New Developments in Methodology for Program Evaluation

该研究项目将为项目评估开发新的统计方法，目标是提高其在实际应用中的可靠性和范围。项目评估方法广泛应用于社会、行为和生物医学科学的各个学科。这些方法是经验性地用来确定政策、干预或治疗对某些感兴趣的结果的因果关系。例如，项目评估方法使政策制定者能够改进社会项目或公共卫生官员，以更好地减少传染病。本项目将为三种现代程序评估方法：binscatter方法、综合控制方法和回归不连续方法开发新的识别、估计和推理结果。本研究结果将为研究人员改进实证工作提供新的方法论工具。研究生将接受新方法的指导和培训，并获得第一手的研究经验。将开发通用软件来实现这些新方法。这个研究项目将为数理统计和计量经济学提供新的发展。对于与机器学习技术密切相关的binscatter方法，将开发新的统一推理方法，这些方法建立在概率论的强大近似工具之上。这些结果将为可能具有异质性处理效果的因果推理模型提供有效的置信区间和关于形状限制的假设检验。对于综合控制方法，将利用高维统计文献中的非渐近概率集中思想开发新的预测区间。本文还将研究这些基于自举的预测区间的实际实现。对于回归不连续设计，该项目将开发新的识别、估计、推理和证伪方法，用于具有持续时间型结果的设置，并允许协变量调整和多个截止点。鲁棒偏置校正推理方法将与调谐参数选择方法一起发展。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This research project will develop new statistical methods for program evaluation with the goal of improving their reliability and scope in real-life applications. Program evaluation methods are widely used in a variety of disciplines in the social, behavioral, and biomedical sciences. These methods are empirically employed to determine the causal effect of a policy, intervention, or treatment on some outcome of interest. For example, program evaluation methods allow policy makers to improve social programs or public health officials to better target the mitigation of infectious diseases. This project will develop new identification, estimation, and inference results for three modern program evaluation methods: binscatter methods, synthetic control methods, and regression discontinuity methods. The results of this research will provide researchers with new methodological tools to improve empirical work. Graduate students will be mentored and trained in the new methodologies and provided with first-hand research experience. General-purpose software will be developed to implement the new methods.This research project will provide new developments in mathematical statistics and econometrics. For binscatter methods, which are closely related to machine learning techniques, new uniform inference methods will be developed that build on strong approximation tools from probability theory. These results will provide valid confidence bands and hypothesis tests about shape restrictions for causal inference models with possibly heterogenous treatment effects. For synthetic control methods, novel prediction intervals will be developed using non-asymptotic probability concentration ideas from the high-dimensional statistical literature. Practical implementation of these prediction intervals based on the bootstrap also will be studied. For regression discontinuity designs, the project will develop new identification, estimation, inference, and falsification methods for settings with duration-type outcomes and allowing for covariate-adjustment and multiple cutoffs. Robust bias correction inference methods will be developed along with tuning parameter selection methods.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.