Volume 34 - Article 38 | Pages 1063–1074  

Convergence in male and female life expectancy: Direction, age pattern, and causes

By Benjamin Seligman, Gabi Greenberg, Shripad Tuljapurkar

Abstract

Background: The cornerstone of mortality- and life-expectancy forecasting in developed nations, the Lee-Carter model relies on assumptions of there being a dominant singular value that captures most of the variance within a matrix of age-specific mortality rates over time and that the time trend captured by this lead singular value is constant. We revisit the model's predictive ability and trends in mortality decline among developed nations since the end of the Cold War.

Objective: To understand the predictive power of the Lee-Carter model with mortality trends since 1990.

Methods: Mortality data were obtained from the Human Mortality Database. Forecasts were made using R with random walk forecasts using the package forecast.

Results: While Lee-Carter forecasts of life expectancy for combined sexes were accurate, sex-specific forecasts tended to somewhat overestimate for females and significantly underestimate for males. Further investigation of the trend for males shows that the first singular value continues to capture the majority of the variation in mortality since 1990, with progress along this dimension moving at a constant rate.

Conclusions: Lee-Carter forecasts have significantly underestimated gains inmale life expectancy without major changes to the model’s assumptions. We believe this represents more rapid progress in tackling male mortality in the G7 countries without major changes to the age pattern of these gains. Curiously, this has not affected combined-sex forecasts, potentially being offset by slight overestimation of female mortality progress.

Contribution: We show that the Lee-Carter model has made inaccurate forecasts of mortality rates unrelated to violations of its underlying assumptions.

Author's Affiliation

• Benjamin Seligman - Stanford University, United States of America EMAIL
• Gabi Greenberg - Stanford University, United States of America EMAIL
• Shripad Tuljapurkar - Stanford University, United States of America EMAIL

Other articles by the same author/authors in Demographic Research

How does the demographic transition affect kinship networks?
Volume 48 - Article 32

Variance in death and its implications for modeling and forecasting mortality
Volume 24 - Article 21

How can economic schemes curtail the increasing sex ratio at birth in China?
Volume 19 - Article 54

Most recent similar articles in Demographic Research

Leveraging deep neural networks to estimate age-specific mortality from life expectancy at birth
Volume 47 - Article 8    | Keywords: death rates, deep neural network, forecasting, life expectancy

Coherent forecasts of mortality with compositional data analysis
Volume 37 - Article 17    | Keywords: coherent mortality forecasting models, compositional data, Lee-Carter model, Li-Lee, life expectancy, mortality forecasts

Divergence without decoupling: Male and female life expectancy usually co-move
Volume 31 - Article 51    | Keywords: Goodman-Grunfeld test, international comparisons, life expectancy, mortality, sex differences

Regional probabilistic fertility forecasting by modeling between-country correlations
Volume 30 - Article 35    | Keywords: aggregation, correlation, forecasting, probabilistic projection, pseudolikelihood, total fertility rate (TFR)

Impact of different mortality forecasting methods and explicit assumptions on projected future life expectancy: The case of the Netherlands
Volume 29 - Article 13    | Keywords: Europe, extrapolation, Lee-Carter model, Netherlands, nonlinear models, smoking