Audit sampling lets an auditor test less than the whole of a population and still reach a supportable conclusion about it. It is not a licence to check a few convenient vouchers. SA 530, Audit Sampling, issued by the ICAI, sets out how a sample must be designed, sized, selected, tested and evaluated so that the result can stand for the population. This guide walks through each step using the definitions and requirements of SA 530 as summarised in the ICAI Board of Studies notes on audit sampling, with worked figures for a test of details and a test of controls.
What counts as audit sampling
SA 530 defines audit sampling as the application of audit procedures to less than 100% of items within a population of audit relevance, such that all sampling units have a chance of selection. The population is the entire set of data from which the sample is selected and about which the auditor wishes to draw conclusions. A sampling unit is an individual item in that population: an invoice, a customer balance, a payment or a single rupee in monetary unit sampling.
The "chance of selection" part is what separates sampling from other ways of choosing items. Picking the ten largest balances is targeted testing of specific items; it tells you about those ten, not about the rest. Checking every item is 100% examination. Each approach supports a different conclusion, and the file should say which one was used.
Terms you need before designing a sample
| Term | Meaning under SA 530 |
|---|---|
| Sampling risk | The risk that the conclusion based on a sample differs from the conclusion that would be reached if the entire population were subjected to the same procedure |
| Non-sampling risk | The risk of reaching an erroneous conclusion for any reason not related to sampling risk, such as using an inappropriate procedure or misinterpreting evidence |
| Anomaly | A misstatement or deviation that is demonstrably not representative of misstatements or deviations in the population |
| Stratification | Dividing a population into sub-populations, each a group of sampling units with similar characteristics |
| Tolerable misstatement | A monetary amount set by the auditor, for which the auditor seeks an appropriate level of assurance that it is not exceeded by the actual misstatement in the population |
| Tolerable rate of deviation | A rate of deviation from prescribed controls set by the auditor, for which the auditor seeks an appropriate level of assurance that it is not exceeded by the actual rate in the population |
Sampling risk cuts two ways. In a test of controls, the dangerous error is concluding that controls are more effective than they are, which leads to over-reliance. In a test of details, the dangerous error is concluding that no material misstatement exists when it does. Both affect audit effectiveness. A larger sample reduces sampling risk; it does nothing for non-sampling risk. If the procedure is wrong, testing more items with it does not help.
Statistical or non-statistical
SA 530 covers both. Statistical sampling chooses and evaluates the sample using mathematical and statistical methods, which lets the auditor measure sampling risk. Non-statistical sampling relies on judgment for size and evaluation, but it is not informal: the sample still has to be large enough to reduce sampling risk to an acceptably low level, and every sampling unit still needs a chance of selection. Haphazard selection, for instance, is acceptable only in non-statistical sampling.
Designing the sample
Start with the objective and the right population
Identify the assertion the procedure addresses and pick a population that points the right way. To test occurrence of recorded sales, start from the sales ledger and vouch to dispatch evidence. To test completeness of sales, start from dispatch records and trace to the ledger. Starting from the ledger can never reveal sales that were left out of it. Confirm that the population is complete, for instance by agreeing the listing total to the trial balance before selecting anything.
Stratify and pull out specific items
Where a population holds a few very large items and many small ones, stratification and separate testing of the large or unusual items usually give better evidence for the same effort. Be clear in the file which part of the population the sample conclusion covers and which part was tested in full.
How large should the sample be
SA 530 requires a sample size sufficient to reduce sampling risk to an acceptably low level. There is no magic figure of 25 or 60 that fits every test. The lower the acceptable risk, the larger the sample. The ICAI notes list the factors that push sample size up:
| Tests of controls: sample size increases with | Tests of details: sample size increases with |
|---|---|
| Greater planned reliance on the control | Higher assessed risk of material misstatement |
| Lower tolerable rate of deviation | Lower tolerable misstatement |
| Higher expected rate of deviation | Higher expected misstatement |
| Higher desired level of assurance | Higher desired level of assurance |
According to the same notes, population size has little, if any, effect on sample size. A firm's methodology or sampling tool can translate these inputs into a number, but the working paper should still record the inputs and the reasoning.
Selecting the items
SA 530 requires selection such that each sampling unit has a chance of being selected. The main methods:
- Random selection: every item, or every item within a stratum, has a known chance, using random number generators or tables.
- Systematic selection: the number of units is divided by the sample size to give an interval. With an interval of 50, the auditor picks a starting point within the first 50 and then every 50th unit.
- Monetary unit sampling: a value-weighted method in which sample size, selection and evaluation lead to a conclusion in monetary amounts, so larger items have a proportionately higher chance of selection.
