Proposal Description

In many sub-Saharan African countries, household health surveys such as the Demographic and Health Surveys (DHS), Multiple Indicator Cluster Surveys (MICS), and national health surveys remain the primary sources of official health statistics. These surveys are carefully designed to produce reliable estimates at national and, in some cases, regional levels. However, they are not designed to support reliable estimation at lower administrative levels such as districts, wards, or villages, where key decisions on health service delivery, budgeting, and intervention targeting are increasingly made. At these levels, direct survey estimates are often unavailable or statistically unreliable due to small sample sizes.

This session addresses this critical gap by demonstrating how Small Area Estimation (SAE) methods can generate precise, disaggregated health indicators using data already available within government systems. SAE methods borrow strength from auxiliary information drawn from population censuses, administrative records, and routine health information systems such as DHIS2, enabling statistically robust estimates for small geographic areas without additional costly data collection.

The session covers both unit-level and area-level SAE approaches, including widely used models such as the Fay–Herriot and Battese–Harter–Fuller constructions. Presentations will focus on key methodological and implementation challenges, including model specification and diagnostics, mean squared error estimation, benchmarking against known aggregates, and communicating uncertainty to policy users.

Particular attention will be given to the realities of statistical production systems in sub-Saharan Africa, including data sparsity, outdated census frames, and heterogeneous data quality across administrative units. The session will also highlight emerging developments that integrate machine learning techniques into SAE frameworks to enable covariate selection and improve prediction performance.

This topic is highly relevant to ongoing efforts to strengthen decentralised health governance across sub-Saharan Africa, including in the host country, Zambia. As governments move toward district-level planning, budgeting, and accountability, the demand for timely and reliable disaggregated health indicators has intensified. National statistical offices and ministries of health are increasingly required to produce small area estimates for SDG monitoring, disease programme targeting (including HIV, tuberculosis, malaria, and maternal and child health), and equitable resource allocation.

This session responds directly to these needs by presenting practical SAE applications and implementation experiences from African national statistical offices and partner institutions. These case studies provide peer learning on how SAE methods can be operationalised within routine official statistics production systems.

By bridging methodological advances in small-area estimation with the practical needs of official statistics systems, this session strengthens the link between statistical theory and the production of policy-relevant evidence. It provides participants with actionable guidance on implementing SAE using existing government data sources in resource-constrained settings.

The session aligns strongly with the World Statistics Congress 2027 theme of “Statistics in the Era of AI” by highlighting the integration of machine learning methods into modern SAE frameworks. It also supports the ISI mandate to strengthen statistical capacity and develop statistical systems. Participants, including researchers, official statisticians, and policy practitioners, will gain practical tools for producing disaggregated health indicators to support evidence-based governance and improved public service delivery.