Effects of the Spanish Influenza Pandemic of 1918-19 on Later Life Mortality of Norwegian Cohorts Born About 1900

Abstract

By using Age-Period-Cohort analysis the paper shows that Norwegian male and female cohorts born about 1900 have experienced significantly higher all-cause mortality in middle and old ages relative to “neighbor” cohorts. In a widely cited study, Horiuchi suggests that only males from belligerent countries who were adolescents during WW I exhibit this cohort effect. The finding in this paper demonstrates that Horiuchi’s explanation may be incomplete.The search for explanations for neutral Norway must clearly go beyond the direct (soldiers wounded physically and mentally) and indirect effects (rationing of food) of WW I on laterlife mortality. This paper suggests that Spanish Influenza 1918-19 is the most important of several possible factors priming the Norwegian cohorts. A large proportion of the cohorts considered contracted Spanish Influenza, but only a small proportion died of it immediately.The net effect on later life mortality is thus assumed to be that of debilitation.

Full text

MEMORANDUM
No 29/2003
Svenn-Erik Mamelund
ISSN: 0801-1117
Department of Economics
University of Oslo
Effects of the Spanish Influenza Pandemic of 1918-19
on Later Life Mortality of Norwegian Cohorts Born
About 1900

This series is published by the
University of Oslo
Department of Economics
In co-operation with
The Frisch Centre for Economic
Research
P. O.Box 1095 Blindern
N-0317 OSLO Norway
Telephone: + 47 22855127
Fax: + 47 22855035
Internet: http://www.oekonomi.uio.no/
e-mail: econdep@econ.uio.no
Gaustadalleén 21
N-0371 OSLO Norway
Telephone: +47 22 95 88 20
Fax: +47 22 95 88 25
Internet: http://www.frisch.uio.no
/
e-mail: frisch@frisch.uio.no
List of the last 10 Memoranda:
No 28 Fedor Iskhakov
Quasi-dynamic forward-looking model for joint household retirement
decision under AFP scheme. 59 pp
.
No 27 Henrik Wiig
The Productivity of Social Capital. 22 pp.
No 26 Tao Zhang
Identifying treatment effects of active labour market programmes for
Norwegian adults. 45 pp.
No 25 Tao Zhang
A Monte Carlo study on non-parametric estimation of duration models
with unobserved heterogeneity. 89 pp.
No 24 Karine Nyborg and Kjetil Telle
The Role of Warnings in Regulation: Keeping Control with Less
Punishment. 31 pp.
No 23 Maria Kalvarskaia
Savings behaviour when households havean access to occupational
pensions. 50 pp.
No 22 Rolf Golombek and Michael Hoel
Climate Policy under Technology Spillovers. 37 pp.
No 21 Jo Thori Lind
Fractionalization and the size of government. 50 pp.
No 20 Michael Hoel, Tor Iversen, Tore Nilssen and Jon Vislie
Genetic testing and repulsion from chance. 20pp.
No 19 Jon Vislie
Domestic Environmental Policy under Asymmetric Information: The
role of foreign ownership, outside options and market power 34 pp.
A complete list of this memo-series is available in a PDF® format at:
http://www.oekonomi.uio.no/memo/

1
Effects of the Spanish Influenza Pandemic of 1918-19 on Later Life
Mortality of Norwegian Cohorts Born About 1900
by Svenn-Erik Mamelund
Address: Department of Economics, University of Oslo
P.O. Box 1095 Blindern, 0317 Oslo, Norway
E-mail: sem@econ.uio.no
Abstract
By using Age-Period-Cohort analysis the paper shows that Norwegian male and female
cohorts born about 1900 have experienced significantly higher all-cause mortality in middle
and old ages relative to “neighbor” cohorts. In a widely cited study, Horiuchi suggests that
only males from belligerent countries who were adolescents during WW I exhibit this cohort
effect. The finding in this paper demonstrates that Horiuchi’s explanation may be incomplete.
The search for explanations for neutral Norway must clearly go beyond the direct (soldiers
wounded physically and mentally) and indirect effects (rationing of food) of WW I on later
life mortality. This paper suggests that Spanish Influenza 1918-19 is the most important of
several possible factors priming the Norwegian cohorts. A large proportion of the cohorts
considered contracted Spanish Influenza, but only a small proportion died of it immediately.
The net effect on later life mortality is thus assumed to be that of debilitation.
Keywords: Spanish Influenza 1918-19; morbidity; mortality; debilitation; selection; APC-
analysis; cohort effects; Norway

