Replication package for Can Digitalization Help Decarbonize? Evidence from Causal Forest Analysis — a panel study of how domestic digital capacity affects CO₂ emissions across 40 economies (2000–2023).
git clone https://github.com/a985783/digital-decarbonization-divide.git
cd digital-decarbonization-divide
bash reproduce.shOr stage by stage:
make setup
make verify
make analysis
make paperSee REPRODUCIBILITY.md for expected outputs and troubleshooting.
| Estimator | ATE (tons CO₂/capita per SD) | 95% CI |
|---|---|---|
| Causal Forest DML | −1.73 | [−2.88, −0.58] |
| IV (OrthoIV) | −1.91 | [−2.37, −1.46] |
Middle-income sweet spot (lower- and upper-middle): −2.17 to −2.29 tons per capita. High-income diminishing returns: −1.26 tons. Placebo signal-to-noise ratio: 23:1.
Primary estimator: CausalForestDML (2,000 trees, honest splitting, GroupKFold cross-fitting, country-cluster bootstrap B=1000). Robustness checks: IV analysis with lagged DCI as instrument, Oster sensitivity (δ=1.01), DragonNet comparison (ATE=−1.95), Leave-One-Country-Out stability.
pip install -r requirements.txtPython 3.8+ required.
streamlit run app.pyFour modules: Data Explorer, Causal Effects, Policy Simulator, Country Comparison.
├── paper.tex / paper_cn.tex # LaTeX source (EN / CN)
├── paper.pdf / paper_cn.pdf # Compiled PDFs
├── references.bib # Bibliography
├── app.py # Streamlit dashboard
├── scripts/ # Analysis pipeline (phases 1–7)
├── data/ # Processed datasets
├── results/ # CSVs and figures
├── policy_toolkit/ # Country classifications, simulator
├── docs/ # Theoretical model, methodology notes
└── tests/ # Reproducibility test suite
All variables drawn from World Bank WDI and WGI (2000–2023). Main dataset: data/clean_data_v5_enhanced.csv. Variable-level documentation: DATA_MANIFEST.md.
Cui, Qingsong (2026). Can Digitalization Help Decarbonize? Evidence from Causal Forest Analysis.
Full citation metadata: CITATION.cff.
Qingsong Cui — qingsongcui9857@gmail.com
MIT. See LICENSE.