|
Abstract:
|
The "95/95 rule" has been used to demonstrate at a 95% confidence level that more than 95% of the tested products meet a success criterion. Under this rule, a binomial method is typically applied which requires testing 60 products with no failures allowed for monthly quality control (QC), or testing 93 products with one failure allowed. A scan statistics method with a moving testing window is recommended by some researchers because its helps to detect cluster of events. However, compared to the binomial test, a smaller sample size is required by the scan statistics method in general. In this abstract, we compare the Type I error rate and power of these two approaches (binomial and scan) of different sampling schemes via simulation. The strength and limitation of these two approaches will also be compared. Our initial findings show that the scan statistics method takes longer to detect change of true failure rate in most scenarios, if a substantial amount of sample size saving is wished, compared to the binomial approach. We will give points of considerations when choosing these approaches.
|
