Principal-Agent Models of Decision Delegation during Systems Design: Integrating Modeling and Behavioral Approaches

现代系统工程项目可能涉及成千上万的工程师，他们分布在数千个独立的组织中。例如，美国空军的F-35“闪电II”战斗机由1400多家供应商提供的30多万个独立零件组成。同样，波音787梦幻客机的开发涉及2000多家供应商，这些供应商分为三个责任层。关于如何在这些组织内部和组织之间委派设计权力的决策可以对项目度量（进度和预算）以及最终工程系统的性能产生重大影响。糟糕的设计授权决策被引用为大型系统工程项目中问题的解释。然而，系统工程中不同的设计委托方法的影响在基础层面上并没有得到很好的理解。对系统工程中设计授权的改进理解将提高系统工程项目的效率和有效性，并导致更好地执行工程系统。本研究的目的是为在系统工程项目的设计阶段发生的决策创建一个新的行为知情建模框架。本研究将把委托代理理论的数学与人类判断和决策行为的新实证结果结合起来。这将使各种决策授权策略的分析能够以一种由有关工程决策者如何行为的数据支持的方式进行，但通过人类研究在规模上是不切实际的。预计新的建模框架将使研究人员能够回答有关设计授权方法的重要问题，例如在特定情况下一种方法是否优于另一种方法，以及方法的性能对问题上下文的敏感程度。这项研究将为工程师在决策过程中如何解释和应对激励因素以及众所周知的设计授权方法如何相互比较提供新的知识。

Modern systems engineering projects can involve tens of thousands of engineers spread across thousands of independent organizations. For example, the Air Force's F-35 Lightning II consists of more than 300,000 individual parts delivered by more than 1,400 suppliers. Similarly, development of the Boeing 787 Dreamliner involved over 2,000 suppliers organized into three tiers of responsibility. Decisions about how to delegate design authority within and among these organizations can have significant impact on project metrics (schedule and budget) as well as the performance of the resulting engineered systems. Poor design delegation decisions have been cited as an explanation for problems on large systems engineering projects. However, the impact of different approaches to design delegation in systems engineering is not well understood at a fundamental level. Improved understanding of design delegation in systems engineering will improve the efficiency and effectiveness of systems engineering projects and result in better-performing engineering systems. The objective of this research is to create a new behaviorally-informed modeling framework for decisions that occur during the design stages of a systems engineering project. The research will combine the mathematics of principal-agent theory with new empirical results about human judgment and decision-making behavior. This will enable the analysis of various decision delegation strategies in a manner that is supported by data about how engineering decision makers behave but at a scale impractical via human studies. It is anticipated that the new modeling framework will enable researchers to answer important questions about design delegation approaches, such as whether one method is superior to another under particular circumstances and the degree to which the performance of a method is sensitive to problem context. The research will contribute new knowledge about how engineers interpret and respond to incentives in their decision making and about how well-known methods for design delegation compare to one another.