Sensitivity analysis of individual treatment effects: A robust conformal inference approach.
Sensitivity analysis of individual treatment effects: A robust conformal inference approach.
复制标题
DOI：
10.1073/pnas.2214889120
复制
发表时间：
2023-02-07
影响因子：
11.1
通讯作者：
Candes, Emmanuel J.
中科院分区：
文献类型：
--
作者：
Jin, Ying;Ren, Zhimei;Candes, Emmanuel J.
关键词：
个体治疗效应（ITE）描述了个体接受治疗与不接受治疗时的结果之间的差异。这种差异可能因个人的特征而异。在观察性研究中，如果忽略同时影响治疗分配和结局的未测量混杂因素，则ITE的推断可能无效。我们提出了一个框架，定量地了解对这些潜在的混杂因素的因果结论的ITE的鲁棒性。这将产生预测带，这些预测带带有严格的不确定性量化工具。无论用于学习治疗效果的机器学习模型有多复杂，也无论样本大小如何，这些工具都适用。我们提出了一个无模型的框架，个人治疗效果（ITE）的敏感性分析，从共形推理的想法的基础上。对于任何单位，我们的程序报告了Γ值，这是一个量化解释ITE证据所需的最小混杂强度的数字。我们的方法依赖于在训练数据混淆的情况下对反事实和ITE的可靠预测推理。在[Z. Tan，J. Am. Stat. Assoc.101，1619-1637（2006）]，我们描述了观测分布和反事实分布之间的变化。首先，我们开发了一个通用的方法，从移位分布的测试样本的预测推理，然后，我们利用这个构建协变量依赖的预测集的反事实。不管移位的值是多少，这些预测集（分别是近似地）实现边际覆盖，如果准确地知道倾向分数（分别地，估计）。我们描述了一个不同的程序，也达到覆盖率，但是，有条件的训练数据。在后一种情况下，我们证明了一个清晰的结果表明，对于某些类别的预测问题，预测区间不可能收紧。我们通过仿真研究验证了方法的有效性和性能，并将其应用于分析真实的数据集。
The individual treatment effect (ITE) describes the difference between an individual’s outcome when receiving a treatment versus not. This difference may vary across individuals conditional on their characteristics. In observational studies, inference for ITEs can be invalid if one ignores unmeasured confounding factors that simultaneously influence the treatment assignment and the outcomes. We propose a framework to quantitatively understand the robustness of causal conclusions on ITEs against such potential confounding factors. This yields prediction bands, which come with rigorous uncertainty quantification tools. These tools apply regardless of the machine learning model employed to learn the treatment effect, however complicated, and regardless of the sample size. We propose a model-free framework for sensitivity analysis of individual treatment effects (ITEs), building upon ideas from conformal inference. For any unit, our procedure reports the Γ-value, a number which quantifies the minimum strength of confounding needed to explain away the evidence for ITE. Our approach rests on the reliable predictive inference of counterfactuals and ITEs in situations where the training data are confounded. Under the marginal sensitivity model of [Z. Tan, J. Am. Stat. Assoc. 101, 1619-1637 (2006)], we characterize the shift between the distribution of the observations and that of the counterfactuals. We first develop a general method for predictive inference of test samples from a shifted distribution; we then leverage this to construct covariate-dependent prediction sets for counterfactuals. No matter the value of the shift, these prediction sets (resp. approximately) achieve marginal coverage if the propensity score is known exactly (resp. estimated). We describe a distinct procedure also attaining coverage, however, conditional on the training data. In the latter case, we prove a sharpness result showing that for certain classes of prediction problems, the prediction intervals cannot possibly be tightened. We verify the validity and performance of the methods via simulation studies and apply them to analyze real datasets.
登录
查看更多内容
DOI：
10.1080/01621459.2017.1307116
发表时间：
2018-01-01
影响因子：
3.7
作者：
Lei, Jing;G'Sell, Max;Wasserman, Larry
通讯作者：
Wasserman, Larry
影响因子：
10.7
作者：
Imbens, GW
通讯作者：
Imbens, GW
DOI：
10.1080/01621459.2022.2069572
发表时间：
2022-05-28
影响因子：
3.7
作者：
Dorn, Jacob;Guo, Kevin
通讯作者：
Guo, Kevin
影响因子：
2.7
作者：
Lei, J.
通讯作者：
Lei, J.
DOI：
10.1111/rssb.12445
发表时间：
2021-10-07
影响因子：
5.8
作者：
Lei, Lihua;Candes, Emmanuel J.
通讯作者：
Candes, Emmanuel J.