Randomization and Social Policy Evaluation
Randomization and Social Policy Evaluation
复制标题
随机化和社会政策评估
DOI：
--
复制
发表时间：
1991
期刊：
影响因子：
--
通讯作者：
J. Heckman
中科院分区：
文献类型：
--
作者：
J. Heckman
本文认为，最近的情况下，随机社会实验，并将其与旧的情况下，社会实验。最近的案例避开了行为模型，假设结果的某些平均差异是评估者感兴趣的参数，并假设随机化不会破坏正在分析的社会计划。程序中断的影响是没有后果的条件下。即使在没有随机化偏倚的情况下，理想的实验数据也不能估计治疗和未治疗的人之间的中位数（其他分位数）差异，而不引用补充的统计假设。最近的情况下，随机实验并没有解决选择适当的阶段，在一个多阶段的计划，随机化应进行。提供了随机化偏倚的证据。
This paper considers the recent case for randomized social experimentation and contrasts it with older cases for social experimentation. The recent case eschews behavioral models, assumes that certain mean differences in outcomes are the parameters of interest to evaluators and assumes that randomization does not disrupt the social program being analyzed. Conditions under which program disruption effects are of no consequence are presented. Even in the absence of randomization bias, ideal experimental data cannot estimate median (other quantile) differences between treated and untreated persons without invoking supplementary statistical assumptions. The recent case for randomized experimentation does not address the choice of the appropriate stage in a multistage program at which randomization should be conducted. Evidence on randomization bias is presented.