What is monetary unit sampling?
Monetary unit sampling treats each individual monetary unit (e.g., each euro or dollar) in the population as a separate sampling unit. The auditor calculates a sampling interval by dividing the population's recorded value by the sample size, then selects every nth monetary unit using systematic selection with a random start. The physical item (invoice, journal entry, balance) that contains the selected monetary unit becomes the item tested.
This structure means high-value items are automatically more likely to be selected — a line item worth ten times as much contains ten times as many sampling units and therefore has ten times the probability of selection. This built-in weighting eliminates the need for separate stratification, which is one reason MUS is the most widely used statistical method in substantive testing of details.
Evaluation uses Poisson-based probability tables. When errors are found, the auditor calculates a tainting percentage (the proportion of the item that is misstated) and uses it to compute an upper error limit. This projected misstatement is compared against tolerable misstatement. A structural limitation applies: MUS is designed for overstatement testing only. Because selection is proportional to recorded value, items that are understated or missing from the population receive little or no coverage.
Key Points
- Automatically selects more from high-value items. Each monetary unit has equal selection probability, so larger items have proportionally greater coverage without separate stratification.
- Designed for overstatement testing only. MUS is structurally unsuitable for completeness (understatement) assertions because selection is proportional to recorded value.
- Compare projected misstatement against both tolerable and expected misstatement. ISA 530.A22 requires two comparisons — one against tolerable misstatement and one to assess whether the nature and cause of errors suggests systematic issues.
- Zero errors does not mean zero risk. A clean sample reduces the upper error limit but does not eliminate sampling risk. The confidence level defines the residual probability that errors exist but were not selected.
Why it matters in practice
The most common practical error is teams performing only one comparison — checking projected misstatement against tolerable misstatement — while omitting the second comparison to expected misstatement. ISA 530.A22 requires the auditor to also consider whether the nature and cause of identified misstatements suggest the actual misstatement in the population may exceed expected misstatement, which would indicate the sample design assumptions were wrong.
A second recurring issue is applying MUS to completeness testing. Because MUS selects proportional to recorded value, items that are missing from the ledger or recorded at zero have no chance of selection. Teams testing the completeness assertion (e.g., unrecorded liabilities, revenue cut-off understatements) should use variables sampling or a non-statistical approach that gives every physical item — not every monetary unit — an equal chance of selection.
Key standard references
- ISA 530.5–15: Core requirements for audit sampling including design, selection, and evaluation.
- ISA 530.A10–A15: Application guidance on statistical and non-statistical approaches, including MUS as a probability-proportional-to-size method.
- ISA 530.A22: Requirement to evaluate the nature and cause of misstatements and compare results against both tolerable and expected misstatement.
Related terms
Related tools
Related reading
Frequently asked questions
Why does MUS automatically weight toward high-value items?
Because each monetary unit is a separate sampling unit. A €500K invoice contains 500,000 sampling units while a €10K invoice contains only 10,000. The larger invoice has 50 times the probability of containing a selected monetary unit, so high-value items are automatically covered without stratification.
Can MUS be used for completeness testing?
MUS is structurally unsuitable for completeness (understatement) testing. It selects proportional to recorded value, giving low-value and zero-value items minimal coverage. If items are missing or understated, MUS will not catch them. Use variables sampling or a non-statistical approach instead.
What happens when a single MUS error pushes the result above tolerable misstatement?
This is common and counterintuitive. Even one error with moderate tainting can push the upper error limit above tolerable misstatement when the tolerance is tight relative to population size. The correct response is to extend the sample, request management adjustment, or both.