Monetary unit sampling is a statistical audit technique that treats every individual dollar in an account balance as its own sampling unit, so an item’s chance of being selected for testing is proportional to its recorded value. Auditors reach for it during substantive testing when the concern is that a balance is overstated, and it concentrates testing effort on the dollars most capable of producing a material error while still giving every dollar in the population a chance of selection.
When to Use It, and When Not To
The method fits accounts where the auditor expects the balance to be fairly stated or to contain only a small number of errors. Accounts receivable, inventory, and loan portfolios are the most common targets because these balances carry inherent risk that recorded values exceed what actually exists. Revenue accounts get the same treatment for the same reason: the worry is that transactions were recorded that did not occur, or were recorded at inflated amounts.
Weighting by dollar value is what makes the technique efficient. A single receivable worth $500,000 is five hundred times more likely to be selected than a $1,000 invoice. That built-in weighting means the auditor does not need to stratify the population into size categories the way classical sampling requires, and any item recorded above the sampling interval is picked up automatically.
The blind spot is understatement. Because selection probability depends on recorded book value, an item recorded at zero has zero chance of being selected. If the audit concern is understated liabilities or expenses, or assets missing from the ledger entirely, monetary unit sampling will not find the problem. Classical variables sampling handles those situations better because it accommodates zero and negative balances.
Negative balances create their own problem. Credit memos in a receivables population usually have to be segregated and tested under separate procedures, or removed from the population entirely and covered another way.
Calculating the Sample Size
Four inputs drive the calculation: the recorded population value, tolerable misstatement, expected misstatement, and the desired confidence level.
The recorded population value comes straight from the trial balance or general ledger. Tolerable misstatement is the maximum dollar error the auditor can accept in the account without concluding it is materially misstated; it is the application of performance materiality to the specific sampling procedure, and it may equal performance materiality or be set lower when the sampling population is smaller than the full account balance.
Expected misstatement is the auditor’s judgment about how much error probably exists, informed by prior-year results, known control weaknesses, or preliminary analytical work. Confidence level, typically 90% or 95%, sets how much assurance is needed that a material misstatement would be detected. A 95% confidence level accepts a 5% risk of concluding a balance is fairly stated when it is not.
Reliability Factors
The confidence level translates into a reliability factor. With zero misstatements expected:
- 95% confidence (5% risk of incorrect acceptance): 3.00
- 90% confidence (10% risk of incorrect acceptance): 2.31
If the auditor expects to find errors, the factor rises. At 95% confidence, one expected misstatement pushes the factor to 4.75 and two push it to 6.30. At 90%, the corresponding factors are 3.89 and 5.33. Higher factors produce larger samples, which is the point: the more error you expect, the more items you need to test.
The Formula
Sample size equals the population value multiplied by the reliability factor, divided by tolerable misstatement. Take a $10,000,000 inventory balance, tolerable misstatement of $500,000, 95% confidence, and zero expected errors:
Sample size = ($10,000,000 × 3.00) ÷ $500,000 = 60 items.
The sampling interval is the population divided by the sample size: $10,000,000 ÷ 60 = $166,667. Every item recorded above $166,667 spans a full interval and will be selected automatically. Those large items get tested individually, and any error found in them enters the evaluation at face value rather than being projected.
Selecting the Items
Selection begins with a random number between 1 and the sampling interval. If the interval is $166,667 and the random start is $45,200, the first selected dollar is the 45,200th dollar in the population. The next is dollar 211,867, then 378,534, and so on through the entire population.
Each selected dollar identifies a logical unit: the invoice, customer account, or inventory record that contains it. The logical unit is what the auditor actually tests. If two selected dollars fall inside the same large invoice, the invoice is tested once but carries the weight of both selections when the results are evaluated. That is a real difference from classical sampling, where each physical item is either in or out of the sample.
Because selections are systematic, the sample spreads evenly across the account’s full monetary value; no cluster of mid-range transactions gets missed because the random start happened to sit elsewhere. Audit software handles the mechanics: computing the interval, generating the start, mapping dollars to logical units, and flagging every item over the interval for 100% testing.
Evaluating Errors and Projecting to the Population
After testing each selected item through confirmation, recalculation, inspection, or whatever procedure fits the assertion, the auditor quantifies any errors and projects them to the full population. How the projection works depends on whether the item was pulled in because it exceeded the sampling interval or because it fell in the routine selection.
