Implement open science practices for a social science study. Five modes: (1) PREREGISTER — OSF / AsPredicted / EGAP / Registered Reports; secondary-data preregistration; deviation reporting; (2) DATA-SHARE — FAIR principles, de-identification, repository selection (Dataverse / ICPSR / Zenodo / QDR), social media policies, computational data (model weights, embeddings), data availability statements; (3) CODE-SHARE — minimum + gold-standard replication packages, README template, renv / conda / Docker / Makefile environments, CITATION.cff, Zenodo DOI via GitHub; (4) FULL-PACKAGE — data management plan (NSF/NIH template), CRediT authorship, COI, IRB statement, preprint strategy, open access + APC waiver guidance; (5) REPLICATION-PACKAGE — end-to-end audit against journal requirements. Required at Nature journals; increasingly expected at ASR, Demography, and Science Advances. Saves preregistration document, data management plan, and replication-package README to disk.
Installation
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Implement open science practices for a social science study. Five modes: (1) PREREGISTER — OSF / AsPredicted / EGAP / Registered Reports; secondary-data preregistration; deviation reporting; (2) DATA-SHARE — FAIR principles, de-identification, repository selection (Dataverse / ICPSR / Zenodo / QDR), social media policies, computational data (model weights, embeddings), data availability statements; (3) CODE-SHARE — minimum + gold-standard replication packages, README template, renv / conda / Docker / Makefile environments, CITATION.cff, Zenodo DOI via GitHub; (4) FULL-PACKAGE — data management plan (NSF/NIH template), CRediT authorship, COI, IRB statement, preprint strategy, open access + APC waiver guidance; (5) REPLICATION-PACKAGE — end-to-end audit against journal requirements. Required at Nature journals; increasingly expected at ASR, Demography, and Science Advances. Saves preregistration document, data management plan, and replication-package README to disk.
tools
Read, Write, Bash, WebSearch
argument-hint
[PREREGISTER|DATA-SHARE|CODE-SHARE|FULL-PACKAGE|REPLICATION-PACKAGE] [study description or journal name]
user-invocable
true
Scholar Open Science
You are an expert in open science practices for social science research. You help
researchers implement transparency, preregistration, data sharing, and reproducibility
in ways that strengthen their work and meet the increasingly stringent requirements
of top journals.
Dispatch Table
If $ARGUMENTS contains
Route to
preregister / prereg / osf / registration
Step 1 (Preregistration)
registered reports / rr / stage 1
Step 1 → Section 1c (Registered Reports)
secondary data / secondary analysis
Step 1 → Section 1f (Secondary-data prereg)
data-share / data share / data availability / deposit
Original data collection (survey, experiment, field study)
Nature Human Behaviour, Nature Computational Science: expected; non-preregistration must be stated in Methods
Science Advances: encouraged; flag as "preregistered" in cover letter
ASR/AJS: growing reviewer expectation; protects against p-hacking allegations
Still valuable (optional):
Secondary data analysis with pre-specified hypotheses (see Section 1f)
Replication studies — use AsPredicted "registered replication" format
Registered Reports (see Section 1c) — submit before data collection; two-stage review
Not applicable (say so explicitly):
Purely inductive qualitative work — note "inductive design; preregistration not applicable"
Archival/historical analysis where hypotheses emerge from sources
Exploratory computational work generating theory (label all analyses "exploratory")
1b. Preregistration Platforms
Platform
Best For
Embargo
DOI
URL
OSF Preregistrations
Social science; most flexible templates
Yes (4 years)
Yes
osf.io/registries
AsPredicted
Quick 9-question; confirmatory studies
Yes (until publication)
No
aspredicted.org
EGAP
Political science / field experiments
Yes
Yes
egap.org/registry
AEA RCT Registry
Economics / social science RCTs
Yes
Yes
socialscienceregistry.org
OSF Registered Reports
Pre-results peer review (any design)
No
Yes
osf.io/registries
1c. Registered Reports (RR) Format
Registered Reports are a special journal format where peer review occurs in two stages:
Stage 1: Methods + Introduction reviewed and conditionally accepted before data collection
Stage 2: Full results submitted; accepted regardless of findings (in-principle acceptance)
Journals accepting RRs in social/computational sciences:
Nature Human Behaviour, Psychological Science, PLOS ONE, Collabra, Advances in Methods and Practices in Psychological Science (AMPPS), Social Psychology — growing list at cos.io/rr
RR Manuscript Structure:
Stage 1 Submission:
├── Introduction (full theory + hypotheses)
├── Methods (complete design, sampling, measures, analysis plan)
├── Power analysis (justify N)
└── No Results or Discussion
Stage 2 Submission (after data collection):
├── Introduction (unchanged)
├── Methods (unchanged; note any deviations)
├── Results
└── Discussion
OSF RR Preregistration workflow:
Write Stage 1 manuscript
Submit to journal; receive In-Principle Acceptance (IPA)
Register on OSF with IPA letter attached
Collect data; write Stage 2 manuscript
Submit Stage 2 to same journal; link OSF registration
Registered Reports — Stage 1 Submission Template
Required sections for Stage 1 (In-Principle Acceptance):
Introduction & Rationale (800--1500 words)
Research question and significance
Theoretical framework
Hypotheses (numbered, directional)
Methods (1500--3000 words)
Study design (with diagram if complex)
Participants: target N, sampling strategy, inclusion/exclusion criteria
Materials/measures: all instruments, scales, variable operationalization
Procedure: step-by-step protocol
Analysis Plan (1000--2000 words)
Primary analyses: exact model specifications for each hypothesis
Secondary/exploratory analyses (clearly labeled)
Stopping rules: minimum N for adequacy; interim analysis plan (if applicable)
Exclusion criteria: pre-specified rules for excluding observations/participants
Multiple comparisons: correction method + which outcomes are primary vs. exploratory
Power analysis: target power (>=0.80), assumed effect size (with justification from prior work), sample size calculation (with code)
Robustness checks: pre-specified alternative specifications
Positive result criteria: what constitutes support for each hypothesis?
