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?