Items Above the Sampling Interval
These were tested individually, so their errors enter at face value with no projection. If a $400,000 receivable is actually worth $370,000, the known misstatement is $30,000 and that figure goes directly into the total.
Items Within the Interval
For each of these, the auditor computes a tainting percentage: the misstatement divided by the recorded book value of the logical unit. An invoice recorded at $5,000 that is actually worth $4,200 has a tainting percentage of ($5,000 − $4,200) ÷ $5,000 = 16%. That percentage is applied to the sampling interval to project the likely error across the slice of the population the item represents. With an interval of $166,667, the projected misstatement for that error is $26,667.
Every error found in the routine sample gets its own projected amount. The auditor ranks them from highest tainting percentage to lowest and applies incremental reliability factors from the confidence factor table to each successive layer. The first error uses the increment between zero and one misstatement (1.75 at 95% confidence), the second uses the increment between one and two (1.55), and so on. The layered result is the upper limit of misstatement, which combines projected errors with an allowance for sampling risk on the untested portion of the population.
The Decision
Compare the upper limit of misstatement against the tolerable misstatement set during planning. If the upper limit falls below tolerable misstatement, the sample provides sufficient evidence that the balance is not materially misstated. If it exceeds tolerable misstatement, the auditor cannot accept the recorded balance as stated.
When the Upper Limit Exceeds Tolerable Misstatement
This is where the hard work of audit sampling actually lives. When the projection says the account may be materially misstated, the auditor has several options and none of them are painless:
- Expand the sample. Testing additional items can shrink the allowance for sampling risk enough to bring the upper limit back below tolerable misstatement.
- Request client adjustments. Ask management to investigate and correct the specific errors identified, then reassess whether the remaining projected misstatement is acceptable.
- Perform alternative procedures. Test the full population using a different method, or apply targeted procedures to the subpopulation where errors were concentrated.
- Modify the audit opinion. If the client refuses to adjust and further testing still indicates material misstatement, the auditor issues a qualified or adverse opinion.
One recurring trap is the treatment of anomalous errors. If an error looks like a one-off event unrelated to the rest of the population, the auditor may conclude it is an anomaly and exclude it from projection. Firm policies vary widely; some prohibit the practice, others allow it routinely. Labeling a systemic error as anomalous can lead directly to accepting a materially misstated balance.
The Qualitative Side of Evaluation
The numerical projection is only half of what the auditor owes on evaluation. Every misstatement also has to be assessed for what it is, not just how big it is. A $2,000 overstatement from a data entry mistake carries different implications than a $2,000 overstatement from someone deliberately inflating a receivable before quarter-end.
When a misstatement appears intentional, it cannot be treated as an isolated event. The auditor has to evaluate whether it indicates broader fraud, reassess the risk of material misstatement across the engagement, reconsider whether internal controls are actually working, and decide whether evidence already gathered elsewhere in the audit remains reliable. If the effect could be material or the scope cannot be readily determined, additional procedures are required to figure out whether fraud has occurred or is likely to have occurred.1Public Company Accounting Oversight Board (PCAOB). AS 2810: Evaluating Audit Results
Possible collusion has to be considered as well. If the misstatement pattern suggests coordination among employees, management, or outside parties, the reliability of other audit evidence may be compromised, and the audit approach may need to be rethought beyond the single account tested through the sample.
Which Standards Apply
Two standards govern audit sampling depending on the type of client.
Private Companies: AICPA AU-C Section 530
For non-issuers, the AICPA’s clarified auditing standard AU-C Section 530 provides the framework. It defines audit sampling, establishes that tolerable misstatement is the application of performance materiality to a specific sampling procedure, and requires the auditor to project sample results to the population. It applies equally to statistical and nonstatistical sampling and leaves the choice of method to professional judgment.
Public Companies: PCAOB AS 2315
For issuers, PCAOB Auditing Standard 2315 governs. It defines audit sampling as applying a procedure to less than 100% of the items in an account balance to evaluate a characteristic of that balance, and it recognizes “sampling with probability proportional to size” as a valid random-based selection method. That is the technical name for what practitioners call monetary unit sampling.2Public Company Accounting Oversight Board (PCAOB). AS 2315: Audit Sampling
Under both frameworks the auditor must plan the sample with specific reference to tolerable misstatement, expected misstatement, and the allowable risk of incorrect acceptance, and must project sample results and compare them against tolerable misstatement. Neither standard dictates the mechanics of monetary unit sampling itself; they set the objectives and leave the selection of the statistical method to the auditor.