Despite global efforts to improve education access, many school-age adolescents and youths remain marginalized due to geographic location, socioeconomic status, cultural norms, and gender. The 2030 Sustainable Development Goals (SDGs) cannot be met without substantial investment in education and addressing inequalities. Lower and upper secondary school ages (12-14 and 15-17 years, respectively) are critical periods influencing future health and development. While investment in secondary education has increased since 2010, out-of-school rates have stagnated for eight years, with sub-national variations poorly understood in many low- and middle-income countries. Distance to school is a known barrier, along with socioeconomic factors like parental education and wealth. This study uses a geospatial approach to examine geographic accessibility to secondary schools and its association with out-of-school rates in Tanzania, Cambodia, and the Dominican Republic, integrating school locations, fine-scale geospatial covariates, and household-level data to estimate access and attendance at a high geographic resolution. This approach is novel in its triangulation of available data sources at a sub-national level.

The introduction cites several sources highlighting the persistent challenge of out-of-school adolescents and youths, the importance of education for achieving the SDGs, and the risk behaviors that can take hold during the secondary school years. It references studies on the barriers to education access, including distance to school and socioeconomic factors. The introduction also notes the limited progress in reducing out-of-school rates and the lack of fine-scale understanding of sub-national variations, which forms the rationale for the present study.

The study utilized a geospatial approach combining data from multiple sources. Geographic location data for secondary schools were obtained from government sources in each of the three selected countries (Tanzania, Cambodia, and the Dominican Republic). These countries were chosen for their geographical diversity and varying distributions of secondary schools. Cluster-level data on school attendance were derived from Demographic and Health Surveys (DHS) for each country, adjusting for sampling and stratification. Ancillary covariates, including land use/land cover (MERIS Glob-Cover), digital elevation models (HydroSHEDS), road data (Open Street Maps, NGA, MapCruzin), night-time lights (VIIRS), and 1x1 km population maps (WorldPop) were assembled. A gridded layer of travel times to secondary schools was estimated using AccessMod, considering various modes of transport and land use impedance values. Adjusted net school attendance rates were calculated using MEASURE DHS guidelines and Stata software, accounting for survey weighting and the timing of the academic year relative to the survey. A Bayesian hierarchical spatial model, implemented in R-INLA, was used to spatially interpolate cluster-level attendance rates, incorporating selected covariates (temperature and night-time lights). Model validation was performed using a 20% holdout data set, assessing goodness-of-fit using the Deviance Information Criterion (DIC), and predictive performance using mean absolute error, root mean square error, and Pearson correlation.

The average straight-line distance to the nearest school was 6.6 km in Tanzania, 3.3 km in Cambodia, and 1.3 km in the Dominican Republic. Travel times were correspondingly longer in Tanzania (0.8 h on average) compared to Cambodia and the Dominican Republic. In the model, only temperature variables and night-time lights (proxy for urbanization) were significant predictors of attendance; travel time was not selected for the predictive model, but was used in post-hoc analyses. Model predictions showed good correlation with observed data. Tanzania had the highest estimated out-of-school rate (57.3%), followed by Cambodia (40%) and the Dominican Republic (10.7%). In Tanzania, eight regions had out-of-school rates exceeding 60%, accounting for over one-third of the total out-of-school population. Maps were produced to visualize the spatial distribution of out-of-school rates and their uncertainty. A strong non-linear association was observed between travel time to school and out-of-school rates (R² >70% for all three countries).

The findings highlight substantial geographic disparities in secondary school access and attendance across the three study countries. The strong association between distance to school and out-of-school rates suggests that improving physical access through increased school provision, especially in remote areas, could significantly reduce the number of out-of-school youth. The results underscore the importance of using high-resolution spatial data to identify areas with the greatest need for educational interventions. The exclusion of travel time from the predictive model may indicate that other factors, not captured in the study, such as indirect costs, household choices, and school quality, play a more significant role in determining attendance. Further research is needed to address these factors and investigate household-level barriers to attendance.

This study demonstrates the effectiveness of geospatial modeling in mapping out-of-school adolescents and youths in low- and middle-income countries. The significant geographic disparities in access and attendance highlight the need for targeted interventions to improve secondary school provision, particularly in areas with low accessibility and high out-of-school rates. Future work could focus on fine-scale optimization models for school location and intervention planning, as well as on exploring the role of household-level barriers to school attendance.

The study's limitations include reliance on DHS data, which may not perfectly capture all out-of-school youth, and the use of straight-line distances and travel time estimations, which may not fully reflect actual travel patterns. The study focused on only three countries and did not consider all possible factors influencing school attendance beyond geographic accessibility.