Do You have a standard operating procedure to define Your sample size?
Sampling can be costly and testing time-consuming, but failing products on the market may be even more expensive.
To balance this out, You need to decide on a sample size. Often this decision is predefined in a corporate standard operating procedure stating that “a sample size of 30 is required for capability studies” or a similar statement. Sometimes there is even a statistical rationale behind this statement.
Douglas Adams wrote that the answer to the question about life, death, the universe and everything is 42; that may be a proper sample size then?
No!
To quote Paul Mathews: “There is not one magic sample size for all situations”.
It is important to consider what question You need answered and how accurate that answer must be. This should be done based on proper risk assessments and by applying statistical tools.
Always consider if You have historical data that can support the decision. If You have a robust and reliable process it may be possible to inspect or monitor the process using a small sample, while a more uncertain situation requires a large sample.
Always consider if You need a “simple” answer or a detailed analysis. Do You need to check that a product is somewhere within specifications or do You need to know exactly where it is?
Always consider the risk. Is a hazard present? Is it likely to cause a hazardous situation? What kind of harm could it do and what is the risk that this will happen?
Remember that large variation and/or high accuracy calls for larger samples.
When You apply statistical methods, think about the confidence level You need and consider if it could be appropriate to use a smaller sample but then apply the limits of the confidence interval rather than the point estimate. (E.g. if the average is 10 and there is a 90% confidence the true mean is 8-12, then compare 8 and 12 with the specifications rather than 10)
Consider the possibility of repeating the test to reduce sampling and/or increase confidence
Good luck with Your sampling
There will be more for the curious readers later on, so stay tuned