SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsmw-replication

mw-replication

Data & analytics

View repository

Updated 10-15 20:09

About this skill

Replication Package Construction (mw-replication)

Overview

Since 2023, 《管理世界》 has implemented theData and Code Availability Policy, and all accepted empirical, simulation, and experimental manuscripts must, after acceptance, submit to the editorial office adata and code package, and accept the editorial office'sreproducibility check. If the check fails, publication will be delayed, and it will be recorded in the author's file.

The goal of this skill: to make the replication package you submitpass the editorial office's review on the first try, without rework.

Trigger Timing

  • Upon receiving an R&R or conditional acceptance from 《管理世界》, the editorial office requires replication materials to be submitted within 30 days
  • Start writing the README on the project kickoff day (strongly recommended; the cost of making it up afterwards is at least 5 times higher)
  • The editorial office's data verification report is returned with a request for supplementary materials
  • Self-check before preparing to submit (《管理世界》 explicitly allows submission together with the manuscript, which can speed up subsequent review)

The Three Pillars of 《管理世界》 Policy

  1. Data archiving — all data used in empirical, simulation, and experimental work; data that cannot be made public must havecomplete provenance documentation
  2. Code archiving — code that can reproduceeverynumber, table, and figure in the paper
  3. README documentation — a Chinese document (English is also acceptable, but Chinese is preferred) that enables an independent replicator to run the entire workflow

Archiving location:the 《管理世界》 data and code hosting platform(uniformly managed by the editorial office; the accessible link will be provided by the editorial office in the acceptance notice).Not accepted statements such as "data available from the authors upon request".

Key Differences from the AER Policy (Most Important for Chinese Journal Authors)

Dimension AER（AEA） 《管理世界》
Primary language English README Chinese README preferred (English may be attached)
Mainstream software Stata + R + Python given equal weight Stata accounts for 90%+, R/Python environments need additional explanation
Restricted commercial data Fewer (FRED/IPUMS are mostly public) CSMAR / Wind / RESSET / GTA account for the majority,and can almost never be publicly redistributed
Variable construction Mostly uses ready-made macro/micro databases A large number ofmanually constructedvariables (policy events, government-business relations, city tiers, patent classification) → must be accompanied byvariable construction documentation
Hosting platform openICPSR 《管理世界》's self-built platform + some authors supplement with GitHub
Timing of replication check After acceptance, before publication (mandatory) After acceptance; editorial office check + spot checks on some manuscripts

Underlying logic: the AER template can be used for reference, butcannot be copied wholesaleThe core pain points of Chinese journals are "restricted commercial databases + many manually constructed variables," which must be addressed with a "data acquisition path document + variable construction manual."

Recommended directory structure

管理世界-复现包-<作者姓-论文短题>/
├── README.md                          # 中文，主入口
├── README_EN.md                       # 可选，英文摘要
├── LICENSE                            # 代码 MIT；数据按原始来源
├── 00_环境说明.txt                     # 软件版本、包版本、硬件、运行时间
├── data/
│   ├── raw/                           # 原始数据（受限数据放占位说明 .txt）
│   ├── intermediate/                  # 清洗后的中间数据集
│   ├── final/                         # 用于回归的最终样本
│   └── codebook/
│       ├── 变量构造说明书.xlsx          # 关键产出，逐变量
│       └── 数据来源清单.xlsx            # 每一份数据：来源、版本、获取日期
├── code/
│   ├── 00_setup.do                    # 路径、包安装、随机种子
│   ├── 01_数据清洗.do
│   ├── 02_变量构造.do
│   ├── 03_主回归.do
│   ├── 04_机制异质性.do
│   ├── 05_稳健性.do
│   ├── 06_表格输出.do
│   ├── 07_图形输出.do
│   └── run_all.do                     # 一键复现总脚本
├── output/
│   ├── tables/                        # 与正文表 1、表 2... 一一对应
│   └── figures/                       # 与正文图 1、图 2... 一一对应
└── docs/
    ├── 数据获取指引.pdf                # 受限数据如何申请/购买
    ├── 论文-表图清单.xlsx              # 表/图 ↔ 脚本行号 映射
    └── 修订对照表.pdf                  # 若为修改稿，列明本次修订项

