Replication Data for: Governance Reform Breaks Decarbonization Dead Zones in Public Procurement

Replication Package

Complete data and code for the Nature Sustainability article analysing competition–carbon relationships in 21.6 million government contracts across 27 countries (2012–2023).

Key Findings

• EU-context carbon premium: −4.3% (d = −0.08, t = −110; N = 13,638,933)
• 22 Decarbonization Dead Zones locking €190–250 billion annually beyond GPP reach
• Staggered DiD (Directive 2014/24/EU): ATT = −5.6 pp (permutation p = 0.042, pre-trend F = 1.32, p = 0.27)
• RDD at €139k threshold: +15.2% bidders, −0.33% carbon intensity
• Six converging within-sector evidence lines: Eurostat sectoral GHG, E-PRTR facility matching, within-supplier paired analysis, EU ETS headroom, UK PPN 06/21, measurement-error sensitivity
• Cross-continental corroboration: US (r = 0.555, p = 0.002), Canada (d = −0.56), Australia (d = 0.19), World Bank GPPD (22 countries, p = 0.016)

Contents

• Data/processed/gprd_with_carbon.parquet: Main analysis dataset (21.6M contracts × EXIOBASE 3.8.2 carbon intensity)
• Data/reference/: CPV-to-EXIOBASE crosswalks, emission factors, country metadata
• Data/external/: SBTi targets, Eurostat GHG, Canada/Australia/Chile procurement, OECD governance
• Data/raw/: EXIOBASE 3.8.2, EU ETS verified emissions, Eurostat sectoral accounts, AusTender
• scripts/: All computation scripts organized in 20 subfolders (read parquet → produce results/*.json)
• scripts/pipeline/: Data processing pipeline (parse → harmonize → link carbon)
• results/: 70 JSON output files from each analysis
• results/figures/: Publication-quality figures
• scripts/reanalysis/: 12-step robustness verification suite
• scripts/lib/: Library modules (causal analysis, data acquisition, mechanism index)
• tests/: Unit tests for core modules

Reproduction

See REPRODUCE.md for step-by-step instructions. All analysis scripts read from Data/processed/gprd_with_carbon.parquet and write to results/. Key entry points:

• python reproduce_manuscript_claims.py — verifies all 6 key manuscript claims
• python reproduce_data_comprehensive.py — comprehensive data audit
• python scripts/run_full_pipeline.py — master orchestration (steps 1–8)

No hardcoded values in primary analysis — all statistics computed from data.

Note on raw data

Raw EU TED XML (>90 GB) must be downloaded separately from https://ted.europa.eu. The processed parquet files contain the complete analysis-ready dataset. See Data/raw/README.md for download instructions.

Related Publication

Ashuraliyev, A. (2025). Governance Reform Breaks Decarbonization Dead Zones in Public Procurement. Nature Sustainability. (Under review)

File Organization

Due to Zenodo file count limits, files are organized as follows:

• replication_code.zip: All analysis scripts (20 categories), tests, and reproduction scripts
• replication_results.zip: All computed results (91 JSON files in 13 subcategories)
• replication_data_part1-3.zip: Reference data, audit files, processed samples, and small external datasets
• canada_contract_history.zip: Canadian government contract data (compressed from 104 MB)
• gprd_with_carbon_part*.zip.*: Main carbon-linked procurement dataset (794 MB, 8 parts)
• gprd_master_part*.zip.*: Full harmonized procurement master dataset (2.5 GB, 23 parts)
• PARQUET_REASSEMBLY.md: Instructions for reassembling split parquet files
• Key documentation files (README, REPRODUCE, CLAIMS_INDEX, etc.) are uploaded individually for direct viewing