Many public-policy studies (Martin and Scott 2020) use randomized field experiments for drawing causal conclusions (e.g., Chen et al. 2020). A typical randomized field experiment involves a control group and a treatment group to which individual units (e.g., consumers, patients) are randomly assigned, after which an intervention is implemented in the treatment group. An intervention could be a marketing program to which only units in the treatment group are exposed. To assess the intervention's efficacy, researchers typically estimate the average treatment effect computed as the mean difference in the outcome between the units in the treatment group and the control group. When applying the results of a randomized experiment, it is assumed that the treatment effect within the manipulated condition is the same for all the units assigned to the treatment condition. This may not always be the case, as the effect may differ for subgroups within a treatment (subgroup differences).

许多公共政策研究（Martin 和 Scott，2020 年）使用随机现场实验来得出因果结论（例如，Chen 等人，2020 年）。一个典型的随机田间实验包括一个对照组和一个治疗组，个体单位（例如消费者、患者）被随机分配到其中，然后在治疗组中实施干预。干预可以是营销计划，只有治疗组中的单位才会接触到该计划。为了评估干预的效果，研究人员通常会估计平均治疗效果计算为治疗组和对照组中单位之间结果的平均差异。在应用随机实验的结果时，假设在操纵条件下的处理效果对于分配给处理条件的所有单元都是相同的。这可能并非总是如此，因为治疗中的亚组效果可能不同（亚组差异）。