Potential outcome model
最早是1920年代內(nèi)曼和費雪提出。幾十年都沒有什么發(fā)展。轉(zhuǎn)折點是Holland（1986）的一篇文章：它highlight(強(qiáng)調(diào))了內(nèi)曼和魯本這個模型的用途。后來很多學(xué)者對它做出貢獻(xiàn)∨撸現(xiàn)在由魯本自己寫了這本書總結(jié)這個領(lǐng)域。
回到 first principles
The discussion is thorough with an effort to build everything from the first principles.
缺點
內(nèi)容不夠廣泛
只介紹potential outcome，沒有介紹其他“對手”的理論嘹吨。比如DAG、比如SEM境氢，都是與potential outcome競爭的理論蟀拷。
超大的篇幅，600多頁萍聊。但是只講了隨機(jī)試驗问芬。沒有講cluster randomized experiments, interference between units, longitudinal data, mediation analysis, nonbinary treatments, regression discontinuity designs, and difference-in-differences designs 這些非常重要的話題。
內(nèi)容不夠易懂
推薦另外2本書做為入門寿桨。
The authors write that their target audience is “researchers in applied fields.” In my view, this book is best suited for applied
researchers who have a solid understanding of basic probability and statistics. Those seeking a nontechnical introduction to
statistics may have a hard time following the detailed mathematical derivation presented in the book. These researchers may
find other textbooks such as Morgan and Winship (2007) and Angrist and Pischke (2009) more accessible. In addition, while
the book includes many real-world applications, it does not provide tips on how to implement the methods. Online materials
including the computer code for the results presented in the book would be a useful supplement to the text.
講觀察性研究的內(nèi)容明顯有不足
沒有覆蓋最新的方法此衅。比如估計propensity score的方法，比如matching methods，都只講了經(jīng)典的方法挡鞍。
有應(yīng)用骑歹，但是沒教實現(xiàn)
In addition, while the book includes many real-world applications, it does not provide tips on how to implement the methods.
