| To achieve higher accuracy in software failure rate estimates in a field environment, software reliability models that were fit from failure data collected during the test interval are often refit using the combined test interval and field interval data. The validity of this analysis depends on the test and field environments being compatible with respect to the manner in which the software is used. In this dissertation, we formulate the hypothesis of compatible test and field environments in terms of a statistical hypothesis and develop appropriate test procedures. We assume the underlying failure process of the software follows a nonhomogeneous Poisson process. The mean value function has two parts, one that applies during the test interval and one that applies during the field interval. Under the hypothesis of compatible environments, the mean value function reduces to a one-part model that applies in both environments. We proposed four test procedures including likelihood ratio test, Cramer-Von Mises test, early detection test and percentile CUSUM test to evaluate the null hypothesis. The test procedures are illustrated by applying them to a real-life software project. Their performance will be evaluated based on the size and power of the tests. |