- Haphazard selection: the auditor selects without a structured technique but avoids conscious bias or predictability. It is not appropriate for statistical sampling.
- Block selection: choosing a run of contiguous items. It cannot ordinarily be used in audit sampling because items in a sequence tend to share characteristics that differ from the rest of the population.
Keep the random seed, the interval and starting point, or the tool output, so that a reviewer can reperform the selection.
Performing the procedure and handling exceptions
Apply the planned procedure to every selected item. Two situations need care.
A selected item is not applicable. The standard's example is a voided cheque selected when testing for evidence of payment authorisation. If the auditor is satisfied that the cheque was properly cancelled, so it is not a deviation, an appropriately chosen replacement is examined.
The procedure cannot be applied. If neither the designed procedure nor a suitable alternative can be applied to a selected item, the auditor treats it as a deviation in a test of controls, or a misstatement in a test of details. Missing documents are not a reason to swap in an easier item.
For every deviation or misstatement, investigate its nature and cause and consider its effect on the audit objective and other areas. Treating an error as an anomaly is allowed only when the auditor obtains a high degree of certainty, through additional procedures, that it is not representative of the population.
Worked example: test of details on purchases
A company's purchases for the year total Rs 5,50,00,000. Eight invoices above Rs 5,00,000 each, totalling Rs 70,00,000, are tested in full as specific items. The remaining Rs 4,80,00,000 is spread over 6,000 routine invoices. The auditor sets tolerable misstatement for purchases at Rs 10,00,000.
- Selection. The firm's methodology gives a sample of 60 invoices. The interval is 6,000 divided by 60, or 100. With a random start at invoice 37, the auditor selects invoices 37, 137, 237 and so on up to 5,937.
- Testing. The 60 invoices have a book value of Rs 12,00,000. Vouching finds pricing errors totalling Rs 18,000 of overstatement. Testing of the eight specific items finds a known misstatement of Rs 50,000.
- Projection. SA 530 requires misstatements in tests of details to be projected to the population. Using a simple ratio projection, the error rate is Rs 18,000 divided by Rs 12,00,000, or 1.5%. Applied to the sampled population of Rs 4,80,00,000, the projected misstatement is Rs 7,20,000.
- Evaluation. Projected misstatement of Rs 7,20,000 plus the known Rs 50,000 gives Rs 7,70,000, against tolerable misstatement of Rs 10,00,000. The remaining margin of Rs 2,30,000 has to cover sampling risk. If the auditor judges it too thin, or if the pricing errors point to a wider process failure, the auditor asks management to investigate and adjust, or extends the testing.
For tests of controls, no explicit projection of deviations is necessary. If 3 of 40 sampled purchase approvals lack evidence of authorisation, the deviation rate is 7.5%. Against a tolerable rate of 5%, the control cannot be relied on as planned, and the auditor revises the risk assessment and expands substantive work.
Evaluating the result
The last requirement is to evaluate the results of the sample and whether the use of audit sampling has provided a reasonable basis for conclusions about the population. If not, the options are to request management to investigate identified misstatements and make adjustments, or to modify the nature, timing and extent of further procedures. The ICAI has also published an Implementation Guide to SA 530, Audit Sampling for practitioners who want fuller practical guidance.
Sampling working paper checklist
- State the audit objective and assertion.
- Define the population and confirm its completeness against the ledger or trial balance.
- Define the sampling unit.
- Separate individually significant and unusual items and record how they were tested.
- Record the sample-size inputs: tolerable and expected misstatement or deviation, assessed risk and desired assurance.
- Record the selection method with seed, interval, start point or tool output.
- Document how voided or inapplicable items were replaced, and any items treated as errors because they could not be tested.
- Investigate every exception for nature and cause; support any anomaly with additional procedures.
- Project misstatements for tests of details and compare with tolerable misstatement.
- Conclude whether the sample gives a reasonable basis for the population conclusion, and record the follow-up where it does not.
Common mistakes in practice
- Using the same sample size every year regardless of risk and expected errors.
- Calling a convenience or block selection a representative sample.
- Drawing a completeness conclusion from a population that starts in the ledger.
- Replacing an item because its documents are hard to find.
- Listing errors without projecting them in tests of details.
- Documenting the selected items but not the objective, population, design inputs and evaluation.
Updated on 27 September 2026: retitled, merged the duplicate guide on selecting, testing and evaluating a sample, and added SA 530 definitions, sample-size factors, the voided-cheque and untestable-item rules, and worked examples for a test of details and a test of controls.