2
1. Introduction
There are studies that have found that German, Austrian, Italian, French, and Polish cohorts
that went through adolescence 1914-18 and young adult cohorts directly engaged in WW I
(primarily men) experienced elevated all-cause mortality in middle and old ages compared to
older and younger cohorts (Horiuchi 1983; Boleslawski 1985; Caselli et al. 1987; Caselli and
Capocaccia 1989; Caselli 1990; Wilmoth et al. 1990). The authors basically attributed their
findings to indirect (shortage of food, malnutrition, and poor hygiene) and direct (physical
and mental scars of combat) long-term effects of WW I. Horiuchi (1983) analyzed the
mortality in belligerent Germany, Austria and, France as well as neutral Sweden and Japan.
The analysis showed that only men born about 1900 from Germany, Austria and France
exhibited relative high mortality in middle and old life. Horiuchi therefore concluded that the
cohort variation in mortality most likely was a long-term effect of the WW I, specifically a
result of malnutrition. Gómez de León (1991), however, found that in neutral Norway the
same male and female cohorts were debilitated relative to younger and older cohorts. What
could be the explanations for neutral Norway? The most obvious is neither direct nor indirect
effects of WW I. Gómez de León (1991) was tempted to explain the cohort differentiation by
Spanish Influenza 1918-19 “but without a clear understanding of the nature of the
mechanisms involved, [he found it] adventurous to attribute them directly to any particular
form of determinacy” (p. 81). One other explanation suggested by Gómez de León,
surprisingly not mentioned by other authors, is that a late wave of predominantly male
overseas emigration, which peaked in the mid 1920s, might have introduced additional
selection if the most selective of the survivors left. Gómez de León showed, although he did
not state it, that Horiuchi’s (1983) hypothesis that only men from belligerent countries exhibit
this cohort effect was wrong. A fourth explanation, in addition to the war 1914-1918, Spanish
Influenza 1918-19, and a large wave of emigration in the mid 1920s, of the differences in
cohort mortality in belligerent and neutral Europe, is cigarette smoking which became
increasingly popular during and after WW I, especially for men (Horiuchi 1983). The four
events occurred when the cohorts born about 1900 was adolescents and young adults, and
they may all be important predictors for the mortality patterns they experienced in middle and
old life in the latter half of the twentieth century.
The aim of the present paper is to have a second look at the Norwegian case, and to
challenge Horiuchi’s WW I hypothesis by following up Gómez de León’s Spanish Influenza
hypothesis. To do this, information on the immediate and long-term effects of Spanish
Influenza on health and mortality is needed. A review of the literature regarding this issue is
presented in section 2. The results of the analysis are presented in section 3. To compare the
results for neutral Norway with Horiuchi’s for belligerent Germany, Austria, and France, a

3
replication of exactly what Horiuchi did to separate the cohort from the age and period effects
is pursued; First, estimation of period rate of mortality change with age, and second
estimation of an Age-Period-Cohort (APC) regression model with dummy variables. Results
are also presented estimating refined and extended versions of the descriptive and statistical
tools applied by Horiuchi. Although it is believed that neutral Norway fits well as a test case to
study effects of Spanish Influenza on later life mortality as it may counterbalance effects of
WW I, in section 4 it is discussed whether shortage (1914-19) and rationing (1918-19) of food
during and shortly after WW I (section 4.1) and possible post-traumatic stress disorders
among the Norwegian neutrality-keeping soldiers and seamen in the merchant fleet, who
survived WW I (section 4.2), may confound the analysis. Also, effects of mass-emigration on
mortality (section 4.3) and cigarette smoking (sections 2.2, 3.2, and 3.3) are discussed as a
possible confounding factors.
2. Spanish Influenza 1918-19 and post recovery-problems
2.1. The incidence and mortality of Spanish Influenza 1918-19
Spanish Influenza affected at least 500 million globally or over one fourth of the population
(Laidlaw 1935). It killed 50-100 million people in less than a year (Johnson and Mueller
2002) ─ five to ten times the death toll of more than four years of war. Unlike other influenza
epidemics or pandemics, the bulk of excess mortality occurred among persons 20-40 years
old. The Spanish Influenza affected 1.2 million Norwegians or little less than half of the
population and it took the life of 15,000 people or 5.7 lives per 1 000 (Mamelund 1998). The
pandemic appeared in three bouts; the first during the summer of 1918, the second during the
fall of 1918, and the third during the first months of 1919. The excess all-cause death rate
during the highly virulent second bout from October to December 1918 (8.6 deaths per 1 000)
was twelve times as high as the corresponding death rate during the first bout of influenza
from July to September 1918 (0.7 deaths per 1 000). Mortality during the third bout of
influenza was relatively low. Spanish Influenza was so serious because of the bacterial
complications, mainly pneumonia, but also meningitis, bronchitis and acute diarrhea
(Mamelund 1998). A little more than one per cent of those who were infected by the disease
in Norway died. Although this lethality may seem relatively low, it is the highest registered
lethality of the four influenza pandemics that came to Norway in the twentieth century
(Mamelund and Iversen 2000).
[Figure 1 approximately here]

