Causal inference in statistics

统计学中的因果推断

• 批准号: 203137-2012
• 负责人: Platt, Robert
• 金额: $ 0.87万
• 依托单位: McGill University
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2015
• 资助国家: 加拿大
• 起止时间: 2015-01-01 至 2016-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=569874
• 关键词: Causal; inference; statistics

I will continue to develop my research program on statistical methods for causal inference in epidemiology, with focus on marginal structural models propensity score. My work to date has focused on the use of the propensity score in marginal structural models (MSMs) and targeted maximum likelihood (TML) methods. The core of my current and proposed work extends these results. I will investigate the properties of various statistical learning-based methods for fitting the propensity score, in the context of MSMs and TML. Via theoretical results and simulations, I will describe the properties of the HDPS in general, and in particular for use in marginal structural models. Close analogues of the propensity score may also be used in randomized trial design. It has been shown that the propensity for selection into a trial could be used to weight results from subjects in the trial to generalize to a larger population. With Elizabeth Stuart, I am investigating the properties of this propensity score for design of randomized trials and their use in meta-analysis. Consider a trial in which the subjects do not adequately reflect the population of interest. We have shown (manuscript in preparation) that if one uses the inverse of the probability of selection into the first trial as a weight for subjects in a second trial, bias is reduced relative to random selection for the second trial. This result will be extended to multiple trials. Finally, I will continue my work on longitudinal models for multivariate outcomes, with applications to pregnancy outcomes. Specifically, I will extend models for longitudinal growth processes to multiple outcomes, and develop a multivariate model extending Cole's SuperImposition by Translation And Rotation model.

我将继续发展我的研究计划在流行病学因果推断的统计方法，重点是边际结构模型的倾向得分。 迄今为止，我的工作主要集中在边缘结构模型（MSM）和目标最大似然（TML）方法中的倾向得分的使用。 我目前和拟议工作的核心是扩展这些结果。 我将在MSM和TML的背景下研究各种基于统计学习的方法来拟合倾向分数的属性。 通过理论结果和模拟，我将描述一般的HDPS的属性，特别是用于边际结构模型。 在随机试验设计中也可以使用倾向评分的近似值。 已经表明，选择进入试验的倾向可以用于对试验中受试者的结果进行加权，以推广到更大的人群。 我与Elizabeth Stuart一起研究了随机试验设计的倾向评分的性质及其在荟萃分析中的应用。 考虑一项试验，其中受试者不能充分反映感兴趣的人群。 我们已经证明（手稿在准备中），如果使用第一次试验中选择概率的倒数作为第二次试验中受试者的权重，则相对于第二次试验的随机选择，偏倚会减少。 这一结果将扩展到多个试验。 最后，我将继续我对多变量结局的纵向模型的研究，并将其应用于妊娠结局。 具体来说，我将扩展模型的纵向增长过程的多个结果，并开发一个多变量模型扩展科尔的叠加平移和旋转模型。