Influence of sample characteristics on probability of detection curves

POD curves are typically constructed as statistical models of responses to known characteristics of the POD demonstration samples. MIL-HDBK-1823A provides some guidance in choosing the number of samples - viz. at least 40 targets for tests that provide a quantitative response (â) and 60 for binary responses (hit/miss) - but says little about the size range or how they should be placed within that range. Here we report on how the number of targets, their size range, and their distribution (uniform, skewed left or right) with respect to a true POD vs. size relationship influences the POD curve that is estimated from these targets. From the parameter estimates we compute confidence bounds to quantity statistical uncertainty. The work is for binary response data and is based on a novel simulation method for computing the confidence bounds which would prohibitively time-consuming using conventional Monte Carlo methods.