My current process is operating at a 0.01 defect frequency (e.g., 3 defects in last 300 units inspected). If I make a process change that could potentially increase the defect frequency, how many units will need to be inspected post-change to determine if a statistically significant difference in defect frequency?
With the current process at 1% defective you have to set a upper defective limit say 2%. Also, the risks have to be assigned, typically 10% for type I & II errors
The sample size is than the sample size and defect value which will accept a 1% or lower defect level 90% of the time while rejecting the product when the relect level is 2% or higher 90% of the time.
An OC curve is the best way to visualize the results
My Excel based calculation says a sample of 948. If there are 13 or less defects, accept that there is no process shift, greater than 13 defects a shift has occurred, given the risks specified.
I suggest you use a Runs Chart and/or Process Behavior Chart (Control Chart), With this type of "rare event" data, both work well. Use "good units inspected between defective units" as your test variable. When you make your process change, you will be able to detect a shift in performance, if there is one. See my article in Quality Digest regarding OSHA Recordable Injuries (same type of data)...http://www.qualitydigest.com/inside/quality-inside
Here is a good formula for the AQL sample size and the RQL sample size The sample size, 947 or 948 is shown.
I would have sent you the JMP table but don't know how to post it here