Replication materials for: Digital Openness, Institutional Innovation, and Urban Economic Resilience: Evidence from China's Cross-Border E-Commerce Pilot Zones

Summary** This deposit contains the data, source archive, processing code and complete numerical results supporting the manuscript "Digital Openness, Institutional Innovation, and Urban Economic Resilience: Evidence from China's Cross-Border E-Commerce Pilot Zones" (authors: Jie Wang and Wei Han). It is the full replication package referenced as "Supplementary Data 1". The study asks whether Chinese cities designated as cross-border e-commerce comprehensive pilot zones before the COVID-19 shock show different short-run output resistance and recovery than cities designated after it. The comparison uses 21 cities directly designated in 2018 and 51 cities directly designated in 2022. Outcomes are built from published comparable-price (constant-price) whole-city GDP growth rates for 2019–2021 and are expressed in log points, defined as 100 × ln(1 + g/100): impact — 2020 annual growth, a(2020); slowdown — change in annual log growth from 2019 to 2020, a(2020) − a(2019); rebound — 2021 annual growth, a(2021); cumulative — cumulative output growth 2019–2021, a(2020) + a(2021). Main results. After adjustment for 2014 log GDP, 2014 secondary/tertiary industry shares and documented 2017 growth, the 2018-vs-2022 cohort difference is −0.28 log points (95% interval −2.28 to 1.72) for the slowdown and −0.64 log points (−3.15 to 1.87) for cumulative 2019–2021 growth. All twelve main confidence intervals include zero and all wild bootstrap p-values exceed 0.05. The unadjusted slowdown difference is +0.17 log points, so the point estimate changes sign across adjustment sets. The estimates therefore describe observational cohort associations, not causal policy effects, and do not establish equivalence between cohorts. Contents of the deposit (1,483 files; approximately 205 MB uncompressed, delivered as Review Materials.zip with Review_Materials/ as the archive root): Data/ — cleaned analysis inputs: Context_280_Cities.csv — 280-city source frame, city identifiers, designation year, 2014 GDP and industry shares, published growth 2018–2023, and derived log-point outcomes. Analysis_72_Cities.csv — the 72-city focused comparison (21 + 51): treatment indicator, province code, baseline covariates, 2017 pre-policy growth with source file/vintage/provenance tier, all growth rates and outcomes, linked pandemic exposure and provincial stringency measures, and an explicit interpretation flag. Panel_72_Cities.csv — long (city × year) version of the growth series with period-use labels (504 city-year records, 2017–2023). Final_Growth_Extract.csv — source-level growth extract (1,678 city-year records) with source file, cell locator, URL, SHA-256, extraction method, validation state, correction URLs and revision status. Results/ — all machine-readable outputs: Main_Estimates.csv, Sensitivity_Estimates.csv, Vintage_Sensitivity.csv, Leave_One_City_Out.csv (288 estimates), Independent_Numeric_Checks.csv (376 coefficient/covariance checks against statsmodels), Official_Vintage_Crosscheck.csv, New_Source_Corrections.csv, New_Official_Source_Checks.csv, Summary_Statistics.csv, Sample_Flow.csv, All_Cohort_Context.csv, and Analysis_Validation.json. Figures/ — Figure_1_Trajectories and Figure_2_Dispersion (PNG and PDF). Code/ — analyze_revision.py (input construction, estimation, diagnostics, figures), write_documents.py (regenerates manuscript tables from result files), validate_final_consistency.py (checks all eight manuscript tables against numeric inputs), verify_new_sources.py, and create_documents.js (Word/PDF rendering, optional). run_all.py — single root runner; executes the source checks, reconstruction, estimation, figures and document generation end to end. Documentation/ — Replication_Guide.md (execution and exhibit map), Variable_Dictionary.csv, Source_and_Rights.md, Analysis_Decision.md, Data_Availability.txt, Runtime.json, plus Historical_Audit/ records of earlier checks. New_Sources/ — additional official bulletins, the public Xuchang API response (source of the 2019 correction from 9.0 to 7.1), institutional policy material and verified reference metadata. Public_Recovery/ — downloaded official economic sources, labelled bulletin reprints, the Tian et al. city-level COVID-19 release, the Oxford COVID-19 Government Response Tracker (OxCGRT) release, and its extraction/diagnostic scripts. Prior_Verification/ — historical: the earlier source reconstructions and broader 165/216-city comparisons, with SHA-256 manifests. These are retained for transparency and are not the current main analysis. Validation/, SHA256SUMS.txt — integrity manifests and clean-rerun verification records. Reproducing the results Use Python 3.11+ in a virtual environment; install requirements.txt (numpy, pandas, scipy, statsmodels, openpyxl, lxml, pypdf, matplotlib). Run python /path/to/Review_Materials/run_all.py from any directory — all paths resolve relative to the scripts and no absolute author directory is required. All source files are bundled, so the computation runs offline. Allow several minutes and approximately 1 GB of working space. Tested configuration: Python 3.12.14 (see Documentation/Runtime.json). Word/PDF rendering additionally needs Node.js with the docx package, or LibreOffice; this is separate from the statistical pipeline. Data sources and provenance Economic inputs come from China City Statistical Yearbook tables, municipal and provincial statistical yearbooks, and statistical bulletins (National Bureau of Statistics; local statistical authorities, 2018–2025). Pandemic exposure is linked from the Tian et al. city-level COVID-19 release (https://github.com/huaiyutian/COVID-19_TCM-50d_China) and provincial response intensity from the Oxford COVID-19 Government Response Tracker (https://github.com/OxCGRT/covid-policy-dataset). Exact repository versions, original attribution and licence notices are retained in the archive. Documented discrepancies (Xiangyang 2020, Shenzhen 2022 decimal corrections; the Xuchang 2019 9.0 → 7.1 correction; alternative publication vintages) are recorded rather than silently overwritten. Missing observations retain their missing-value codes and are never imputed as zero. Licence and rights The raw materials are statistical publications and existing research data releases, not new surveys; public accessibility does not imply transfer of third-party copyright. The authors assert no blanket licence over government publications or third-party datasets — the OxCGRT materials retain their CC BY 4.0 notice and the Tian et al. code retains its MIT statement, which covers code only. The cleaned datasets, derived variables, documentation and code in this deposit are released under CC BY 4.0 (see the licence field), on the condition that the underlying sources are cited as specified in Documentation/Source_and_Rights.md. Scope and limitations The analysis is observational. It uses aggregate secondary data only, contains no confidential human records, and has no dependency on restricted databases. Selection into the pilot-zone programme, heterogeneous pandemic exposure, limited covariate overlap (effective sample sizes of roughly 7–10 per group under overlap weights) and mixed publication vintages constrain interpretation; the design is cross-sectional and is not a difference-in-differences estimator. A successful rerun demonstrates computational reproducibility, not causal identification. The earlier composite measure and mediation findings from the original submission are retired from the revised manuscript and their upstream replication is not claimed by this package.