📄 public and global health

Small-area estimation of district-level fertility in 36 countries in sub-Saharan Africa 2000-2025

This study utilizes a spatiotemporal Bayesian hierarchical model to generate district-level fertility estimates across 36 sub-Saharan African countries from 2000 to 2025, revealing significant subnational heterogeneity that national trends fail to capture and highlighting the critical need for localized data to guide effective public health resource management.

Original authors: Stevens, O., Imai-Eaton, J. W.

Published 2026-07-20

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Stevens, O., Imai-Eaton, J. W.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine trying to understand the heartbeat of a massive, bustling city by only listening to the average noise level of the entire metropolis. You might hear a general hum, but you'd miss the frantic jazz of the downtown district, the quiet lull of the suburbs, or the sudden silence of a neighborhood under construction. This is the challenge demographers face when studying populations in sub-Saharan Africa. For a long time, scientists have looked at "national" fertility rates—the average number of children a woman has in an entire country. It's like looking at a blurry, low-resolution photo of a crowd; you can see the general shape, but you can't tell who is holding a baby or who is dancing. To truly understand how populations grow, shrink, or change, you need a high-definition zoom lens that shows what's happening in every single neighborhood, or "district," and for every specific age group. This paper dives into that granular world, using a massive amount of survey data to map out exactly how many babies are being born, where, and when, across 36 countries. It's not just about counting heads; it's about understanding the unique rhythm of life in every corner of the region, which is crucial for planning schools, hospitals, and family support services.

The researchers behind this study decided to stop guessing and start mapping. They gathered data from 194 different household surveys—think of them as thousands of conversations with families across 36 sub-Saharan African nations, covering the years 2000 to 2025. Instead of just taking the national average and spreading it out like peanut butter on toast, they built a sophisticated mathematical "time machine" (a spatiotemporal Bayesian model) that could stitch together these scattered pieces of information. This model was smart enough to handle messy data, like when people forget exactly when a baby was born or when surveys only ask about the last few years. It corrected for these "hiccups" in memory and reporting to reveal the true picture of fertility at the district level.

What they found was a story of incredible diversity. While the national averages suggested a slow, steady decline in the number of children per woman across the region, the district-level map told a much more complex tale. In many countries, the difference in fertility rates between neighboring districts was actually larger than the difference between the countries themselves. For instance, in Ethiopia, the number of children per woman varied wildly from district to district, ranging from as low as 0.91 to as high as 7.42. This means that if you only looked at the national average, you would be completely wrong about what's happening in specific towns. In fact, in 29 out of the 36 countries studied, the national average was lower than the median of the districts, largely because big, populous cities (which tend to have fewer children) dragged the national number down, hiding the higher fertility rates in rural areas.

The study also revealed that the "when" of having children is changing, but not in a uniform way. While the total number of births is dropping in most places, the decline isn't happening equally for all ages. The biggest drops are seen among older women (ages 40-49), while teenage fertility (ages 15-19) has remained stubbornly unchanged in several countries, including Rwanda, Tanzania, and Zimbabwe. This suggests that the "fertility transition"—the shift from having many children to having fewer—is happening at different speeds and in different ways depending on exactly where you are and how old the mother is.

Perhaps most importantly, the authors argue that relying on national averages to plan for the future is a recipe for disaster. If a government uses a single national number to decide how many schools to build or how much medicine to stock in a district, they will likely get it wrong. In some areas, they might build too many resources for a shrinking population, while in others, they might be dangerously underprepared for a boom. The paper suggests that these "fertility stalls," where a country seems to stop having fewer children, might actually be a mix of some districts declining while others stay the same, a nuance that gets lost in national statistics. By using these new, high-resolution maps, planners can finally see the real landscape of family life in sub-Saharan Africa, ensuring that resources go exactly where they are needed most.

Technical Summary: Small-area estimation of district-level fertility in 36 countries in sub-Saharan Africa 2000-2025

Problem Statement
Policymakers and program planners in sub-Saharan Africa (SSA) require district-level fertility estimates to guide local decision-making, resource allocation, and health programming (e.g., maternal services, HIV prevention). However, existing approaches typically rely on national-level trends or age-structures from higher administrative levels (e.g., provinces), which fail to capture district-level heterogeneity. Furthermore, direct estimates from household surveys at the district level suffer from large sampling variability due to small sample sizes. Additionally, survey data often contain systematic reporting biases, such as birth displacement (shifting births to the five-year threshold) and heaping (reporting births in specific years like 2000 or the survey year), which distort fertility trends. There is a lack of district-level fertility estimates for SSA that analyze district-level birth data while correcting for these biases.

