When Ms. Iyer is auditing a client with 12,000 sales invoices, checking every single one is impractical. So she picks a sample — a carefully chosen subset — and draws conclusions about all 12,000 based on what she finds in that sample. This is audit sampling, governed by SA 530 (Audit Sampling) under the ICAI curriculum. The core idea: if your sample is representative, your conclusion about the full population is reliable.
Sampling falls into two families. Statistical sampling uses probability theory — every item has a known, non-zero chance of selection, and the auditor can mathematically measure sampling risk (the risk that the sample conclusion differs from what a 100% check would reveal). Non-statistical sampling relies on professional judgement — no mathematical projection is made. Both are permitted under SA 530, but statistical sampling is more defensible and objective. Within these families, four specific methods matter for your exam:
Random sampling — every item has an equal chance; uses random number tables or computer tools. Most unbiased. Systematic (Interval) sampling — select a random start point, then pick every nth item. If the population is 6,000 and sample size is 100, the interval is 60. Fast and practical, but dangerous if the population has a repeating pattern matching your interval. Haphazard sampling — the auditor selects items without a structured method but also without conscious bias. This is not the same as random sampling — there's no equal probability of selection. Used only in non-statistical contexts. Stratified sampling — split the population into layers (strata), e.g., invoices above ₹10 lakhs and below ₹10 lakhs, then sample each layer separately. High-value items get heavier coverage. Extremely common in practice.
Two risks always accompany sampling. Sampling risk — the gap between your sample conclusion and the true population position — decreases as sample size increases. Non-sampling risk — errors from applying the wrong procedure or misreading a document — is unaffected by sample size; it's controlled through training and supervision. Sample size itself goes up when: tolerable misstatement is low, expected misstatement is high, confidence level required is high, or population size is large. These four drivers are a favourite MCQ trigger in Paper 5.