Identifying the ruptures shaping the segmented line of the secular trends in maximum life expectancies

Jacques Vallin, Institut National d'Études Démographiques (INED)
Carlo G. Camarda, Max Planck Institute for Demographic Research
France Meslé, Institut National d'Études Démographiques (INED)

In their 2002 paper, Oeppen and Vaupel showed that the trend in the highest life expectancy reached each year in the world follows a straight line with a 25% gradient, since 1841 (the start year of their database). If we had been nearing a limit, they claimed, we would have seen a flattening of the curve over recent decades. As this is not the case, there is no reason to believe that progress in life expectancy will stall in the near future. More recently, Vallin and Meslé (2009), by looking at more data (more life tables for the 1841-2000 period and additional tables before 1841 and after 2000) and, more importantly, by checking better data quality, recognized that maximum life expectancies follow a succession of four segments with associated different slopes: 0,5%, 11%, 32%, and finally 23%. This indicates that, at each stage, new tools for health improvements trigger new paces in life expectancy increase. They related the three cuts to important changes in the history of the health transition: the years 1790s, at the time of Jenner and of the French Revolution, the time of Pasteur around the year 1885 and at the end of the 1960s when started the so-called Cardiovascular Revolution. However, Vallin and Meslé’s identification of the time cuts was intuitive, simply suggested by the general shape of the cloud of points. Therefore it is still controversial whether or not humans experienced a continuous and constant life expectancy increase. The objective of this paper will be to check the existence of eventual cuts, their number and their calendar. Segmented regression provides an elegant framework and will be employed to statistically identify possible ruptures in the time series of maximum life expectancies. Further analysis by age will shed light on determinants of the eventual variations in gradient.

  See extended abstract

Presented in Session 8: Life table analysis