随机化实验和观察性研究中处理效应的识别与估计

Identifying and Evaluating Treatment Effects in Randomized Experiments and Observable Studies

文章考虑处理效应评估的统计方法。在许多社会科学中,人们经常关心处理效应的评估问题。这个问题最初来源于评价一种新药有没有疗效,而后人们把它扩展到很多领域。人们评估处理效应的最直接的方法是比较处理组与对照组的平均输出。然而,这种方法通常会导致处理效应估计产生选择性偏差。在随机化试验中,处理变量的随机化分配能够克服这种偏差。然而,随机化实验经常是不实用的,研究者们用的更多的是观测研究。在观测研究中,无混杂分配能够克服选择性偏差。并且在这个假设下,回归、匹配和倾向得分等方法都可以用于评估处理效应。

This paper considers the statistical methods for evaluating the treatment effects. In many social sciences, the evaluation problem of the treatment effect has been often concerned. The question comes from the concern about whether a new drug is effective or not. And then it is extended to many fields. Simple comparisons of program participants with non-participants often lead to the selection bias. In a randomized trial, randomization eliminates selection bias. However, a randomized trial is often im- practical. Much of the research we do, attempts to exploit observational studies. Under the unconfoundedness, the observational studies overcome selection bias. And with the assumption, many methods such as the regression, matching and propensity score may be used to evaluate the average treatment effect.