Interpretation of null results: planned equivalence test bounds (if applicable)
Timeline & Feasibility
Data collection start/end dates
IRB status
Funding status
Data access confirmed (for secondary data)
Stage 2 Deviation Documentation (template for later)
Any deviations from Stage 1 protocol must be documented with justification
Classify deviations: minor (does not affect interpretation) vs. major (affects primary analyses)
Journals accepting Registered Reports: Nature Human Behaviour, Cortex, Royal Society Open Science, European Journal of Social Psychology, Political Science Research and Methods, others (see COS registry: cos.io/rr/).
1d. OSF Preregistration Workflow
Step 1: Create OSF account → New Project → Add collaborators → Organize folders
OSF Project structure:
├── Preregistration/ ← Lock registration here
├── Data/ ← Deposit analysis dataset
├── Code/ ← Link to GitHub repository
└── Papers/ ← Preprint + published version
Step 2: Write the preregistration document (use template below — save as PDF and upload)
═══════════════════════════════════════════════════════════
PREREGISTRATION: [Paper Title]
Date: [Today's date]
Authors: [Names + affiliations]
OSF Project: [URL]
═══════════════════════════════════════════════════════════
STUDY INFORMATION
─────────────────
Title: [Matches planned paper title]
Research questions:
1. [RQ1 — specific, answerable]
2. [RQ2 — if applicable]
Hypotheses (numbered, directional):
H1: [Variable X] is positively/negatively associated with [Outcome Y]
among [Population Z].
Rationale: [2–3 sentences from theory section]
Direction: Positive / Negative / Non-linear [specify shape]
H2: The effect of X on Y is [stronger/weaker] for [Group A] vs. [Group B].
Rationale: [...]
H0 (null): [If testing null — e.g., "We expect no difference between groups
if the mechanism requires [condition] that is absent here."]
DESIGN PLAN
───────────
Study type: Observational survey / RCT / Field experiment / Secondary data analysis
Blinding: [Yes — double-blind / No — observational / Partial]
Study design: Cross-sectional / Longitudinal (waves: N=?) / Experimental
Randomization: [How assigned to conditions — N/A if observational]
SAMPLING PLAN
─────────────
Existing data: Yes / No
If yes: [Describe what you know; confirm no confirmatory analyses run yet]
Data collection: [Detailed procedure]
Sample size: N = [X]; justify via power analysis
Power analysis: We need N = [X] to detect d = [Y] / β = [Z] with 80%
power at α = .05 (two-tailed), using [pwr::pwr.t.test / simr / MDES].
Stopping rule: [Pre-specified; e.g., "close survey after N = 1,000 completes
or 90 days, whichever comes first"]
VARIABLES
─────────
Primary dependent variable: [Name], [operationalization], [scale/range]
Secondary DVs: [If any]
Primary independent variable: [Name], [operationalization]
Moderators: [If testing H2]
Covariates: [List each; one-phrase justification per covariate]
- Age: [Controls for life course variation in X]
- Female: [Controls for gender gap in Y]
ANALYSIS PLAN
─────────────
Primary model: [OLS / FE / logistic / Cox / multilevel] with
[robust SEs / clustered SEs at level Z / bootstrapped SEs]
Equation: Y_i = β₀ + β₁X_i + β₂C₁ + β₃C₂ + ε_i
Test for H1: β₁ significantly positive (p < .05, two-tailed)
Test for H2: β₃ (X × Group) significantly different from zero
Sample restrictions:
Include: [Age range; employment status; geography]
Exclude: [Missing on outcome (N ≈ ?); extreme outliers > 3 SD]
Transformations: [Log income; standardize continuous predictors (mean=0, SD=1)]
Missing data: [Listwise deletion / Multiple imputation (m=20, mice package)]
Multiple comparisons: [We test [N] hypotheses; apply [Bonferroni / BH-FDR /
none — hypotheses derived from single theory and are not independent]]
Exploratory analyses (not pre-specified; will be labeled "exploratory"):
- Heterogeneity by [demographic group]
- [Other post-hoc analyses]
SOFTWARE AND VERSIONS
─────────────────────
Primary: R version [X.X.X] / Stata [version] / Python [version]
Key packages: [fixest, marginaleffects, mice, etc. with versions]
Seed: set.seed([fixed number, e.g., 42])
OTHER
─────
IRB approval: [Protocol number / pending / exempt — see Section 4d]
Funding: [Grant number if applicable]
Embargoed: [Yes — publish upon paper submission / No — make public now]
═══════════════════════════════════════════════════════════
Once locked, cannot be modified — deviations must be reported in paper
Step 4: Cite in manuscript
"This study was preregistered prior to data collection at the Open Science
Framework (https://osf.io/[ID], registered [date])."
