Sequencing Military Expenditure Across AI Readiness Pathways to SDG Performance in Africa - Dataset

Description

This dataset is a cleaned cross-country panel covering African countries from 2020 to 2024. The 2025 SDG score is used only to construct the one-year-forward SDG outcome and is not an additional estimation year. It links AI readiness, military expenditure and sustainable development performance, alongside macroeconomic controls capturing industrial structure, income level and trade openness. The dataset supports country-level analysis of how African economies prepare for AI adoption and how this readiness relates to SDG outcomes under different military-expenditure conditions. The .csv file can be used for estimations in Stata, while the .xlsx file contains the supporting details below. Country and time identifiers: - cid: Unique numeric country identifier for panel estimation. - countryname: Country name. - regionname: World Bank regional classification. - incomelevelname: World Bank income-level classification. - year: Observation year; the estimation panel covers 2020 to 2024. AI readiness indicators: - ai: Overall AI Readiness Score, capturing national preparedness to adopt and deploy AI through government capacity, technology-sector maturity, and data and infrastructure readiness. - govai: Government AI readiness pillar, measuring public-sector preparedness through strategy, governance and ethics, digital capacity, and adaptability. - techai: Technology-sector AI readiness pillar, capturing sector maturity, innovation capacity and human capital. - dataai: Data and infrastructure AI readiness pillar, capturing infrastructure, data availability and data representativeness. Security, development and economic indicators: - milex: Military expenditure as a percentage of GDP, capturing defence expenditure relative to national output. - lnMILEX: Natural logarithm of military expenditure, used in the empirical models and interaction terms. - ind: Industry value added as a percentage of GDP, including manufacturing, mining, construction, electricity, water and gas. - sdg: Official aggregate SDG Index Score, ranging from 0 to 100. - sdgno9no17: Reduced-overlap SDG measure, calculated after excluding SDGs 9 and 17. - sdg_f1: One-year-forward aggregate SDG outcome, matching each 2020 to 2024 observation with the following year's SDG score. - gdppc: GDP per capita in constant 2015 US dollars, capturing income and broader development capacity. - lnGDPPC: Natural logarithm of GDP per capita. - trade: Trade as a percentage of GDP, measuring exports plus imports relative to national output. - lnTRADE: Natural logarithm of trade openness. - sample_main: Indicator identifying observations included in the matched main estimation sample. Mapping and visualisation variables: - AI readiness z-score: Country-level mean z-score for AI readiness over 2020 to 2024. - Military expenditure z-score: Country-level mean z-score for military expenditure over 2020 to 2024. - SDG performance z-score: Country-level mean z-score for SDG performance over 2020 to 2024.

Steps to reproduce

Step 1: Indicator sources used in the merged dataset • ai, govai, techai, dataai: Oxford Insights, Government AI Readiness Index, covering the overall score and three pillar scores for 2020–2024. • milex: SIPRI military expenditure series, accessed through World Bank WDI, measured as % of GDP. • ind, gdppc, trade: World Bank, WDI. • sdg, sdgno9no17, sdg_f1: Sustainable Development Report 2026 backdated series. Step 2: Assemble and verify the panel • Import the CSV and declare the panel using xtset cid year. • The working dataset contains 220 observations for 44 African countries over 2020–2024. • Check country IDs, years, income groups and regions. • Correct Morocco and Mozambique 2020 AI-readiness values against Oxford Insights. • Keep Djibouti and Guinea-Bissau 2021 AI-readiness values missing rather than interpolating them. Step 3: Treatment of trade gaps • For Burundi and Nigeria, replace missing trade values with year-specific means from countries in the same income group, excluding the country being imputed. • For Lesotho, Malawi and Togo, estimate remaining gaps using within-country linear trends based only on available years. • Replace values only where trade is missing; do not alter observed values. Step 4: SDG outcome construction • Use only the SDR 2026 backdated series, not mixed annual editions. • sdg is the official aggregate SDG Index. • sdgno9no17 is the mean of the 15 goals remaining after excluding SDGs 9 and 17. • sdg_f1 matches each 2020–2024 observation with the following year’s SDG score. The 2025 SDG value is used only to construct the 2024 forward outcome. • Seychelles has no comparable SDR 2026 SDG observations and remains missing. Step 5: Military expenditure and estimation sample • Do not impute missing milex because it is the key moderator. • Djibouti has no usable milex observations; other gaps are retained as missing. • Complete-case matching across model variables gives the main sample of 205 observations across 42 countries. • Use the same matched sample for the principal TWFE, reduced-overlap and one-year-forward models. Step 6: Post-processing and validation • Generate lnMILEX, lnGDPPC and lnTRADE for strictly positive values. • Mean-centre AI-readiness pillars and lnMILEX only when constructing interaction terms; retain original variables as lower-order terms. • Re-check missingness, summary statistics, panel IDs and the 205-observation sample flag. • Generate country-level mean z-scores for AI readiness, military expenditure and SDG performance for mapping. • Save the processed dataset separately from the raw files.

Institutions

• Asia Pacific University of Technology & Innovation

  Kuala Lumpur, Kuala Lumpur
• Universiti Putra Malaysia

  Selangor, Seri Kembangan

Categories

Licence