A Bayesian model for age at death with cohort effects

Background: Ongoing mortality trends affect the distribution of age at death, typically described by parametric models. Cohort effects can markedly perturb the distribution and reduce the fit of such models, and this needs to be specifically taken into account. Objective: This study examines the integration of cohort effects in a three-component parametric model for the age-at-death distribution, applying it to data with significant cohort effects. Methods: We employed a mixture model with a half-normal and two skew-normal components, adapted to a Bayesian framework to include multiplicative cohort effects. The model was applied to data from five Italian regions, with cohort effects estimated for the 1915–1925 cohorts. Results: Incorporating cohort effects significantly improved the model’s fit. A notable finding of the comprehensive model is the shift in Italy from premature to middle-age mortality components over time. Our results also demonstrate the tendency for mortality structures to spatially homogenize over time in Italy. Conclusions: The study underscores the importance of including cohort effects in mortality models in order to provide a more detailed picture of mortality trends. Contribution: This work introduces a novel application of a Bayesian mixture model with cohort effects, offering enhanced tools for demographic analysis and new insights into the evolution of mortality components in Italy. This approach is general but fully formalized and hence it can be readily used for demographic studies in other regions as well.