How to Use LQAS to Classify Programme Coverage

Lot Quality Assurance Sampling classifies whether a coverage indicator meets or falls below a threshold, rather than estimating a precise coverage rate. It is widely used in humanitarian and health programme monitoring, where a fast pass or fail classification across many districts or lots matters more than a precise percentage in any one of them. This guide first explains what LQAS is and when it applies, then how the classification method works, then walks through how to apply it in AnalyZ Solutions.

01What LQAS is, and when it applies

LQAS originated in industrial quality control, where a manufacturer needed a fast, cheap way to decide whether a batch, or lot, of products met an acceptable quality standard without inspecting every item. Public health and humanitarian programmes adapted the same logic to monitoring: rather than estimating the exact coverage of a service in every district, LQAS draws a small sample from each district, or lot, and applies a decision rule to classify that lot as meeting or falling below a threshold. The result is a pass or fail classification for each lot, not a coverage percentage.

LQAS is a strong fit when three conditions hold together.

• The decision that matters is a threshold, not a precise number. The programme needs to know whether coverage in a given area is high enough, not exactly how high it is. A decision such as does this district need additional support is naturally a threshold question.
• Many lots need to be assessed at manageable cost. LQAS is designed to be repeated across a large number of districts, facilities, or catchment areas using a small, standard sample size in each, making it far cheaper per lot than a precision-based survey repeated at the same scale.
• Fast, actionable results matter more than statistical precision. LQAS trades away the ability to report a precise coverage figure in exchange for a quick classification that can drive an operational decision, such as where to direct additional resources.

Why LQAS is efficient for monitoring many lots

A precision-based survey, such as the Cross-sectional tool, generally needs several hundred respondents per area to produce a reasonably tight coverage estimate. Repeating that across dozens or hundreds of districts is rarely affordable or timely. LQAS deliberately gives up precision, it does not attempt to estimate the exact coverage rate, in exchange for a much smaller sample per lot, often as few as 19 people, while still keeping the chance of a wrong pass or fail classification acceptably low. This trade-off is what makes it practical to monitor many areas on a recurring basis, which a precision-based approach applied at the same scale usually cannot support.

Common pitfalls

• Reporting LQAS results as if they were precise coverage estimates. A lot classified as passing has not been shown to have any specific coverage percentage, only that its coverage is reliably above the threshold given the sample drawn. Reporting a specific coverage figure derived from an LQAS sample overstates what the design actually supports.
• Comparing LQAS classifications across lots as though they were precise rankings. Two lots that both pass are not necessarily equally far above the threshold. LQAS was not designed to rank or compare lots against each other, only to classify each one independently against the same threshold.
• Using LQAS thresholds that do not reflect an operationally meaningful distinction. The upper and lower bounds should represent coverage levels that genuinely correspond to different actions the programme would take. Bounds set arbitrarily close together produce a design that struggles to reliably distinguish clearly acceptable from clearly unacceptable lots.
• Treating the standard lot size of 19 as universally correct. It is the right answer for the commonly used threshold and error rate combination it was derived from, not a fixed constant. A different threshold, or tighter error tolerances, changes the required lot size, which is why AnalyZ Solutions calculates it directly from the inputs provided rather than defaulting to 19 automatically.

02How the classification method works

Every other tool in Sample Size Estimation answers a version of how many people do I need to estimate a value precisely. LQAS answers a different question: how many people do I need to sample from a lot, and what decision rule should I apply, to reliably classify that lot as meeting or not meeting a threshold, while keeping the chance of misclassification acceptably low.

Two error rates govern the design.

• Alpha error. The chance of classifying a lot as failing when it actually meets the threshold, a false alarm.
• Beta error. The chance of classifying a lot as passing when it actually falls below the threshold, a missed problem.

Standard LQAS designs keep both error rates at or below 10%. Given a coverage threshold and the upper and lower bounds that define what counts as clearly acceptable and clearly unacceptable, AnalyZ Solutions searches for the smallest lot size and decision rule that satisfies both constraints, using binomial probabilities.

Each input plays a specific role in the search.

• Coverage threshold. The value the lot is being classified against, for example 80% immunisation coverage. This is the dividing line between pass and fail in principle, though the actual decision rule accounts for sampling variability around it.
• Upper bound. A coverage level clearly high enough that a lot performing at or above it should almost always be classified as passing. Set this at a level the programme would confidently consider a success.
• Lower bound. A coverage level clearly low enough that a lot performing at or below it should almost always be classified as failing. Set this at a level the programme would confidently consider in need of attention.

Choosing your bounds

The gap between the upper and lower bounds matters as much as the threshold itself. A wide gap, for example an upper bound well above and a lower bound well below the threshold, allows a smaller lot size to reliably tell the two apart. A narrow gap demands a larger lot size to achieve the same error rates, since distinguishing two coverage levels that are close together is inherently harder from a small sample. Set the bounds to reflect coverage levels that would genuinely lead the programme to different decisions, not values chosen arbitrarily close to the threshold.

Choosing your error rate tolerance

The 10% and 10% convention is a widely used default, not a fixed requirement. A programme making a high stakes decision, such as whether to divert emergency resources to a district, may prefer to tighten one or both error rates below 10%, at the cost of a larger lot size. A programme running frequent, lower stakes monitoring rounds across many districts may find the standard 10% and 10% convention gives an efficient balance between accuracy and the practical burden of repeated data collection.

1. Set your coverage threshold. The value you are classifying against, for example 80% immunisation coverage. This is the dividing line your decision is built around, so it should reflect an actual programmatic standard, not an arbitrary round number.
2. Set the upper bound. The coverage level clearly high enough that a lot performing at or above it should almost always be classified as passing. Set this at a level your programme would confidently consider a success, as discussed above, since the gap between this and the lower bound directly affects how large a lot size the tool will require.
3. Set the lower bound. The coverage level clearly low enough that a lot performing at or below it should almost always be classified as failing. Set this at a level your programme would confidently consider in need of attention. Avoid placing the upper and lower bounds too close to the threshold, since this forces a much larger lot size to reliably tell the two apart.
4. Calculate. The tool searches for the minimum lot size and decision rule, the maximum number of failures the lot can show and still pass, that keeps both the alpha and beta error rates at or below 10%, and reports both alongside the result.

03Worked example

A programme wants to classify districts by whether vaccination coverage meets 80%. Coverage of 90% or above should clearly be classified as acceptable, and coverage of 70% or below should clearly be classified as unacceptable.

Interpret the result as pass or fail for each district sampled, not as a precise coverage percentage. A district can be classified as passing without knowing its coverage is exactly 84% or 87%, only that it is reliably above the threshold given the sample drawn.

04Common mistakes to avoid

• Reporting a coverage percentage derived from an LQAS sample. The design supports a pass or fail classification, not a precise estimate. A separate precision-based survey is needed if a specific coverage figure must be reported.
• Setting the upper and lower bounds too close to the threshold. This forces a much larger lot size to achieve the same error rates, and if the lot size is not increased to match, produces a design that cannot reliably distinguish the two.
• Reusing a lot size of 19 without checking it fits the actual threshold and bounds chosen. The correct lot size depends on the specific threshold and bounds. Nineteen is common, not universal.
• Treating a passing classification as evidence of high performance rather than adequate performance. A lot that just clears the threshold and a lot performing well above it can both be classified as passing. LQAS does not distinguish between them.