Model-Based and Design-Based Approaches to Longitudinal Causal Decomposition Analysis

该项目将开发基于模型和基于设计的方法，以确定导致社会差异的风险因素。尽管在各个领域取得了进展，但基于种族/民族等各种特征，美国社会群体在认知、经济和健康结果方面仍然存在巨大差异。由于不可能改变归因于的特征，差异研究人员和政策制定者的兴趣往往集中在确定可能在减少这种差异中发挥作用的可塑风险因素上。然而，目前识别危险因素的方法存在局限性。该项目将开发一个综合框架，利用纵向观察数据和顺序随机实验设计来确定风险因素，从而解决这些限制。将向研究人员提供如何使用这些方法的方法学指导。将开发一个R包、示例代码和视频教程。研究生将得到指导，并将在少数民族服务机构开设社会科学定量因果推理的本科课程。该项目将开发基于模型和基于设计的因果分解方法，以考虑时变的风险因素（称为“中介”）和结果。现有的因果分解模型存在局限性。它们只考虑了时间固定的中介和结果，而没有深入了解采取干预措施以减少差距的最佳时间。即使进行了彻底的敏感性分析，也有可能遗漏变量偏差，因为结果是基于研究人员对合理混杂程度的主观判断。为了建立一个分析纵向观测数据的框架，该项目将使用一个广义线性混合模型，将年龄作为时钟。这种方法将能够确定介质的效果何时达到峰值、消退或保持平稳。这项分析的结果可以为干预的时机提供信息，包括顺序干预或同时干预是否可以更有效地减少随着时间的推移的差距。该项目还将开发随机顺序实验设计，以确定其效果（即减少视差和剩余视差）。实验设计依赖较少的假设，并允许研究人员测试实际的（而不是假设的）干预措施，以减少差异。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 project will develop model-based and design-based approaches to identify risk factors that contribute to social disparities. Despite progress in various fields, large disparities in cognitive, economic, and health outcomes persist across social groups in the US based on various characteristics such as race/ethnicity. Since it is not possible to modify ascribed characteristics, the interests of disparity researchers and policymakers often center on identifying malleable risk factors that may play a role in reducing such disparities. However, there are limitations with the current methods for identifying risk factors. This project will address those limitations by developing a comprehensive framework that leverages longitudinal observational data and sequential randomized experimental designs to identify risk factors. Methodological guidance on how to use these approaches will be provided to researchers. An R package, sample code, and video tutorials will be developed. Graduate students will be mentored, and an undergraduate course on quantitative causal reasoning for the social sciences will be created at a Minority Serving Institution.This project will develop model-based and design-based approaches to causal decomposition that allow for time-varying risk factors (referred to as 'mediators') and outcomes. Current causal decomposition models are restricted. They only consider time-fixed mediators and outcomes and do not provide insight into the optimal time for interventions to reduce disparities. There also is the possibility of omitted variable bias, even with thorough sensitivity analysis, since the results are based on the researcher's subjective judgment on what constitutes a reasonable level of confounding. To develop a framework for analyzing longitudinal observational data, the project will use a generalized linear mixed model, incorporating age as a clock. This approach will enable the determination of when the effects of mediators' peak, fade, or remain flat. The results from this analysis could inform the timing of interventions, including whether sequential or simultaneous interventions may be more effective for reducing disparities over time. The project also will develop randomized sequential experimental designs that identify their effects (i.e., disparity reduction and remaining disparity). Experimental designs rely on fewer assumptions and allow researchers to test actual (rather than hypothetical) interventions to reduce disparities.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.