4
The influenza-censuses carried out in 1918-19 for a number of cities in the United States and
England, and for the city of Bergen, Norway, give a fairly reliable picture of the age-sex
pattern of influenza incidence (Great Britain Ministry of Health 1920; Vaughan 1921;
Hanssen 1923; Collins 1931; Sydenstricker 1931; Britten 1932). These studies clearly show
an unusually high influenza incidence during the summer wave of 1918 for those between 10-
39 years of age, especially men, and a rapidly falling incidence by age for those above 40
years (see Figure 1 for Bergen, Norway). The age-specific incidence curve during the fall
wave was similar to that seen during the summer wave, but those hardest hit during the first
wave seem to have been less hit during the second wave in 1918, probably due to relative
immunity acquired. The cross over in sex-difference in incidence at ages 10-39 is therefore
also apparent. The W-shaped age pattern of influenza and pneumonia death rates for Norway
1918-19 presented in Figure 2 is representative for the global picture. However, the average
influenza and pneumonia death rates for the two years 1918-19 stand in great contrast to the
U-shaped age distribution of influenza and pneumonia death rates for the average of the two
non-pandemic years of 1917 and 1921 (Figure 2). The relatively high lethality and the
peculiar excess age-specific influenza and pneumonia mortality of Spanish Influenza have
received substantial research attention. Despite recent and extensive molecular and
paleomicrobiological research efforts, however, these issues are still a mystery (Basler et al.
2001; Davis et al. 1999; Gibbs et al. 2001; Reid et al. 1999, 2000, 2001, 2002; Tauenberger et
al. 1997, 2001; Tumpey et al. 2002). Those born about 1900 as was identified as high
mortality cohorts by Gómez de León (1991) are represented among those with the highest
influenza incidence (Figure 1). The Norwegian cohorts born 1904-09 (10-14 years 1918-19),
1899-1904 (15-19 years 1918-19), however, experienced relatively low mortality in 1918-19
compared to the cohorts born 1880-1899 (20-40 years 1918-19) (Figure 2). The disease thus
marked a large proportion of the cohorts born 1899-09, but only a small proportion died of it
immediately (see Figure 3). In other words, the cohorts that were adolescent in 1918 and 1919
may have experienced large debilitation effects of morbidity and small selection effects of
mortality due to Spanish Influenza. The interplay of the two mechanisms is assumed to lead
to a net or average cohort effect caused by high relative mortality in later life.
[Figures 2 and 3 approximately here]
The young adults, those born 1890-99 (20-29 years 1918-19), however, experienced high
incidences and death rates in 1918-19. In this situation there are two effects that work in
opposite directions. On the one hand it is likely that those with the poorest health died while
those with the best health survived. This is assumed to result in low future mortality for the
cohorts considered. On the other hand it is possible that Spanish Influenza weakened the

