced near-misses and to managers whose decisions resulted in successes, and both ratings are significantly different from ratings of managers who experienced failures. We suggest that the failure to hold managers accountable for near-misses is a foregone learning opportunity for both the manager and the organization. In our second set of studies, we confirm that near-miss information leads people to choose a riskier alternative because of a lower perceived risk following near-miss events. We explore several alternative explanations for these findings, including the role of Bayesian updating in processing near-miss data. Ultimately, the analysis suggests that managers and organizations are reducing their perception of the risk, although not necessarily updating (lowering) the statistical probability of the failure event. We speculate that this divergence arises because perceived risk is the product of associative processing, whereas statistical risk arises from rule-based processing.