Tim Folkerts
Trusted Information Resource
I have created a "beta version" of an Excel spread sheet to help determine the sample size for sampling plans. I've complained enough about the lousy statistics of most sampling plans, so I figured I ought to try something constructive.
You have to input
* the maximum percentage of defective units that you would like (aka AQL)
* alpha (the odds of rejecting a lot that just barely meets expectations)
* the percentage of defective units that you will NOT accept (let's call it UQL for unacceptable quality level)
* beta (the odds of accepting a lot that fails to meet this minimum standard)
Then you adjust a slider until it tells you you have an acceptable solution (color-coded and everything)! (With a little cleverness, you could work backwards to create c=0 plans or find out what kind of statistics are actually involved with your current plan.)
For example, the default settings as saved are AQL = 1%, UQL =5%, and alpha = beta = 5%.
Within 1-2 minutes, anyone can find that the acceptable plans for these numbers would be
SAMPLE.SIZE....ACCEPT NUMBER
124-137.........2
153-180.........3
199-207.........4
(The reject number is 1 more than the accept number.)
This version is based on the binomial distribution, meaning 1) only 1 defect per unit is possible and 2) the sample size is small compared to the lot size. I could presumably work out similar versions for the Poisson distribution and for the hypergeometric distribution, but I haven't gotten that ambitious yet.
I think the calculations are all correct, but so far there has been no independent confirmation. Use at your own risk!
Any comments/suggested improvements/donations would be appreciated!
Tim F
You have to input
* the maximum percentage of defective units that you would like (aka AQL)
* alpha (the odds of rejecting a lot that just barely meets expectations)
* the percentage of defective units that you will NOT accept (let's call it UQL for unacceptable quality level)
* beta (the odds of accepting a lot that fails to meet this minimum standard)
Then you adjust a slider until it tells you you have an acceptable solution (color-coded and everything)! (With a little cleverness, you could work backwards to create c=0 plans or find out what kind of statistics are actually involved with your current plan.)
For example, the default settings as saved are AQL = 1%, UQL =5%, and alpha = beta = 5%.
Within 1-2 minutes, anyone can find that the acceptable plans for these numbers would be
SAMPLE.SIZE....ACCEPT NUMBER
124-137.........2
153-180.........3
199-207.........4
(The reject number is 1 more than the accept number.)
This version is based on the binomial distribution, meaning 1) only 1 defect per unit is possible and 2) the sample size is small compared to the lot size. I could presumably work out similar versions for the Poisson distribution and for the hypergeometric distribution, but I haven't gotten that ambitious yet.
I think the calculations are all correct, but so far there has been no independent confirmation. Use at your own risk!
Any comments/suggested improvements/donations would be appreciated!
Tim F