5
average survivors compared to the original cohort. This is assumed to give higher future
cohort mortality. The net cohort effect for the 1890-99 cohort may therefore not necessarily
deviate from that of neighbor cohorts as the effects of selection may have cancelled out the
effects of debilitation.
2.2. Later-life health and mortality of Spanish Influenza survivors after 1918
Spanish Influenza survivors were reported to have problems with sleeping, depressions,
mental distractions, low blood pressure, dizziness and to cope at work and with everyday life
for weeks, months or even years after 1918-19 (Mamelund 1998). According to data from the
Norwegian asylum hospitals (calculated from Medisinaldirektøren 1916, 1917, 1920a-b,
1921, 1923a-b, 1924,1925a-c, 1927, 1928, 1929, 1930, and 1931), there was an excess in the
number of first time hospitalized patients with mental diseases caused by influenza and
pneumonia each year from 1918 to 1923 when compared to the average of the years 1915-17
and 1924-26 (80 men and 78 women). The number of persons, who suffered mental
distractions after Spanish Influenza, is probably much higher than this calculation show as it
is likely that people affected by milder or temporary post influenza melancholia did not see a
psychiatrist. Hepatitis, ear illnesses, deafness, blindness, and baldness (esp. young girls) are
other after-effects that have been linked to Spanish Influenza. It has also been reported that
one-third of the influenza survivors have experienced heart problems, lung tuberculosis and
kidney disease in later life (Collier 1974).
Those who struggled with one or several post-recovery problems and diseases
associated with Spanish Influenza may have experienced higher mortality immediately in
1918-19 or later in life. There are at least three examples from the literature that gives support
to this view: First, Wasserman (1992) has found that excess in influenza death rates 1918-20
was significantly and positively related to suicide in the United States independent of factors
like alcohol consumption and the number of casualties during WW I. The suggested
explanations were a drop in social integration (closing of schools, churches, theatres, banning
of large public meetings and so on) and fear induced by the pandemic (the infected could die
within three days). Several cases of suicide may also have occurred after 1920 either due to
the above-mentioned psychological health problems of some survivors (direct effect) or to the
unbearable loss of a spouse, children or close relatives (indirect effect) (Rice 1988; Crosby
1989; Phillips 1990).
The second example of higher later-life mortality of Spanish Influenza survivors that
may be an after-effect of Spanish influenza is mortality associated with the Sleeping Sickness
Pandemic (encephalitis lethargica or VON ECONOMOS’s disease) of 1919-28. The Sleeping
Sickness Pandemic caused hundreds of survivors to slip into a bizarre rigid paralysis with
similarities to advanced Parkinson's disease. These patients, only occasionally able to

6
communicate or move, were nearly all institutionalized for life (Sacks 1999). It was largely
the work by Ravenholt and Foege (1982) that established the link between the Spanish
Influenza, the Sleeping Sickness Pandemic and increased risk of developing and dying from
Parkinson’s disease in later life. The hypothesis of a causal link was built up on the basis of
two observations. The first is that the pandemics seemed to share etiology. The fact that the
incidence of Spanish Influenza (see Figure 1) and the Sleeping Sickness (see Lund 1997) was
highest among adolescents and young adults (10-30 years), and that the incidence was higher
in men than in women in both pandemics, support this view. The second of the observations
are that the Sleeping Sickness Pandemic followed the Spanish Influenza Pandemic in time
and space. The time sequence given by Ravenholt and Foege (1982), however, has recently
been questioned. In a review of the literature, Casals et al. (1998) found reports of Sleeping
Sickness in several European countries three years before Spanish Influenza broke out in
pandemic dimensions in 1918. Recent genetic as well as archival research, however, now also
place the origin of the Spanish Influenza virus and the first influenza cases around 1915 (Reid
et al. 1999; Oxford et al. 1999; Shortridge 1999). This gives additional support to the view
that the Sleeping Sickness Pandemic was causally linked to Spanish Influenza.
Globally it is estimated that more than one million were infected and that half a
million died from the Sleeping Sickness Pandemic in the period 1919-1928 (Ravenholt and
Foege 1982). In Norway 268 cases and 52 deaths were reported (Mamelund 1999). When
compared to the reported cases per 1 000 in the two neutral countries of Sweden (3.0) and
Denmark (5.9) the figure for Norway (1.0) seems curiously small (Swedish and Danish
figures are from Matheson Commission 1929). The likely explanation is diagnostic
difficulties in Norway with a widespread population and shortage of doctors (Lund 1997).
Poskanzer and Schwab (1963) analyzed 1 000 patients at Massachusetts General, the United
States, 1920-1959, and found that those experiencing a severe case of the Sleeping Sickness
developed Parkinsonism immediately, whereas those experiencing a mild or undiagnosed
attack developed Parkinsonism later in life. For England and Wales, Martyn (1997) has found
that the generation born 1889-1908 was two to three times as likely to die of Parkinson’s
disease in the period 1950-92 than those born before 1888 or after 1924. Ben-Shlomo et al.
(1993) have found similar cohort-effects for the mortality of Parkinson’s disease in the
Republic of Ireland. There are no similar studies on mortality of Parkinson’s disease in
Norway.
The third example of relatively high later-life mortality of Spanish Influenza
survivors that may be linked to Spanish Influenza is from coronary heart disease (CHD).
Azambuja and Duncan (2002) using cross sectional data have found that Spanish Influenza
mortality is a good predictor for the rise (1920-67) and the fall (1968-85) in CHD mortality in
the United States. The analysis revealed that cohorts born about 1900, who had the highest

