Statistical vs non-statistical audit sampling

Under ISA 530, sampling is statistical when it meets two conditions: items are selected randomly, and results are evaluated with probability theory, including measuring sampling risk. If either is missing, it is non-statistical sampling. Both are acceptable and the choice is a matter of judgement, but both still have to project misstatements and allow for sampling risk.

Updated 10 November 2026

What statistical sampling means in audit

The standard's definition says nothing about formulas or sizes. It sets two requirements:

  1. Random selection. Every sampling unit has a known probability of selection: simple random, systematic with a random start, or monetary unit selection.
  2. Evaluation with probability theory. The result is expressed as a limit (on misstatement or deviations) at a confidence level, which measures sampling risk.

The usual statistical methods in financial audit are monetary unit sampling (MUS) for tests of details, attribute sampling for tests of controls, and classical variables sampling (mean per unit, difference, ratio), often stratified. That is also the answer to "attribute sampling vs statistical sampling": attribute sampling is one kind of statistical sampling. Attributes versus variables is about what you measure, a rate or an amount (see attribute vs variable sampling).

What non-statistical sampling is

Any sampling that does not meet both requirements. Typically the size comes from a firm table or the auditor's judgement, items are chosen haphazardly (without a structured technique), and the result is evaluated by comparing projected misstatement with tolerable misstatement, with no computed limit.

Non-statistical does not mean anything goes. ISA 530 applies in full:

  • The sample must reduce sampling risk to an acceptably low level (paragraph 7), and the same factors drive its size.
  • Every unit must have a chance of selection and the sample must be representative, free of bias (paragraph 8).
  • Misstatements in tests of details are projected to the population (paragraph 14).
  • Sampling risk is considered in the evaluation: if projected misstatement comes close to tolerable misstatement, or exceeds the expected misstatement in the design, the sample may not give a reasonable basis (paragraph 15 and its application material).

Do not confuse non-statistical sampling with what is sometimes called judgemental sampling: picking the largest, unusual or specific items. That is a valid means of selection under ISA 500, but it is not sampling. Nothing is projected and it says nothing about the rest of the population.

Size does not decide. ISA 530 states that sample size is not a valid criterion to distinguish the two approaches. Twenty-five items selected randomly and evaluated statistically are a statistical sample; two hundred picked by eye are not.

Comparison table

AspectStatisticalNon-statistical
SelectionRandom, systematic with random start, or MUSAny of those or haphazard; never block selection to conclude on the whole
SizeFormula or table with an explicit confidence levelJudgement, firm table or simplified formula
ProjectionRequiredRequired
Sampling riskMeasured: an upper limit at a confidence levelConsidered by judgement, no figure
ConclusionUpper limit against tolerable misstatementProjected misstatement against tolerable misstatement, with a reasoned margin
Defending it to a reviewerAnyone can reproduce itDepends on how well the judgement is documented
NeedsA tool and training in the methodExperience and firm guidance

When to use each

Advantages of statistical sampling

  • It quantifies sampling risk, so the conclusion does not rest on an impression.
  • It forces you to write down confidence, tolerable misstatement and expected misstatement.
  • It is reproducible: another auditor gets the same size and the same limit.
  • With MUS, large items are picked automatically and the sample is usually efficient when few misstatements are expected.

Drawbacks of statistical sampling

  • It needs the complete population in electronic form and a reliable tool.
  • Applied badly (non-random selection, no projection) it gives false precision.
  • In small or very uneven populations, testing the key items may be the better route.

Where non-statistical sampling fits

  • Small populations or low-risk tests where the sample is short anyway.
  • Paper or scattered records, where strict random selection is impractical.
  • Supporting tests, when most of the evidence comes from other procedures.

For significant areas with large electronic populations, the statistical approach is easier to defend to a reviewer or an inspector.

