Are Classical Experiments Needed for Manpower Policy
比较了随机实验与非实验方法在评估人力政策效果上的优劣，指出实验的主要成本是费用，但能提供更可靠的效果估计，并给出了判断实验是否值得的简单框架。
The critical element that distinguishes classical experiments from all other modes of analysis is the random assignment of treatment to enrollees in a study. This paper examines the major methodological advantages of random assignment for the purpose of estimating the effectiveness of current manpower policy. It also reviews the claimed methodological and ethical objections to experiments. The main valid objection to an experiment is neither methodological nor ethical, it is the experiment's cost in relation to that of nonexperimental methods of analysis. The authors argue that the offsetting gain from experimentation is the inherent reliability of experimental estimates of treatment effects. The paper offers a simple framework for deciding whether the improved reliability of treatment-effect estimates is worth the added cost of experimentation. It concludes with an assessment of the actual value of experiments for evaluating current manpower policies.