双重差分法的安慰剂检验:一个实践的指南  

Placebo Tests for Difference-in-differences:A Practical Guide

作　　者：陈强[1] 齐霁 颜冠鹏 Chen Qiang;Qi Jian;Yan Guanpeng(School of Economics,Shandong University;School of Economics,Shandong University of Finance and Economics)

机构地区：[1]山东大学经济学院 [2]山东财经大学经济学院

出　　处：《管理世界》2025年第2期181-203,共23页Journal of Management World

基　　金：山东省自然科学基金青年项目(基金号:ZR2024QG240)的资助。

摘　　要：双重差分法所依赖的平行趋势假定本质上不可检验,而处理效应也可能由不可观测的时变混杂因素所驱动,导致潜在偏差。为此,在DID实证研究中,越来越多地使用安慰剂检验进行证伪检验,或作为替代的统计推断方法;包括时间、空间、混合及外部安慰剂检验。但目前文献中仍存在一些认识与操作的误区。本文系统梳理了文献中DID安慰剂检验的不同方法,针对不同的DID设计(含标准DID、交叠DID、一般DID、连续DID、DDD及队列DID),详细介绍了安慰剂检验的基本原理与操作过程,并提出规范化建议。最后,使用团队开发的Stata命令didplacebo,结合经典案例演示了标准DID与交叠DID的安慰剂检验操作。The estimation of difference-in-differences(DID)models relies on the parallel trends assumption,which is fundamentally untestable.Moreover,treatment effects may be driven by other unobservable time-varying confounding events,which results in potential bias.To address these issues,placebo tests for estimating DID models have become increasingly popular in empirical work as falsification tests or alternative methods of statistical inference,which include in-time,in-space,mixed and external placebo tests.However,there are still some misunderstandings and dubious implementations in the literature.This paper systematically reviews different approaches of placebo tests for DID estimation.We introduce the basic principles and implementations for different DID designs including standard DID,staggered DID,general DID,continuous DID,DDD and cohort DID,and offer some practical guidelines.Using a Stata command didplacebo developed by our team,we demonstrate the implementations of placebo tests for standard and staggered DID designs using classic examples.

关 键 词：双重差分法 安慰剂检验 证伪检验 

分 类 号：O212.1[理学—概率论与数理统计]