Audit sampling risk: what it is and how to control it
Sampling risk is the risk that the conclusion you draw from a sample differs from the one you would reach by applying the same procedure to the whole population. You control it through sample size: 95% confidence means accepting a 5% risk of incorrect acceptance. Non-sampling risk does not fall however much you enlarge the sample.
Updated 10 November 2026
Sampling risk vs non-sampling risk
ISA 530 separates two sources of error. Sampling risk exists because you only look at part of the population: even if you do everything right, the sample can come out better or worse than the population. Non-sampling risk is every other reason for a wrong conclusion. The standard gives three examples: unsuitable procedures, misinterpreting evidence, and failing to recognise a misstatement or deviation that is in front of you.
| Sampling risk | Non-sampling risk | |
|---|---|---|
| Source | Testing part rather than all | Auditor or procedure errors |
| Example | The sample picks up few misstated invoices although the population has many | Confirming a balance at an address the client supplied, without checking it |
| Reduced by | A larger sample or higher confidence | Good design, supervision, review and training |
| Measurable? | Yes, with statistical sampling | No |
| Gone at 100% testing? | Yes | No |
Selecting specific items, such as everything over 50,000, carries no sampling risk because nothing is projected, but it does carry non-sampling risk, as ISA 500 points out.
The four sampling risks
The standard defines sampling risk through two possible wrong conclusions, which apply to tests of details and tests of controls alike. That gives four names:
| Test | Risk | What you wrongly conclude | Affects |
|---|---|---|---|
| Details (substantive) | Risk of incorrect acceptance | The balance is not materially misstated when it is | Effectiveness |
| Details (substantive) | Risk of incorrect rejection | The balance is materially misstated when it is not | Efficiency |
| Controls | Risk of overreliance | The control works better than it does | Effectiveness |
| Controls | Risk of underreliance | The control works worse than it does | Efficiency |
The two effectiveness risks are the dangerous ones: they can lead to an inappropriate opinion. The efficiency risks cost hours, because you will normally do extra work until you find the first conclusion was wrong. Overreliance and underreliance are the AICPA guide's terms; PCAOB AS 2315 calls them assessing control risk too low and too high.
Risk of incorrect acceptance and the audit sampling confidence level
The confidence level is the complement of the risk of incorrect acceptance (or overreliance) you accept: 95% confidence is a 5% risk. In monetary unit sampling it becomes the confidence factor in the formula. If the population's actual misstatement equalled tolerable misstatement, a sample designed at 95% would lead you to accept it at most 5 times in 100.
With a balance of 4,812,350 and tolerable misstatement of 120,000, this is how the MUS sample changes with confidence:
| Confidence | Risk of incorrect acceptance | CF | EF | Items, EM 15,000 | Items, EM 0 |
|---|---|---|---|---|---|
| 80% | 20% | 1.61 | 1.3 | 78 | 65 |
| 85% | 15% | 1.90 | 1.4 | 93 | 77 |
| 90% | 10% | 2.31 | 1.5 | 115 | 93 |
| 95% | 5% | 3.00 | 1.6 | 151 | 121 |
| 99% | 1% | 4.61 | 1.9 | 243 | 185 |
CF: confidence factor; EF: expansion factor; EM: expected misstatement. Factors from the AICPA Audit Sampling Guide; sizes from n = BV × CF ÷ (TM − EM × EF), rounded up.
Tests of controls behave the same way. With a 5% tolerable and 1% expected deviation rate, attribute sampling needs 59 items at 80%, 67 at 85%, 77 at 90%, 93 at 95% and 165 at 99% (the last one allows two deviations instead of one).
Dropping from 95% to 90% saves 36 items in the MUS example; moving to 99% adds 92. ISA 530 does not set the level: firm methodologies tie it to assessed risk and to the evidence from other procedures.
Risk of incorrect rejection: what confidence does not control
The confidence level only limits the effectiveness risks. The efficiency risks depend on how much expected misstatement or deviation you build into the design. This controls example, at 95% confidence and a 5% tolerable rate, shows it:
| Expected deviation rate in the design | Items | Deviations allowed | P(rely) if the control fails 5% | P(reject) if it fails 1% | P(reject) if it fails 2% |
|---|---|---|---|---|---|
| 0% | 59 | 0 | 4.85% | 44.7% | 69.6% |
| 1% | 93 | 1 | 5.00% | 23.8% | 55.7% |
| 2% | 181 | 4 | 4.92% | 3.6% | 29.7% |
Exact binomial probabilities for each sample size.
