AI Human Capital and Government AI Readiness: Country-Level Data

Abstract

Title AI Human Capital and Government AI Readiness: Country-Level Data and Stata Analysis Code for 166 Countries Data type Quantitative, cross-sectional, country-level secondary data, accompanied by Stata analysis code and Excel documentation. Each observation represents one country. Data accessibility The dataset and accompanying analysis files are intended for public access through Zenodo. Files are provided in Stata (.dta), Excel (.xlsx), and CSV (.csv) formats, with Stata do-files (.do) documenting the analyses. The underlying indicators originate from Oxford Insights and the World Bank. Users should acknowledge these sources and the repository record and observe the applicable source-data and repository licenses. Data description This dataset supports research on the relationships among economic development, AI human capital, national internet use (%), and adjusted government AI readiness across 166 countries. It combines Oxford Insights’ 2025 Government AI Readiness Index scores with GDP per capita from the World Bank’s 2025 column and national internet use from its 2024 column. Economic development is measured as the natural logarithm of GDP per capita in current US dollars. AI human capital is Oxford Insights’ published Human Capital dimension. National internet use (%) represents population connectivity, serving as a proxy for the connectivity component of AI infrastructure. Adjusted government AI readiness combines four Oxford Insights pillars: Adjusted AI readiness = 0.20 × Policy capacity + 0.30 × Governance + 0.30 × Public sector adoption + 0.20 × Resilience. The AI Infrastructure and Development and Diffusion pillars are excluded from this outcome to reduce overlap with the explanatory variables. The adjusted score is a study-specific measure, not Oxford Insights’ official overall index. The archive preserves the 166-country sample used in the manuscript. It is neither a random sample nor an exhaustive set of eligible countries. Subsequent country-matching checks identified additional eligible cases that were not included in the reported analyses. The cross-sectional design supports analysis of associations rather than causal effects. File descriptions Format Contents and purpose Stata dataset (.dta) The current analysis dataset contains 166 observations and 43 variables, including source indicators, the adjusted readiness composite, standardized predictors, the interaction term, and estimation-sample indicators. Excel workbook (.xlsx) Includes four worksheets: AI_Human_Capital_Stata, Matched_Data, Dropped_Missing_WB, and ReadMe. These provide analysis data, the original 29-variable matched extract, archival exclusion records, and documentation. CSV dataset (.csv) Provides the 166-country, 29-variable matched extract in a broadly accessible tabular format. Stata do-file (.do) Documents data checks, descriptive statistics, correlations, internal-consistency calculations, regressions, confirmatory factor analysis, moderation, and latent moderated-mediation analyses. Analysis output (.txt or .log) Records estimated coefficients, uncertainty intervals, model diagnostics, and conditional associations for comparison with the manuscript. Software and analysis The analysis code specifies Stata 19.0. Regression models use HC3 heteroskedasticity-consistent standard errors. The latent models use maximum likelihood with Satorra–Bentler correction; conditional associations and indirect associations use delta-method confidence intervals. To run the code, download the Stata dataset and do-file, update the data-directory path in the do-file, and execute it in Stata 19. Excel and CSV files can be inspected without Stata. Archival analyses retained in the code should be distinguished from the models reported in the final manuscript.