Example: non-statistical size (AICPA) vs statistical

For non-statistical sampling in tests of details, the AICPA Audit Sampling Guide offers a simple formula:

n = (book value ÷ tolerable misstatement) × assurance factor

The assurance factor comes from a table, based on the assessed risk of material misstatement and the assurance other procedures give on the same assertion. The top factor, 3.0, matches the MUS confidence factor at 95% with no errors (3.00), and 2.3 matches the 90% factor (2.31). With a balance of 4,812,350 and tolerable misstatement of 120,000:

ApproachInputsCalculationItems
Non-statistical (AICPA)Factor 3.04,812,350 ÷ 120,000 × 3.0 = 120.31121
Statistical MUS95%, no expected misstatement4,812,350 × 3.00 ÷ 120,000121
Statistical MUS95%, expected misstatement 15,0004,812,350 × 3.00 ÷ (120,000 − 15,000 × 1.6)151
Non-statistical (AICPA)Factor 2.34,812,350 ÷ 120,000 × 2.3 = 92.2493
Statistical MUS90%, no expected misstatement4,812,350 × 2.31 ÷ 120,00093
Statistical MUS90%, expected misstatement 15,0004,812,350 × 2.31 ÷ (120,000 − 15,000 × 1.5)115

With no expected misstatement both approaches give the same size: the AICPA formula is MUS with zero expected errors. The difference shows when you expect to find misstatements. MUS builds that into the formula (151 items instead of 121); in the non-statistical approach it is up to the auditor to enlarge the sample. If they do not, the 121-item sample leaves less room to absorb errors without exceeding tolerable misstatement.

The real difference: evaluation

Suppose the 151-item sample turns up two misstatements: a balance of 12,400 that should be 11,160 (10% tainting) and one of 3,050 that should be 2,135 (30%), with an interval of 31,869.87.

  • Statistical evaluation. Projected misstatement 12,747.95, plus 95,609.60 basic precision and 8,923.56 incremental allowance: an upper limit of 117,281.11. The allowance for sampling risk, 104,533.17, is computed. As it does not exceed 120,000, you conclude at 95% confidence that misstatement does not exceed tolerable misstatement, if only just.
  • Non-statistical evaluation. You have the same projection, 12,747.95, and compare it with 120,000. It looks like plenty of headroom, but you do not know how much of it sampling risk uses up. ISA 530 asks you to judge it: here projected misstatement is below the expected misstatement in the design (15,000), which supports the conclusion.

The example shows why the statistical answer is stricter: a projection of 12,748 against tolerable misstatement of 120,000 leaves, at 95% confidence, less than 3,000 of headroom. The full calculation is in evaluating MUS results, and the risk concept in sampling risk.

Common mistakes

  • Calling a haphazard sample statistical: ISA 530 says haphazard selection is not appropriate for statistical sampling.
  • Not projecting in a non-statistical sample and just calling the differences "not significant".
  • Selecting from a population that does not reconcile to the ledger.
  • Mixing specific items and sample items in the same evaluation.
  • Swapping awkward items for convenient ones halfway through the test.

Frequently asked questions

What is statistical sampling in audit?

Under ISA 530, sampling that selects items randomly and evaluates the results with probability theory, including measuring sampling risk. If either feature is missing, it is non-statistical sampling.

Is non-statistical sampling allowed under ISA 530?

Yes. The standard accepts both approaches and leaves the choice to the auditor's judgement. Non-statistical samples must still be representative, misstatements must be projected and sampling risk considered in the evaluation.

Is monetary unit sampling statistical?

Yes, when items are selected with a random start and results are evaluated with the upper misstatement limit. If you selected by amount but only compared the projection with tolerable misstatement, the evaluation would be non-statistical.

Is attribute sampling statistical?

It is when selection is random and the upper deviation limit is computed from the binomial distribution or the AICPA tables. "Attribute" describes what you measure (a deviation rate), not the approach.

What is judgemental sampling in audit?

The term is used for non-statistical sampling, where size and evaluation rest on the auditor's judgement. It should not be confused with selecting specific items such as the largest or unusual ones, which is not sampling and is not projected.

  • IAASB, ISA 530 Audit Sampling
  • IAASB, ISA 500 Audit Evidence
  • AICPA, AU-C 530 Audit Sampling
  • AICPA, Audit Guide: Audit Sampling

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.