7
incidence and mortality of Spanish Influenza, also had the highest CHD mortality in later life.
Preceding and succeeding cohorts had subsequently lower CHD mortality. The higher
incidence of Spanish Influenza 1918-19 among men than among women was also used to
explain why men always have had higher CHD mortality than women after 1920. The same
temporal rise and fall in CHD mortality has also been observed in Norway. However,
traditional risk factors like smoking, unhealthy diet, and little physical activity have hitherto
been used to explain the cyclical mortality pattern (see for example Huserbråten 1993 or Aase
and Storm-Furru 1996). Since traditional risk factors in most studies of CHD mortality do not
explain more than half of the variance in CHD mortality, Azambuja and Duncan (2002)
believes that including whether or not a person was exposed to Spanish Influenza in 1918-19
in the analysis will add more understanding to the pattern of CHD mortality.
3. Results
3.1. An empirical tool to trace cohort differentiation in mortality
Horiuchi (1983) formulated the hypothesis that only male members of cohorts who were
adolescents and young adults in belligerent countries during WW I have experienced elevated
mortality or debilitation relative to other cohorts as they passed through middle and old ages.
The purpose of the analysis in sections 3.1 and 3.2 is to show that Horiuchi’s hypothesis was
wrong and that his explanation was incomplete. Hourichi explained the relatively high
mortality of men born about 1900 in Germany by high mortality of cardiovascular diseases in
middle and old age caused by malnutrition experienced in adolescence during WW I. The
hypothesis proposed here is that Spanish Influenza debilitated the same cohorts of males and
females in neutral Norway and that the Spanish Influenza hypothesis is more plausible than
other possible explanations of the differences in cohort mortality like WW I, emigration, and
cigarette smoking.
The first thing Horiuchi (1983) did to separate the effects of period from effects of
cohort and age was to calculate the log of the period rate of mortality change K(x) from one
five-year age group to the next (starting at age 40) for the single calendar years of 1959, 1964,
1969, and 1974. K(x) for a given calendar year is defined as
log( (x))
K(x) =
d
dx
µ
, (1)
where
(x)
µ
is the force of mortality at exact age x. K(x) is estimated as
ˆ
K(x) = log M(x+5) - log M(x)
, (2)

8
where M(x) is a five-year age specific death rate. After age 40, mortality increases relatively
smoothly with age. Therefore, when a high-mortality cohort follows a low-mortality cohort,
K(x) tends to be large. In the opposite case, K(x) tends to be small. Thus, right-way moving
peaks and hollows in the K(x) curve for consecutive periods indicate that there are high
mortality cohorts between the peak and the hollow. K(x) is calculated for all-cause mortality
for males and females aged 40 and above for calendar years of 1959, 1964, 1969, and 1974
(data from Mamelund and Borgan 1996).
[Figures 4 and 5 approximately here]
For males, those born between 1895 and 1910 are identified as high mortality cohorts (using
model (2)), seen by the downward turning and upward turning arrows in the period 1959-74
in Figure 4. K(x) curves estimated for smoothed one-year age groups confirm that that men
born between 1895 and 1910 had higher mortality than neighbor cohorts, not only in the
period 1959-1974, but also in the 1950s (results not shown). No striking cohort effect was
found for other male cohorts or for the 1895-1910 cohorts before 1950 or after the mid 1970s.
This is at odds with Horiuchi (1983) who suggested that this cohort effect is exclusive to men
from belligerent countries. The K(x) curves for females presented in Figure 5, however, do not
display visible cohort differentiation in mortality. This seems to be in accordance with what
Horiuchi found for females in belligerent countries.
A weakness of K(x) is that it only can be used to detect age and cohort variation in
the death rates. Thus, the effect of age and cohort on the variation in the death rates
independent of period is not accounted for. This lead Horiuchi to apply a statistical method,
(on data for men only), which controls for age, cohort, and period simultaneously. This is
important, as cohort effects derived from a K(x) analysis may be different for different time
periods.
3.2. A statistical tool to trace cohort differentiation in mortality
In this section, the assumptions and structure of the APC-model applied by Horiuchi (1983) are
first described. Secondly, a refined and extended version of Horiuchi’s model is presented. And
thirdly, the results of both models are displayed.
Horiuchi (see pp. 81-84) wrote the log-transformed death rate M
ij
, for each sex as
ij ij
G = log M, (3)

