随机影响评估存在的问题及应对策略：一个研究综述

Problems and Countermeasures for Randomized Impact Evaluations: A Research Review

摘要: 随机影响评估的“反事实”设计可以获得最可信的因果推断，是公共政策或项目评估工具箱中的标准工具。然而，现实中开展随机影响评估面临着诸多问题，其中一些问题会导致较小的障碍，造成估计结果存在偏倚，而另一些问题会导致实验失败，使随机影响评估无法对感兴趣的假设提供有效的检验。缺乏对这些问题的深入认识及其应对策略的全面把握，阻碍了随机影响评估在实际政策或项目评估中的运用。本文通过对重要文献进行系统综述，围绕“存在问题–问题来源–潜在后果”，全面、深入地揭示了随机影响评估中存在的损耗、不依从、溢出效应、驱动效应、伦理、随机偏倚、外部有效性等七种问题，并对每一种问题的应对策略进行了系统梳理和概括。本研究有助于对评估者运用随机对照实验开展政策评估提供一定的指导，推进随机影响评估的开展。

Abstract: The “counterfactual” design of randomized impact evaluations can obtain the most credible causal inference, which is a standard tool in the toolbox of public policy or project evaluation. However, in reality, there are a number of problems with conducting random impact assessments, some of which can lead to minor barriers that can bias the estimation results, while others can lead to experimental failures that prevent random impact assessments from providing a valid test for hypotheses of interest. The lack of an in-depth understanding of these issues and their countermeasures hinders the use of randomized impact evaluations in the evaluation of actual policies or projects. This paper systematically reviews the important literature, and comprehensively and deeply reveals seven problems in randomized impact evaluations, including attrition, non-compliance, spillover effect, driving effect, ethics, random bias, and external effectiveness, focusing on “existing problems, problem sources, and potential consequences”, and systematically sorts out and summarizes the countermeasures for each problem. This study will help to provide some guidance for evaluators to use randomized controlled trials to conduct policy evaluations and promote the development of randomized impact evaluations.