What can we learn from functional clustering of mortality data? An application to HMD data
In most cases, mortality is analysed considering summary indicators (e. g. e_0 or e^†_0) that either focus on a specific mortality component or pool all component-specific information in one measure. This can be a limitation, when we are interested to analyse the global evolution of mortality patterns without loosing sight of specific components evolution. The paper analyses whether there are different patterns of mortality decline among developed countries, identifying the role played by all the mortality components. We implement a cluster analysis using a Functional Data Analysis (FDA) approach, which allows us to consider age-specific mortality rather than summary measures as it analyses curves rather than scalar data. Combined with a Functional Principal Component Analysis (PCA) method it can identify what part of the curves (mortality components) is responsible for assigning one country to a specific cluster. FDA clustering is applied to 32 countries of Human Mortality Database and years 1960–2010. The results show that the evolutions of developed countries follow the same pattern (with different timing): (1) a reduction of infant mortality, (2) an increase of premature mortality, (3) a shift and compression of deaths. Some countries are following this scheme and recovering the gap with precursors, others do not show signs of recovery. Eastern Europe countries are still at stage (2) and it is not clear if and when they will enter into phase (3). All the country differences relates the different timing with which countries undergo the stages identified by clusters. The cluster analysis based on FDA allows therefore a comprehensive understanding of the patterns of mortality decline for considered countries.
READ FULL TEXT