What is stratification in audit sampling?
A receivables listing with 4,000 items ranging from EUR 12 to EUR 1.2M produces a sample size that makes reviewers wince. The variability is too wide. You end up pulling 60+ confirmations when 30 would give the same assurance (if you split the population first). That split is stratification, and skipping it is one of the most common reasons audit teams over-sample.
ISA 530 .A6 defines stratification as dividing a population into discrete sub-populations so that each stratum shares similar characteristics. A population with wide variability in item values requires a larger sample to reach a given confidence level. Split that population into strata with narrower variability, and each stratum requires a smaller sample. The sum of those stratum-level samples is typically smaller than the single unstratified sample would have been.
For tests of details, the most common approach is monetary stratification. Items above a certain threshold (often performance materiality or a percentage of it) go into a "test all" stratum. Items between that threshold and a lower cutoff form a middle stratum sampled at a moderate rate. Items below the lower cutoff form a residual stratum, sampled lightly or excluded if their total value is immaterial.
ISA 530 .A10(e) lists population stratification as a factor the auditor must consider when determining sample size. If the population is highly variable and the auditor does not stratify, the resulting sample size must be large enough to compensate. In our experience, stratifying first is more efficient on every engagement where the population has a long tail.
Key Points
- Stratification reduces variability within each sub-population, which allows a smaller total sample to reach the same confidence level.
- The most common stratification criterion is monetary value, but risk characteristics and transaction type are also valid bases.
- ISA 530 .A6 states that stratification can produce a lower sample size without a proportional increase in sampling risk.
- Poor stratification (or none at all) is one of the fastest ways to end up with an over-sized or under-targeted sample.
Why it matters in practice
Teams frequently stratify by value but fail to document why the stratum boundaries were set where they were. ISA 530 .A6 requires the strata to have similar characteristics. A boundary at EUR 50K needs a rationale ("EUR 50K represents 2% of PM" or "items above EUR 50K are individually significant to the assertion"). An arbitrary cutoff without documented reasoning is a review finding. This is the PIOOMA problem: the boundaries look "pulled out of mid-air" unless you tie them back to PM or individual significance.
Some teams apply MUS to an unstratified population and assume the monetary weighting inherent in MUS is equivalent to stratification. It is not. MUS weights selection probability by value, but it does not ensure adequate coverage of the mid-range stratum. ISA 530 .A6 treats stratification as a separate design step that precedes sample selection, not as a feature built into MUS. This is the finding that generates the most review notes on sampling files we see.
Worked example: Transportes Iberica SL
Client: Spanish logistics company, FY2024, revenue EUR 96M, Spanish GAAP (PGC) reporter.
The population is 4,120 trade receivable balances totalling EUR 22.4M, tested for existence.
Step 1: the team exports the receivables listing and sorts by value. 18 balances exceed EUR 200K each, totalling EUR 8.1M (36% of population value). The remaining 4,102 balances range from EUR 12 to EUR 195K, with a mean of EUR 3.5K and high variability.
Step 2: the team defines strata. They create three:
- Stratum 1 (individually significant): all 18 items above EUR 200K. Tested 100%. No sampling.
- Stratum 2 (moderate value): 314 items between EUR 30K and EUR 200K, totalling EUR 9.8M. Sampled using MUS.
- Stratum 3 (low value): 3,788 items below EUR 30K, totalling EUR 4.5M. Sampled using MUS with a lower required assurance because Strata 1 and 2 cover 80% of population value.
Step 3: calculate stratum-level sample sizes. PM is EUR 2.9M. Tolerable misstatement for the full test is EUR 2.9M, allocated across strata based on value:
- Stratum 1: 18 items tested in full (no sampling risk).
- Stratum 2: TM allocation EUR 2.0M. At 95% confidence (R-factor 3.0), expected error EUR 50K. Sample = (9,800,000 x 3.0) / 2,000,000 = 15 items.
- Stratum 3: TM allocation EUR 0.9M. At 90% confidence (R-factor 2.3), expected error EUR 20K. Sample = (4,500,000 x 2.3) / 900,000 = 12 items.
Step 4: execute and evaluate per stratum. One error of EUR 14K found in Stratum 2, projected within that stratum. No errors in Stratum 3. Stratum 1 items confirmed without exception. Total projected misstatement across all strata: EUR 73K, below PM of EUR 2.9M.
Stratification allowed the team to test 45 items while covering all high-value balances in full and sampling the mid-range and low-value strata separately. An unstratified MUS approach would have pulled predominantly high-value items, leaving the mid-range (where the actual error was found) under-represented.
Key standard references
- ISA 530 .A6 covers stratification guidance: dividing a population into sub-populations with similar characteristics to reduce variability and improve sample efficiency.
- ISA 530 .A10(e) lists stratification as a factor the auditor must consider when determining sample size for tests of details.
- ISA 530 .A11 allows the auditor to use statistical or non-statistical methods for sample design, including stratified approaches.
- ISA 530.5 requires the auditor to determine a sample size sufficient to reduce sampling risk to an acceptably low level, which stratification directly supports.
Related terms
Related tools
Related reading
Frequently asked questions
Why does stratification reduce the required sample size?
By lowering variability within each stratum, the auditor can achieve the same confidence level with fewer items per stratum.
Is MUS the same as stratification?
No. MUS weights selection probability by value but does not ensure adequate coverage of mid-range items. Stratification is a separate design step that precedes sample selection.