Projection of mortality rates as

illustrated by Weibull distribution

Daniel Ryan
Watson Wyatt

Building mortality models

Possible forms of mortality models:

Projection of past trends without adjustment
Projection of past trends with adjustment
Adjust for changes in risk factors
Application to alternative populations
Projections from medical cohort data by
disease condition, allowing for effects of
treatment and disease interaction

Development of actuarial mortality models

Gompertz - Bcx
Makeham A + Bcx
Kannisto Bcx / (1 + Bcx )
Other variants by Beard, Perks
Each model approximates actual mortality
experience either in specified calendar year or
over lifetime of individual

1
Weibull distribution

Published in 1939 by Waloddi Weibull

Probability density function

Cumulative distribution function

is scale parameter, and is shape parameter

3 parameter version available

Uses of Weibull distribution

Weibull distribution normally used as a reliability

distribution to model:
Material strength
Times-to-failure of electronic and mechanical
components
More recent application to model various
phases of mortality experience

Value of slope or shape parameter

2
Value of slope or shape parameter

Effect of value of slope parameter:

<1 Infant mortality or burn-in
1 Random failures
1-4 Wear out failure
3.5 Similar to Normal distribution
>4 Old age or rapid wear out

Weibull distribution in mortality models

Oldest-old mortality in China Demographic

Research (2003) Vaupel & Yi
Comparison of Weibull to Kannisto
Application of mortality models to Japan
SOA s Living to 100 (2005) Ozeki
Use of Mixed Weibull distributions at different ages
Use in construction of Japanese life tables

Weibull distribution in mortality models

Complementarity between survival and

mortality - Weon
Weon model derived from Weibull distribution using
age-dependent shape parameter
Analysis of trends in the age-specific shape of mortality
curves for populations in the US and Japan SOA s
Living to 100 (2005) Dugan et al
Analysis of trends in mortality near or during retirement
for four European countries ICA (2006) Humble et al

3
Source of mortality experience and choice
of measure of mortality

Population mortality experience by individual age and

for both sexes taken from Human Mortality Database
over the period 1960 to 1999 for:
England & Wales
USA
Japan
France
Italy
Probability of death from starting age chose age 50 -
for each subsequent age from vertical life table

Method for fitting Weibull mortality models

Method of percentiles
Fit Weibull cumulative probability distribution to
actual experience at selected percentiles
Chose 50th and 95th percentiles to highlight old age
mortality
Comparison of parameter values from Weibull
mortality models in different calendar years

E&W males appropriateness of fit

Fit of Weibull distribution to probabilty of death at specified age
- E&W males in calendar year 1999

0.0450

0.0400
Probability of death at specified age

0.0350

0.0300

0.0250
Actual
Estimated
0.0200

0.0150

0.0100

0.0050

0.0000
50 55 60 65 70 75 80
Attained age

4
E&W females appropriateness of fit
Fit of Weibull distribution to probabilty of death at specified age
- E&W females in calendar year 1999

0.0450

0.0400
Probability of death at specified age

0.0350

0.0300

0.0250
Actual
Estimated
0.0200

0.0150

0.0100

0.0050

0.0000
50 55 60 65 70 75 80
Attained age

Specimen parameter values for Weibull

mortality models in different calendar years
Co-efficients of Weibull distribution fitted to USA male population
mortality experience for period 1960-1999
8.5 82.0

8.3 81.0

8.1 80.0

7.9 79.0
Value of co-efficient A

Value of co-efficient B

7.7 78.0

7.5 77.0

7.3 76.0

7.1 75.0

6.9 74.0

6.7 73.0

6.5 72.0
1960 1965 1970 1975 1980 1985 1990 1995
Calendar year
Co-eff A Co-eff B Linear (Co-eff A) Linear (Co-eff B)

Investigation period affecting comparisons

between actual and projected mortality
Male life expectancy at age 50 - Actual vs Modelled (based 1960-1979)