1e. AsPredicted Quick Format (9 Questions)
Use when you want a quick preregistration before accessing data:
1. Data collection. Have any data been collected already?
→ [No / Yes, but I have not conducted any confirmatory analysis]
2. Hypothesis. What's the main question / hypothesis?
H1: [Directional prediction]
3. Dependent variable. How measured?
[Name; scale; reference period; e.g., "Income in USD, continuous, past year"]
4. Conditions. How many conditions?
[Observational: N/A / Experimental: [list treatment arms]]
5. Analyses. Exact model?
[OLS / logistic / FE with robust SEs; controls: age, gender, race, education]
6. Outliers and exclusions. Criteria?
[Exclude missing on DV; winsorize income at 99th percentile]
7. Sample size. How many obs / what determines it?
[N = 1,200; secondary data PSID 2019 wave]
8. Other. Anything else to preregister?
[Exploratory: heterogeneity by race; will be labeled exploratory in paper]
9. Title. Give a title.
[Preregistration of: Paper Title]
1f. Secondary Data Preregistration
Even with existing data, preregistering before confirmatory analysis retains value:
Three scenarios and how to handle:
Scenario
Preregistration approach
Downloaded but not looked at
Full preregistration; note data downloaded but analyses not run
Looked at descriptives only
Partial preregistration; note "examined descriptive statistics; no confirmatory analyses run"
Fully analyzed exploratory
Cannot preregister confirmatory analysis; label all results "exploratory"
Language for secondary data preregistration:
Methods section:
"Although this study uses secondary data ([Dataset, year]), we preregistered
our hypotheses and analysis plan before conducting any confirmatory analyses
(OSF: https://osf.io/[code], registered [date]). Prior to registration, we
had examined [descriptive statistics / variable distributions] but had not
tested any directional hypotheses."
Language for un-preregistered secondary data:
"This study uses secondary data and was not preregistered. All reported
analyses should be interpreted as exploratory. To guard against inflated
Type I error, we [applied Bonferroni correction / report exact p-values
and encourage replication / conducted a pre-specified sensitivity analysis
with a held-out validation sample (N = [X])]."
1g. Reporting Deviations
Every deviation from the preregistration MUST be disclosed:
"Our preregistered analysis plan specified [X]. We deviate from this plan
in the following ways:
(1) [Deviation]: [We added a covariate (household size) not in the original
plan because reviewers of a related paper flagged it as a potential
confounder.] The preregistered specification (without this control) is
reported in Appendix Table A[X] and yields [consistent / slightly
different] results.
(2) [Deviation]: [We changed the SE clustering level from individual to
county after discovering within-county clustering in the residuals.]"
See references/preregistration.md for templates and platform comparisons.
Step 2: DATA SHARING
2a. FAIR Data Principles
FAIR = Findable, Accessible, Interoperable, Reusable.
Required for Nature journals; NHB/NCS reporting summary asks about each dimension.
Principle
What it means
How to implement
Findable
Data has persistent identifier (DOI) and rich metadata
Deposit in Zenodo/Dataverse with DOI; fill title, authors, subject, description
Accessible
Data retrievable via standard protocol; access conditions clear
Use HTTPS-accessible repository; specify who can access restricted data + how
Interoperable
Uses standard formats and vocabulary
Save as CSV/TSV (not Excel); use standard codebook with variable names + labels; apply domain ontologies where relevant
Reusable
Clear license; rich provenance; data described fully
Attach LICENSE (CC-BY or CC0); write codebook documenting all variables, value codes, missing codes, transformations
FAIR compliance checklist for manuscript:
Dataset deposited with DOI
Metadata: title, authors, abstract, keywords, date, version filled
Data in open format (CSV, TSV, JSON — not SPSS .sav or Stata .dta without conversion)
Codebook documents all variables (name, label, type, range, missing code)
LICENSE file present (CC0 for data; CC-BY or MIT for code)
Provenance documented (where data came from, transformations applied)
FAIR Principles — Operationalization Checklist
Findable:
Dataset has a globally unique persistent identifier (DOI via Zenodo/Dataverse/ICPSR)
Citation instructions provided (BibTeX entry for the dataset)
2b. Data Sensitivity Classification
Can you share the data?
│
├─ ORIGINAL COLLECTION
│ ├─ Fully de-identified + consent allows sharing
│ │ └─ → Deposit openly (Harvard Dataverse / Zenodo / OSF)
│ ├─ Sensitive (health, income, location) → de-identify first (Step 2d)
│ │ └─ → Deposit with access controls (ICPSR restricted / embargoed Dataverse)
│ └─ Cannot share (IRB restriction, no consent)
│ └─ → Code only; restricted-access data availability statement
│
├─ SECONDARY DATA
│ ├─ Public-use (ACS, GSS, CPS, HRS public, NLSY public)
│ │ └─ → Include processed data in package OR point to official download
│ ├─ Restricted-use (NLSY Geo, FSRDC, linkage files)
│ │ └─ → Code only; describe access pathway in paper
│ └─ Licensed/proprietary (Gallup, Nielsen, LexisNexis)
│ └─ → Code + derived aggregates if permitted; data access statement
│
├─ SOCIAL MEDIA (see Step 2f for platform-specific rules)
│ ├─ Twitter/X: Cannot redistribute raw; can share tweet IDs (rehydrate)
│ ├─ Reddit: pushshift data sharing restricted since 2023; share post IDs
│ └─ Derived data (DTM, embeddings, aggregated counts): usually shareable
│
└─ COMPUTATIONAL / ML (see Step 2e)
├─ Model weights: share via HuggingFace Hub or Zenodo
└─ Embeddings / feature matrices: share as CSV/numpy array via Zenodo
2c. Repository Selection
Repository
Field
Max size
Access control
Best for
Harvard Dataverse
Social science
2.5 GB/file, 1 TB/dataset
Public / restricted / embargo
Survey data, population datasets, sociology
ICPSR
Social/behavioral
Unlimited
Restricted-use support
Large surveys, administrative data, longitudinal
OSF
Broad
50 GB total
Public / private
Small datasets; integrates with preregistration
Zenodo
Broad/computational
50 GB/record
Public (versioned)
Code + data; NCS requirement; GitHub integration
QDR
Qualitative SS
Moderate
Project-based
Interview transcripts, field notes, mixed methods
UK Data Service
UK social science
Large
Restricted
UK-based population data
Harvard Dataverse deposit steps:
1. dataverse.harvard.edu → Log in → "Add Data" → "New Dataset"
2. Fill metadata: title, author, contact, description, subject, keywords
3. Upload files (data + codebook.pdf + README.md + code/)
4. Set access:
- Public: available immediately on publication
- Restricted: requestors must apply + sign DUA
- Embargo: public after [date] (use for pre-publication)
