Attribute vs variable sampling in audit: differences and examples

Attribute sampling measures how often something happens (a control failing) and is used for tests of controls. Variable sampling measures how much (the misstatement in a balance) and is used for substantive tests of details; monetary unit sampling sits between the two.

Updated 10 November 2026

The basic difference: rate or amount

Every audit sample answers a question. If the question is "how often does this control fail?", the answer is a rate and the method is attribute sampling: each item either complies or is a deviation, whatever its amount. If the question is "how much misstatement could there be in this balance?", the answer is an amount and the method is a form of variable sampling: you measure the difference between book value and audited value for each item and project it to the population.

ISA 530 (AU-C 530 in the US) covers both and prescribes neither. What it requires is consistency: the sample has to fit the purpose of the procedure, and the results have to be evaluated in the terms the sample was planned in. A control deviation is not projected in currency, and a misstatement in currency is not evaluated as a rate.

Tests of controls and substantive tests

In a test of controls the auditor gathers evidence that a control operated through the period, so as to rely on it and do less other testing. What matters is the deviation rate, compared with a tolerable rate such as 5%. That is attribute sampling.

In a substantive test of details the auditor looks for misstatement in a balance or class of transactions: receivables, inventory, expenses. What matters is the amount of misstatement, compared with tolerable misstatement, which derives from performance materiality. That is variable sampling in one of its forms.

Dual-purpose tests check the control and the amount on the same items. They are allowed, but each result is evaluated with its own method: deviations as a rate, differences as an amount. The sample size is the larger of the two.

MUS: variable sampling built on attribute theory

Monetary unit sampling (MUS, also called PPS or dollar unit sampling) is a hybrid. The sampling unit is each currency unit, and the underlying question is an attribute one: what proportion of the currency units in the population is wrong? Its confidence factors come from the Poisson distribution, an approximation to the binomial, and the result is turned into an amount by multiplying by the sampling interval. It is attribute sampling applied to monetary units in order to conclude on an amount.

That explains its strengths and weaknesses. It favours large items without any stratification, needs no estimate of the standard deviation and gives small samples when few misstatements are expected. On the other hand, omitted and zero-balance items cannot be selected, understatements are handled poorly and the sample grows very quickly if many misstatements are expected.

Classical variables sampling

Classical variables sampling relies on the normal distribution. The sampling unit is the item, selection is random (almost always within strata by value) and the total misstatement is estimated with a precision interval. There are three estimators:

  • Mean-per-unit. Estimates total audited value as the number of items times the mean audited value in the sample, and misstatement as the difference from book value. It needs no differences, but its precision depends on the spread of audited values, which is usually wide.
  • Difference. Projects the average difference per item: number of items times the mean sample difference. Very precise when differences are small relative to item values.
  • Ratio. Projects the ratio of differences to book value in the sample onto the population book value. Better than difference estimation when misstatements grow with item size.

Difference and ratio estimation need a reasonable number of differences in the sample: with very few, the calculated precision is not reliable. The details, with a full example, are in stratified sampling in audit.

Comparison table

AttributeMonetary unit (MUS)Classical variables
Type of testTest of controlsSubstantive test of detailsSubstantive test of details
What it measuresDeviation rateMisstatement in currency (overstatement)Misstatement in currency (both directions)
Sampling unitItem or transactionEach currency unitItem
DistributionBinomial (hypergeometric for small populations)PoissonNormal
SelectionSimple random or systematicSystematic over the cumulative listingRandom within strata
ResultUpper deviation limit against tolerable rateUpper misstatement limit against tolerable misstatementProjected misstatement ± precision against tolerable misstatement
Many misstatements expectedDo not rely on the controlSample explodesWorks well
Zero or negative balancesNot relevantHandled separatelyIncluded
Needs a spread estimateNoNoYes (standard deviation per stratum)

How to choose

  1. Are you testing a control or a balance? A control means attribute sampling. A balance or class of transactions, read on.
  2. Is the risk overstatement? Receivables, revenue and inventory are usually tested for overstatement, which suits MUS. If the risk is understatement (payables, unrecorded liabilities), a sample drawn from the recorded balance cannot find what was never recorded: you need another population, such as subsequent payments, and variables sampling usually fits better.
  3. Do you expect few misstatements? With few, MUS gives smaller samples. With many, or many small differences in both directions, stratified variables sampling is more efficient.
  4. Are there many zero, negative or opposite-sign balances? Classical variables sampling handles them naturally; under MUS they are split out and tested another way.
  5. Do you have an estimate of the spread? Classical variables sampling needs one to size the sample. In a first-year audit without data, MUS is easier to plan.