32

31

30
Life expectancy at age 50

29
E&W(a)
28 E&W(m)
USA(a)
27
USA(m)
26 Japan(a)
Japan(m)
25

24

23

22
1960 1965 1970 1975 1980 1985 1990 1995
Calendar year

5
Investigation period affecting comparisons
between actual and projected mortality
Female life expectancy at age 50 - Actual vs Modelled (based 1960-1979)

36

35

34
Life expectancy at age 50

33
E&W(a)
32 E&W(m)
USA(a)
31
USA(m)
30 Japan(a)
Japan(m)
29

28

27

26
1960 1965 1970 1975 1980 1985 1990 1995
Calendar year

Comparisons of male life expectancy from

Weibull mortality model
Projected male life expectancy at age 50 -
Weibull mortality model (1960-1999)

50

45
Projected life expectancy at age 50

40

35
E&W
30
USA
25 Japan
France
20
Italy
15

10

0
2000 2010 2020 2030 2040 2050
Calendar year

Comparisons of female life expectancy from

Weibull mortality model
Projected female life expectancy at age 50 -
Weibull mortality model (1960-1999)

50

45
Projected life expectancy at age 50

40

35
E&W
30
USA
25 Japan
France
20
Italy
15

10

0
2000 2010 2020 2030 2040 2050
Calendar year

6
E&W males - mortality improvements from
Weibull mortality model
England & Wales Male population mortality improvements
- Actual (1960-1999), Projected (2000+)
Age 90

Age 80
4.0%-4.5%
3.5%-4.0%
3.0%-3.5%
Age 70 2.5%-3.0%
2.0%-2.5%
1.5%-2.0%
1.0%-1.5%
Age 60 0.5%-1.0%
0.0%-0.5%

Age 50
1961 1971 1981 1991 2001 2011 2021 2031 2041
Calendar year

E&W females - mortality improvements

from Weibull mortality model
England & Wales Female population mortality improvements
- Actual (1960-1999), Projected (2000+)
Age 90

Age 80

3.5%-4.0%
3.0%-3.5%
Age 70 2.5%-3.0%
2.0%-2.5%
1.5%-2.0%
1.0%-1.5%
Age 60 0.5%-1.0%
0.0%-0.5%

Age 50
1961 1971 1981 1991 2001 2011 2021 2031 2041
Calendar year

USA males - mortality improvements from

Weibull mortality model
USA Male population mortality improvements
- Actual (1960-1999), Projected (2000+)
Age 90

Age 80

3.0%-3.5%
2.5%-3.0%
Age 70 2.0%-2.5%
1.5%-2.0%
1.0%-1.5%
0.5%-1.0%
Age 60 0.0%-0.5%

Age 50
1961 1971 1981 1991 2001 2011 2021 2031 2041
Calendar year

7
USA females - mortality improvements from
Weibull mortality model
USA Female population mortality improvements
- Actual (1960-1999), Projected (2000+)
Age 90

Age 80

3.5%-4.0%
3.0%-3.5%
Age 70 2.5%-3.0%
2.0%-2.5%
1.5%-2.0%
1.0%-1.5%
Age 60 0.5%-1.0%
0.0%-0.5%

Age 50
1961 1971 1981 1991 2001 2011 2021 2031 2041
Calendar year

Further areas for consideration over

application of Weibull mortality model

Differing views over scope for further improvements in

life expectancy
Application to insured population experience
Quantifiable differences between future expectation and
past trends for prevalence of known risk factors e.g.
smoking
Adjustment where appropriate for publicly available
disease specific models, with recent attention on
cardiovascular mortality

Broken limits to life expectancy

95
UN

World Bank
90
Olshansky et al.
UN

Fries, Olshansky et al.,

85 Coale Coale & Guo
World Bank, UN

Bourgeois-Pichat, UN
80 Siegel
Bourgeois-Pichat
UN, Frejka

75

Dublin
70 Dublin & Lotka

65 Dublin

60

55

50

45
1840 1860 1880 1900 1920 1940 1960 1980 2000 2020 2040

Science 2002: Vaupel & Oeppen.