技能 数据科学 学术数据和结果复现指南

学术数据和结果复现指南

v20260724
rof-replication-and-data-policy
本指南详细介绍了学术论文提交时的数据复现和代码共享政策。它指导用户如何准备完整的复制包,包括处理专有数据(使用伪数据集和日志),撰写数据可用性声明,确保所有研究结果的完全可追溯性,从而降低出版风险。
获取技能
476 次下载
概览

Review of Finance Replication And Data Policy

Use this before submission and after acceptance. Reopen the current RoF code-sharing and data-availability policy before final packaging.

Package checklist

  • Prepare replication programs with comments, software names, versions, package dependencies, and execution order.
  • Provide actual data when non-proprietary. If data are proprietary, prepare a pseudo-dataset plus log files or other evidence required by current policy.
  • Draft a Data Availability Statement and formatted dataset citations.
  • Request any exception in the cover letter at initial submission when policy requires early disclosure.
  • Expect publication to wait until code is received when the policy applies.
  • Package materials for OUP supplementary hosting rather than inventing a separate archive.

RoF replication packet

Build the packet around editorial risk:

  • README states execution order, software, package versions, expected runtime, and output locations.
  • data_manifest names every raw, intermediate, pseudo, and restricted dataset, with access status.
  • run_all recreates each table, figure, and appendix exhibit or clearly marks non-runnable restricted steps.
  • logs prove that the submitted outputs were generated from the submitted programs.
  • DAS and dataset citations match the manuscript, cover letter, and supplementary files.

For proprietary data, the pseudo-dataset is not decoration. It should let the editor see variable names, file structure, merge keys, and code flow, while logs or other evidence document the confidential run. Request exceptions at the earliest required stage; late exceptions create publication risk.

OUP supplementary package blueprint

rof_replication/
  README.md            run order, software + versions, runtime, output map
  data/raw/            non-proprietary inputs only, with dataset citations
  data/pseudo/         simulated rows mirroring restricted files (schema-true)
  code/01_build        cleaning: return screens, merges, winsorizing
  code/02_main         every body table and figure, numbered as in the paper
  code/03_appendix     internet-appendix exhibits
  logs/                full execution logs from the confidential run
  output/              regenerated exhibits, traceable to the submitted PDF

Finance-data shipping reality

Source Ships in the package? What ships instead
CRSP / Compustat (WRDS) No — licensed exact query, extraction date, variable list, pseudo-rows
Datastream / Refinitiv No request scripts, screen list incl. dead-stock handling, pseudo-data
Bankscope / Orbis (BvD) No; vintages also vanish vintage date, consolidation filters, ID list if the license allows
ECB SDW / Eurostat Usually yes series codes plus download date, or the snapshot itself
Hand-collected (filings, prospectuses) Yes — expected the dataset plus the collection protocol
Proprietary (exchange, supervisor, bank) No pseudo-dataset, logs, access procedure, exception named in the cover letter

Confirm the current shipping rules for each source against the journal's current author guidelines and your own license terms before promising anything in the DAS.

Vintage and version control

  • BvD products overwrite history: pin the Orbis/Bankscope vintage in the README and store per-table row counts so the editor's office can audit drift in any re-pull.
  • WRDS tables are backfilled; record extraction dates next to every query.
  • Pin software environments (Stata version plus ado dates, renv or requirements files); the expectation behind RoF's policy is that the editorial side can execute the package, not reverse-engineer it.

Pre-acceptance dry run

  • Clone the package to a clean machine or container and run only what the README says — no tribal knowledge allowed.
  • Trace every number in the submitted PDF to a line in output/; anything untraceable is either a stale exhibit or a missing script.
  • Run the pseudo-data through the full code path: identical merge keys, variable types, and column order, so structure errors surface before the editor finds them.
  • Reconcile four documents — manuscript DAS, cover letter, README, dataset citations (Chicago style, matching the manuscript's reference format) — into one consistent story.

Worked exception case

Illustrative. A microstructure paper combines proprietary order-book data from a European exchange with public Datastream prices. The packet ships: complete code; a pseudo order-book with the true schema and fabricated quantities; logs from the confidential run; Datastream request scripts; a cover-letter paragraph requesting the exception at initial submission — the stage RoF requires — citing the exchange's redistribution restriction; and a DAS routing readers to the exchange's access procedure. Publication then waits only on packet verification, not a post-acceptance scramble. Verify the exact exception mechanics against the journal's current author guidelines.

Output format

[Replication status] ready / needs fixes / exception needed
[Data status] public / proprietary / restricted / simulated
[Code status] complete / missing pieces / undocumented
[DAS] <draft or missing>
[Publication risk] <low/medium/high>
信息
Category 数据科学
Name rof-replication-and-data-policy
版本 v20260724
大小 5.67KB
更新时间 2026-07-29
语言