5. Publish → receives DOI (e.g., https://doi.org/10.7910/DVN/...)
Zenodo + GitHub auto-archive:
# 1. Go to zenodo.org/account/settings/github# 2. Toggle ON the repository# 3. Create a GitHub release:
git tag v1.0.0
git push origin v1.0.0
# → Zenodo auto-creates DOI for this release# Add badge to README:# []
2d. De-Identification Procedures
Direct identifiers — remove entirely:
Full name → participant ID (store ID-name lookup table separately, encrypted)
Exact address → state / county / 3-digit ZIP
Date of birth → age in years (or 5-year age bands if N < 100 per cell)
Exact event dates → year or year-quarter
Phone numbers, email addresses, SSNs → remove
Employer name → NAICS/SIC industry code
School name → institution type/level
Photos, voice recordings → store separately; do not deposit publicly
Quasi-identifiers — suppress or aggregate:
Apply k-anonymity: every combination of quasi-identifiers (sex × race × age × geography)
must appear at least k=5 times.
Cell suppression: Replace cells with N < 5 with NA in public tables
Top/bottom-coding: Income > $500K → "500K+" (report threshold in codebook)
Noise injection: Add small random noise to continuous variables (document in codebook)
2e. Computational Data Sharing (Model Weights, Embeddings)
For computational papers (NCS, Science Advances, CSS venues):
Artifact
Format
Repository
Notes
Fine-tuned model weights
HuggingFace Hub (safetensors)
HuggingFace + Zenodo for DOI
License: Apache-2.0 if based on Apache model
Pre-trained embeddings (dense vectors)
numpy .npy / CSV
Zenodo
Include column/row metadata
Document-term matrix
scipy sparse .npz / CSV
Zenodo
Include vocabulary file
Topic model (STM/LDA)
.rds (R) / .pkl (Python)
OSF or Zenodo
Include run script
Annotation labels (LLM/human)
CSV with doc_id, label, annotator
Zenodo
Include codebook + IAA scores
Network edgelist
CSV (source, target, weight)
Zenodo or Harvard Dataverse
Include node attribute file
HuggingFace Hub upload:
from huggingface_hub import HfApi
api = HfApi()
api.create_repo(repo_id="username/paper-model", repo_type="model")
api.upload_folder(
folder_path="./model_weights",
repo_id="username/paper-model",
repo_type="model"
)
# Add model card README with paper citation, training details, intended use
Linking to Zenodo for persistent DOI:
# After uploading to Zenodo manually or via API:# In paper: "Model weights and embeddings are archived at [DOI]"# In HuggingFace model card: "Archived version: [Zenodo DOI]"
2f. Social Media Platform Policies
Platform
Raw data
Tweet/Post IDs
Derived features
Notes
Twitter/X
Cannot redistribute
Can share (rehydrate via API)
DTM, embeddings: OK
Use twarc to rehydrate IDs
Reddit
Cannot redistribute
Can share post/comment IDs
Aggregated metrics: OK
Pushshift no longer public post-2023; use Reddit API
Facebook/Meta
Cannot redistribute
Cannot share
Aggregated only
CrowdTangle shut down 2024; use Meta Content Library (academic)
TikTok
Cannot redistribute
Cannot share
Aggregated only
TikTok Research API (apply)
YouTube
Cannot redistribute
Video IDs: OK
Aggregated metrics: OK
YouTube Data API v3
Mastodon
Federated — check instance ToS
Can share post URLs
Derived: OK
More permissive than corporate platforms
LinkedIn
Cannot redistribute
Cannot share
Aggregated only
Very restricted; academic access via approved partners
Tweet ID archive deposit:
# Save tweet IDs only (compliant with Twitter ToS)import pandas as pd
df_ids = df[['tweet_id', 'created_at']].copy() # IDs + timestamps only
df_ids.to_csv('tweet_ids.csv', index=False)
# Deposit tweet_ids.csv on Zenodo# Rehydration instruction in README:# twarc2 hydrate tweet_ids.csv --output hydrated_tweets.jsonl
Data availability statement for social media:
"The raw social media data cannot be redistributed per the platform's Terms of
Service. Tweet/post IDs sufficient for data reconstruction are available at
[Zenodo DOI]. The derived dataset (document-term matrix / aggregated engagement
counts) used in this analysis is also archived at [Zenodo DOI]. Data collection
and processing code is available at [GitHub URL]."
2g. Data Availability Statement Templates
Fully open (deposited):
"The data that support the findings of this study are openly available at
[Harvard Dataverse / Zenodo] at https://doi.org/[DOI], reference number [X]."
Restricted (apply for access):
"The data used in this study are available through [ICPSR / institutional
repository] under a restricted data use agreement. Instructions for obtaining
access are available at [URL]. The analysis code is available at [GitHub URL]."
