
Abstract
This thesis is devoted to studying some methods for estimating life expectancy.
The thesis consists of 3 chapters.
In Chapter 1, we review the preparatory knowledge including some knowledge in
probability theory, and survival analysis theory (survival analysis model, Kaplan-
Meier estimation). In addition, we present two methods of estimating average life
expectancy (Chiang method, Silcocks method) that are currently widely used, as
well as some existing problems with these two methods. Besides, we introduce
the Bootstrap method, which is a simple and highly applicable modern statistical
method. Finally, we introduce the FilaBavi data set.
In Chapter 2, we propose two new methods for estimating life expectancy. The
first method (named the Kaplan-Meier method) is built based on the Kaplan-
Meier estimate for the survival function. It is applied to the semi-cohort dataset,
extracting complete information from data fully recorded birth date and death
date of all death individuals, providing the most accurate estimation of life ex-
pectancy. Therefore, that method can be adopted as a "standard" in the accuracy
investigation of other life expectancy estimations. The second method, called the
local parametric method, is tailored according to the theoretical background of the
survival process with local parametric Weibull distributions and can be applied to
abridged datasets containing only a pair of number of deaths and persons in each
age group. With this method, we have built a formula to estimate the average
life expectancy and variance of the estimate, proving that the estimate has an
approximately normal distribution to then provide a formula for the confidence
interval of the life expectancy.
In Chapter 3, we present the results of applying the methods (Kaplan-Meier
method, local parametric method, Chiang method, Silcocks method) on the real
FilaBavi data set. The calculation results show that the local parameter method
provides a more accurate and higher effective life expectancy estimation than the
Chiang and Silcocks methods.
iii