Carlo Giovanni G. Camarda , Institut National d'Études Démographiques (INED)
Importance of describing mortality at the limits of life span has lately lead to relevant as well as controversial articles. Whereas considerable effort has been made in collecting data and estimating models on oldest-old individuals, testing statistical confidence about eventual conclusions has been largely neglected. How certain can we be saying that risk of dying increases, levels out, or paradoxically decreases over age 105? Can we detail particular mortality age-patterns at those high ages? In this paper, it is shown that very little can be affirmed when we venture in describing mortality at extreme ages. Instead of analyzing actual data, we perform a series of simulation studies mimicking actual scenarios. By knowing the true underlying age-patterns, we generate lifetimes which are either fully observed or censored/truncated from controlled mechanisms. Our findings are thus robust with respect to factors such particular observation schemes, heterogeneity and data quality issues. Given sample sizes currently available and levels of mortality experienced in present populations, we show that before age 110 only a gompertzian increase of mortality can be eventually detected. Afterwards a plateau will be regularly recognized as the most suitable pattern, regardless the complexity of the true underlying mortality.
Presented in Session 15. Mortality Models