Multiple time scales in survival analysis

Abstract

In many survival analysis applications, there may not be a unique plausible scale in which to measure time to failure or assess performance. This is especially the case when several measures of usage or exposure are available on each unit or individual. For example, the age, the total number of flight hours, and the number of landings are usage measures that are often considered important in aircraft reliability. This thesis considers the definition of a "good" time scale, along with models that combine different usage or exposure measures into a single scale. New semiparametric inference methods for these models are developed and their efficiency in moderate size samples in investigated. Datasets are used to illustrate the implementation of these methods and of some model assessment procedures. Ideas for further applications and research are also given.