Secondary public data (no deposit):
"The [General Social Survey / American Community Survey / NLSY] data used in
this study are publicly available and can be downloaded from [official source URL].
The derived analytic dataset and replication code are archived at [Zenodo/Dataverse
DOI]."
Qualitative (anonymized excerpts only):
"Interview transcripts cannot be publicly shared to protect participant
confidentiality per IRB protocol [number]. Anonymized excerpts supporting the
main findings are provided in the Online Supplementary Appendix. Requests for
additional access should be directed to the corresponding author."
Social media (IDs + derived):
[See Section 2f template above]
See references/data-sharing.md for repository setup, de-identification code, and FAIR
implementation.
Data Sharing Statements for Restricted Data
When data cannot be shared publicly:
Template 1 (Licensed data — PSID, NLSY, NHANES restricted):
"The data used in this study are available from [source] under a restricted-use Data Use Agreement. Researchers can apply for access at [URL]. Our analysis code is available at [repository URL] and can be applied to the data once access is granted."
Template 2 (IRB-restricted qualitative data):
"Interview transcripts contain identifying information and cannot be shared per our IRB protocol ([institution], #[number]). De-identified thematic summaries and the codebook are available at [repository URL]. Interested researchers may contact [email] to discuss data access."
Template 3 (Partner/proprietary data):
"The data were provided by [organization] under a data sharing agreement that prohibits redistribution. Interested researchers should contact [organization contact] to negotiate access. Our analysis code is available at [repository URL]."
If output/[slug]/scripts/ exists from a prior /scholar-analyze or /scholar-compute run, use it as the primary source for replication package code:
Copy and renumber scripts from output/[slug]/scripts/ into the replication package code/ directory. Preserve the original numbering where possible; renumber only if gaps or collisions exist.
Copy coding-decisions-log.md into the package root as DECISIONS.md — this provides reviewers and replicators with the rationale behind every analytic choice.
Use script-index.md to populate the README run-order table and paper-element correspondence table (Step 3b).
This eliminates the need to reconstruct scripts from memory or scratch — all code is already saved, self-contained, and documented with decision rationale.
If output/[slug]/scripts/ does not exist, proceed with the standard Step 3a approach below.
3a. Replication Package Structures
Minimum standard (ASR, AJS, Demography — Stata/R papers):
replication-[author-year]/
├── README.md ← Software versions; data sources; run order
├── data/
│ ├── raw/ ← Raw data files OR download instructions
│ └── analysis/ ← Processed analytic dataset (if shareable)
├── code/
│ ├── 01_clean.R ← Data cleaning → writes analysis/analysis_data.rds
│ ├── 02_main_models.R ← Tables 2–3
│ ├── 03_figures.R ← All figures → output/figures/
│ └── 04_appendix.R ← Supplementary analyses
└── output/
├── tables/
└── figures/
Gold standard (NCS, Science Advances — computational papers):
replication-[author-year]/
├── README.md
├── LICENSE ← MIT (code) + CC-BY or CC0 (data)
├── CITATION.cff ← Machine-readable citation metadata
├── Makefile ← Orchestrates full pipeline: make all
├── environment.yml ← Conda environment (exact package versions)
├── renv.lock ← R environment (if R code included)
├── Dockerfile ← Full container (optional but valued by NCS)
├── data/
│ ├── README.md
│ ├── raw/
│ └── processed/
├── code/
│ ├── 01_data_prep.py
│ ├── 02_model_train.py
│ ├── 03_analysis.py
│ ├── 04_figures.py
│ └── utils/
├── output/
│ ├── figures/
│ ├── tables/
│ └── models/ ← Saved model weights / checkpoints
├── paper/
│ └── manuscript.tex ← LaTeX source (if available)
└── tests/ ← Unit tests for key functions (pytest)
3b. README Template
# Replication Package for: [Paper Title]## Citation
[Author(s)]. ([Year]). [Paper Title]. *[Journal]*, *[Vol]*([Issue]), [Pages].
https://doi.org/[DOI]
Replication package archived at: https://doi.org/[Zenodo DOI]
## Overview
[2–3 sentences describing what this package reproduces: which tables, figures,
and analyses from the paper.]
## Software Requirements### R- R version 4.3.2 (or later)
- Restore package environment: `Rscript -e "renv::restore()"`- Key packages: fixest (>= 0.12.1)), modelsummary (1.4.5), marginaleffects (0.18.0)
### Python (if applicable)- Python 3.11
- Install: `conda env create -f environment.yml && conda activate [env-name]`- OR: `pip install -r requirements.txt`### System dependencies- [pdftotext, if used] — install via `brew install poppler` (macOS)
## Data Sources
| File | Source | Access | Notes |
|------|--------|--------|-------|
| `data/raw/gss_2022.csv` | GSS (NORC) | Download from gss.norc.org | Wave 2022 |
| `data/raw/acs_sample.rds` | ACS via tidycensus | Included (public) | 5-year 2019–2023 |
| `data/processed/analysis_data.rds` | Derived | Included | Output of 01_clean.R |
## Run Order
```bash
# Option A: Run full pipeline via Makefile
make all
# Option B: Run scripts manually in order
Rscript code/01_clean.R
Rscript code/02_main_models.R # → output/tables/table2.tex, table3.tex
Rscript code/03_figures.R # → output/figures/figure1.pdf – figure3.pdf
Rscript code/04_appendix.R # → output/tables/tableA1.tex – tableA4.tex
Output Files
Script
Output
Corresponds to
02_main_models.R
table2.tex, table3.tex
Tables 2–3 in paper
03_figures.R
figure1.pdf, figure2.pdf
Figures 1–2 in paper
04_appendix.R
tableA1.tex – tableA4.tex
Appendix Tables A1–A4
Expected Runtime
Full replication: ~[X] minutes on a standard laptop ([RAM] RAM, [CPU]).
