Bayesian Changepoint Detection in the Log-Power Model

Abstract

We consider testing and estimation of the changepoint problems for a reliability growth model; log-power model (Xie and Zhao, 1993). The mean value function of the log-power model have an important tool of the graphical interpretation. We select one model between the no-changepoint model and the change model by the Bayes factor. If we accept the no-chagepoint model, we carry out a multiple test with the intrinsic priors of Berger and Pericchi(1996) and determine whether a system is improving or deteriorating over time. We demonstrate our results with the real datasets of Magurire and Pearson (1952) and some simulated datasets.