From: Stan Hilliard
Date: 21 May 2002
Time: 16:55:07
Yoann,
What is in the bottles -- liquid, powder, capsules, etc. What are the characteristics/variables to be tested? What is lyophylisation?
What is the approximate number of bottles per lot? Are the bottles grouped in any way, like in boxes?
Do the bottles all come from one homogenous lot? Can there be real bottle to bottle differences? Are your lots made from more that one batch, so that there would be batch to batch differences within a lot?
The basic statistical concept that you need to understand for the factors that effect the sample size is the operating characteristic curve (oc curve). See
www.samplingplans.com/usingoccurves.htm
I suspect that you are testing the properties of a homogenous fluid rather than individual parts. In that case, you would use a sampling plan for the mean of each characteristic tested. For examples of sampling plans for the mean developed by software program TP414, see
www.samplingplans.com/outputvariablesmean.htm
Here, the statistical sample size is not the quantity of material but is the number of measurements to take. The quantity of material that you take per bottle depends on the amount of composite material consumed by testing and retaining.
In a sampling plan for the mean, you make the accept/reject decisions by comparing the average of the measurements to decision limits. The decision limit(s) and the sample size are the result of designing the sampling plan. The software program TP414 does the calculations for the sampling plan and the oc curve.
There is a tutorial on the general concepts of sampling for variables at
www.samplingplans.com/modern3.htm
Once you have determined a sample size in terms of the number of measurements, you need to determine how to construct the composite blend. Assuming that there is no bottle to bottle variability, or that bottle to bottle differences are unimportant to quality, the number of bottles that you need to sample to make the composite blend is a matter of engineering judgement.
On the other hand, if there could be bottle to bottle variability, and if this effects the quality, maybe you would test individual unblended bottles. Can you describe more about this for your case?
It would be useful to determine the standard deviation that represents between-bottle variability and also the standard deviation that represents the testing the same bottle repeatedly.
Stan Hilliard