What it means
A company often cannot inspect every unit, so it chooses a sample and uses the results to decide whether to accept or investigate a larger lot. That decision relies on how the units were selected: if an inspector checks only boxes at the front of a pallet, goods at the centre or rear may have different damage or production history.
Bias can be deliberate or accidental, since a supplier might present its best items first, a busy receiving team might choose the nearest cartons and a manager might sample only the morning shift because it is convenient. Even a large sample can be misleading if all units come from one unrepresentative part of the process.
The sampling plan should define the population, selection method and risk. Random selection can help for a broadly uniform batch, while stratified selection may be needed across shifts, machines, suppliers, dates or pallet locations.
Safety-critical defects can require a different or more complete control, and a glossary cannot prescribe one acceptance number for every product. Record where sampled units came from.
If three of 50 units fail, the raw 6% result tells a manager something about those units, but it does not by itself prove an exact defect rate across all 10,000 units, and a biased sample weakens the inference further. Report uncertainty and investigate patterns before extrapolating costs or releasing goods.
Good sampling does not replace process control. If a machine alarmed during one production hour, sample deliberately from the affected time and hold those units while investigating, which is a targeted check, not a random estimate of the whole run, so label it accordingly.
For managers, the practical question is which units were looked at and whether the choice of units could have hidden the problem. A polished pass-rate dashboard cannot answer that without a sound selection method.
In practice
Real-world examples.
Example
An inspector opens only top-layer cartons on a pallet. Damage caused by poor strapping in the middle layers is missed, so the sample overstates the shipment's condition.
Example
A factory takes ten units from each of three shifts instead of 30 from the daytime shift alone, allowing it to see whether problems cluster overnight.
Example
After a machine fault at 14:00, quality checks the affected production window specifically. It does not call that targeted test a random sample of the whole day's output.
Formula
Calculation
Observed sample defect share = Defective units in inspected sample / Units inspected x 100
Coverage by stratum = Units sampled in a defined group / Units available in that group x 100
Worked example. A factory makes 10,000 units across two shifts. An inspector checks 100 units from the day shift and finds two defects.
- Observed day-shift sample defect share = 2 / 100 x 100 = 2%.
- The night shift has not been sampled, so the business cannot claim the entire day's production has a 2% defect rate from this evidence.
The next step depends on the risk and agreed sampling plan, not on rounding the 2% into certainty.Case study
Seen in the real world.
This illustrative and entirely fictional example follows Cedar Packaging, an invented carton maker. Its incoming material inspection showed consistently low defects. Yet the production line complained about weak sheets. A supervisor watched the receiving process and saw that inspectors pulled material from the top of each stack, where sheets were easiest to reach and usually undamaged. The quality team reviewed the lot structure and changed its selection method to include locations across a delivery.
It also recorded which supplier and production shift made the sheets. The next review found that defects clustered in one part of particular stacks. Cedar held the affected material and discussed the cause with the supplier. The new method did not magically make quality worse. It made the information more honest, allowing Cedar to fix a problem it had previously failed to see.
Watch out
Common mistakes.
- Assuming a bigger sample is unbiased when every unit was chosen from the same convenient location or time.
- Treating a targeted investigation sample as a random estimate of the entire batch. Both can be useful, but answer different questions.
- Reporting a precise batch defect rate without describing the population and selection method. A percentage without context can mislead acceptance and cost decisions.
Questions
People also ask.
Is random sampling always best?
It helps when the population is suitable for it, but known differences between shifts, machines or pallet layers may require planned coverage of those groups. Use a method that fits the risk.
Can a passing sample prove there are no defects?
No. Uninspected units can still have problems, and selection bias can hide them. The conclusion depends on the sampling plan and other process evidence.
What should a manager ask before accepting a result?
Ask which batch was sampled, how units were selected, where and when they came from, what was tested and what the agreed acceptance rule says.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%