An age–period–cohort model in a Dirichlet framework: A coherent causes of death estimation

Though pivotal in longevity studies, multi-outcome modelling is largely neglected in the associated statistical literature. Here, we focus on the case of compositional data, especially relevant in longevity analysis, where overall mortality can be described as the composition of several causes of death. We propose an age–period–cohort model within the Dirichlet framework with a specific interest in its use for modelling longevity with multiple causes of death. We introduce a flexible approach to incorporating the Dirichlet distribution into the age–period–cohort framework. Then, using US cause-specific mortality data, we provide a comprehensive discussion and comparison of alternative modelling approaches.