ML training step (02_model_train.py): ~[X] hours on GPU; pre-trained weights
provided in output/models/ to skip this step.
Random Seeds
All random seeds are set at the top of each script: set.seed(42) (R);
np.random.seed(42); torch.manual_seed(42) (Python).
Contact
[Corresponding author name] — [email]
### 3c. Reproducibility Environments
#### R — renv
```r
# Initialize renv in project directory
renv::init()
# After installing all packages, snapshot:
renv::snapshot()
# → Creates renv.lock with exact package versions
# To restore on another machine:
renv::restore()
Python — Conda environment
# Create from scratch:
conda create -n paper-env python=3.11
conda activate paper-env
pip install -r requirements.txt
# Export to environment.yml:
conda envexport > environment.yml
# Prune to essentials (remove build strings for portability):
conda envexport --no-builds > environment.yml
# To recreate on another machine:
conda env create -f environment.yml
conda activate paper-env
# Dockerfile
FROM rocker/tidyverse:4.4.1
# Install system dependencies
RUN apt-get update && apt-get install -y \
libgdal-dev libproj-dev libgeos-dev \
&& rm -rf /var/lib/apt/lists/*
# Copy project
COPY . /paper
# Restore R environment
WORKDIR /paper
RUN Rscript -e "renv::restore()"
# Default command
CMD ["make", "all"]
# Build and run:
docker build -t paper-replication .
docker run --rm -v $(pwd)/output:/paper/output paper-replication
3d. Code Quality Standards
Script header template (use at top of every script):
# ============================================================# Script: 02_main_models.R# Purpose: Estimate main OLS + FE models (Tables 2–3)# Input: data/processed/analysis_data.rds# Output: output/tables/table2.tex, output/tables/table3.tex# Author: [Name]# Date: [YYYY-MM-DD]# Notes: Clustered SEs at state level; AME via marginaleffects# ============================================================
library(fixest)
library(modelsummary)
library(marginaleffects)
set.seed(42)# For reproducibility
data <- readRDS("data/processed/analysis_data.rds")
Anti-patterns to avoid:
Hardcoded absolute paths (/Users/myname/Desktop/...) — use project-relative paths
No script-level comments
Single 2,000-line omnibus script
No run order documentation
Results that require manually sourcing intermediate objects
rm(list=ls()) at top of script (destroys reproducibility)
3e. CITATION.cff
cff-version:1.2.0message:"If you use this code, please cite it as below."type:softwareauthors:-family-names:Zhanggiven-names:Yongjunorcid:"https://orcid.org/0000-0000-0000-0000"affiliation:"University Name, Department"title:"Replication code for: [Paper Title]"version:1.0.0doi:10.5281/zenodo.XXXXXXXdate-released:2025-01-15url:"https://github.com/username/repository"repository-code:"https://github.com/username/repository"license:MITpreferred-citation:type:articletitle:"[Paper Title]"authors:-family-names:Zhanggiven-names:Yongjunjournal:"[Journal Name]"year:2025doi:10.XXXX/XXXXXXXXXX
Add citation prompt to README:
## Citation
If you use this code or data, please cite the paper:
> [Author(s)]. ([Year]). [Title]. *[Journal]*. https://doi.org/[DOI]
And the replication package:
> [Author(s)]. ([Year]). *Replication code for: [Title]* (v1.0.0).> Zenodo. https://doi.org/10.5281/zenodo.XXXXXXX
3f. Zenodo DOI via GitHub
Step-by-step GitHub → Zenodo integration:
1. Go to zenodo.org → Log in → "GitHub" tab
2. Toggle ON the repository you want to archive
3. On GitHub: create a Release (tag v1.0.0)
→ Releases → "Draft a new release" → Tag: v1.0.0 → Publish release
4. Zenodo auto-creates a DOI for this release (within minutes)
5. Copy DOI badge and add to README:
[](...)
6. Every new Release = new Zenodo version with same concept DOI
Step 4: FULL OPEN SCIENCE PACKAGE
Run after Steps 1–3 to assemble the complete open science package required for
journal submission (especially Nature journals).
4a. Data Management Plan (NSF/NIH DMP)
Required for NSF SBE proposals; increasingly expected by NIH; good practice for all.
═══════════════════════════════════════════════════════════════
DATA MANAGEMENT PLAN — [Project Title]
PI: [Name], [Institution], [email]
Grant Program: [NSF SBE / NIH NICHD / Unfunded academic study]
Date: [Date]
═══════════════════════════════════════════════════════════════
1. TYPES OF DATA, SAMPLES, COLLECTIONS, SOFTWARE, AND OTHER
MATERIALS TO BE PRODUCED IN THE COURSE OF THE PROJECT
Primary data:
- [Dataset name]: [Brief description]. Format: CSV / SPSS / Stata .dta.
Estimated size: [X] MB / GB. Collection period: [dates].
- [Survey microdata / interview transcripts / scraped text corpus]: ...
Derived data:
- Analytic datasets (cleaned, de-identified): [format, size]
- Model outputs (embeddings, predictions, topic distributions): [format, size]
Software and code:
- Analysis scripts: R / Python / Stata. Version-controlled on GitHub.
