A General Empirical Workflow For DID

通用 DID 实证 Workflow

A General Empirical Workflow for Difference-in-Differences
作者 / Author：许文立 Wenli Xu

机构 / Affiliation：澳门城市大学金融学院 / Faculty of Finance, City University of Macau

邮箱 / Email：wlxu@cityu.edu.mo
中文 | 适用场景：交错处理时间（Staggered DID）、多期面板数据、存在异质性处理效应的政策评估研究。

English | Applicable to: staggered treatment timing, multi-period panel data, and policy evaluation with heterogeneous treatment effects.
参考文献 / References：Callaway & Sant’Anna (2021), Sun & Abraham (2021), Borusyak et al. (2024), de Chaisemartin & D’Haultfœuille (2020), Wooldridge (2021), Roth (2022), Roth & Sant’Anna (2023), Oster (2019).

目录 / Table of Contents

1. 数据与描述性统计 / Data and Descriptive Statistics
1. 识别策略与前提检验 / Identification Strategy and Prerequisites
1. 基准回归 / Baseline Regression
1. 平行趋势检验：事件研究图 / Parallel Trends: Event Study Plots
1. 平行趋势敏感性：HonestDiD / Parallel Trends Sensitivity: HonestDiD
1. 排除备择解释 / Ruling Out Alternative Explanations
1. 稳健性检验 / Robustness Checks
1. 异质性处理效应估计量 / Heterogeneous Treatment Effect Estimators
1. 机制分析 / Mechanism Analysis
1. 异质性分析 / Heterogeneity Analysis
1. 福利与政策含义 / Welfare and Policy Implications
1. 完整 Workflow 一览 / Complete Workflow Overview
1. 软件包清单 / Software Packages
1. 参考文献 / References

1. 数据与描述性统计 / Data and Descriptive Statistics

1.1 样本构建 / Sample Construction

中文

- 明确面板单元（省份 / 城市 / 企业）、时间跨度、处理变量定义
- 报告数据来源、缺失值处理、winsorize 方案
- 明确 cohort 结构：各批次处理组的数量与启动时间分布

English

- Define the panel unit (province / city / firm), time span, and treatment variable
- Document data sources, missing value treatment, and winsorization scheme
- Describe the cohort structure: number of treated units and policy onset dates per cohort

1.2 描述性统计表 / Descriptive Statistics Table

中文写作提示：若基期处理组与控制组存在显著差异，需在正文中说明识别策略如何处理，而非回避。

Writing tip: If pre-treatment differences…