1 Introduction
Zero-dose children, defined as children who have not received any routine vaccinations, remain a central concern for immunization planning in Nigeria. As of 2022, an estimated 2.2 million children in the country were unvaccinated (Gavi Zero-Dose Learning Hub 2023), and Kano State continues to carry one of the highest zero-dose burdens in the country. Yet it remains difficult to pinpoint where zero-dose children are concentrated and why they are missed, because actionable information is often limited at local levels, even as routine monitoring data continue to improve. This creates a familiar tension for measurement: estimates must be representative and credible, but also timely and affordable.
For vaccination coverage, multistage cluster probability surveys are widely treated as the gold standard because they are designed to support valid inference to the target population. In practice, representativeness is achieved by sampling clusters such as enumeration areas and then sampling households within them. This clustering is operationally efficient, but it inflates sampling variance relative to simple random sampling and therefore increases the sample size needed to reach a given precision. In settings without a unified registry of children, these designs also require probabilistic household selection, listing and screening to find age-eligible children, travel beyond convenient areas, and repeat visits to address non-contact and non-response. Recent World Health Organization (WHO) guidance further emphasizes probability designs tailored to explicit inferential objectives and precision targets, rather than the historically common Expanded Programme on Immunization (EPI) \(30 \times 7\) template (WHO 2018). In many contexts, that shift implies larger sample sizes than past practice, and it makes the time and cost burden of methodical field logistics harder to avoid.
In late 2024, the Gates Foundation (GF) commissioned Mindset to conduct a baseline study of zero-dose children in Kano State using this gold-standard approach (Mindset 2025). The study covered 15 Local Government Areas (LGAs) identified by government as priority areas for improving immunization uptake. Three LGAs were designated as sentinel strata (Gaya, Gabasawa, and Nassarawa) and sampled to support standalone, representative estimates. The remaining 12 LGAs were combined into a fourth, non-sentinel stratum. The baseline was intentionally large, even by vaccination coverage survey standards, and field data collection took three months to complete. The study was powered not only to estimate baseline coverage, but to support inference about change in the sentinel LGAs, including detection of relatively small differences at a subsequent assessment. Because implementing partners are running interventions in these areas, the GF required a baseline precise enough to support hypothesis testing of program impacts, which increased the required sample size beyond what a conventional EPI-style design would suggest.
The large sample size for the Mindset baseline study exemplifies why gold-standard surveys are not always a realistic default for ongoing monitoring. Even when they are the right tool for establishing a benchmark, their timelines and costs can exceed what is feasible for routine decisions. This gap between what is methodologically ideal and what is operationally feasible has sharpened interest in alternatives, especially as new open-source data streams and maturing technologies expand the set of plausible options for measuring coverage. The practical question is therefore not whether the gold standard is valuable, but when it is feasible, and how other measurement approaches compare to it in terms of accuracy, timeliness, cost, and operational burden.
There is therefore strong interest in systematically comparing faster, lower-cost approaches to the gold-standard benchmark. To build that evidence base, Mindset implemented five alternative methods alongside the baseline probability survey as part of a head-to-head comparison study. The methods included the Network Scale-Up Method (NSUM), an Adaptive Sampling (AS) design, Lot Quality Assurance Sampling (LQAS), an enhanced administrative method, and Rapid Convenience Monitoring (RCM). The sections below briefly describe each method before presenting the comparison results.
NSUM: This method uses respondents’ reports about their personal networks to infer the prevalence of a characteristic in the broader population. In practice, respondents are asked how many people they know (their alters) and how many of those alters meet the criterion of interest — in this study, whether any of an alter’s children aged 12–23 months are unvaccinated. The method then scales these alter counts to estimate prevalence, effectively increasing information per interview relative to direct measurement. Its performance depends on strong assumptions about mixing patterns, visibility, and reporting accuracy, which can introduce bias if violated.
AS: This method begins with an initial probability sample and then expands the sample using a pre-specified rule based on information observed during data collection. In networked or clustered settings, this often means adding linked individuals or nearby households when a trigger condition is met — for example, when the number of zero-dose children observed in an initial cluster exceeds a pre-set threshold. The objective is to increase efficiency for rare or spatially concentrated outcomes by reallocating effort toward areas with higher yield. Inference requires careful accounting for the adaptive selection mechanism to avoid biased estimates and to quantify precision.
LQAS: This method is designed primarily for classification, not point estimation. It has been widely used in global health and development for vaccination coverage monitoring and supervision. It uses small samples within lots (for example, administrative wards or health facility catchments) to determine whether coverage is likely above or below a threshold level. The method is attractive operationally because it can be fast and resource-light, and it maps naturally to supervisory decisions. Weighted point estimates can be produced at aggregate levels, but these are typically secondary and may be unstable unless sample sizes are increased beyond standard LQAS practice.
RCM: This method is a rapid field approach intended to identify missed populations and operational gaps, rather than to produce representative estimates. It typically relies on convenience selection of locations and households, often guided by local knowledge and implementation priorities. The main output is actionable intelligence about the children and areas encountered along the walk, such as surfacing reasons for non-vaccination or flagging apparent service gaps where they happen to be observed. Because selection is non-probabilistic, population-level prevalence estimates are not interpretable without strong and usually unjustified assumptions.
Administrative approaches: This family of methods estimates coverage using routine service statistics reported by Primary Health Centers (PHCs) on platforms like District Health Information System (DHIS2), typically by combining counts of doses delivered with target population denominators. They can be produced frequently and at fine geographic resolution, which makes them operationally useful for monitoring. However, both numerators and denominators can be unreliable due to reporting error, duplicate counting, migration and cross-area service use, and outdated population projections. These issues can produce implausible values (for example, coverage above 100%) and can make trends hard to interpret without validation.
While much of the comparison centers on estimated prevalences of coverage and zero-dose status, not all methods are designed to estimate in the same way. LQAS is primarily a classification tool with estimation as a secondary objective, and RCM is explicitly not representative and is not intended for population-level coverage estimation. Even among methods that do target prevalence, sample sizes, sampling frames, and inferential assumptions differ substantially, leading to different precision and different vulnerability to bias. For that reason, the report emphasizes structured, fair comparisons, including normalization and design-aware adjustments where needed, rather than treating any single set of raw estimates as directly commensurate.
Finally, the report includes a cost analysis that decomposes survey activities into their operational components. This provides a more granular view of what drives cost and duration across methods, and generates practical inputs for future planning. Together, the accuracy, precision, and operational findings are intended to inform method selection for future coverage measurement, with an explicit focus on the trade space between statistical performance, cost, and speed.