Multipopulation mortality analysis: bringing out the unobservable with latent clustering

Mortality patterns experienced in closely related populations show similarities in some aspects and differences in others. Indeed, if a decline in mortality rates among low-mortality countries is observed, it is possible that populations experience different trends through which this decline occurs. Observing mortality rates for ages and over specific time windows, it is evident that the different interactions between the variables age and time influence longevity trends. Therefore, to grasp the complexity of the phenomenon, the similarities or differences in mortality need to be analyzed by considering three dimensions: age, year, and country, simultaneously. With this aim in mind, we propose applying a multidimensional latent clustering approach to multipopulation mortality data in this paper. We investigate some similarities between the mortality experience of different countries, searching for latent structure across these groups. Starting from the observation units represented by single countries, we nest them in higher-level units of clusters. We apply the proposed model to the mortality rates of 20 developed countries using data from 1965 to 2019 from the Human Mortality Database. We present detailed results for the lower mortality cluster, which collects ages from 50 to 60 among all countries of the selected dataset and highlights different mortality trends between the countries.