AI Governance Readiness Index for Africa (AIGRI) 2026 Dataset and Replication Package

  This repository contains the complete dataset and replication package for the AI Governance Readiness Index for Africa (AIGRI) 2026, a composite indicator framework that assesses AI governance capacity across 30 African nations. The index evaluates countries across 16 governance indicators organised into four thematic pillars: Regulatory Safeguards, Enforcement Capacity, Infrastructure Sovereignty, and Talent & Civic Space. Pillar weights are derived empirically via principal component analysis (PCA) implemented in zero-dependency pure Python 3 using power iteration, ensuring full reproducibility without proprietary statistical software.   The dataset was created to address a significant gap in the comparative AI governance literature: the absence of a rigorously weighted, Africa-specific readiness metric that reflects the continent's heterogeneous institutional, infrastructural, and socio-political contexts. By making both the raw indicator data and the full computational pipeline openly available, this repository enables scholars to (1) replicate all index scores and tier classifications reported in the associated paper; (2) substitute alternative indicator specifications or weighting schemes; (3) extend coverage to additional African Union member states; and (4) update scores as governance environments evolve. An embedded interactive dashboard (HTML/CSS/JavaScript) provides immediate visual access to country rankings, pillar profiles, and sensitivity analyses without requiring any software installation beyond a modern web browser. All underlying data were retrieved in May 2026 from publicly accessible institutional databases.    Software Requirements Component Requirement Python version 3.8 or higher (tested on 3.8, 3.10, 3.12) External Python packages None. All computation uses Python standard library only. Standard library modules used os, csv, math, json Operating system Platform-independent (Linux, macOS, Windows) Web browser (dashboard) Any modern browser supporting ES6: Chrome 88+, Firefox 85+, Safari 14+, Edge 88+ LaTeX (optional) Any current TeX distribution (TeX Live 2023+ or MiKTeX 23+) with Pandoc 3.0+ for compiling paper/draft.md to PDF Known Limitations Readers, reviewers, and users who build upon this dataset should be aware of the following limitations: Ordinal coding: Twelve of the 16 governance indicators are measured on researcher-assigned 0–10 ordinal scales, not cardinal or ratio scales. Differences between adjacent values (e.g., a score of 6 vs. 7) do not imply equal substantive differences in governance capacity. Min-max normalisation treats the 0–10 scale as if it were cardinal; this is a pragmatic approximation standard in composite index methodology but should be borne in mind when interpreting inter-country differences near tier boundaries. Incomplete African Union coverage: The sample of 30 countries represents 55.6% of the 54 African Union member states. Countries were selected based on data availability and regional representativeness. Findings and tier classifications should not be extrapolated to unsampled AU members. Cross-sectional snapshot: All data reflect governance conditions as of May 2026. AI governance environments are evolving rapidly across Africa; policy changes, new legislation, and institutional developments occurring after this date are not reflected in the index. Broadband penetration exceeding 100: The broadband_penetration variable can and does exceed 100 subscriptions per 100 inhabitants in some countries where multiple fixed broadband contracts per household or business premises are common. This is an artefact of the ITU's measurement methodology and does not indicate a data error. Values above 100 are treated as valid for min-max normalisation purposes. Observational econometrics: The OLS regression linking AIGRI scores to AI adoption (Section 5 of the associated paper) is observational. The positive and statistically significant coefficient on AIGRI should not be interpreted as establishing a causal effect of governance readiness on AI adoption. Omitted variable bias, reverse causality, and measurement error in the dependent variable remain potential threats to causal identification. Single-year PCA weights: Pillar weights are derived from the cross-sectional variance structure of the 30-country sample in 2026. These weights may not be stable across different country samples, time periods, or regional subsets. Ethical Statement This research does not involve human subjects, clinical trials, animal research, or any form of data collection from individuals. All 20 variables in the dataset are derived exclusively from publicly available, institutionally published databases and government documents. No personally identifiable information (PII), sensitive personal data, or individual-level microdata are present in any file in this repository. The dataset contains only country-level aggregate indicators. The research design and data collection protocol did not require ethics committee review under institutional guidelines governing social science research without human subjects. Nonetheless, the research team has adhered throughout to the principles of transparency, reproducibility, and responsible use of data as set out in the FAIR Data Principles (Wilkinson et al., 2016) and the Declaration on Research Data Management of the African Academy of Sciences. Code The analytical pipeline (src/pipeline.py) and dashboard code (dashboard/) are also released under CC BY 4.0. Users wishing to apply a code-specific open licence may treat these files as equivalent to MIT Licensed with attribution required. Third-party Data Constituent data sourced from third-party providers (e.g., ITU, World Bank, V-Dem, Tortoise Media) remain subject to the terms and conditions of their respective publishers. Users are responsible for compliance with those terms when redistributing or publishing derivative works that contain or reproduce third-party data. Consult the URLs listed in Section 8 for current licensing information. Acknowledgements The authors gratefully acknowledge the following institutions and initiatives for making their data freely accessible, without which the construction of the AIGRI 2026 would not have been possible: United Nations Conference on Trade and Development (UNCTAD) for the UNCTAD Technology and Innovation Report and e-Government readiness data informing the desk-coding framework. International Telecommunication Union (ITU) for the Global Cybersecurity Index (4th Edition), the World Telecommunication/ICT Indicators Database (broadband penetration), and the ICT Development Index 2024. World Bank Group for the World Development Indicators (WDI) 2025 release, including GDP per capita estimates. UNESCO Institute for Statistics (UIS) for data on tertiary education enrolment, STEM gender parity, and national digital literacy programme data. V-Dem Institute (University of Gothenburg) for the Digital Society Survey, version 14, and the associated civic space indicators. African Union (AU) Commission for publicly archiving treaty ratification records under the Malabo Convention on Cyber Security and Personal Data Protection. Organisation for Economic Co-operation and Development (OECD) for the AI Policy Observatory framework and comparative AI strategy taxonomy that informed the ai_strategy desk-coding rubric. Masakhane NLP Project for the open catalogue of African language NLP resources that informed the localized_nlp_resources coding. Tortoise Media for the Global AI Index 2025, used as the primary external validation variable for AI adoption. Stanford Human-Centered AI (HAI) Institute for the AI Index Report 2025, including private AI investment data by country. Cloudscene for the commercial data centre market intelligence used to inform the local_datacenters coding. Any errors, omissions, or misclassifications in the derived indicators are the sole responsibility of the authors.