Sampling fundamentals

Audit sampling (ISA 530) means applying a procedure to fewer than 100% of the items in a population so that every sampling unit has a chance of selection, and using the results to conclude on the whole population. Picking only the largest items, or every item over a threshold, isn't sampling: it's selecting specific items, and the results can't be projected to the rest.

Sampling risk is the risk that the conclusion from the sample differs from the conclusion had the whole population been tested. In tests of controls it means concluding controls are more effective than they are (or less); in tests of details, concluding there's no material misstatement when there is (or the reverse). Non-sampling risk is everything else — the wrong procedure, misreading evidence, failing to recognise a deviation. Larger samples reduce sampling risk; only good audit work reduces non-sampling risk.

Sample size rises with a higher assessed risk of material misstatement, a lower tolerable misstatement (usually linked to performance materiality), a higher expected misstatement, and the more assurance the auditor wants from this test. It falls when other procedures already address the same assertion or controls are relied on. For large populations, the number of items in the population has little effect.

Selection methods: - Random — every item has an equal chance (random number generator). - Systematic — a constant interval from a random start: population ÷ sample size. - Monetary unit sampling (MUS) — every currency unit has an equal chance, so larger items are more likely to be picked; the interval is population value ÷ sample size. For a 2,450,000 population and 49 selections, every 50,000th currency unit is picked. - Haphazard — no conscious bias or pattern; acceptable for non-statistical sampling but not for statistical sampling. - Block — a run of consecutive items; rarely appropriate, because most populations are structured so items in a block are alike.

Projection. A misstatement found in a sample is projected to the population — commonly by the ratio method: sample misstatement × population value ÷ sample value. When a population is stratified, items tested 100% (key items) aren't projected: their misstatement is simply added. Only the sampled stratum is projected.

The projected misstatement is then compared with tolerable misstatement. If it is close to or above tolerable misstatement, the auditor can't conclude that the population isn't materially misstated — the allowance for sampling risk could push the true figure over. ISA 530 also allows a misstatement to be treated as an anomaly (not representative of the population) only in extremely rare cases, with a high degree of certainty supported by additional procedures.