{"title":"嵌套工具变量设计:转换者平均治疗效果、识别、有效估计和普适性","authors":"Rui Wang, Ying-Qi Zhao, Oliver Dukes, Bo Zhang","doi":"arxiv-2405.07102","DOIUrl":null,"url":null,"abstract":"Instrumental variables (IV) are a commonly used tool to estimate causal\neffects from non-randomized data. A prototype of an IV is a randomized trial\nwith non-compliance where the randomized treatment assignment serves as an IV\nfor the non-ignorable treatment received. Under a monotonicity assumption, a\nvalid IV non-parametrically identifies the average treatment effect among a\nnon-identifiable complier subgroup, whose generalizability is often under\ndebate. In many studies, there could exist multiple versions of an IV, for\ninstance, different nudges to take the same treatment in different study sites\nin a multi-center clinical trial. These different versions of an IV may result\nin different compliance rates and offer a unique opportunity to study IV\nestimates' generalizability. In this article, we introduce a novel nested IV\nassumption and study identification of the average treatment effect among two\nlatent subgroups: always-compliers and switchers, who are defined based on the\njoint potential treatment received under two versions of a binary IV. We derive\nthe efficient influence function for the SWitcher Average Treatment Effect\n(SWATE) and propose efficient estimators. We then propose formal statistical\ntests of the generalizability of IV estimates based on comparing the\nconditional average treatment effect among the always-compliers and that among\nthe switchers under the nested IV framework. We apply the proposed framework\nand method to the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer\nScreening Trial and study the causal effect of colorectal cancer screening and\nits generalizability.","PeriodicalId":501323,"journal":{"name":"arXiv - STAT - Other Statistics","volume":"24 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Nested Instrumental Variables Design: Switcher Average Treatment Effect, Identification, Efficient Estimation and Generalizability\",\"authors\":\"Rui Wang, Ying-Qi Zhao, Oliver Dukes, Bo Zhang\",\"doi\":\"arxiv-2405.07102\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Instrumental variables (IV) are a commonly used tool to estimate causal\\neffects from non-randomized data. A prototype of an IV is a randomized trial\\nwith non-compliance where the randomized treatment assignment serves as an IV\\nfor the non-ignorable treatment received. Under a monotonicity assumption, a\\nvalid IV non-parametrically identifies the average treatment effect among a\\nnon-identifiable complier subgroup, whose generalizability is often under\\ndebate. In many studies, there could exist multiple versions of an IV, for\\ninstance, different nudges to take the same treatment in different study sites\\nin a multi-center clinical trial. These different versions of an IV may result\\nin different compliance rates and offer a unique opportunity to study IV\\nestimates' generalizability. In this article, we introduce a novel nested IV\\nassumption and study identification of the average treatment effect among two\\nlatent subgroups: always-compliers and switchers, who are defined based on the\\njoint potential treatment received under two versions of a binary IV. We derive\\nthe efficient influence function for the SWitcher Average Treatment Effect\\n(SWATE) and propose efficient estimators. We then propose formal statistical\\ntests of the generalizability of IV estimates based on comparing the\\nconditional average treatment effect among the always-compliers and that among\\nthe switchers under the nested IV framework. We apply the proposed framework\\nand method to the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer\\nScreening Trial and study the causal effect of colorectal cancer screening and\\nits generalizability.\",\"PeriodicalId\":501323,\"journal\":{\"name\":\"arXiv - STAT - Other Statistics\",\"volume\":\"24 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-05-11\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - STAT - Other Statistics\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2405.07102\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - STAT - Other Statistics","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2405.07102","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
摘要
工具变量(IV)是从非随机数据中估计因果效应的常用工具。IV 的一个原型是具有非遵从性的随机试验,其中随机治疗分配可作为所接受的不可忽略的治疗的 IV。在单调性假设下,有效的 IV 可以非参数地识别不可识别的违规者亚群中的平均治疗效果,其普遍性往往受到争议。在许多研究中,可能存在多个版本的 IV,例如,在一项多中心临床试验中,不同的研究地点对采取相同治疗方法的不同劝告。这些不同版本的静脉注射可能会导致不同的依从率,为研究静脉注射估计值的可推广性提供了一个独特的机会。在本文中,我们引入了一个新颖的嵌套 IV 假设,并研究了在两类人群中平均治疗效果的识别问题:始终遵从者和转换者,这两类人群是根据二元 IV 的两个版本下共同接受的潜在治疗来定义的。我们推导出转换者平均治疗效果(SWATE)的有效影响函数,并提出了有效的估计值。然后,我们在比较嵌套 IV 框架下始终遵守者和转换者之间的条件平均治疗效果的基础上,对 IV 估计值的可推广性提出了正式的统计检验。我们将提出的框架和方法应用于前列腺癌、肺癌、结直肠癌和卵巢癌(PLCO)筛查试验,研究结直肠癌筛查的因果效应及其可推广性。
Nested Instrumental Variables Design: Switcher Average Treatment Effect, Identification, Efficient Estimation and Generalizability
Instrumental variables (IV) are a commonly used tool to estimate causal
effects from non-randomized data. A prototype of an IV is a randomized trial
with non-compliance where the randomized treatment assignment serves as an IV
for the non-ignorable treatment received. Under a monotonicity assumption, a
valid IV non-parametrically identifies the average treatment effect among a
non-identifiable complier subgroup, whose generalizability is often under
debate. In many studies, there could exist multiple versions of an IV, for
instance, different nudges to take the same treatment in different study sites
in a multi-center clinical trial. These different versions of an IV may result
in different compliance rates and offer a unique opportunity to study IV
estimates' generalizability. In this article, we introduce a novel nested IV
assumption and study identification of the average treatment effect among two
latent subgroups: always-compliers and switchers, who are defined based on the
joint potential treatment received under two versions of a binary IV. We derive
the efficient influence function for the SWitcher Average Treatment Effect
(SWATE) and propose efficient estimators. We then propose formal statistical
tests of the generalizability of IV estimates based on comparing the
conditional average treatment effect among the always-compliers and that among
the switchers under the nested IV framework. We apply the proposed framework
and method to the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer
Screening Trial and study the causal effect of colorectal cancer screening and
its generalizability.