9
where i = 1,2,…,10 are five-year age groups 35-39 to 80-84 and j = 1,…,4 are the single
calendar years 1959, 1964, 1969, and 1974. This implies that k = 1,…,13 are five-year cohorts
born between 1870-74 and 1935-39, with k = j - i. The suggested model for G
ij
was
ij
G = + + + +
ijkij
α
βγ δ ε
, (4)
where
α
is a constant, and
β
,
γ
and
δ
are age, period and cohort effects respectively. It is the
simplest APC model as the effects of age, period and cohort on G
ij
are assumed to be additive,
while there is no interaction effects. Horiuchi selected k = 13 as reference category for the
cohort effects, i = 10 as reference category for the age effects, and j = 4 as the reference
category for the period effects. In addition, he assumed that the effect for the next youngest
cohort (1930-34, k = 12) was equal to that of the reference cohort (1935-39, k = 13). This way
he broke the perfect linear relationship between age, period, and cohort, so that the model
could be identified. The model was estimated using ordinary least squares (OLS).
By including only ages up to 84 years and thereby assuming that no debilitation
occurred after this age, and by including only four calendar years in the analysis, each
estimated cohort effect in model (4) might result in too large a standard deviation due to few
observations included. For this reason, model (4) needs to be extended. Model (5), which will
be presented below, is an extension of Horiuchi’s model (4) and differs in several ways. The
major difference is the number and selection of age-groups, periods, and cohorts. I use i =
1,2,…,14 for the five-year age groups 35-39,…,100-104 and j = 1,…,17, for the single
calendar years 1917, 1922, 1927, …, 1992, and 1997. I also use k = 1,…,15 for the five-year
cohorts born between 1865-69 and 1935-39. I use death rates for single calendar years 1917-
97 and one-year age groups 35-104 to estimate the model. The total number of observations
in the data material is 238. The all-cause death rates from 1917 to 1994 are from Mamelund
and Borgan 1996, while the all-cause death rates from 1995 to 1997 are from Statistics
Norway’s (SSB) online database (see http://www3.ssb.no/statistikkbanken/). The average
number of observations for each five-year cohort is 11.2, with a maximum of 14 observations
(followed from 35-39 years to 100-104 years) for the four five-year cohorts born between
1880 and 1899, and with a minimum of observations for the cohorts born 1925-29 (8), 1930-
34 (7), and 1935-39 (6).
The suggested model is
2
ij 1 2
G = + i+ i + + +
jkij
α
ββ γ δ ε
, (5)

10
where α is a constant, and γ and δ are period and cohort effects respectively. When estimating
Horiuchi’s model (4) on the Norwegian data it was found that the age effects increased
linearly with age for males and females. Thus the age effects could be parameterized as a
linear function of age. In other words, it was assumed that mortality increases exponentially
with age (Gompertz curve). To take into account the possibility of slower increase,
stabilization or even a decline in mortality among the oldest old due to selection effects, a
quadratic effect of age was added. Hence the term
β
2
i
2
in model (5). Given this
parameterization the perfect relationship between the three variables age, period, and cohort
was broken, and the model could be identified without identification constraints. Thus, model
(5) estimates only two coefficients for the age-effects as compared to 10 in Horiuchi’s model
(4), thereby sharpening all parameter estimates in model (5). In model (5), j = 1 (1917) is
selected as reference category for the period effects, and k = 1 (1865-69) as the reference
category for the cohort effects. Because the G
ij
’s are not observed, but estimated, weighted
least squares (WLS) estimation was used to estimate model (5), using the inverse value of the
estimated variance for each G
ij
as weight (Horiuchi used OLS).
1
The parameter estimates for
the effect of age, period or cohort on G
ij
in model (5) did not change significantly whether the
analysis started with one of the calendar years 1917, 1918, 1919, 1920, or 1921 (and ended
with one of the calendar years 1997, 1998, 1999, 2000, or 2001).
The estimated 5-year cohort effects (δ’s) of model (4) follow a straight line from one
cohort to the next for males and females, and do not confirm the hypothesis of high relative
mortality of cohorts born about 1900 (Figure 6). Not a single of the cohort effects, however,
differed significantly from zero as model (4) estimated relatively many parameters (23)
compared to the number of observations (40). The age effects were all negative and
statistically different from that of the reference category (i = 80-84 years), and increased as
expected with age (estimates not shown here). The period-effects (γ’s) are in accordance with
a previous study on Norwegian mortality (Mamelund 1996); mortality declined for women,
but stagnated for men in the 1960s (estimates not shown here).
In Figure 7 the estimated 5-year cohort effects (δ’s) from model (5) are presented.
This time, the cohort effects do not follow a straight line from one cohort to the next. Instead,
1
An approximate expression for the variance of the logarithm of the death rate is found as follows.
Assume that the survival probability l(t)= 1+bt is a linear function of time t through a given calendar
year, where 0<t<1 and b<0. The Maximum Likelihood estimator for b and the death rate m are
ˆ
/bDN=−
and
1
ˆ
/( )
2
mDN D=−respectively, where D is the observed number of deaths in a
given calendar year, and N is the number of persons alive at t=0. The variance of
ˆ
b
equals

