| 注册
首页|期刊导航|中国人口·资源与环境|因果推断经验研究中双重差分法的局限与改进

因果推断经验研究中双重差分法的局限与改进

陈南旭 李益 韩君

中国人口·资源与环境2026,Vol.36Issue(6):217-234,18.
✕
中国人口·资源与环境2026,Vol.36Issue(6):217-234,18.DOI:10.12062/cpre.20260101

因果推断经验研究中双重差分法的局限与改进

Limitations and improvements of the difference‑in‑differences method in causal inference empirical studies:evidence from sustainable development policies

陈南旭 1李益 2韩君3

作者信息

  • 1. 兰州大学经济学院,甘肃 兰州 730000||兰州大学数量经济研究所,甘肃 兰州 730000
  • 2. 兰州大学经济学院,甘肃 兰州 730000
  • 3. 兰州财经大学统计与数据科学学院,甘肃 兰州 730020
  • 折叠

摘要

Abstract

The difference-in-differences(DID)method is widely used to evaluate the causal effects of policies.However,when the re-search samples fail to satisfy the complete randomization assumption,its estimation results are often biased.To reduce possible biases,this study first systematically reviewed the evolution and econometric logic of the DID method.It found that under non-random sample conditions,the DID method faced sample selection bias that undermined randomness,the parallel trends assumption that was difficult to satisfy,and identification dilemmas caused by policy type differences.The study then proposed an improved strategic framework cen-tered on approximating random experiment conditions,comprising three stages:① pre-evaluation—constructing a basic randomized structure through scientific sample definition and randomization experiments to ensure homogeneity between treatment and control groups across multiple dimensions;② during-evaluation—systematically approximating randomized experiment conditions through methods such as enhanced propensity score matching and extended breakpoint regression designs;and ③ post-evaluation—testing ro-bustness of the conclusion through sensitivity analysis.Furthermore,considering that the implementation targets of the sustainable de-velopment policy pilot areas were not randomly selected and usually required consideration of factors such as environmental conditions,economic conditions,geographical location,and historical experience,this study classified sustainable development policies into envi-ronmental regulation type,economic incentive type,and social participation type,and proposed corresponding improvement strategies.For environmental regulatory policies,the extended regression discontinuity design was employed to alleviate the threshold dependence and spatial spillover problems.For economic incentive policies,the enhanced propensity score matching method was recommended to alleviate self-selection bias and treatment effect heterogeneity.For social participatory policies,the enhanced propensity score matching method was referred to integrate multi-dimensional information such as geographical distance,socio-economic characteristics,and cul-tural identity,followed by extended regression discontinuity design for evaluation.Practical application showed that the above-men-tioned improved methods enhanced the credibility of estimation results.

关键词

因果推断/双重差分法/选择偏差/政策评估/可持续发展政策

Key words

causal inference/difference-in-differences method/selection bias/policy evaluation/sustainable development policy

分类

管理科学

引用本文复制引用

陈南旭,李益,韩君..因果推断经验研究中双重差分法的局限与改进[J].中国人口·资源与环境,2026,36(6):217-234,18.

基金项目

中央高校基本科研业务费专项资金项目"能源约束下算力基础设施建设对西北地区城市社会承载力的影响研究"(批准号:2025lzu-jbkyxs004). (批准号:2025lzu-jbkyxs004)

中国人口·资源与环境

1002-2104

访问量0
|
下载量0
段落导航相关论文