Data considerations for Attributes Acceptance Sampling

To ensure that your results are valid, consider the following guidelines when you collect data, perform the analysis, and interpret your results.

The sample should be selected randomly
The units to be inspected should be selected randomly and should be representative of all the items in the lot. This may require extra effort, such as numbering each item and drawing random numbers or stratifying the lot and sampling from each strata or layer. However, this process is necessary for the effectiveness of the sampling process.
The data must be counts of defectives or number of defects
Each item that you inspect must be classified as either acceptable or not acceptable (defective), or you must be able to count the number of defects per item. If your data are measurements instead of counts of defects or defectives, you must create a variables sampling plan.
Individual lots should be homogeneous
Lots represent the entire population of units that the sample will be taken from. Lots should be homogeneous. They should be packaged and shipped in sizes that are well-managed by both the consumer and supplier, and in a way that allows easy selection of samples. Inspecting larger lots is usually more economical than inspecting a series of smaller lots.
The consumer and supplier should agree on target quality levels
The consumer and supplier should agree to the highest defective rate or defect rate that is acceptable (average quality level, AQL). The consumer and supplier should also agree to the highest defective rate or defect rate that the consumer is willing to tolerate in an individual lot (rejectable quality level, RQL).
The AQL describes what the sampling plan will accept, and the RQL describes what the sampling plan will reject. You want to design a sampling plan that accepts a particular lot of product at the AQL most of the time, and rejects a particular lot of product at the RQL most of the time.
Use the hypergeometric distribution for isolated lots of finite size
By default, Minitab uses the binomial distribution to create sampling plans and compare sampling plans for go/no go data. To correctly use the binomial distribution, Minitab assumes that the sample comes from a large lot (the lot size is at least ten times greater than the sample size) or from a stream of lots randomly selected from an ongoing process. Many of your sampling applications may satisfy this assumption.
If the lot of products that you sample from is an isolated lot of finite size, then the exact distribution for calculating the probability of acceptance is the hypergeometric distribution. For example, you receive one special order shipment of 500 labels.

The hypergeometric distribution is only available when you have go/no go (defectives) data and when you specify the lot size. Minitab uses the Poisson distribution when you count the number of defects.