11
there is clear cohort differentiation with maximum cohort effects or highest relative mortality
occurring for women born 1890-99 and for men born 1900-09, and with gradually decreasing
cohort effects for older and younger male and female cohorts. The cohort effect for men born
1905-1909 is statistically different from the cohort effects of those born during the 1870s (at
0.01 level), the 1880s (at 0.05 level), the five-year period of 1890-94 (at 0.1 level), as well as
those born in the 1930s (at 0.05 level). The cohort effects for those born 1895-1904 and 1910-
1929 do not differ significantly from the maximum cohort effect at 0.1 level (calculations not
shown here). Nevertheless, the results show that it is highly unlikely that the concave curve
for the cohort effects with maximum occurring for men born about 1900 are due to mere
chance. Of the oldest female cohorts, only those born 1870-74 (0.05 level) differ relatively
strongly from that of the maximum cohort effect estimated for those born 1890-94 (the 1875-
79 cohort is significantly different at 0.2 level). All cohort effects for women born after 1904,
however, differ strongly from that of the maximum cohort effect for women born 1890-94 (at
0.01 level or lower).
The reason why the K(x) analysis for women did not show a similar cohort
differentiation as that identified in the APC-analysis might be that it only controls for age and
cohort, and not for the three dimensions of age, period, and cohort simultaneously. Cohort
effects may for instance be counterbalanced by the general decline in period mortality
throughout the period studied.
[Figures 6, 7, 8, and 9 approximately here]
The estimates of the parameters β
1
and β
2
that parameterize the age-effects for males and
females were all statistically significant at 0.01 level. Figure 8 gives the estimated effects β
1
i
+ β
2
i
2
; The death rates increases from one five-year age group to the next by an average of 20
and 23 per cent for men and women respectively at ages below 70, while the average increase
in the death rates at ages 70 and above is 23 and 35 per cent for men and women respectively.
When removing the constant (α) when estimating model (5), the crossover in the age-effects
by sex seen in Figure 8 disappears. The period-effects (γ’s) are in accordance with what we
already know (Figure 9). Men on the one hand, experienced significant mortality decline from
1917 until the late 1950s. The decline stagnated in the 1960s (the differences between the
1950s and the 1960s not significant at 0.1 level), but then it continued into the late 1970s,
3
()/()NDDN− (Keilman and Gill 1986). By the Delta-method we find an approximate expression
for the variance of the log of the estimated death rate
ˆ
m
as
2
()
ˆ
(log )
(/2)
NN D
Var m
DN D
−
≈
−

12
1980s, and 1990s. Females on the other hand experienced a significant decline in mortality
from 1917 until 1977, while mortality increased significantly from 1977 and into the 1980s
and 1990s (at 0.01 level). Because of this, the difference between male and female life
expectancy decreased in the late 1980s and the 1990s (Mamelund 1996). One of the reasons
why this happened may be that women have adapted the smoking behavior of men, while
fewer young men started to smoke and long-time smokers was able to quit, with the result that
CHD mortality declined faster for men than for women (Pampel 2002).
As discussed in
section 2.2, however, Azambuja and Duncan (2002) have suggested that the degree in which
Spanish Influenza 1918-19 infected the different cohorts might be a competing or additional
factor explaining the changing risks of CHD mortality.
3.3. Discussion of results
The analysis showed that Norwegian males and females born about 1900 experienced
significantly higher mortality in middle and old ages than neighbor cohorts. The results
confirm the finding of another study on the Norwegian data (Gómez de León 1991). The
same male and females cohorts were also identified as high mortality cohorts in Germany
(only men), Austria, Hungary, Italy, France, and Poland, all belligerent countries during WW
I (Horiuchi 1983; Boleslawski 1985; Caselli et al. 1987; Caselli and Capocaccia 1989; Caselli
1990; Wilmoth et al. 1990). The results in this paper show that Horiuchi (1983) was wrong
insisting that only male cohorts born around 1900 in belligerent countries have experienced
high relative mortality in middle and old life. Clearly, the search for causes that can explain
the cohort differentiation in neutral countries, and probably also belligerent countries, must go
beyond the indirect and direct effects of WW I. Hourichi explained the relatively high average
mortality of men born about 1900 by high mortality of cardiovascular diseases in old age caused
by malnutrition experienced in adolescence. In a country such as Germany, which suffered
severe malnutrition (see last part of section 4.2), it is also reasonable to believe that
malnutrition played a relatively important role in shaping the identified pattern of male cohort
differentiation. However, there are also other candidates, for instance CHD mortality related
to cigarette smoking, mentioned by Horiuchi, but considered to be of less importance
compared to effects of WW I. Anyway, there seems to be no reason to believe that only
teenagers and young adults at the end of WW I should become the heaviest smokers, and that
the death rates of cardiovascular diseases would be boosted for a particular group of cohorts
relative to that of preceding and succeeding cohorts. Spanish Influenza is a third candidate,
relevant for both neutral and belligerent countries.
In his K(x) analysis, Horiuchi did not find cohort differentiation for German females.
It seems as if this was the reason why Horiuchi did not further investigate female mortality by
controlling for age, period, and cohort effects simultaneously. Instead, Horiuchi suggested

