mortality change the details are messy year to year decline irregular persistent, puzzling...
TRANSCRIPT
MORTALITY CHANGE
THE DETAILS ARE MESSY
•Year to year decline irregular
•Persistent, puzzling differentials
•Cause of death structure difficult to understand & to predict
•Poor understanding of causal relationship to driving forces
•Startling reversibility -- the Former Soviet Union
BUT…
IN THE AGGREGATE (i.e., age/sex)
OVER THE LONG-TERM ( >40 years)
IN HIGHLY INDUSTRIALIZED NATIONS
THERE APPEARS TO BE A
Simple, general (?) pattern of decline
m(x,t) = central death rate
log m(x,t) = s a(x) k(t) + r b(x) g(t) + …
Singular Values s > r > … > 0
IF s >> r > …
DOMINANT TEMPORAL PATTERN IS
k(t)
% VARIANCE EXPLAINED IS
s2/(s2 + r2 + …)
AGGREGATE AGE/SEX MORTALITY CHANGE
log m(x,t) = a(x ) + b(x) k(t) + e(t)
G-7 = Canada, France, Germany, Italy, Japan, UK, US
Period = 1950 TO 1994
Lee-Carter; Tuljapurkar-Li-Boe
IN ALL THE G 7
LEAD TEMPORAL COMPONENT
k(t)
EXPLAINS OVER 92 % OF
VARIATION IN log m(x,t)
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995-20
-15
-10
-5
0
5
10
15
20
CanadaFranceGermanyItalyJapanUKUS
1990 2000 2010 2020 2030 2040 205074
76
78
80
82
84
86
USA
95%
75%
50%
25%
5%
H
M
L
xe
Year
g
1990 2000 2010 2020 2030 2040 205076
78
80
82
84
86
88
UK
95%
75%
50%
25%
5%
H
M
L
xe
Year
f
IMPROVING AGGREGATE FORECASTS
1. Take the uncertainty seriously.
2. Use higher components of temporal structure to improve forecasts.
3. Examine and forecast adult mortality separately from infant/child mortality.
4. Decompose differences (e.g., by sex) as deviations from the aggregate.
PROBLEMS WITH CAUSE OF DEATH (COD) METHODS
1. DEPENDENCE BETWEEN CAUSES
Risk factors with multiple effects, complex states of health
2. CAUSE STRUCTURE SHIFTS OVER TIME
Causes appear, peak, disappear.
3. LIMITED CAUSAL UNDERSTANDING OF RISK FACTOR DYNAMICS
At population level, hard to use even smoking prevalence, intensity to predict
4. INACCURACY IN COD ASSIGNMENT