Inspection sample size help

[rkremser]

So I have a box with 2000 lets call them marbles. I need to determine how many marbles I need to sample to make sure i get at least one from each of 4 500 marble subsets. How would the number change if i only needed a 95% confidence interval for getting one from each subset.

Thoughts?

 

[rkremser]

So its 1501 to 100% ensure you have one of each. But I keep coming back to only really needing a 95% certainty which is where things get fuzzy

 

[tomwalz]

maybe 1419.

Don't use this if you have a grade riding on it. I tried a couple of approaches and ended up using autofill in Excel.

However something feels funny about a couple of my assumptions.

Lunch is over so back to work.

Thanks for the entertainment.

Tom

Thomas J. Walz
Carbide Processors, Inc.

http://www.carbideprocessors.com

Good engineering starts with a Grainger Catalog.

 

[Compositepro]

I'm not going to spend much time on this so my guess is you multiply the probability of not getting a certain lot by the number of samples until that probability falls to less than 5%. There is a 25% chance of getting a certain lot. So 0.75 to the 11th power is 0.042. 11 samples should have a 96% chance of having at least one marble from each lot. To be precise the probabilities change slightly each time you remove a marble.

 

[rkremser]

Yep, its not a graded thing, unfortunately a real world sampling issue.
Compositepro,
The more i've been thinking about it the closer I think you are. I was going down a similar path just from the other side of things but your solution seems correct.

Anyone else have any thoughts before we go forward?