Learning, Private Information, and the Economic Evaluation of Randomized Experiments
通过分析受试者退出行为揭示其对治疗的私人效用，评估随机实验中的处理效应，发现副作用和疗效学习过程共同驱动退出，并以艾滋病临床试验ACTG 175为例说明AZT因副作用小而效用最高。
Many randomized experiments are plagued by attrition, even among subjects receiving more effective treatments. We estimate the subject's utility associated with the receipt of treatment, as revealed by dropout behavior, to evaluate treatment effects. Utility is a function of both "publicly observed" outcomes and side effects privately observed by the subject. We analyze an influential AIDS clinical trial, ACTG 175, and show that for many subjects, AZT yields the highest level of utility despite having the smallest impact on the publicly observed outcome because of mild side effects. Moreover, although subjects enter the experiment uncertain of treatment effectiveness (and often the treatment received), the learning process implies that early dropout in ACTG 175 is primarily driven by side effects, whereas later attrition reflects declining treatment effectiveness.