Methodology
The authors developed a spatiotemporal Bayesian hierarchical model to estimate age-specific fertility rates (ASFR) and total fertility rates (TFR) for 36 SSA countries between 2000 and 2025. The analysis involved three primary steps:

Data Extraction and Preparation:
- Data was extracted from 194 nationally representative household surveys (including DHS, MICS, MIS, AIS, and PHIA) conducted since 1995.
- The dataset included 17.7 million person-years and 3.3 million births from women aged 15–49.
- Survey clusters were assigned to administrative districts using geographic coordinates (offset by 2–10km for confidentiality) or administrative strata.
- Birth and person-year data were aggregated by district, five-year age group, single calendar year, and "time preceding the survey" (TIPS) to account for recall periods.
- Population denominators were derived from WorldPop gridded data, calibrated to UN World Population Prospects national totals.
Process Model (True Fertility):
- The true fertility rate ( $\lambda_{ait}$ $λ_{ai t}$ ) was modeled as a log-linear function comprising:
  - An intercept and a typical age pattern ( $\alpha_a$ ) modeled via a first-order random walk.
  - A linear national trend ( $\beta_1 t$ ) and stochastic national variation ( $\gamma_t$ ) modeled using cubic B-splines with ARIMA(1,1,0) penalization.
  - District-level variation ( $\delta_i$ ) modeled via an intrinsic conditional autoregressive (ICAR) model.
  - Two-way interactions ( $\eta$ ) allowing for variation in trends by age/time, district/time, and district/age, modeled using separable Kronecker products of AR1 and ICAR priors.
- Fixed effects were included to capture specific country-level shocks (e.g., famine in Malawi, fertility boom in Zimbabwe) and urban/rural constraints in Ethiopia.
Observation Model (Bias Correction):
- The model explicitly adjusted for systematic reporting biases by relating observed survey data to the true fertility rate.
- Birth Displacement: Fixed effects and random survey-specific interactions were applied for TIPS = 5 and 6 years to correct for the displacement of births beyond the five-year recall threshold.
- Heaping: Fixed effects were applied for TIPS = 0 (survey year) and TIPS = 10 (ten-year recall) to correct for heaping. Additional time fixed effects were included for years 1999, 2000, and 2001 to address specific heaping around the millennium.
- An observation-level random effect accounted for overdispersion from unmodeled errors.
Implementation:
- The model was implemented in C++ using the Template Model Builder (TMB) package in R.
- Posterior distributions were sampled to generate mean, median, and 95% credible intervals for ASFR and TFR.

Key Contributions

Granular Estimation: Produced the first district-level estimates of ASFR and TFR for 36 SSA countries, moving beyond the limitations of province-level or national-level aggregation.
Bias Correction: Integrated a robust observation model that explicitly corrects for birth displacement and heaping, which are known systematic deficiencies in retrospective survey data.
Spatiotemporal Reconciliation: Developed a framework that reconciles point data (survey clusters) and areal data (districts) while borrowing strength across space, time, and age groups to stabilize estimates in data-sparse areas.
Heterogeneity Characterization: Quantified the extent to which district-level dynamics deviate from national trends and age patterns.

Results

TFR Trends: Between 2000 and 2020, national TFR declined in the majority of countries (median -22%). However, declines stalled in specific countries (Mozambique, Namibia, Rwanda, Zimbabwe) during certain periods.
Subnational Heterogeneity: Significant heterogeneity in TFR was observed within most countries, often exceeding the variation between national averages. In Ethiopia, district-level TFR ranged from 0.91 to 7.42. In 29 of 36 countries, the national TFR was lower than the median district-level TFR, largely because populous urban districts have lower fertility.
Age Patterns: While total fertility declined, the decline was not uniform across age groups. Fertility declined most sharply among women aged 40–49. Conversely, teenage fertility (15–19 years) remained unchanged or declined minimally in several Eastern and Southern African countries.
Spatial Variation: Peak fertility shifted from age 20–24 in Eastern/Southern Africa to 25–29 in Western/Central Africa. Urban areas generally exhibited older mean ages of childbearing compared to surrounding rural provinces.
Bias Estimates: The observation model confirmed substantial reporting biases: survey-year fertility was estimated 3% higher than true fertility, displacement at TIPS=6 was 5% higher, and ten-year heaping was 5% higher.

Significance and Claims
The paper claims that district-level fertility dynamics often differ significantly from national TFR trends and age patterns. Consequently, using national fertility rates to estimate births at the district level will produce inaccurate results in the majority of SSA districts. This inaccuracy leads to inappropriate resource management for key global public health objectives, including family planning, maternal health, and disease prevention. The authors assert that their high-resolution, bias-corrected estimates provide a necessary foundation for local decision-making and monitoring progress toward Sustainable Development Goal 3.7. They further suggest that "fertility stalls" identified at the national level may mask diverging subnational trends, necessitating subnational analysis to distinguish between true stalls and masked dynamics.

More like this