New Statistics for Demand-Based Software Testing

Abstract Current statistical failure probability estimators for software are based on failure probability models which are too simple to be generally applicable. New failure probability models are required which account for the complexities of software structure, or at least allow the significance of such complexities to be understood since it may be the case that current models are sufficient in some circumstances — this question is open. An example of such a new failure probability model is developed for a simple program structure, and shows how current models can lead to inadequate failure probability estimation.