Essential README sections (Chinese template)

1. Paper information

  • Paper title / authors / corresponding author email
  • Planned issue number (to be filled in after notification from the editorial office)
  • Version number of this replication package + last updated date

2. Data availability statement

Foreachdataset, the following must be stated:

  • Data source (institution / database / URL)
  • Access method (public download / commercial subscription / application-based access)
  • Access date and version number (e.g., CSMAR September 2024 version)
  • Whether it is included in this package; if not,reason + alternative access path

Restricted data standard statement template(apply directly):

The CSMAR Chinese listed company financial database used in this study is commercially subscribed data. According to the user agreement of Shenzhen CSMAR Data Technology Co., Ltd., the raw data may not be redistributed externally. This replication package provides: (a) complete variable construction code; (b) a variable construction manual; (c) a de-identified sample dataset (the first 100 observations) for code smoke testing; (d) data acquisition contact information, see docs/数据获取指引.pdf。

3. Data list

Table: file name | description | source | time span | number of observations | whether included

4. Computing environment

操作系统：Windows 11 / macOS 14
统计软件：Stata 18.0 MP（必需）
Stata 外部包：reghdfe 6.12, ftools 2.49, did 1.3.1, csdid 1.7
辅助软件：Python 3.11（用于第 7 节文本分析），见 requirements.txt
硬件：标准笔记本（16 GB 内存）
总运行时间：约 45 分钟

5. Code description

Describe each script in order: what it does, what it reads in, and what it outputs. Clearly specify the main entry script (e.g., run_all.do）。

6. Replication operation guide (written for unfamiliar replicators)

1. 解压本复现包到工作目录。
2. 打开 Stata 18，执行 do "code/00_setup.do" 完成包安装。
3. 编辑 00_setup.do 第 8 行 global path "你的路径"。
4. 执行 do "code/run_all.do"，约 45 分钟。
5. 全部表格、图形输出至 output/ 目录。
6. 与论文正文核对：见 docs/论文-表图清单.xlsx。

7. Table and figure replication cross-reference table (must be delivered)

Paper location Script file Starting line Output file
Table 1 06_表格输出.do 15 output/tables/tab1.tex
Table 2 06_表格输出.do 72 output/tables/tab2.tex
Figure 1 07_图形输出.do 12 output/figures/fig1.pdf
Appendix Table A1 06_表格输出.do 188 output/tables/tabA1.tex

8. Variable construction description

Foreachmain variable (dependent, explanatory, mechanism, control), explain in one line:

  • Variable name (consistent with the code and the paper)
  • Chinese meaning
  • Calculation formula
  • Data source field
  • Unit / logarithmic transformation / winsorize interval

Place in data/codebook/变量构造说明书.xlsx。

Coding discipline (must be followed from day one)

General

  • One master script(run_all.do) that runs end to end
  • All relative paths：data/raw/foo.dta，Neverwrite D:/我的论文/...
  • Explicitly set the random seed：set seed 20260525
  • Fixed package versions: in 00_setup.do , ssc install reghdfe, replace, and state the tested versions in the README
  • Every runwrite a log：log using logs/run_$S_DATE.log, replace

Stata (mainstream for Management World)

  • First line of each do file:version 18.0
  • Before any randomization:set seed
  • Avoid set more off and other display settings in shared scripts
  • Always use high-dimensional fixed effects reghdfe, do not xtreg, fe pair with i.year i.firm
  • Comments in Chinese are also acceptable, butensure UTF-8 encoding, otherwise cross-platform garbled text will occur
  • Use Chinese filenames with caution — the editorial review machines are mostly in Chinese Windows environments, but cross-platform compatibility is poor;it is recommended to name code files in English / numbers, data files may use Chinese

R / Python (auxiliary)

  • R: use renv::snapshot() to lock versions
  • Python：requirements.txt use == to lock versions
  • Any ML pipeline must set.seed() / numpy.random.default_rng(seed=...)

