Space-Time Smoothing of Complex Survey Data: Small Area Estimation for Child Mortality

Area of Science:

• Demography
• Public Health
• Biostatistics

Background:

• Civil registration and vital statistics (CRVS) systems are often incomplete in low- and middle-income countries (LMICs).
• Household sample surveys and demographic surveillance systems (DSS) are frequently used to estimate health indicators, but these methods have limitations.
• Small area estimation is crucial for understanding geographical disparities in health outcomes when comprehensive vital statistics are unavailable.

Purpose of the Study:

• To develop and apply a spatio-temporal statistical model for producing small area estimates of child mortality over time.
• To address data sparsity and geographical heterogeneity in health indicators in data-scarce regions.
• To account for complex survey weighting and potential biases in statistical models.

Main Methods:

• Integration of data from household sample surveys (e.g., Demographic and Health Surveys) and demographic surveillance system sites.
• Application of hierarchical models incorporating spatio-temporal smoothing and random effects for area, time, and survey.
• Development of a variance estimator for under-five child mortality that accommodates complex survey designs.
• Utilizing Integrated Nested Laplace Approximation (INLA) for efficient model implementation.

Main Results:

• The study successfully produced small area estimates of child mortality rates in five-year intervals for regions in Tanzania.
• The proposed variance estimator effectively accounts for complex survey weighting, reducing potential bias.
• Spatio-temporal smoothing proved beneficial in managing data sparsity and improving estimation accuracy.
• Model comparison using measures like the conditional predictive ordinate (CPO) guided the selection of appropriate hierarchical models.

Conclusions:

• Combining diverse data sources with advanced statistical modeling enables robust small area estimation of child mortality in LMICs.
• The developed methodology provides a valuable tool for monitoring child health and identifying geographical inequalities.
• Accurate estimation of child mortality is essential for targeted public health interventions and policy development in regions with weak vital statistics systems.