A standard household survey built to estimate vaccination coverage within a few percentage points can require 600 to 900 respondents. A field team trying to decide, this week, whether a specific district needs a follow-up campaign rarely has that kind of time or budget. Lot Quality Assurance Sampling (LQAS) was built for exactly that gap: a small, statistically defensible sample that answers one narrow question — does this area meet the threshold, or does it not — instead of trying to estimate a precise number.
A method borrowed from the factory floor
LQAS did not start in public health. It originated in industrial quality control in the 1920s, where a factory needed a fast way to decide whether a batch, or "lot," of manufactured goods was good enough to ship without inspecting every unit. The public health adaptation followed decades later, gaining traction in the 1980s and 1990s as a way to assess local health systems in lower-resource settings. The World Health Organization introduced the technique into public health survey practice around 1991 and has recommended it for surveys measuring childhood immunization coverage since.
The underlying idea transfers cleanly. Instead of a factory batch, the "lot" is a geographic area — a health zone, a district, a supervision area within a vaccination campaign. Instead of asking "what percentage of this batch is defective," the field team asks "does this area meet or fall short of our coverage target." That reframing is what allows the sample size to shrink so dramatically.
Why 19 observations can be enough
A typical LQAS sample per lot runs 15 to 30 observations, with 19 a commonly used figure. That number is not arbitrary. It comes from the binomial distribution: given a target coverage level and an acceptable margin of decision error, statisticians can calculate the smallest sample size and the exact "failure count" that separates a passing lot from a failing one. A frequently cited example: with a sample of 19 households and a target coverage of 80 percent, the decision rule might be that if fewer than 13 of those 19 are vaccinated, the lot is classified as not meeting target.
Two kinds of error are managed deliberately rather than ignored. An alpha error means a genuinely adequate area gets wrongly flagged as failing; a beta error means a genuinely inadequate area gets wrongly passed. LQAS tables are built so that both error rates, combined, stay under roughly 10 percent — a level program managers can live with in exchange for a survey that takes a fraction of the staff-days a full coverage estimate would require.
What LQAS trades away, and what it buys
The honest way to describe LQAS is as a trade. A conventional cross-sectional survey, sized to produce a coverage estimate with a tight confidence interval, might need 600 to 900 completed questionnaires across a region. An LQAS design covering five supervision areas at 19 observations each totals roughly 95 — a tenth of that effort. What is lost is precision: LQAS does not tell a manager that coverage in a given area is "73 percent." It tells them the area passed or failed against a defined threshold, full stop.
For an operational manager, that trade is usually the right one. A single, precise, region-wide estimate hides exactly the variation that matters most in the field — the pocket of a city or the corner of a district where a campaign quietly under-delivered while the aggregate number looked fine. Sampling many small lots instead of one large area, even with less precision in each, is what surfaces that pocket.
The method’s place in operational research
LQAS sits squarely in the tradition this desk covers under operational research: it is a method built to answer a program’s own question about its own performance, on a timeline the program can actually act on, rather than a method optimized for publishable precision. Médecins Sans Frontières used LQAS during a mass vaccination campaign in N’Djamena, Chad, in 2005, to evaluate coverage across smaller geographic areas of the city than a single aggregate estimate would have captured — identifying areas with meaningfully lower coverage than the citywide figure suggested, and directing follow-up vaccination activity there specifically. That is the pattern LQAS is designed to enable: not a report filed after the fact, but a decision made while the campaign teams are still deployed.
The sampling mechanics behind it are also worth naming, because they are what keep the method statistically honest despite its small size. Locations within a lot are typically selected using probability proportional to estimated size, so that more populous sub-areas have a correspondingly higher chance of being chosen; households within those locations are then selected through systematic random sampling. Skipping either step — convenience-sampling the nearest households, for instance — breaks the statistical guarantees the small sample size depends on.
Where the method fits and where it doesn’t
LQAS is not a replacement for a proper coverage survey when a program genuinely needs a precise, defensible number — for a donor report, for a published evaluation, for comparing coverage year over year with statistical confidence. It is a monitoring tool, built for a specific and recurring operational question: which of our supervision areas need attention right now. Used for that purpose, in the settings it was built for, it lets a small evaluation team cover far more ground, far more often, than a full survey design would ever allow — which is precisely why it has remained a standard part of the operational research toolkit for field immunization and health-system programs for over three decades.