Side-by-side comparison
| Dimension | Statistical sampling | Non-statistical sampling |
|---|---|---|
| Selection method | Random or systematic with a random start (every item has a known, non-zero probability of selection) | Judgmental or haphazard selection (ISA 530.A13) |
| Sampling risk measurement | Quantified mathematically (confidence level and precision) | Not quantified; assessed through auditor judgment |
| Sample size determination | Calculated using statistical formulas based on confidence level, expected error, tolerable misstatement, and population size | Determined by auditor judgment, often guided by firm methodology |
| Evaluation of results | Projects misstatements to the population with a measurable confidence interval | Projects misstatements judgmentally; no confidence interval available |
| Documentation burden | Higher: must document the statistical parameters, selection method, evaluation model, and projection calculation | Lower in mechanics, but higher in justification: must explain why the approach is appropriate |
Key Points
- ISA 530 treats both methods as equally valid when properly designed and documented.
- Statistical sampling quantifies sampling risk; non-statistical sampling does not, which makes the conclusion harder to defend under inspection.
- The choice depends on population characteristics and the engagement team's ability to justify the approach.
- Switching between methods mid-procedure without documenting the rationale is a common inspection finding.
When the distinction matters on an engagement
The practical difference surfaces at evaluation, not at selection. Both methods require you to project found misstatements to the population under ISA 530.14. With statistical sampling, the projection is mathematical: the sample result plus an allowance for sampling risk produces a range, and you compare that range to tolerable misstatement. With non-statistical sampling, the projection relies on your judgment, and you cannot attach a confidence level to the conclusion.
This matters when a reviewer or inspector asks: "How do you know the untested portion of the population is not materially misstated?" With a statistical sample, you point to the confidence level. With a non-statistical sample, you point to your documentation of why the sample size was sufficient and why the selection method was appropriate. ISA 530.A11 acknowledges that both approaches can provide sufficient appropriate audit evidence. But the non-statistical path requires stronger qualitative justification in the file, and that justification is exactly what inspectors report as missing.
Worked example: Hartmann Maschinenbau GmbH
Client: German machine manufacturer, FY2024, revenue €67M, HGB reporter.
The engagement team needs to test the accuracy of trade receivables (€9.4M, 312 line items). Performance materiality is €200K. Expected misstatement based on prior year experience: €15K.
Statistical approach applied
The team used monetary unit sampling with a confidence level of 95%. Based on the population (€9.4M), tolerable misstatement (€200K), expected misstatement (€15K), and a 95% confidence level, the calculated sample size was 58 items.
Documentation note: "MUS applied to trade receivables accuracy. Parameters: population €9.4M, tolerable misstatement €200K, expected misstatement €15K, confidence 95%. Sample size: 58 sampling units. Selection: systematic selection with a random start point. Ref: ISA 530.7."
One misstatement found: an invoice for €4,200 overstated by €380 (tainting factor: 9.05%). Projected misstatement: €850K in the worst case at the sampling interval. The team evaluated this against tolerable misstatement and concluded further investigation was required.
Non-statistical approach applied to the same population
The team selected 40 items using a combination of all items above €150K (12 items, covering €3.1M) and 28 items selected haphazardly from the remaining population.
Documentation note: "Non-statistical sample of trade receivables. Rationale for approach: the population contains a small number of individually significant items that can be tested in full, with the remainder consisting of homogeneous, low-value items where haphazard selection provides reasonable coverage. Sample size determined based on firm methodology and prior year experience. Ref: ISA 530.A11."
Same misstatement found in the haphazard portion (€380 overstatement). The team projected the error to the untested portion but could not attach a confidence level to the projection. The file relied on a qualitative argument that the error was isolated. If the reviewer asks for the statistical basis of that conclusion, there is none. The qualitative justification must stand on its own.
What reviewers get wrong
The AFM's inspection findings have repeatedly flagged files where teams described their approach as "statistical" but used haphazard selection, which is by definition non-statistical. ISA 530.A13 is explicit: a statistical sample requires every sampling unit to have a known, non-zero probability of selection. Calling a haphazard selection "random" does not make it statistical, and the mislabelling creates a documentation mismatch that inspectors treat as a methodology failure.
Teams frequently do not document the rationale for choosing non-statistical over statistical sampling (or vice versa). ISA 530.A10 notes that the decision is a matter for the auditor's judgment. But "judgment" still requires documentation. A file that contains a non-statistical sample with no explanation of why that approach was chosen gives the reviewer no basis to assess whether the sample is sufficient.
Key standard references
- ISA 530.5–8: Requirements for sample design, including the choice of sampling approach.
- ISA 530.A10: States that the decision between statistical and non-statistical sampling is a matter for auditor judgment.
- ISA 530.A11: Confirms both methods can provide sufficient appropriate audit evidence.
- ISA 530.A13: Defines statistical sampling as requiring every unit to have a known, non-zero probability of selection.
Related terms
Related tools
Related reading
Frequently asked questions
Does ISA 530 require statistical sampling?
No. ISA 530.A10 states that the decision between statistical and non-statistical sampling is a matter for the auditor's judgment. Both methods can provide sufficient appropriate audit evidence when properly designed. However, the non-statistical path requires stronger qualitative justification in the file because sampling risk cannot be quantified mathematically.
What is the most common inspection finding related to sampling method choice?
The AFM and other regulators have repeatedly flagged files where teams described their approach as statistical but used haphazard selection. ISA 530.A13 is explicit: a statistical sample requires every sampling unit to have a known, non-zero probability of selection. Haphazard selection is by definition non-statistical, and mislabelling it creates a documentation mismatch that inspectors treat as a methodology failure.