A case for new statistical software testing models

There is growing interest in statistical software testing (SST) as a software assurance technique. While the approach has major attractions, there is a need for new statistical models to infer failure probabilities from SST. The paper constructs a simple but realistic case in which the traditional binomial model does not work. The paper shows that if possible test failure dependencies are neglected, could the failure probability would be underestimated. The paper compares the results of our new probability model based on pairwise failures with results achieved when applying the traditional single-urn model, i.e., assuming no dependencies in the failure process