Why does LQAS use a sample of 19?

Because 19 is close to the smallest sample at which a binomial decision rule can separate a common upper target (such as 80 percent coverage) from a lower threshold (such as 50 percent) while keeping both classification errors below roughly 10 percent. Other sample sizes work; 19 became the field convention because it is the cheapest that still classifies reliably.

Can LQAS tell me the coverage rate in an area?

Not precisely. With 19 observations the margin of error on a rate would be enormous. LQAS answers a classification question, reaching the benchmark or not, for each small area. To estimate an actual rate, combine the LQAS samples from many areas or run a proper coverage survey.

When is LQAS better than a regular survey?

When you supervise many small units (blocks, facilities, villages) and need to know which ones need attention this round, cheaply and repeatedly. A survey estimates the average well but hides which units are failing; LQAS flags failing units directly with a tiny per-unit sample.

Evaluation methods

LQAS, explained for program teams

LQAS (Lot Quality Assurance Sampling) is a monitoring method that classifies each area as reaching or missing a target, instead of estimating a precise rate. You draw a small fixed sample per area, classically 19 people, and apply a simple decision rule: if at least d of them meet the standard, the area passes; if fewer, it is flagged for attention. It lets one team check dozens of blocks every monitoring round at a fraction of survey cost.

LQAS began on factory floors in the 1920s, where inspectors judged whole production lots by testing a handful of items against an accept-or-reject rule. Public health teams adapted it in the 1980s and 1990s, with WHO and others using it to supervise immunisation coverage area by area. The insight transfers perfectly to education and social programs: a block is a lot, and the question is not "what exactly is the rate here?" but "is this block on track or does it need help?"

How the decision rule works

An LQAS design fixes two benchmarks: an upper standard you want areas to reach (say 80 percent of children reading at level) and a lower threshold that clearly signals trouble (say 50 percent). The sample size n and decision rule d are then chosen so that binomial probability keeps both mistakes rare:

Alpha risk: flagging an area that is actually at or above the standard.
Beta risk: passing an area that is actually at or below the lower threshold.

Sample n per area, pass if successes ≥ d Example rule: n = 19, d = 13 for an 80 percent standard against a 50 percent threshold, holding both risks near or below 10 percent. Sampling Shala's engine searches the binomial tables and picks the (n, d) pair that minimises the larger of the two risks for your own benchmarks.

The verdict is a classification, pass or flag, not an estimate of the rate.

What LQAS buys you, and what it cannot

Buys: per-area verdicts every round, tiny per-area cost (19 interviews per block), results field teams can act on the same week, and honest control of both false-alarm and missed-problem risks.
Cannot: tell you the coverage rate in any single area with useful precision. Nineteen observations carry a margin of error of over twenty points. If the headline rate matters, pool the samples across all areas, or run a proper coverage survey.

A practical pattern for education programs: run LQAS every quarter to flag which blocks need support, and a full clustered survey once a year to estimate the program-wide rate. The two answer different questions and neither replaces the other.

Frequently asked questions

Why 19?

It is close to the smallest sample where a binomial rule can separate a standard like 80 percent from a threshold like 50 percent with both classification errors held near 10 percent. Nothing is magic about it; other (n, d) pairs suit other benchmarks, and the engine computes the right pair for yours.

Can LQAS estimate the coverage rate?

Not per area; the sample is deliberately too small for that. It classifies. If you also need a rate, combine samples across areas (19 x 20 blocks is 380 observations, a respectable survey) or design a survey alongside.

When should I choose LQAS over a survey?

When your real question is "which of my many units needs attention now?" asked repeatedly. Surveys average over units and hide the failing ones; LQAS names them, cheaply, every round.

Design your LQAS decision rule

Set your own standard and threshold in Build mode; the engine finds the honest (n, d) pair and writes up the plan.

Open Build mode

LQAS, explained for program teams

How the decision rule works

What LQAS buys you, and what it cannot

Frequently asked questions

Design your LQAS decision rule

Keep reading