知识卡片

普通读书笔记卡 · 6584 · 生成模型:GPT-6

已处理效应针对实际接受者

理解等级:中级

ATT=E[Y_1−Y_0|X=1],目标人群是自然分配中实际接受处理者;ATE则对整体人群加权。当分配更偏向某类人且效应异质时,两者可不同。随机试验中的总体平均收益不能直接替代部署后接受者的收益。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      average treatment effect on the treated). We write this as counterfactual query E (AT=+t – AT=–t |T = +t ), as in “for people who saw policy +t, how much more attrition do they have relative to what they would have if they had seen –t ?” Decomposing for binary A as we did with the ATE, we can write this as P(AT=+t=+a|T = +t)

    参考来源

    - 位置:《Causal AI》第10章《识别与因果层级》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-10.xhtml) - 结论依据:摘录中关于已处理效应针对实际接受者的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:average treatment effect on the treated). We write this as counterfactual query E (AT=+t – AT=–t |T = +t ), as in “for people who saw policy +t, how much more attrition do they have relative to what they would have if they had seen –t ?” Decomposing for binary A as we did with the ATE, we can write this as P(AT=+t=+a|T = +t)