具有不依从现象的临床试验中平均处理效应的上下界估计

在随机处理一对照的临床试验中，经常出现不依从或部分依从的现象，此时，由于所涉及到的“虚拟事实”变量，即不能观察到的潜在变量太多而不易估计其平均因果效应ACE．在仅出现完全依从和不依从情况时，Balke and Pearl利用线性规划的方法获得了ACE估计量的上下界，利用他们所提供的方法，有时会出现下界为负数，显然，这样的下界没什么实际意义．根据Angrist，Imbebns＆Rubin讨论工具变量时所提出一些假设条件，导出了在不同情况下，计算ACE估计量的上下界的方法，并证明了其下界一定是非负的，所得到的上下界区间比Balke and Pearl的区间要小．同时，还讨论了部分依从情况下，ACE估计量的上下界的计算方法，并得到了相应的结果．

Estimation of boundary of average treatment effects in clinical trials with noncompliance

Noncompliance or partial compliance often occurs in a clinical trial.In this case,average treatment effects,or called average causal effect(ACE) of treatment,may not easily estimated because there are too many potential variables which can not be simultaneously observed.Balke and Pearl have obtained the upper and lower boundary of ACE under the assumption of non-compliance or total compliance.By their method,lower boundary may sometimes be negative,which is abnormal.This paper proposes a method to calculate...