All three designs hold the risk of overreliance at 5%. But with 59 items and no deviations allowed, a control that fails only 1% of the time (well under the tolerable rate) would be rejected almost half the time. With 93 items, finding two deviations gives an upper limit of 6.62% and also leads to rejection, even if the true rate is 1%. Allowing for some error in the design costs items but cuts false negative conclusions and the extra work they bring. In tests of details, expected misstatement plays the same role.
Sampling risk and the audit risk model
Audit risk is the risk of material misstatement (inherent and control risk) times detection risk. For a sampling-based test of details, detection risk is largely its risk of incorrect acceptance. The appendix to PCAOB AS 2315 and the AICPA guide put it like this:
TD = AR ÷ (RMM × AP)
- TD: allowable risk of incorrect acceptance for the test of details.
- AR: audit risk you accept for the assertion, for example 5%.
- RMM: assessed risk of material misstatement (inherent × control risk).
- AP: risk that analytical procedures and other substantive tests fail to detect the misstatement.
| Risk of material misstatement | Risk from other procedures | Risk of incorrect acceptance | Confidence | MUS items |
|---|---|---|---|---|
| 100% (no reliance on controls) | 100% (no other procedures) | 5% | 95% | 151 |
| 50% | 100% | 10% | 90% | 115 |
| 100% | 50% | 10% | 90% | 115 |
| 50% | 50% | 20% | 80% | 78 |
Audit risk 5%; same balance, tolerable misstatement 120,000, expected misstatement 15,000.
The ISAs do not require you to put numbers on these risks, and most methodologies use tables that turn "high, medium or low risk" into a confidence level. The logic is the same: the more assurance controls and other procedures give you, the more sampling risk you can accept in the test of details and the fewer items you need. The other drivers of size are in audit sample size.
Allowing for sampling risk when you evaluate
In statistical sampling you quantify it. In MUS, the upper misstatement limit adds an allowance for sampling risk (basic precision plus incremental allowance) to projected misstatement and compares the total with tolerable misstatement; see evaluating MUS results. In attribute sampling, the upper deviation limit is compared with the tolerable rate.
In non-statistical sampling there is no figure, but ISA 530 still requires you to consider it: the closer projected misstatement comes to tolerable misstatement, the likelier actual misstatement exceeds it, and if it exceeds the expected misstatement used in the design, sampling risk may be unacceptable. The two approaches are compared in statistical vs non-statistical sampling.
Reducing non-sampling risk
- Reconcile the population to the ledger before selecting, so you are not sampling an incomplete listing.
- Write down what counts as a misstatement or deviation before you start.
- Use a procedure that fits the assertion: a confirmation tests existence, not valuation.
- Test every item to the same standard and keep the evidence on file.
- Have someone experienced in the area supervise and review.
Frequently asked questions
What is audit sampling risk?
The risk that the conclusion based on a sample differs from the one you would reach by applying the same procedure to the whole population. It is controlled through sample size and, in statistical sampling, measured.
What is non-sampling risk in audit?
The risk of a wrong conclusion for reasons unrelated to sampling: unsuitable procedures, misinterpreted evidence or misstatements that go unrecognised. Unlike sampling risk, it does not fall when the sample grows and does not disappear at 100% testing.
What is the risk of incorrect acceptance?
The risk of concluding from the sample that a balance is not materially misstated when it is. It is the complement of the confidence level: at 95% confidence you accept a 5% risk. It affects audit effectiveness.
What confidence level is used in audit sampling?
ISA 530 does not set one. Firm methodologies commonly use 80% to 95% for tests of details depending on assessed risk and other evidence, and 90% or 95% for tests of controls the auditor relies on.
Which sampling risks affect audit effectiveness?
The risk of incorrect acceptance in tests of details and the risk of overreliance in tests of controls. Incorrect rejection and underreliance affect efficiency, because they lead to extra work.
- IAASB, ISA 530 Audit Sampling
- IAASB, ISA 500 Audit Evidence
- PCAOB, AS 2315 Audit Sampling
- 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.