One example of each

Attribute: purchase order approval

Example population of 2,500 purchase orders. At 95% confidence, a 5% tolerable rate and a 1% expected rate, the sample is 93 orders with 1 deviation allowed. If 2 deviations turn up, the upper deviation limit is 6.62%, above 5%: you cannot rely on the control as planned and you extend substantive testing of purchases.

Monetary unit: trade receivables

Example population of 1,200 open invoices with a book value of 16,967,214.41. At a tolerable misstatement of 375,000, expected misstatement of 50,000 and 95% confidence, the MUS sample is 173 selections: an interval of 98,076.38 (book value ÷ 173), and every invoice larger than the interval is certain to be selected.

Classical variables: the same receivables

On the same 1,200 invoices, with the same tolerable and expected misstatement and the same confidence, stratified variables sampling (four strata plus a top stratum holding the 7 invoices of 150,000 or more, examined 100%) needs 336 items. Without stratification, mean-per-unit would need 1,153 of the 1,200. After testing the stratified sample, difference estimation projects misstatement of 203,445.25 with precision of 118,654.39: an upper limit of 322,099.64, within tolerable misstatement.

In this population MUS is cheaper because few misstatements were expected at planning and value is concentrated in large invoices. Classical variables sampling tests more items, but it returns a centred estimate of misstatement with its interval, measures understatements too and copes better if misstatements turn out to be many.

Not to be confused with acceptance sampling. Many searches for "attribute vs variable sampling plan" come from quality control: the acceptance sampling plans of ISO 2859 (attributes) and ISO 3951 (variables), and the older ANSI Z1.4 and Z1.9, with tables by acceptable quality level (AQL). The statistics overlap, but they answer a different question (accept or reject a lot) and are not used to audit financial statements.

Common mistakes when choosing

  • Evaluating a test of controls in currency, or a substantive test as an error rate. Each is evaluated in the terms it was planned in.
  • Using MUS to search for unrecorded liabilities. What is not in the listing cannot be selected.
  • Using unstratified variables sampling on a population with widely spread values: the sample approaches the whole population.
  • Applying difference or ratio estimation with hardly any differences found and trusting a precision figure that is not reliable.
  • Combining tests with very different sizes in one dual-purpose sample without checking that it covers the larger of the two.

Frequently asked questions

What is the difference between attribute sampling and variable sampling?

Attribute sampling measures a rate, the deviation rate of a control, and is used for tests of controls. Variable sampling measures an amount, the misstatement in a balance, and is used for substantive tests of details.

Is monetary unit sampling attribute or variable sampling?

A hybrid: it is used to conclude on an amount, like variable sampling, but it rests on attribute sampling theory applied to each currency unit in the population, with factors from the Poisson distribution.

What is variable sampling in audit?

Statistical sampling that estimates the monetary misstatement in a balance or class of transactions from the difference between book and audited value of the sampled items. The classical techniques are mean-per-unit, difference and ratio estimation, nearly always with stratification.

When is classical variables sampling better than MUS?

When you expect many misstatements, when there are understatements or zero and negative balances, and when you need a centred estimate of misstatement with a precision interval. With few expected misstatements and an overstatement risk, MUS is usually more efficient.

Can attribute sampling be used for a substantive test?

To estimate the proportion of items in error, yes, but not to conclude on an amount: an error of 10 and one of 100,000 count the same. To conclude in currency, use MUS or variables sampling.

  • IAASB, ISA 530 Audit Sampling (Appendix 1, stratification and value-weighted selection)
  • AICPA, AU-C Section 530 Audit Sampling
  • AICPA, Audit Guide: Audit Sampling
  • ISO 2859-1 and ISO 3951-1, sampling procedures for inspection by attributes and by variables

From the calculation to the working paper

AuditSampler Pro plans, selects and evaluates monetary unit, attribute and stratified samples under ISA 530, and produces the working paper as a PDF. Client data stays on your computer.