Inspection sample planner
Work out how many sites or meters to check, how to spread the checks across groups, and how to read the result afterwards. The sums assume a random sample. A sample chosen because it looks suspicious is useful for follow-up but cannot describe the whole population.
HOW MANY TO CHECK TO ESTIMATE A RATE
For example, the share of checked sites where an issue is confirmed
How sure you want to be that the answer falls within the margin.
Your choice. Smaller margins need larger samples.
Use 50 if you have no idea. It gives the most cautious sample.
Sample size n0 = z squared x p x (1 minus p) / e squared, reduced for a finite population by n = n0 / (1 + (n0 minus 1) / N). Using an expected rate of 50% gives the largest, most cautious sample. The sample must be drawn at random from the whole population for the margin to hold.
HOW MANY TO CHECK TO HAVE A CHANCE OF FINDING AT LEAST ONE
Sometimes called discovery sampling
Your choice. Rarer issues need larger samples.
How sure you want to be that the answer falls within the margin.
If a share q of the population has the issue, a random sample of n has at least the confidence level you chose of containing at least one. With a population size the exact hypergeometric chance is used; without one, n = ln(1 minus confidence) / ln(1 minus q). If the sample finds none, you can say the share is probably below q. You cannot say there are none.
HOW MANY METERS TO TEST TO ESTIMATE AVERAGE ERROR
For a bench or field accuracy test
Your estimate from earlier tests. Starting value is illustrative.
Your choice.
How sure you want to be that the answer falls within the margin.
n0 = (z x sigma / E) squared, reduced for a finite population as above. Sigma is your expected spread of meter error (standard deviation, in percentage points), taken from earlier tests or a pilot. Formal acceptance testing of meter batches follows national metrology rules and standards such as the ISO 2859 series: check which rule applies to you and confirm with a metering engineer.
SPREAD THE CHECKS ACROSS GROUPS
For example regions, tariff classes or feeders
From the first planner or your own budget.
Where the group has that many.
The second option puts more checks where the expected rate is closer to 50%, because those groups vary most.
| Group | Population | Expected rate (%) | Checks | Share of checks |
|---|
Allocation is rounded so the checks add up to your total, groups are never given more checks than they hold, and each group gets at least the minimum you set where it has that many. Draw the checks at random inside each group.
WHAT DOES THE RESULT TELL ME
After the checks are done
Leave 0 to skip the population range.
How sure you want to be that the answer falls within the margin.
The range uses the Wilson score interval, which behaves better than the simple formula for small samples and for rates near 0% or 100%. It applies only if the checked sites were chosen at random from the population. If they were chosen because they looked unusual, the result describes that group only and must not be scaled up. The population count range ignores the finite-population correction, so it is slightly wide when the sample is a large share of the population.
This planner applies standard sampling formulas. They assume random sampling from the whole population. Starting values are illustrative placeholders, not benchmarks. The tool does not detect or confirm loss, fraud or theft, and a sample size is not a quota or a target for inspectors. For formal acceptance testing of meters, follow the rules and standards that apply to you and confirm with a metering or statistics specialist. Results are indicators that require verification and human review.