Exploring the high-resolution mapping of gender-disaggregated development indicators.

Area of Science:

• Geospatial analysis
• Development indicators
• Health equity

Background:

• National/subnational development indicators can mask crucial inequities, particularly for rural populations.
• Targeting limited resources effectively requires understanding geographical variations in health and wealth.
• Spatial interpolation offers potential for high-resolution mapping of demographic and health data.

Purpose of the Study:

• To assess the accuracy of spatial interpolation methods for generating gender-disaggregated, high-resolution maps.
• To evaluate the predictability of literacy, stunting, and modern contraceptive use across diverse settings.
• To compare Bayesian geostatistical and machine learning approaches for mapping socio-economic indicators.

Main Methods:

• Utilized geolocated Demographic and Health Surveys (DHS) cluster data and geospatial covariates.
• Applied Bayesian geostatistical and machine learning modeling across four low-income countries.
• Generated 1x1 km resolution maps with uncertainty estimates for key gender-disaggregated indicators.

Main Results:

• Achieved high predictive accuracy (up to 75% explained variance) for female literacy in Nigeria and Kenya.
• Demonstrated moderate accuracy (50-70% explained variance) for several other socio-economic indicators.
• Observed significant variability in mapping accuracy (2-30% explained variance) across countries and variables, with both methods showing limitations.

Conclusions:

• Spatial interpolation shows potential for creating detailed maps to support geographically stratified decision-making and monitor development goals.
• Substantial variations in accuracy highlight challenges in universal application across diverse countries and indicators.
• Emphasized the critical importance of rigorous validation and uncertainty quantification when applying these methods.