Newtonstein-creator/ssa-green-infrastructure-financing-gap

Replication code and data for a Sub-Saharan Africa green infrastructure financing-gap analysis (World Bank PPI, Beta regression, logit, XGBoost/SHAP) — submitted to Springer Discover Cities. # Green Finance Gap — Project Folder Instructions Companion codebase to *Financing the Unbuilt City: Evidence of a Government Support Gap in Sub-Saharan Africa's Green Infrastructure Transition* (Discover Cities submission). Everything in here is real, tested, and reproduces the exact numbers reported in the manuscript — nothing is a placeholder or a stand-in. ## Folder structure ``` green_finance_gap/ ├── README.md ← this file ├── REAL_RUN_SUMMARY.md ← full honest write-up of every result, including the ones that changed across model versions ├── src/ │ ├── config.py single source of truth: sectors, years, paths, variable schema │ ├── build_real_panel.py builds panel_real.csv and projects_real.csv from the raw PPI/WGI/WDI files │ ├── panel_regression.py Beta regression — the primary cell-level model (Table 2) │ ├── financing_gap_logit.py project-level logistic regression — the paper's strongest result (Table 3) │ ├── robustness.py dominant-economy exclusion + lagged-covariate checks │ ├── ml_model.py XGBoost + SHAP, bounded via logit transform (Figs. 1–4) │ └── compute_kappa.py Cohen's kappa between two completed coding sheets ├── data/ │ ├── raw/ │ │ └── urbanization_level.xlsx the real WDI urbanization data pulled via DataBank │ └── processed/ │ ├── panel_real.csv 212 country-sector-year cells — feeds panel_regression.py, robustness.py, ml_model.py │ ├── projects_real.csv 356 classified projects — feeds financing_gap_logit.py │ ├── validation_sample_raw.csv the 115-project stratified sample used for dual-coding │ └── AUTOMATED_CLASSIFICATION.csv the codebook's automated calls — see note below ├── dual_coding_exercise/ │ ├── Classification_Codebook.docx the written rules both coders applied │ ├── Coder_A_coding_sheet_completed.xlsx real, completed │ ├── Coder_B_coding_sh …