Replication package for: Institutions, Education Quality and the Slow Closure of Human Capital Gaps: Global Evidence for Sustainable Development

This record contains the data, code and outputs needed to reproduce all tables and figures of the article “Institutions, Education Quality and the Slow Closure of Human Capital Gaps: Global Evidence for Sustainable Development”. The study examines two limits to education's contribution to the Sustainable Development Goals: a conversion gap, where institutional quality is associated with how much learning schooling produces, and an inertia gap, where the adult population's human capital adjusts only through cohort replacement. The package is a single flat folder. All scripts read their inputs from, and write their outputs to, the same folder. Running 02_build_panel.py, 03_analysis.py, 04_figures.py and 05_figure1_workflow.py in this order reproduces every result reported in the article and its appendices; the outputs are also included. Input data (downloaded from primary sources on 26 October 2026): wb_long.csv (World Bank API: Human Capital Index, World Development Indicators and Worldwide Governance Indicators); hlo_database.xlsx (Harmonized Learning Outcomes database); hdr25.csv (UNDP Human Development Report 2025 composite indices time series); pwt110.xlsx (Penn World Table 11.0); wcde_ssp1_bmys.rds, wcde_ssp2_bmys.rds, wcde_ssp3_bmys.rds and wcde_ssp2_mys.rds (Wittgenstein Centre Data Explorer, version 3); vdem_v16_subset.csv (selected variables from the V-Dem Country–Year Core dataset, version 16); sdr2026.xlsx (Sustainable Development Report 2026 database); wb_country_meta.csv (World Bank country metadata and income groups). The scripts 01_download_worldbank.py and 01b_download_other_sources.sh document how these files were obtained. Scripts: 02_build_panel.py builds the harmonised country–year panel (panel_cy.csv) and auxiliary files (hlo_cy.csv, hlo_5y.csv, wic_long.csv, sdr_goals.csv, coverage.csv). 03_analysis.py estimates the learning-loss regressions, the stochastic learning frontier with institution-dependent inefficiency, the Phillips–Sul convergence tests and club clustering, the β-convergence models, the absolute-gap trends and the time-to-parity analysis under the Shared Socioeconomic Pathways, and writes all tables (T3–T7 for the main text; A–E files for the appendices) and a summary of key statistics (keys.json). 04_figures.py and 05_figure1_workflow.py produce Figures 1–5. Software: Python 3.13 with pandas 3.0, NumPy 2.5, statsmodels 0.15, SciPy 1.18, matplotlib 3.11, pyreadr 0.5 and pycountry 26.2. The stochastic frontier, the Phillips–Sul test and the club-clustering algorithm are implemented in 03_analysis.py. The input data are redistributed under the terms of their original providers; users should cite the original sources as listed in the article.