Collaborative Research in Decision, Risk, and Management Science

在分析核电站、液化天然气储存设施、危险材料运输和航天飞机等复杂技术系统的风险时，确定风险分析模型的详细程度是一个重要问题。过于详细和过于简单的模型都可能导致错误的风险估计。然而，尽管风险分析在过去15年里得到了广泛的发展，但这个问题几乎没有得到研究。该项目的一名调查人员过去的工作确定了风险分析中的三个潜在错误来源：建模错误、参数错误和聚合错误。通过简单的模型与详细的模型将不同类型的误差降至最低。特别是，在聚合错误的情况下，使用过于简单的模型可能会忽略可能在更详细的级别上可用的信息。例如，在Challenger灾难发生时，仅基于23个航班的0起事故的综合分析就会表明，每一次飞行的可靠性都在增加。然而，包括23个航班中6个O形圈被侵蚀的证据在内的分类分析将得出结论，仍然存在相当大的风险水平。该项目的主要研究人员已经确定了在简单的风险分析模型中聚集误差消失的充要条件。然而，这些条件在实践中不太可能实现。因此，当没有实现完全聚合时，需要了解聚合误差的决定因素。拟议的工作将就聚合误差严重的情况提供指导，以便在这种情况下可以通过使用更详细的模型来避免大误差。

Determining how detailed risk analysis models should be is an important question in analyzing the risks of complex technological systems such as nuclear power plants, liquid natural gas storage facilities, hazardous material transport, and the space shuttle. Both overly detailed and overly simplistic models can lead to erroneous risk estimates. However, this issue has gone virtually unstudied despite the widespread growth of risk analysis in the last 15 years. Past work by one of the investigators on this project has identified three potential sources of error in risk analyses: modeling error; parameter error; and aggregation error. Different types of errors are minimized by simple versus detailed models. In particular, in the case of aggregation error, the use of overly simple models can neglect information that may be available at more detailed levels. For example, at the time of the Challenger disaster, an aggregate analysis based only on 0 accidents in 23 flights would have indicated increasing reliability with each flight. However, a disaggregate analysis including the evidence of O-ring erosion on 6 of the 23 flights would have concluded that a substantial level of risk remained. The principal investigators for this project have identified necessary and sufficient conditions for aggregation error to vanish in simple risk analysis models. However, these conditions are unlikely to be achieved in practice. Therefore, there is a need to understand the determinants of aggregation error when perfect aggregation is not achieved. The proposed work will provide guidance on the circumstances under which aggregation error will be severe, so that large errors can be avoided through the use of more detailed models in such cases.