- Configuration files: renv.lock / environment.yml
2. DATA FORMAT AND METADATA STANDARDS
- Primary data: CSV with UTF-8 encoding; SPSS .sav for survey data
- Metadata standard: [Dublin Core / DDI Codebook for surveys / schema.org]
- Codebook: Variable name, label, type, range, missing code, source for
each variable. Deposited with data.
3. POLICIES FOR ACCESS AND SHARING, REUSE, REDISTRIBUTION
- De-identified public-use data: deposited at [Harvard Dataverse / ICPSR /
Zenodo]; available publicly upon publication
- Restricted data (PII / health / linked administrative): deposited at
[ICPSR restricted] under DUA; available upon approved request
- Social media data: tweet/post IDs deposited; raw data not redistributed
per platform ToS
- Code: deposited on GitHub + Zenodo under MIT license
- Embargo period: [None / Until [date] / Until publication]
4. POLICIES AND PROVISIONS FOR RE-USE, RE-DISTRIBUTION, DERIVATIVES
- License: CC-BY 4.0 for data; MIT for code
- Attribution requirement: cite original paper using CITATION.cff
- Prohibited uses: re-identification of individuals; direct commercial use
without consent of PI
5. PLANS FOR ARCHIVING AND PRESERVING ACCESS TO DATA AND CODE
- Repository: [Harvard Dataverse (social science data) + Zenodo (code)]
- Preservation period: Minimum 10 years post-publication; indefinite for
open repositories
- Version control: Git; GitHub releases tagged at submission and publication
- Backup: [Institutional server / cloud backup policy]
- Format migration: CSV/TSV are open formats; no migration risk
- DOI registration: Both datasets and code will receive persistent DOIs
6. DATA SECURITY AND PRIVACY PROTECTIONS
- PII handling: all direct identifiers removed before analysis; ID lookup
table stored encrypted on institutional server; not deposited
- IRB compliance: IRB protocol [number] governs data collection;
consent form includes data sharing language
- Restricted data: HIPAA-compliant storage; DUA required for sharing
- Secure storage during project: [institutional secure research storage /
encrypted local drive]
═══════════════════════════════════════════════════════════════
NSF-specific DMP notes (SBE 2-page DMP):
Must address: types of data; data format; metadata standards; policies for access/sharing; policies for re-use; plans for archiving
NSF Public Access Plan (2023): NSF-funded papers must be deposited in an approved repository; data must have a DOI
NIH-specific DMP notes (DMSP — Data Management and Sharing Plan):
Required for all NIH submissions since January 2023
Must specify: data type, consent and IRB, de-identification, repository, timeline for sharing, responsible party
NIH prefers: NIMH Data Archive; dbGaP (genetics); ICSPR; Zenodo
Maximum 2 pages
4b. CRediT Authorship Table
CRediT (Contributor Roles Taxonomy) is required by NHB, NCS, and PNAS.
List all 14 roles; mark each author's contribution.
Template 1 (No conflict):
"The authors declare no competing interests."
Template 2 (Funding conflict):
"[Author X] has received research funding from [organization] for projects
related to [topic]. This work was not funded by [organization]. The remaining
authors declare no competing interests."
Template 3 (Advisory role):
"[Author X] serves on the scientific advisory board of [organization], which
works in a domain related to this research. [Author X] was not involved in
[specific aspect]. The remaining authors declare no competing interests."
Template 4 (Equity/employment):
"[Author X] holds equity in [company] that produces products related to the
topic of this research. All analyses were conducted independently. The remaining
authors declare no competing interests."
4d. IRB Statement
Template 1 (IRB approved):
"All procedures were approved by the [University] Institutional Review Board
(protocol [number], approved [date]). [Participants / Subjects] provided
[written / verbal / electronic] informed consent prior to participation."
Template 2 (IRB exempt):
"The [University] Institutional Review Board determined this research to be
exempt from full review under 45 CFR 46.104(d), category [2/4]
(protocol [number])."
Template 3 (Secondary data — IRB exempt):
"This study uses publicly available secondary data with no individual
identifiers. The [University] IRB determined this research to be exempt
(protocol [number])."
Template 4 (Waived — social media public data):
"Data were collected from public social media profiles. The [University]
IRB reviewed this protocol and waived the requirement for informed consent
for publicly available data (protocol [number])."
4e. Open Access Strategy
Scenario
Recommended action
NSF- or NIH-funded
Deposit preprint on SocArXiv/SSRN before submission; journal version goes through publisher OA after acceptance
Science Advances target
APC required (~$4,950); verify NSF/NIH public access compliance; APC waivers available for AAAS members + low-income affiliations
Nature journal (NHB/NCS)
Hybrid OA (~€9,500 APC); apply for APC waiver if low-income affiliation; check if institution has Springer Nature agreement
ASR/AJS/Demography target
Post preprint to SocArXiv (all allow preprints); journal version paywalled unless paying hybrid OA
Institutional read & publish deal
Check library portal — many universities have Springer/Elsevier/Wiley agreements that cover APCs
No funding for APC
SocArXiv/SSRN preprint provides open access; sufficient for most community visibility
Preprint server guide:
Server
Field
URL
SocArXiv
Sociology, social science
osf.io/preprints/socarxiv
SSRN
Social science, law, economics
ssrn.com
arXiv (cs.SI, cs.CY, stat.AP)
Computational social science
arxiv.org
PsyArXiv
Psychology
psyarxiv.com
EarthArXiv / ESSOAr
Environmental / earth science
eartharxiv.org
bioRxiv / medRxiv
Biology / health
biorxiv.org
APC waiver procedures:
Nature journals waiver:
1. Submit paper; if accepted, go to OA billing page
2. Search author institutional affiliation
3. If institution has Springer Nature deal → covered
4. If LMIC affiliation → automatic waiver option
5. If NSF/NIH funded → upload grant number; publisher checks compliance
AAAS / Science Advances:
1. After acceptance, AAAS membership number can reduce fee
2. Waivers: submit written request to oa@aaas.org before APC invoice
Step 5: REPLICATION PACKAGE AUDIT
Run this step to verify an existing replication package meets journal standards before submission.
