Bayesian probabilistic projections of life expectancy for all countries

Area of Science:

• Demography
• Statistical modeling
• Epidemiology

Background:

• Accurate life expectancy forecasts are crucial for global population projections.
• Existing methods may not fully capture demographic uncertainties.
• Probabilistic forecasting is essential for robust planning.

Purpose of the Study:

• To develop and evaluate a Bayesian hierarchical model for probabilistic male life expectancy forecasts.
• To provide forecasts for all countries up to the year 2100.
• To inform probabilistic population projections for the United Nations.

Main Methods:

• Bayesian hierarchical modeling approach.
• Out-of-sample cross-validation using data from 1950-1995.
• Forecasting male period life expectancy at birth.

Main Results:

• The model achieved a mean absolute error of approximately 1 year for 10-year predictions.
• Forecasts demonstrated a 40% improvement in accuracy compared to the UN methodology.
• Probabilistic forecasts were well-calibrated, with 80% prediction intervals containing the true values 80% of the time.

Conclusions:

• The proposed Bayesian model provides reliable probabilistic forecasts of male life expectancy.
• The method is applicable globally and demonstrates superior performance over existing approaches.
• Freely available R packages (bayesLife, bayesDem) facilitate the implementation of this method.