Measuring and forecasting progress towards the education-related SDG targets

The study addresses how countries have progressed toward education-related SDG targets—particularly universal primary and secondary completion and equitable access to tertiary education—while centering educational inequality. Building on the legacy of the MDGs, which emphasized primary education to 2015, the SDGs broaden the agenda to include reducing inequalities. The authors model within-country distributions of years of schooling from 1970 to 2018 and forecast to 2030, focusing on attainment among young adults (ages 25–29) to assess recent cohort progress, gender disparities, and overall inequality. The purpose is to provide a quantitative framework to benchmark national and regional progress toward SDG 4 (quality education) and SDG 5 (gender equality), highlight persistent gaps at secondary and tertiary levels, and evaluate trends in inequality using distributional metrics.

The paper situates education as both a human right and a driver of economic development, health, fertility, political participation, social empowerment, and human capital, citing evidence across multiple decades and regions. It notes the MDGs’ focus on primary education and the SDGs’ expanded emphasis on equity and inter-goal interactions. Prior work has mapped local variation in educational attainment (notably in Africa), harmonized global attainment datasets, and explored links between education and outcomes like child mortality and human capital indices. Studies have documented persistent gender and socio-economic gaps in many low- and middle-income countries, and in contrast, a trend in OECD and other countries where boys increasingly lag behind girls as development advances. The paper also references broader inequality frameworks (e.g., Kuznets curve) and calls to track disparities beyond gender, including wealth, ethnicity, race, ability, and urban-rural divides.

Following GATHER guidelines, the authors estimate mean years of schooling and full single-year distributions of educational attainment for ages 25–29 from 1970–2018 and project to 2030 for 195 countries/territories (GBD 2017 set), using 3,180 nationally representative censuses and surveys. Key components:

• Data compilation and harmonization: Data describing distributions of years of schooling by age and sex. Sources reporting single-year schooling used directly; binned categories (e.g., some primary) are probabilistically split into single-year proportions via a published crosswalk model. Attainment is top-coded at 18 years.
• Data adjustment for source bias: A mixed-effects regression adjusts for systematic provider biases regionally and within countries. Gold-standard sources (primarily IPUMS censuses; DHS where IPUMS absent) anchor adjustments. Model: logit(P) = β0 + β1age + β2sex + β3location + β4year + u_dataprovider (region-specific) + u_location:dataprovider (nested, country-specific). Regional effects align all data to regional gold standards; country effects correct local biases where gold standards exist.
• Stage 1: Estimation of mean years of schooling and proportion with no education (1970–2018) using a cohort extrapolation model followed by an age-period model with Gaussian process regression (GPR) to synthesize data and quantify uncertainty.
• Cohort extrapolation: Leveraging stability of cohort attainment after age 25, repeat cohort observations (≤10 years apart, both post-1990) are used to estimate age-related changes. Differences are normalized so the average change at age 65 is zero to mitigate inter-wave survey bias. The modeled logit differences include natural spline over age with random intercepts for location nested within super-region.
• Age-period modeling: Fit on adjusted, cohort-extrapolated values to interpolate/extrapolate single-year attainment P for 1970–2018 separately by sex and GBD region groupings, yielding country-age-year-specific estimates.
• Stage 2: Distribution modeling: An ensemble K-nearest neighbours algorithm estimates the full within-country distribution of years of schooling (1970–2018), drawing on stage-1 outputs (e.g., mean, proportion with no schooling).
• Forecasting to 2030: Trends in distributions are projected using a rate-of-change approach; mean years for 2019–2030 are derived from projected distributions.
• Uncertainty and validation: All analyses use 1,000 draws to propagate data/model uncertainty. Out-of-sample predictive validation informs model selection and hyperparameter optimization.
• Inequality metrics: Within-country inequality is summarized via the average interpersonal difference (AID), the average difference in years of schooling between two randomly selected individuals. Alternative inequality measures (e.g., Gini) are explored in Supplementary Information.

