Identifying the parsimonious model for adult and old-age human mortality: statistical evidence, applications and consequences

Németh, L.

Rostock, University of Rostock (2016)


The dissertation aims to identify the parsimonious model for adult and old-age human mortality and tries to answer the following research questions: How much does life-table aggregation at oldest ages influence the mortality measures based on life tables? How biased are mortality measures if the fitted mortality model is misspecified? Do the largest high-quality mortality databases cope successfully with the aggregation of data? Is it essential to reconstruct adult human mortality within the ΓGM framework for the published tables? How strong life expectancy and lifespan inequality is related for various populations in these databases? Can the qualitative pattern of adult and old-age human mortality be captured by frailty models? The standard way of “closing” a life table might lead to erroneous conclusions about mortality measures, therefore the thesis proposes fitting the Gamma-Gompertz-Makeham model and shows that model-based measures are less sensitive to the age at censoring. In addition, the thesis studies the impact of neglecting statistically significant extrinsic mortality or frailty on human mortality measures and presents the arising bias. The dissertation compares the life table “closing” methods used by the largest high-quality life-table databases with the ΓGM estimates. As a result reconstructing adult mortality within the ΓGM framework is essential for some of the published tables. The dissertation also tries to understand why the strength of the relationship between life expectancy and lifespan inequality varies across publications based on these life-table databases. The eventual levelling off of human death rates has been hypothesized for more than seven decades and partial evidence has been found. If a human mortality plateau exists then individual risks of death from senescent causes are proportional and people age essentially at the same rate. Given currently available mortality data, the dissertation tests the existence of a human mortality plateau. Results speak in favor of a levelling off and show that the plateau-generating mechanisms, the gamma-Gompertz model and its Makeham extension, ΓGM, explains the most adequately the qualitative pattern of human mortality rates at the front end.