Comparison of Statistical Methods for Detector Testing Programs
A typical goal for any detector testing program is to ascertain not only the performance of the detector systems under test, but also the confidence that systems accepted using that testing program’s acceptance criteria will exceed a minimum acceptable performance (which is usually expressed as the minimum acceptable success probability, p). A similar problem often arises in statistics, where we would like to ascertain the fraction, p, of a population of items that possess a property that may take one of two possible values. Typically, the problem is approached by drawing a fixed sample of size n, with the number of items out of n that possess the desired property, x, being termed successes. The sample mean gives an estimate of the population mean p ≈ x/n, although usually it is desirable to accompany such an estimate with a statement concerning the range within which p may fall and the confidence associated with that range. Procedures for establishing such ranges and confidence limits are described in detail by Clopper, Brown, and Agresti for two-sided symmetric confidence intervals.