How to convince customer with sampling check on lot quantity?

[YKT]

I have a question. A client had indicated that he wants to know the how their supplier determine the confidence level for the quantity.

The client want to know how sampling plan is conducted to ensure that number of quantity of products per pallet is within their specs.

I've asked the supplier to perform a random sampling, based on the AQL level. For instance, if the supplier ship 1000 pallets to the customers, and let say based on the certain AQL, the number of pallets to be inspected is 10 and acceptance level = 0,1.

Can I say that if the inspection shows that the 10 pallets randomly picked has the right quantity (as per the spec), the 1000 pallets can be statistically considered as having the right quantity per specifciation ?

 

[Tim Folkerts]

Yes, you can infer something about the 1000 pallets in the whole shipment, but with the numbers you give , you can't infer very much.

• The Z1.4 tables do not have 10 as a possible sample size, but if you go up to 13 for the sample size and have Ac=0, Re=1, then you have a sampling plan that is in the table. (It is actually the plan for AQL=1). Then the OC curves and tables in Z1.4 will tell you probabilities.
  You would accept lots with 0.39% (or less) defective 95% of the time, and reject lots with 20.6% (or more) 95% of the time. Those are not very tight limits on the quality!
• For normal inspection, Z1.4 gives a sample size of 125 for AQL = 0.1. With this plan, you would accept lots with 0.04% (or less) defective 95% of the time, and reject lots with 2.4% (or more) 95% of the time.
  Those are much tighter limits on defect rates. But obviously it is a great deal more inspection.

So if you are content being able to usually (95% of the time) catch defect rates of 20%, then a sample of 13 will do.

So if you are content being able to usually (95% of the time) catch defect rates of 2.4%, then a sample of 125 will be required.

If even that is not good enough you either need a bigger sample or a better way to prevent & detect defects.


Tim F

 

[bobdoering]

Sometimes I ponder if customers dwell more on statistical notions than reality. Unless you have some very conscientious inspectors, what would keep them from grabbing samples from the easiest skids to reach - rather than a true randomization? How about randomization of the boxes on a skid, and within the box (not just the product on top?). Perhaps the shipping and receiving folks randomized the skids when pulling on and off the truck, hard to say. I only offer this cynical interlude to ponder the statistical basis for sampling plans - random sampling - and how much time needs to be spent on the question of confidence level.

How random is random? Does quality level "TCE" suffice?

Many customers specify the plan they expect to be used - thereby specifying the risk they are willing to accept, rather than letting the supplier guess.