依从者平均因果效应的统计推断

Statistical Inference of Complier Average Causal Effect

本文主要研究临床试验中出现不依从现象时依从子总体的平均因果效应的统计推断。考虑在实际服药状态是二值变量且有相关性,治疗响应变量连续取值的情形,同时考虑协变量对治疗效果有影响的情况下,给出了治疗效果参数的可识别性及近似条件极大似然估计,且构造了记分检验统计量。对特殊模型进行了模拟,结果表明与理论预期一致。

In this paper,we study the statistical inference of compiler average causal effect when noncompliance appears in clinical trials.Assuming that the actual treatment variables are binary and correlated,and the response variables are continuous,the estimation of treatment effect adjusted for covariates is considered.We propose the identifiability and an approximate likelihood estimators of the treatment causal parameters,and construct a score test.We carry out simulations based on a special model,and the results reveal that the test appears as anticipated theoretically.