From: Stan Hilliard
Date: 08 Jul 1999
Time: 12:11:40
Lisa, When you mention predicting future samples, do you mean estimating how much change will occur over time? If so, I think that you would need to have data taken over time, and perform a time series analysis.
Would you settle for an interval estimate of the population fraction "defective" as of the time that the sample was taken (based on n=20). The binomial distribution will provide one-sided confidence limits (lower or upper) or two sided limits.
For example, with a sample size of 20 and 2 defectives, the sample is 10% defective, but the population could be as high as 28% defective. The 28% is the one sided upper 95% binomial confidence limit for the fraction defective of the population.
In your case, you could convert that into estimated number of items in your database:
Point estimate: 0.10 X 6523 = 652 defective items.
Upper confidence limit: 0.28 X 6523 = 1825 defective items.
I calculated the confidence limits with software program TP105, choosing "Kind of Input = Sample", and entering n=20, X=2. See www.samplingplans.com/programtp105.htm
Sincerely, Stan Hilliard