New Weighting Methods for Causal Inference

在社会、统计、经济和医学科学中，进行因果推理的能力对于解决广泛的研究问题是很重要的。从社会的角度来看，在缺乏因果推理的情况下，很难确定政策干预是否有效。这个研究项目将开发因果推理的新方法。由此产生的方法将改进因果推理的统计分析，从而从许多学科的实验和观察性干预研究中得出更准确的结论。该研究开发的开源R/Matlab软件将为科学界提供有价值的数据分析和教育工具。从重要的实际应用，如医疗保健服务中的种族差异，本研究将开发新的，通用的，可访问的加权方法，从观察数据中得出因果和协变量平衡的描述性比较。基于人群的观察性研究越来越多地用于得出因果结论，但复杂的（通常是未知的）潜在数据生成系统对有效的因果推断提出了巨大挑战。源于调查研究的加权方法是灵活而强大的因果推理工具，协变量平衡描述性比较分析也是如此，但与回归和匹配方法相比，这些方法可以说是欠发达的。研究重点将包括超越传统逆概率权重的基于权重的因果推理的统一框架，以及多层结构复杂调查数据的倾向得分加权方法。这项研究将把理论和方法的发展与在卫生政策、流行病学、社会科学和经济学等领域的激励应用相结合。

The ability to make causal inferences is important for addressing a wide range of research questions in the social, statistical, economic, and medical sciences. From a societal perspective, it is difficult to determine if a policy intervention is effective in the absence of causal inference. This research project will develop new methods for causal inference. The resulting methods will improve statistical analyses for causal inference, thereby enabling more accurate conclusions from experimental and observational intervention studies across many disciplines. The open source R/Matlab software developed from the research will provide valuable data analysis and educational tools for the scientific community.Motivated from important real applications like racial disparity in health care service, this research will develop new, general, and accessible weighting methods for drawing causal and covariates-balanced descriptive comparisons from observational data. Population-based observational studies have been increasingly used for drawing causal conclusions, but the complex (and usually unknown) underlying data-generating system poses great challenges to valid causal inference. Weighting methods that originated in survey research are flexible and powerful tools for causal inference, as are covariates-balanced descriptive comparative analysis, but these are arguably less developed compared to regression and matching methods. Research foci will include a unified framework for weighting-based causal inference that extends beyond the traditional inverse probability weights, and propensity score weighting methods for complex survey data with multilevel structure. This research will blend theory and methodological developments with motivating applications in areas including health policy, epidemiology, social sciences, and economics.