Abstract
In this paper, we present a new imprecise statistical inference method for accelerated life testing data, where nonparametric predictive inferences at the normal stress level is integrated with a parametric power-Weibull model. The method includes imprecision based on the likelihood ratio test, which provides robustness with regard to the model assumptions. The imprecision leads to observations at increased stress levels being transformed into interval-valued observations at the normal stress level, where the width of an interval is larger for observations from higher stress levels. Simulation studies are presented to investigate the performance of the proposed method.