Ch20 Treatment Confounder Feedback

The previous chapter identified sequential exchangeability as a key condition to identify the causal effects of time-varying treatments. Suppose that we have a study in which the strongest form of sequential exchangeability holds: the measured time-varying confounders are sufficient to validly estimate the causal effect of any treatment strategy. Then the question is what confounding adjustment method to use. The answer to this question highlights a key problem in causal inference about time-varying treatments: treatment-confounder feedback. When treatment-confounder feedback exists, using traditional adjustment methods may introduce bias in the effect estimates. That is, even if we had all the information required to validly estimate the average causal effect of any treatment strategy, we would be generally unable to do so. This chapter describes the structure oftreatment-confounder feedback and the reasons why traditional adjustment methods fail.

上一章将顺序可交换性确定为确定时变处理的因果关系的关键条件。假设我们有一项研究，其中最强形式的顺序互换性成立：所测得的时变混杂因素足以有效估计任何治疗策略的因果关系，那么问题是要使用哪种混杂调整方法。这个问题的答案突出了随时间变化的治疗的因果推论中的一个关键问题：治疗混杂反馈。当存在治疗混杂反馈时，使用传统的调整方法可能会在效果估计中引入偏差。也就是说，即使我们拥有有效估计任何治疗策略的平均因果效应所需的所有信息，我们也通常无法做到这一点。本章介绍了处理混杂反馈的结构以及传统调整方法失败的原因。

Table of Contents

1. The element of treatment confounder feedback

2. The bias of traditional methods

3. Why traditional methods fails

4. Why traditional methods cannot be fixed

5. Adjusting for past treatment