What the editorial office actually checks

  1. Is the README complete and does it conform to the template?
  2. Can the master script run from scratch on a clean machine?
  3. Do the numbers produced match the paper?(especially the main table coefficients and standard errors)
  4. Does every dataset have a source description?
  5. Do restricted data have complete provenance + alternative access paths?

If any item fails → return for revision, at most two rounds; if it still fails after two rounds → reject.

Common failure modes that get sent back (high-frequency items in editorial feedback):

  • Hard-coded absolute paths (E:/师姐论文/）
  • Calling a Stata package directly without installing it
  • The sample size of the main regression is inconsistent with the paper (intermediate data filtering not archived)
  • "Data comes from CSMAR" mentioned in one sentence, with no version and no fields
  • Tables were manually adjusted in Excel, and the script output does not match the main text
  • Missing variable construction documentation — reviewers cannot tell how "government-enterprise connection degree" is calculated

Special handling for restricted data

Chinese commercial databases (CSMAR / Wind / RESSET / Guotai'an / CEInet / Tonghuashun iFinD, etc.)almost all prohibit redistribution of raw data. Handling path:

  1. Do not provide raw data, but provide:
    • Complete data cleaning and variable construction code
    • For each external field, thefield name + database location(e.g., CSMAR / 公司研究系列 / 治理结构 / FS_Combas.A001000000）
    • A de-identified sample (such as the first 100 companies, a random sample, or synthetic data) so the code can be smoke-tested
  2. Provide in the READMEaccess guidance: database name, subscribing institution, application process, estimated cost, estimated timeline
  3. Consider submittingintermediate aggregated data(such as industry × year averages), allowing partial reproduction

Government microdata (CFPS / CHIP / CHARLS / CGSS):

  • Generally can be applied for access, butcannot be redistributed
  • Provide links to the complete application process (such as PKU ISSS, RUC NSD)
  • Providing cleaning code is sufficient

Related resources

Load on demand from this repository or plugin package:

  • Directory skeleton:examples/replication-skeleton/
  • Variable construction documentation template:templates/codebook-template.csv
  • Data access guidance template:templates/data-acquisition-guide.md

Pre-submission self-check list

  • The master script runs from scratch on another clean machine (a colleague's computer)
  • All paths are relativized
  • Random seeds have been set
  • All software + package versions are written into the README
  • README is in Chinese and covers the 8 required sections
  • Every table and every figure in the paper can be found in output/ the corresponding file
  • Paper-Table and Figure List xlsx has been filled in
  • Variable Construction Manual xlsx covers all main regression variables
  • Every dataset has a source description (including version and acquisition date)
  • Restricted data: the full set of three items—standard statement + access guidance + de-identified sample—is complete
  • Code file names in English/numbers, UTF-8 encoding without BOM
  • LICENSE file has been added
  • Directory structure is consistent with mw-replication recommendations

Output Format

【复现包类型】公开数据 / 受限数据 / 混合
【README 状态】完成 / 待补章节：...
【可重复性自测】已在另一台机器跑通 / 未测
【受限数据处理】N/A / 已附标准声明+获取指引+脱敏样本
【变量构造说明书】完成 / 缺：变量 X、Y、Z
【下一步】mw-submission

Anti-patterns

  • Waiting until the acceptance notice is received to start organizing — the editorial office's 30-day deadline is stressful
  • Writing the README as "notes for yourself" rather than "a guide for unfamiliar replicators"
  • "Data available from the authors upon request" — directly rejected
  • Manually editing the script output tables and pasting them into Word, then saying "the code is just like this" — the editorial office will find it does not match when they run it
  • Running with Stata 16 but writing Stata 18 in the README, or vice versa
  • Submitting only the main regression code, not the robustness / mechanism / heterogeneity code — those are also in the paper and all must be reproducible
  • Using Chinese paths or Chinese file names, which directly causes errors when run cross-platform
  • Packaging and uploading the original CSMAR dta files — this violates the database user agreement and may lead to liability