5a. Audit Checklist
Structure (required by all journals):
README.md present and complete (software versions, data sources, run order, expected output)
Scripts numbered and named descriptively (01_clean, 02_analyze, 03_figures)
All relative paths — no hardcoded absolute paths
Output files generated by scripts (not manually placed)
Random seeds set and documented
Run test (do this before submission):
# Clone into a fresh directory and run from scratch
git clone [repo] test-replication
cd test-replication
# Restore environment
conda env create -f environment.yml # OR: Rscript -e "renv::restore()"# Run full pipeline
make all # OR: run scripts in numbered order# Check: all output files generated? All tables match paper?
Data compliance:
No raw data with PII included
Data availability statement matches what is actually deposited
License file present
CITATION.cff present
Script archive integration (if output/[slug]/scripts/ was produced by scholar-analyze/compute/full-paper):
Scripts from output/[slug]/scripts/ incorporated into code/ directory (not reconstructed from scratch)
DECISIONS.md present in package root (copied from coding-decisions-log.md)
All scripts have standard headers (purpose, input, output, date, seed)
Script-index run-order table used to populate README
Process Logging (REQUIRED) — Reasoning · Action · Observation trace:
This skill emits an append-only RAO trace at ${OUTPUT_ROOT}/logs/trace-scholar-open-<date>.ndjson — the source of truth. The human-readable process-log-scholar-open-<date>.md is rendered from it. Full protocol + privacy rule: _shared/process-logger.md.
At each meaningful step (a decision, a script/tool run, a gate call, a subagent dispatch), append one record. emit-trace.sh derives seq from the file, so no state is tracked across the stateless Bash blocks:
Privacy (C-01 / LOCAL_MODE): the trace carries aggregate metrics, verdicts, counts, and file refs ONLY — never raw data rows, verbatim quotes, or PII.
Version Collision Avoidance (MANDATORY)
Before EVERY Write tool call below, run this Bash block to determine the correct save path. Do NOT hardcode paths from the filename templates — they show naming patterns only.
# MANDATORY: Replace [slug] and [YYYY-MM-DD] with actuals before running
OUTPUT_ROOT="${OUTPUT_ROOT:-output}"# BASE pattern: ${OUTPUT_ROOT}/[slug]/preregistration-[slug]-[YYYY-MM-DD]
OUTDIR="$(dirname "${OUTPUT_ROOT}/[slug]/preregistration-[slug]-[YYYY-MM-DD]")"
STEM="$(basename "${OUTPUT_ROOT}/[slug]/preregistration-[slug]-[YYYY-MM-DD]")"mkdir -p "$OUTDIR"
bash "${SCHOLAR_SKILL_DIR:-.}/scripts/gates/version-check.sh""$OUTDIR""$STEM"mkdir -p "$(dirname "$BASE")"echo"SAVE_PATH=${BASE}.md"echo"BASE=${BASE}"
Use the printed SAVE_PATH as file_path in the Write tool call. Re-run this block (with the appropriate BASE) for each additional file. The same version suffix must be used for all related output files (.md, .docx, .tex, .pdf).
Preregistration: Study preregistered on OSF / AsPredicted / EGAP with directional hypotheses and exact analysis plan; OR non-preregistration statement prepared
IRB approval: Protocol approved / exempt determination obtained; consent form includes data sharing language
DMP written: NSF/NIH DMP completed (if funded) or internal DMP for data governance
Pre-analysis plan: Hypothesis, estimator, SE type, covariates, exclusions, and exploratory analyses pre-specified
Data Sharing
FAIR compliance: Dataset deposited with DOI; open format (CSV/TSV); codebook attached; license specified (CC-BY / CC0)
De-identification: All 18 HIPAA identifiers removed; k-anonymity (k≥5) verified with sdcMicro
Repository: Data deposited at Harvard Dataverse / ICPSR / Zenodo / QDR as appropriate for data type
Platform policies: Social media data — raw not distributed; IDs or derived features only; platform ToS reviewed
Computational artifacts: Model weights / embeddings / annotation labels archived at Zenodo or HuggingFace Hub
Code & Reproducibility
Replication package structure: Numbered scripts, README, output/, data/ with README
Environment pinned: renv.lock (R) or environment.yml (Python) with exact package versions
Seeds documented: All random seeds set and noted in scripts and README
Clean-run verified: Package runs from scratch in fresh environment; all output files regenerated
CITATION.cff: Present; includes paper DOI and package DOI
In the Manuscript
Preregistration citation: OSF URL + registration date in Methods section
Data availability statement: Matches actual repository; includes DOI
Code availability statement: GitHub URL + Zenodo DOI; includes version tag
Confirmatory vs. exploratory labeled: All non-preregistered analyses explicitly marked "exploratory"
CRediT statement: All 14 roles assigned (required by NHB, NCS, PNAS)
COI declaration: "No competing interests" or specific conflicts named
Preprint posted: SocArXiv / SSRN / arXiv preprint linked in cover letter
See references/preregistration.md for preregistration templates and platform comparisons.
See references/data-sharing.md for FAIR implementation, de-identification code, and repository setup.
