---
name: ecj-replication-package
description: Use when assembling the data and code replication package for a The Economic Journal (EJ) manuscript to the RES / EJ Data Editor standard (DCAS-endorsed, Zenodo deposit, reproducibility check before final acceptance). Builds the package and README; it does not run the analysis itself.
---

# Replication Package (ecj-replication-package)

## When to trigger

- The paper is heading toward acceptance and the **EJ Data Editor** needs a reproducible deposit
- You want the package to pass the EJ Data Editor's reproducibility check on the first pass
- Some data are proprietary or restricted and you must request an exemption and document access
- You are setting up the project early so reproducibility is not a last-minute scramble

> Verify the current policy on the EJ Data Editor site (ejdataeditor.github.io) and the OUP
> Instructions before depositing. EJ runs **pre-acceptance reproducibility checks**: the paper is
> accepted for final publication only after results have been checked for reproducibility. The
> package is posted to the journal's Zenodo repository or another trusted repository and linked from
> the paper. It is essential to **request a data exemption at the point of first submission** if you
> face any access restrictions.

## What a passing package contains

1. **README** (the centerpiece) following the DCAS / Social Science Data Editors README template:
   - Overview of what the code does and the mapping from code → every exhibit and in-text number.
   - Data availability statement: source, terms, whether each dataset is public / restricted / proprietary, and exact access steps (registrations, memberships, monetary and time costs). State clearly if data cannot be shared and why, referencing the exemption requested at first submission.
   - Computational requirements: software + versions, packages + versions, OS, memory, and approximate run time.
   - Instructions to run: a single master script ordering everything end to end.
   - List of every table/figure/in-text number with the script and line that produces it.
2. **Data**: raw inputs (when license permits) and the code that builds analysis files from them. Provide complete documentation of all variables; if data are in a proprietary format (e.g., Stata `.dta`), also provide an ASCII/plain-text copy such as `.csv`. If raw data are restricted, include construction code plus a synthetic/simulated dataset that lets the pipeline run.
3. **Code**: a `master` script that reproduces every number, table, and figure from raw inputs, with relative paths and fixed seeds.
4. **Output**: log files and generated exhibits, so the editor can diff against the paper.

## Reproducibility discipline

- One master script; no manual steps, no hard-coded absolute paths, no "run cell 4 then cell 2."
- Set and record random seeds for any simulation, bootstrap, or ML step.
- Pin software and package versions; record them in the README and, where possible, in a lockfile/environment file.
- Every exhibit and in-text number in the paper is regenerated by the code — no hand-edited tables.
- Directory layout is clean: `data/` (raw, derived), `code/` (build, analysis), `output/` (tables, figures, logs).

## Restricted / proprietary data (the EJ exemption route)

- Request the exemption **at first submission**, not at acceptance — the EJ Data Editor stresses this timing.
- You may not need to deposit the data, but you must deposit the **code** and a precise access path so a third party with the same license can reproduce results.
- Provide a Data Availability Statement and, where feasible, a small simulated dataset matching the schema so the pipeline is executable.
- Confidential-data results may require a verification arrangement with the EJ Data Editor; document it.

## Checklist

- [ ] README follows the DCAS template (overview, data availability, requirements, run instructions, exhibit map)
- [ ] Deposit goes to the journal's Zenodo repository (or another trusted repository) with a license allowing replication
- [ ] Package layout matches EJ guidance: `1-paper`, `2-appendices`, `README.pdf`, `3-replication-package.zip`, and optional `4-confidential-data-not-for-publication.zip`
- [ ] Single master script reproduces every table, figure, and in-text number from inputs
- [ ] Software and package versions pinned and recorded
- [ ] Random seeds set and documented
- [ ] Relative paths only; runs on a clean machine in a fresh directory
- [ ] All variables documented; proprietary-format data also provided as ASCII/plain text
- [ ] Data availability statement covers each dataset (public / restricted / proprietary) with access steps and costs
- [ ] Restricted data: exemption requested **at first submission**
- [ ] Package re-run from scratch and output diffed against the paper, ready for the EJ Data Editor
- [ ] Current EJ/RES data policy (DCAS, Zenodo, EJ Data Editor) verified on the official pages

## Anti-patterns

- A zip of scripts with no README and no code → exhibit mapping
- Absolute paths (`/Users/me/...`) that break on any other machine
- Unset seeds so bootstrap/simulation numbers do not reproduce
- "Data available on request" with no construction code and no access detail
- Requesting a restricted-data exemption only at acceptance instead of at first submission
- Proprietary-only data with no ASCII/plain-text companion and no variable documentation
- Hand-edited tables that the code does not actually generate
- Submitting without re-running the package on a clean environment

## Output format

```
【Policy verified】EJ/RES data policy (DCAS, Zenodo, EJ Data Editor) checked on official pages [y/n]
【README】DCAS template sections present? [y/n each]
【Deposit】Zenodo (or trusted repo) + replication license attached? [y/n]
【Master script】reproduces all exhibits + in-text numbers from raw? [y/n]
【Versions + seeds】pinned/documented? [y/n]
【Data status】public / restricted (exemption at first submission) + access path; ASCII companion? [y/n]
【Clean-machine test】passed, ready for EJ Data Editor? [y/n]
【Next】ecj-submission
```