Bridging Randomized Experiments and Observational Studies

尽管完美的随机实验是对治疗效果进行因果推断的黄金标准，但它们通常不可能解决重要问题。 随机实验的局限性包括人类受试者的伦理考虑，实施时间的长度以及诸如不依从等并发症。 最近在非随机研究中进行因果推断的工作，以创造性的方式结合了统计和传统的计量经济学方法，显示出在完全受控的随机实验和非受控的观察性研究之间建立新的重要桥梁的巨大希望。 继续这项工作将导致重要的新技术，用于分析不完善的因果关系研究。 目前大量的应用程序包括推理的有效性“选择”方案相对于标准方案小学教育。

Although perfect randomized experiments are the gold-standard for drawing causal inferences about the effects of treatments, they are often impossible to conduct to address important questions. Limitations of randomized experiments include ethical considerations with human subjects, length of time to implement, and complications such as noncompliance. Recent work for drawing causal inferences in non-randomized studies that combines both statistical and traditional econometric approaches in creative ways shows great promise for building new and important bridges between the perfectly controlled randomized experiment and the uncontrolled observational study. Continuing this work should lead to important new techniques for the analysis of imperfect studies for causal effects. Substantial current applications include inference for the effectiveness of `choice` programs relative to standard programs for elementary school education.