CAREER: Advances in Randomization Inference for Causal Effects: Heterogeneity, Sensitivity, and Complexity

理解因果效应在各种社会，生物医学和工业研究中具有重要意义，因为它在决策和政策制定中起着至关重要的作用。该项目旨在创建创新的统计方法，提供对因果效应异质性的更全面的理解，对观察性研究中因果结论对不可测量混杂的敏感性的更可靠的评估，以及对现代复杂实验的稳健推断。这项研究有可能回答政治学、教育学和社会学等多种学科的问题。例如，除了通常的平均治疗效果外，该项目还可以帮助解决有关在任何程度上受益于特定政策的个人比例的问题。PI打算通过出版物、演示和分发开放源码软件来传播研究成果。此外，教育和外联活动将系统地纳入研究议程，旨在加强本科教育，向更广泛的受众传播因果关系知识，并使研究生具备关键技能，使他们成为深入的研究人员和人类-主要研究者计划开发新的工具，提供一个更全面和强大的理解因果关系的影响，随机实验和观察性研究。这些工具将建立在随机化推断的基础上，或受到随机化推断的启发，随机化推断使用治疗分配的随机化作为推理基础。该项目有三个主要目标。首先，PI将为个体因果效应的分布开发推断技术，这在实践中是一个重要的问题，但由于其无法从观察到的数据中识别而难以推断。其次，该项目将提供新的敏感性分析，可以适应观察性研究中的极端隐藏混杂，这可以加强因果关系的结论。第三，PI将为复杂的随机实验开发强大的推理方法，这些实验超出了简单的随机化或涉及同行影响。最后，该项目将提供新的计算效率高的算法，并将创建公开可用的R软件包，这将有助于在应用程序中使用这些新工具。该奖项反映了NSF的法定使命，并已被认为是值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估的支持。

Understanding causal effects holds significant importance in various social, biomedical, and industrial studies, as it plays a vital role in decision making and policy formulation. This project aims to create innovative statistical methodologies that provide a more comprehensive understanding of causal effect heterogeneity, a more reliable assessment of the sensitivity of causal conclusions to unmeasured confounding in observational studies, and robust inference for modern complex experiments. The research has the potential to answer questions in such a diverse set of disciplines, as political science, education, and sociology. For instance, the project can help address inquiries regarding the proportion of individuals who benefit from a specific policy to any extent, in addition to the usual average treatment effects. The PI intends to disseminate the research outputs through publications, presentations, and the distribution of open-source software. Additionally, the educational and outreach activities will be systematically integrated to the research agenda, aiming to enhance undergraduate education, spread causality knowledge to the broader audiences, and equip graduate students with the critical skills allowing them to become in-depth researchers and human-centered educators.The Principal Investigator plans to develop new tools that provide a more comprehensive and robust understanding of causal effects in both randomized experiments and observational studies. These tools will be built upon or inspired by the randomization inference, which uses the randomization of treatment assignments as the reasoned basis. The project has three primary objectives. First, the PI will develop inference techniques for the distribution of individual causal effects, which is an important concern in practice, yet difficult to infer due to its unidentifiability from the observed data. Second, the project will deliver new sensitivity analyses that can accommodate extreme hidden confounding in observational studies, which can strengthen the causal conclusions. Third, the PI will develop robust inference methods for complex randomized experiments that go beyond simple randomization or involve peer influence. Finally, the project will provide new computationally efficient algorithms and will create publicly available R software packages that will facilitate the use of these new tools in applications.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.