From: Stan Hilliard
Date: 03 Jan 2000
Time: 23:00:50
Greetings Lloyd,
=== You said: ===== 1: If the above approach is acceptable how many results would I ideally need to return? Obviously 30 is not acceptable in light of what I have read in other literature?
2: Is the above approach acceptable when considering the multitude of sampling techniques available? === end of quote =====
The central limit theorem is not the right criterion to determine the sample size. Anyhow, the distribution of sample averages reaches normally at way less than n=30. It's more like n=5.
The sample size should be related to how precise you want the sample estimate to be. This will depend on the intended use of the sample estimate. What is the characteristic that you plan to measure?
The approach depends on how you plan to use the sample estimate.
If you plan to make an accept/reject decision then you should use the acceptance sampling approach based on the operating characteristic curve. See:
www.samplingplans.com/usingoccurves.htm
If you don't make an accept/reject decision, but require that the margin of error of the sample estimate be sufficiently small, then the sample size should be large enough to produce a margin of error of the required size.
Another factor that will make a large difference in the sample size is the type of data. Is the characteristic a measured variable or is it attributes data?
Stan Hilliard