A Bayesian model for age at death with unequal adult mortality and cohort effects

The conventional approach to modeling age at death through parametric models often employs a three-component mixture as a standard. However, recognizing the need to more accurately reflect the diverse trends in old age mortality, this study introduces a five-component Bayesian mortality model. This model is designed to address data exhibiting cohort effects and transitions in the old age mortality component, which moves from a skew-normal distribution to a mix of Gaussian and skew-normal distributions as the period progresses. The proposed model comprises a point probability for infant mortality, a uniform distribution for background mortality, a Gaussian distribution for premature mortality, and two components for old age mortality (Gaussian and skew-normal). Applied to data from 10 Italian regions spanning 1974–2022, including cohort effects for cohorts born between 1913 and 1947, the model demonstrates a stable interpretation of mortality components over time. In contrast to simpler models, it reveals a significant shift in old age mortality from a skew-normal to a combined Gaussian and skew-normal distribution around the millennium’s turn. This study not only sheds light on the evolution of old age mortality in Italy, highlighting a non-uniform shift and compression of mortality across the population, but also lays the groundwork for future forecasting models by maintaining a consistent parameter interpretation throughout the period.