Re: Attribute Data - What is the required sample size
I am trying to resolve the same questions myself; I would love to see further discussion on this. Many articles recommend the binomial direct calculation or the normal approximation to the binomial without describing the assumptions. Upon investigation, my process(es) do not meet the assumptions. When using AQL as a required performance level to choose the lot tolerance percent defective at which to evaluate, I get sample sizes that are larger than my lot size or close to lot sizes, but the binomial requires the sample to be a) small when compared to lot size and b) to be from a stream of production. Many of the processes under evaluation can be considered streams, but some are not; they are manufactured in the same sized lots and then may not be made again for weeks.
I think, because the desired LTPD or 'reject quality level' is so small, by desire and nature of the product and history (prior to validation of the particular processes that are within many in the manufacture of the product), that the Poisson approximation to the binomial should be used, but the quantities turn out to be very similar to the binomial.
Anybody else resolved these issues? Dodge-Romig tables based upon lot sizes and LTPD?
Thanks very much.