Probabilistic Projection of Subnational Life Expectancy

Area of Science:

• Demography
• Statistical modeling
• Public health

Background:

• Subnational mortality projections are crucial for demographers.
• Existing national models lack regional specificity and ease of use.

Purpose of the Study:

• To develop a probabilistic method for projecting subnational life expectancy.
• To adapt the UN's national Bayesian hierarchical model for regional use.
• To create user-friendly projection tools.

Main Methods:

• Proposed three probabilistic projection methods for subnational life expectancy.
• Two methods use simple scaling.
• The third method employs a heteroskedastic first-order autoregressive (AR(1)) process for regional life expectancy deviations from national averages.

Main Results:

• The proposed methods, particularly the AR(1) model, outperformed comparative methods in out-of-sample projections.
• Models demonstrated good calibration for individual regions.
• The AR(1) method showed superior performance in capturing crossover patterns between regions.

Conclusions:

• The developed methods provide accurate and well-calibrated probabilistic forecasts for subnational life expectancy.
• While generally effective, limitations exist in precisely modeling within-country variation.
• The AR(1) method remains the best available option for addressing these regional projection challenges.