Mean-per-unit estimation is a classical variables sampling method in which an auditor calculates the average audited value of a randomly selected sample, then multiplies that average by the total number of items in the population to estimate the full account balance. A confidence interval built around that estimate lets the auditor test whether the client’s recorded book value is materially misstated. Unlike difference or ratio estimation, mean-per-unit (MPU) works only with the audited values themselves and ignores book values entirely, which is both its defining trait and the reason it usually requires larger samples than the alternatives.
When Auditors Reach for MPU
MPU earns its place when reliable recorded book values are not available for every item in the population. That includes first-time valuations, populations where book values exist but correlate weakly with audited amounts, and situations where the ledger simply cannot be trusted at the item level. Where per-item book values are dependable, difference or ratio estimation almost always produces a tighter answer with fewer items tested, because those methods exploit the correlation between book and audited amounts to shrink variability. MPU has no such lever, so it leans harder on sample size and, as covered below, on stratification.
What You Need Before You Calculate
Three inputs have to be in hand before any arithmetic begins.
The population size, N, is the total count of discrete items in the account: every receivable invoice, every inventory line, every disbursement. The count comes from the client’s ledger or a physical listing and should be verified before sampling.
The sample size, n, is how many items you will examine directly. Items must be selected randomly, because MPU’s statistical validity depends on every item having an equal chance of being chosen.1Public Company Accounting Oversight Board. AS 2315: Audit Sampling Any nonrandom selection introduces bias that invalidates the projection.
For each selected item you then obtain an audited value: the verified amount supported by evidence such as shipping documents, vendor invoices, bank statements, or confirmation letters. If a receivable shows a book value of $500 but the customer confirms $450, the audited value is $450. Those verified figures feed every calculation that follows.
Calculating the Point Estimate
Add up the audited values of the sampled items. Divide the total by the sample size to get the sample mean:
Sample mean = Sum of audited values ÷ n
If you tested 200 inventory items and their verified values total $40,000, the sample mean is $200 per unit. That figure is your best estimate of what a single item in the population is worth, based on the evidence you gathered.
Multiply the sample mean by the population size to project the total:
Point estimate = Sample mean × N
With a population of 5,000 items and a sample mean of $200, the point estimate for the full account is $1,000,000. That number is the baseline you will compare against the client’s recorded balance.
Building the Confidence Interval
A point estimate on its own is not enough, because a sample never mirrors the full population perfectly. The allowance for sampling risk, sometimes called achieved precision, quantifies how far off the estimate might be and creates a range around it. Auditing standards require this risk to be assessed when drawing conclusions from sample-based evidence.1Public Company Accounting Oversight Board. AS 2315: Audit Sampling
The formula is:
Allowance = N × z × (s ÷ √n)
N is the population size, z is the z-score for the chosen confidence level (1.96 for 95% confidence), s is the standard deviation of the audited values in the sample, and n is the sample size. Standard deviation measures how spread out the sampled values are. More variation produces a wider interval and less certainty about the estimate.
If the point estimate is $1,000,000 and the calculated allowance is $50,000, you can say with 95% confidence that the true population value lies between $950,000 and $1,050,000. When the client’s recorded book value falls inside that interval, you have statistical support that the account is not materially misstated. When it falls outside, you need to expand the sample, perform alternative procedures, or request an adjustment.
Finite Population Correction
When the sample exceeds roughly 5% of the population, a finite population correction factor tightens the interval:
FPC = √((N − n) ÷ N)
With N of 1,000 and n of 200, the FPC is √(800 ÷ 1,000) ≈ 0.894, so the adjusted allowance is about 89% of the uncorrected figure. In most large-population audits the sample is well under 5% of the total, and the correction is skipped without meaningful effect.
What Drives Sample Size
Four inputs determine how many items you need to test. Miscalibrating any of them produces a sample that is either wastefully large or too small to support a conclusion.
- Confidence level. Higher confidence demands more items. Moving from 90% (z = 1.645) to 95% (z = 1.96) increases the required sample substantially.
- Tolerable misstatement. This is the largest dollar error you can accept without concluding the account is materially misstated, and it must always be set below overall materiality for the financial statements. A tighter tolerable misstatement requires a larger sample. If circumstances change during the engagement, reassess it.2Public Company Accounting Oversight Board. AS 2105: Consideration of Materiality in Planning and Performing an Audit
- Expected variability. Populations whose items differ widely in value require larger samples. Receivables ranging from $10 to $500,000 need far more testing than receivables ranging from $90 to $110.
- Expected misstatement. When prior-year results or known control weaknesses suggest meaningful errors, you need more items to separate real misstatement from sampling noise.
Stratification: The Main Efficiency Lever
High variability is MPU’s biggest practical weakness, and stratification is the standard fix. Divide the population into subgroups of similar items and sample each subgroup separately.1Public Company Accounting Oversight Board. AS 2315: Audit Sampling Common bases include recorded dollar value, the internal controls that processed the items, and specific risk characteristics. A receivables population might split into balances over $50,000, balances between $10,000 and $50,000, and balances under $10,000.
Because items within a stratum resemble each other more than they resemble items in other strata, the standard deviation inside each group drops and overall precision improves without adding to the total sample size. Project the misstatement results for each stratum independently and then sum them for a population-wide conclusion. A well-stratified MPU application can hit the same precision as an unstratified one with far fewer items tested. The catch: the variable used to build the strata must actually correlate with the values being measured. Stratifying by an unrelated characteristic provides no benefit.
MPU vs. Difference and Ratio Estimation
MPU is one of three classical variables sampling approaches. Picking the wrong one can mean testing hundreds of extra items.
- Difference estimation calculates the average dollar difference between audited and book values in the sample and projects it across the population. It fits when errors are roughly constant in dollar amount regardless of item size.
- Ratio estimation calculates the ratio of total audited values to total book values in the sample and applies it to the population’s total book value. It fits when errors are proportional to item size.
- MPU ignores book values entirely and projects total value from the average audited value alone.
Difference and ratio estimation almost always require smaller samples than MPU at the same precision, because they take advantage of the correlation between book and audited amounts. Reach for MPU when book values are unreliable or unavailable across the population, when the account is being valued for the first time, or when the book-to-audited correlation is too weak for the other two methods to help.
Evaluating the Results
Compare the confidence interval to the recorded book value and the projected misstatement to the tolerable threshold. If the book value sits inside the interval and the projected misstatement is under tolerable, you have reasonable support that the account is not materially misstated. If not, expand the sample, apply alternative procedures, or ask management to adjust.
Dollar totals do not settle the question by themselves. Auditing standards require you to weigh qualitative factors alongside the numbers.3Public Company Accounting Oversight Board. AS 2810: Evaluating Audit Results A numerically small error can still be material if it turns a reported loss into income, affects loan covenant compliance, increases management compensation, or points to intentional manipulation. An error that looks trivial in one period may also be material if it reflects a pattern likely to compound in future periods.
The cause of any error found matters as much as its size. A misstatement traceable to a systematic processing flaw suggests the untested portion of the population may carry a higher error rate than the sample shows. A misstatement caused by an isolated one-time event is less troubling from a projection standpoint, though it may still call for disclosure depending on the circumstances.3Public Company Accounting Oversight Board. AS 2810: Evaluating Audit Results