What Is Variable Sampling? Estimating Monetary Misstatement the Audit Way

Variable sampling is where audit meets estimation theory. Where attribute sampling asks a yes/no question (‘did the control operate?’), variable sampling asks a how-much question: by how many dirhams is this balance misstated? The auditor draws a sample, measures the monetary error in each item (book value minus audited value), and projects that error across the population using methods matched to the population's character.

The dominant method in modern practice is monetary unit sampling (MUS) — also called probability-proportional-to-size. By defining the sampling unit as the individual currency unit, MUS automatically weights selection toward large balances: the AED 2 million key customer is virtually certain to be tested, the AED 300 staff advance almost certainly not. This mirrors audit risk perfectly — big balances carry big misstatement risk — and it simplifies projection mathematics.

The classical variables methods — mean-per-unit, difference estimation, ratio estimation — treat the logical item (the invoice, the balance) as the sampling unit and use the sample's average error or error ratio to project. They shine where misstatements are numerous and small, or where MUS's assumptions break (many credit balances, expected understatement). ISA 530 does not mandate a method; it mandates the discipline — proper design, genuine randomisation, mathematically sound projection, and evaluation against tolerable misstatement with an explicit allowance for sampling risk.

How variable sampling works, end to end, on a UAE receivables audit: Step 1 – define the population. All 2,800 customer balances totalling AED 31,000,000 at 31 December — completeness checked (all branches, all divisions, credit balances segregated for separate treatment).

Step 2 – set parameters. Tolerable misstatement AED 700,000 (below overall materiality, leaving room for other undetected errors); expected misstatement AED 80,000 based on prior-year findings; 95% confidence (5% sampling risk). The software computes an MUS sample of 95 items.

Step 3 – select and test. Random selection via the tool — no judgemental cherry-picking. Each selected balance is confirmed or alternatively vouched (subsequent receipts, shipping documents). Errors are measured in dirhams: a AED 120,000 balance confirmed at AED 105,000 is a AED 15,000 overstatement.

Step 4 – project and evaluate. Suppose total sample misstatement is AED 42,000. MUS projection scales this to roughly AED 380,000 for the population; the upper error limit (projection plus sampling-risk allowance) comes to AED 590,000 — under the AED 700,000 tolerable threshold. Conclusion: receivables are fairly stated. Every number in that chain — population, parameters, projection, allowance — is documented, because the review partner and any inspector will re-perform the logic.

In the UAE, variable sampling meets populations with local character. Re-export traders carry receivables spread across dozens of countries with wide value ranges — textbook MUS populations, but auditors must confirm the population file captures every sales ledger (mainland plus free zone entities sometimes run separate systems that do not reconcile). Contracting companies have retentions and work-in-progress balances where classical difference estimation can outperform MUS, because errors tend to be systematic percentage-type mismeasurements rather than isolated overstatements.

Credit balances need care: UAE distributors often carry customer credit balances (advance payments) inside receivables — MUS populations should exclude or separately stratify them, since sampling for overstatement in a balance that is really a liability produces nonsense. Related-party balances are typically removed from the statistical population and tested 100% — their risk profile (and ISA 550 requirements) demands individual attention, not sampling.

A growing UAE trend is stratification by risk: high-value and related-party items tested fully, the remaining ‘clean’ population sampled with MUS. This hybrid — 100% of what matters individually, statistics for the rest — is what ISA 530's framework intends, and it focuses partner attention where judgement actually lives.

Worked example: payables of AED 18,500,000 across 4,100 supplier balances at Noor Retail LLC. The auditor's concern is understatement (unrecorded liabilities) — so MUS, designed for overstatement, is the wrong tool; the team uses difference estimation on a random sample of 80 balances, testing each by vouching to supplier statements and subsequent payments.

Results: the sample's book values total AED 1,900,000; audited values total AED 1,862,000 — an average understatement of AED 475 per item. Projected to 4,100 items: AED 1,947,500 of population understatement. The upper limit including sampling risk: AED 2,600,000 — against tolerable misstatement of AED 800,000. The population fails.

Response: investigation reveals a systematic cause — December supplier invoices from one logistics provider (about AED 2,200,000) were never entered; the AP clerk was on leave and no cover was arranged. Management books the full AED 2,200,000; the auditor re-performs the projection on the corrected population and it passes. The example shows variable sampling's real power: not the elegant maths, but catching a AED 2.2 million understatement that judgemental testing of ‘large suppliers’ had missed for years — because the missing invoices were, by definition, not on anyone's list.

Using MUS for understatement. MUS is built for overstatement; testing completeness with it understates the risk — choose difference/ratio estimation or directional substantive procedures instead.

Leaving credit balances in the population. Customer advances inside receivables distort MUS selection and projection — stratify them out.

Sampling related parties statistically. Connected balances need 100% ISA 550 attention, not a random sample.

Incomplete populations. Missing branches, missing months, missing free zone ledgers — projection from a partial file is fiction.

Ignoring the allowance for sampling risk. Comparing raw projection (not the upper error limit) to tolerable misstatement overstates comfort.

No documentation of parameters. An undocumented sample size is indefensible — record tolerable misstatement, confidence and method.

Choosing classical methods for low-error populations. When few errors are expected, attribute-style MUS is more efficient — classical variables methods earn their keep in error-prone populations.

Forgetting the normality assumption. Classical methods assume roughly normal error distributions — skewed populations (a few huge errors) break the maths; stratify first.

Mixing methods in one evaluation. Combining MUS and classical projections without a coherent framework — pick the method that fits the population, or stratify and treat each stratum on its own terms.

Statistical sampling is the ISA 530 umbrella variable sampling sits under. Attribute sampling is its controls-testing sibling. Monetary unit sampling (MUS) is the dominant variables method in UAE practice. Sampling risk and tolerable misstatement are the two parameters every projection is judged against. Difference estimation is the classical alternative for understatement and systematic errors. And non-statistical sampling covers the judgemental selection of key and suspicious items that statistics should never replace. Together they let auditors give quantified comfort over balances no human could test in full.

Mean-per-unit estimation projects the sample's average error across the population — the simplest classical method. Ratio estimation scales book values by the sample's error ratio — efficient when errors correlate with balance size. Stratified variables sampling combines both worlds: homogeneous strata with classical projection inside each. And the standard error is the precision measure that turns a point projection into a defensible confidence interval.