• Primary attainment (6+ years, ages 25–29): Global proportion increased from 50.1% (95% UI 49.3–51.0) in 1970 to 83.2% (82.1–84.0) in 2018; projected 89.4% (87.4–91.0) by 2030. Some countries will still fall short by 2030, often due to female schooling gaps.
• Secondary attainment (12+ years): Heterogeneous and far from universal; high-income and Eastern Europe/Central Asia already near 50% by 1970 while many regions were ≤10%. No major region achieves near-universal secondary completion by 2018 or by 2030; inter-regional disparities remain large, narrowing only slightly.
• Tertiary completion (15+ years): Substantial global scale-up since 1970, accelerating after 2000; by 2030, high-income and Eastern Europe/Central Asia projected near 45–46% completion; other regions show gains but remain much lower. Regional disparities in tertiary completion are widening over time.
• Regional projections for 2030 (ages 25–29): • 6+ years: Eastern Europe & Central Asia 99%; High-income 98%; Southeast/East Asia & Oceania 97%; Latin America & Caribbean 92%; North Africa & Middle East 92%; Global 89%; South Asia 86%; Sub-Saharan Africa 80%. • 12+ years: Eastern Europe & Central Asia 85%; High-income 85%; Southeast/East Asia & Oceania 76%; North Africa & Middle East 76%; Global 61%; Latin America & Caribbean 50%; South Asia 49%; Sub-Saharan Africa 45%. • 15+ years: Eastern Europe & Central Asia 46%; High-income 45%; North Africa & Middle East 42%; Southeast/East Asia & Oceania 35%; Global 32%; Latin America & Caribbean 22%; Sub-Saharan Africa 18%.
• Gender gaps (ages 25–29): Mean years of schooling gap favoring men declined from 1.7 (1.6–1.8) years in 1970 to 0.3 (−0.2–0.8) years in 2018; projected to reverse by 2030. Countries with a statistically significant male advantage fell from 142 (1970) to 27 (2018) to 4 (2030). By 2030, women are projected to have significantly higher attainment than men in 18 countries. Primary-level gender gap fell from 0.9 (0.9–1.0) additional male years (1970) to 0.3 (0.2–0.4) (2018). Women overtook men in global tertiary attainment in 2012 and are forecast to do so for secondary in 2026; however, large gaps against women persist in parts of sub-Saharan Africa and North Africa/Middle East.
• Inequality (AID): Global within-country inequality peaked in 2017 at 4.6 years (4.5–4.7) and is projected to decline through 2030, exhibiting Kuznets-like inverse-U trajectories across most regions. South Asia had the highest inequality in 2018 (AID 6.0, 5.7–6.3). Latin America & Caribbean had the highest inequality in 1970 (AID 4.5, 4.4–4.6) but follows a comparatively lower Kuznets curve relative to regions at similar mean attainment. Southeast/East Asia & Oceania show the flattest Kuznets curve (least unequal development trajectory among LMIC regions); Eastern Europe & Central Asia achieved high mean attainment with relatively low inequality.

The findings indicate substantial global progress toward SDG 4.1 for primary education, while highlighting persistent and projected shortfalls in secondary completion and widening disparities in tertiary completion. By focusing on young adults (25–29), the study demonstrates that the global gender gap has largely closed and is reversing in many settings, though significant disadvantages for women remain in parts of sub-Saharan Africa and North Africa/Middle East. The distributional approach and AID metric contextualize average gains by revealing within-country inequality trajectories, generally following a Kuznets pattern with inequality rising and then falling as average attainment increases. These results provide actionable benchmarks for countries to gauge progress relative to peers at similar development levels, identify where disparities are most entrenched, and inform targeted interventions. Policy implications include prioritizing measures that improve access and completion at secondary and tertiary levels and that address intersecting disparities (gender, wealth, geography, ethnicity, ability). Reducing educational inequalities is expected to yield broader societal benefits in human capital, productivity, health, and well-being, reinforcing synergies across SDGs.

This study contributes a validated, distribution-based framework for measuring and forecasting educational attainment and inequality for 195 countries from 1970 to 2030. It shows the world is largely on track for near-universal primary education by 2030, but faces substantial challenges in secondary completion and growing disparities in tertiary attainment. The global gender gap in schooling among young adults has nearly closed and is projected to reverse in many countries, although regional pockets of disadvantage persist. Inequality in education peaked around 2017 and is projected to decline, broadly following Kuznets-like trajectories that vary by region. The framework and metrics enable countries to track progress toward SDG targets and benchmark inequality over time. Future research could enhance subgroup disaggregation (e.g., wealth, ethnicity, disability, urban–rural), integrate school quality and learning outcomes, refine projections under alternative policy scenarios, and improve data coverage and harmonization for underrepresented countries and time periods.

• Reliance on secondary survey and census data may introduce measurement error and residual biases despite regional and country-level source adjustments.
• Binned attainment categories are crosswalked to single-year distributions, which may add uncertainty relative to directly observed single-year data.
• Top-coding attainment at 18 years may underrepresent very high levels of education in some settings.
• Cohort extrapolation assumes stability of educational attainment after age 25 and uses normalization based on ages 60–70 to correct survey-wave biases; deviations from these assumptions (e.g., late-life migration or mortality differentials) could affect estimates.
• Forecasts to 2030 are based on historical rates of change and may not capture future shocks, policy shifts, or crises.
• Inequality is summarized primarily via AID; while complementary measures (e.g., Gini) are explored in supplementary materials, subgroup-specific inequalities (wealth, ethnicity, disability) are not directly estimated in the main analysis.
• Standard experimental design statements (no randomization/blinding, no predetermination of sample size) reflect observational modeling and do not eliminate potential confounding inherent in the source data.