What is variables sampling?
Ask a senior to pick a sampling approach for a completeness test and you will almost always hear "just use MUS." The problem is that MUS is structurally biased toward detecting overstatement because it weights selection by recorded value; items recorded at zero have zero chance of selection. Variables sampling exists precisely for the situations where MUS falls short, and knowing when to reach for it is one of the quieter skill gaps on many audit teams.
Variables sampling treats each physical item (invoice, balance, transaction) as the sampling unit, not each monetary unit. The auditor selects items, determines their audited value, and uses one of four estimation approaches to extrapolate the result to the full population: mean-per-unit, difference estimation, ratio estimation, or a stratified combination of these. Mean-per-unit calculates the average audited value of sampled items and multiplies by the population count to estimate the total audited value. Difference estimation calculates the average difference between recorded and audited values and multiplies by the count. Ratio estimation computes the ratio of total audited values to total recorded values in the sample and applies that ratio to the population's recorded total. In our experience, difference estimation is the workhorse for most receivables and payables populations because differences between recorded and audited values tend to be small and consistent.
Variables sampling requires an estimate of the population's standard deviation to calculate sample size. Highly skewed populations (a few very large items and many small ones) produce large standard deviations, driving sample sizes up to impractical levels. Stratification (dividing the population into more homogeneous subgroups) reduces within-stratum variability and keeps sample sizes manageable. Unlike MUS, variables sampling works for both overstatement and understatement because every physical item has equal selection probability regardless of its recorded value.
Key Points
- Estimates monetary amount, not rate. Variables sampling produces a projected monetary misstatement or total audited value, making it the counterpart to attribute sampling (which estimates deviation rates).
- Works for both over- and understatement. Because each physical item has equal selection probability, variables sampling is appropriate for completeness and existence assertions alike.
- Needs population standard deviation estimate. Sample size calculations require a standard deviation input, typically obtained from prior-year working papers (WPs) or a pilot sample.
- Stratification is usually necessary for skewed populations. ISA 530 .A9 identifies stratification as a means to improve audit efficiency by reducing variability within each stratum.
Why it matters in practice
The most common practical error is skipping stratification when the population is heavily skewed. Nobody enjoys the extra setup, but skipping it is how sample sizes blow out to 100+ items and the whole approach gets abandoned in favour of PIOOMA ("pulled it out of my arse") sample sizes that have no statistical basis. Without stratification, the standard deviation of the full population drives the sample size formula and the resulting number feels impractical. Rather than increasing the sample, teams reduce it judgmentally, which defeats the purpose of statistical sampling entirely. ISA 530 .A9 specifically identifies stratification as the solution: by grouping similar-sized items together, each stratum has lower variability and therefore requires a smaller sample.
A second issue is confusing the estimation methods. Mean-per-unit works best when there is little correlation between recorded and audited values (for example, when many items have significant differences). Difference estimation is more efficient when differences are small and consistent. Ratio estimation works well when misstatements are proportional to recorded values. Applying the wrong method does not invalidate the sample, but it can produce unnecessarily wide confidence intervals or sample sizes large enough that someone on the team will suggest you "just roll it forward" from last year. This doesn't mean the method choice is arbitrary. It means you need to look at the population before picking, not after.
Key standard references
- ISA 530.5 -15 sets out the core requirements for audit sampling: design, selection, evaluation of results, and documentation of the sampling approach.
- ISA 530 .A9 identifies stratification as a means to improve efficiency and reduce variability in sampling.
- ISA 530 .A10-A15 provides application guidance on statistical and non-statistical sampling approaches, including the variables sampling methods discussed above.
Related terms
Related tools
Related reading
Frequently asked questions
What are the three estimation methods in variables sampling?
Mean-per-unit (audits each item, calculates average, multiplies by population count), difference (calculates average difference between recorded and audited values, multiplies by count), and ratio (computes ratio of audited to recorded values, applies to population). The choice depends on population characteristics.
When should variables sampling be used instead of MUS?
When the risk includes understatement (completeness), when both over- and understatement are concerns, or when the population has many zero-value or low-value items. Variables sampling gives every item equal selection probability, making it appropriate where MUS is structurally unsuitable.
Why is stratification usually necessary?
If the population is highly skewed (few large items, many small), the standard deviation is large, driving up sample size. Stratification reduces within-stratum variability, keeping sample sizes practical. ISA 530.A9 specifically points to stratification as a means of improving efficiency.