Sometimes it is necessary to estimate the mean and/or percentile of a certain population by limited sampling, in order to compare with target values. Done properly, the method should account for random chance in the sampling so that the two risks can be considered:
- The risk that the check is failed, even though the population meets the required value. This is known as the “producer’s risk” or “manufacturer’s risk”.
- The risk that the check is passed, even though the population is an unacceptable difference from the required value. This is known as the “consumer’s risk”.
One example is the “acceptance procedure for verification of a lot” found in the standard EN 14358. This procedure is intended to be used when checking a certain population of timber (of initially unknown mean, variation and distribution) against requirements that are defined by only a mean or percentile value. This is also the procedure used for the grading machine installation check according to EN 14081-2, to evaluate if a grading machine is passing timber with the required grade determining properties for the strength class being produced. Unfortunately, the procedure is confusing, with the standard giving incomplete information. This paper provides some explanation, suggests some improvements, and gives some examples with model and real data.