Replication Package (qje-replication-package)

When to trigger

• The paper is heading toward acceptance and a data/code deposit is required
• You want a reproducible pipeline early, to pre-empt referee replication requests
• A reproducibility check or data editor asked for materials that regenerate every exhibit
• Raw data are restricted/proprietary and you need a disclosure-compliant plan

Why this matters at QJE

QJE's policy: it publishes papers only if the data are clearly documented and readily available for replication, and authors of accepted empirical/simulation/experimental papers must, prior to publication, provide the data, programs, and computation details sufficient to replicate the results. These are posted to the QJE Dataverse (the journal's repository on Harvard Dataverse), not openICPSR — a concrete difference from the AEA journals, even though QJE explicitly adopted the AER data availability policy. Source: academic.oup.com/qje/pages/data_policy. Proprietary-data exemptions exist but are narrow: you must flag proprietary data in the cover letter at submission, and even with an approved exemption you must still deposit the programs (the exemption criterion is that another investigator could in principle obtain the data independently). A README PDF documenting each file and how to run replication is required. The durable requirement is fixed: a stranger should regenerate every number, table, and figure from your deposit. Funding/conflict disclosure follows the AEA Disclosure Policy.

Package anatomy

replication/
  README.{md,pdf}        # the contract: data sources, software, run order, runtime
  data/
    raw/                 # as obtained (or access instructions if restricted)
    derived/             # built by code, never hand-edited
  code/
    00_setup            # installs/pins packages, sets paths
    01_build            # raw -> analysis sample
    02_analysis         # analysis sample -> estimates
    03_exhibits         # estimates -> every table & figure
  output/
    tables/  figures/    # regenerated, byte-checked against the paper

README must specify

• Data provenance: every source, version/vintage, access date; for restricted data, exact application steps and who can obtain it.
• Software environment: Stata/R/Python versions and pinned package versions (e.g., a requirements.txt, renv.lock, or a list of ssc/net packages with versions).
• Run instructions: a single master script (run_all) and the order; expected wall-clock runtime and hardware.
• Mapping: a table linking every exhibit in the paper to the script and line that produces it.
• Seeds: set and report random seeds for any simulation, bootstrap, or randomization inference.

Restricted-data plan

• State clearly which results can be reproduced with public data and which require restricted access. Per QJE policy, flag proprietary data in the cover letter at submission; even with an exemption you still deposit the programs, and the access path must let another investigator obtain the data in principle.
• Provide code that runs against the restricted data plus a synthetic or public subsample so the pipeline is checkable end-to-end.
• Include the data-use agreement constraints and a disclosure-review note where applicable.

Checklist

• One master script regenerates every table and figure from raw inputs
• Software and package versions are pinned and documented
• Every exhibit maps to a specific script/line in the README
• Random seeds set and reported for all stochastic steps
• Restricted/proprietary data have documented access steps + a runnable proxy
• No hand-edited derived data; all derived files are code-produced
• Absolute paths removed; runs from a single configurable root
• Output regenerated and checked against the manuscript numbers
• Deposit staged for the QJE Dataverse (not openICPSR); README PDF included
• Proprietary data flagged in the cover letter; programs deposited even if data exempt

Anti-patterns

• A zip of do-files with no README and no run order
• Hard-coded local paths (/Users/me/Desktop/...) that break on any other machine
• Derived datasets edited by hand and not reproducible from raw data
• Unpinned packages, so results drift with the next package update
• "Data available on request" with no access procedure for restricted sources
• Missing seeds, so bootstrap/RI numbers cannot be reproduced
• Assuming the deposit is optional — at QJE it gates publication of accepted empirical papers

Output format

【Master script】run_all present? [Y/N]
【Env pinned】Stata/R/Python versions + package versions documented? [Y/N]
【Exhibit map】every table/figure mapped to code? [Y/N]
【Seeds】set + reported? [Y/N]
【Restricted data】access steps + runnable proxy? [Y/N / NA]
【Reproduced output】matches paper? [Y/N]
【Next step】qje-referee-strategy