13
that women, who on average have more body-fat than men, were better able to bear a period
of malnutrition without experiencing long-term effects on health and mortality. Wilmoth et al.
(1990) were not convinced by Horiuchi’s explanation; The authors believed that German
women also were debilitated by indirect effects of WW I, including malnutrition, but that the
weakening was obscured by female mortality decline in the post WW II period. For men, they
further suggested that the cohort differentiation in mortality was maintained or even made
more marked in Germany as mortality for men stopped to decline and slightly increased in the
1960s and 1970s (as was the case in most industrialized countries). This development support
an argument for not only using K(x) analysis, but that further analysis should be carried out
using APC-analysis. As seen in section 3.2 the period effects estimated for females in Norway
from 1917 until the late 1970s was not strong enough to counterbalance the cohort
differentiation found for the women born about 1900. Horiuchi (1983) suggests that the
increase in CHD mortality of German men in the 1960s and 1970s may have been caused by
malnutrition experienced in the years 1914-18. In this paper it is suggested that Spanish
Influenza 1918-19 may be the most important cause why CHD mortality among men also
stopped to decline and slightly increased in Norway in the 1960s. According to the latter
assumption, the increase in CHD mortality is thus not a factor, which “maintain” the
weakening of the cohorts born about 1900, but instead a direct result of living through a
period with Spanish Influenza and malnutrition in adolescence.
Horiuchi (1983) did not find a clear pattern of cohort differentiation among men in
neutral Sweden and Japan using the K(x) tool (women were not analyzed). The reason why
Horiuchi did not find cohort differentiation for men in Sweden and Japan nor cohort
differentiation for women in Germany might be that he assumed that the same male cohorts
that were debilitated in belligerent Germany, especially that of the five-year cohort of 1899-
1904, also were debilitated in all other countries, and that German males and females were
equally affected. There are several problems with this assumption. First, the center of high
mortality generations may vary from country to country and between the sexes. As seen in
section 3.2, the center of the high mortality generations of Norwegian females was ten years
later than that of their male counterpart. Wilmoth et al. (1990) identified the same sex
difference as was seen in Norway for France. The reason why the debilitated female cohorts
(1890-1899) were older than their male counterpart (1900-1909) may be that women were
extremely vulnerable to complications in connection with pregnancy and birth when infected
by Spanish influenza (Mamelund 2003). Also, the reason why the cohort differentiation in all-
cause mortality identified for Germany, France and for Norway is more pronounced for men
than for women might be that the incidence of Spanish influenza 1918-19 was greater among
men than among women, which is believed to be the reason why CHD mortality in the latter
part of the twentieth century was greater for men than for women. Second, although Horiuchi

14
found no visible cohort effects in the data before 1959 for German males, that is 40-45 years
after the cohorts of interest experienced deteriorating living conditions and hardship due to
WW I (and Spanish Influenza), it is a dubious assumption that the expected negative effects
on health and mortality do not show up in the data before 1959 for German females or for
males in neutral countries (i.e. not including data in the analysis before 1959). This is
confirmed by the smoothed one-year K(x) analysis in the present paper (results not shown),
which shows that Norwegian men born between 1895 and 1910 had higher mortality than
neighbor cohorts, not only in the 1960s, but also the 1950s. Third, by including only ages up
to 84 years and thereby assuming that no debilitation occurred after this age, and by including
only four calendar years in the analysis, each estimated cohort effect might have too large a
standard deviation due to few observations included. (The author of the present paper is aware
that 20 years have passed since Horiuchi published his paper, and that observations in period
1983-2003 were not available to him (see section 3.2)). As mentioned in section 3.1, however,
no striking cohort effect was found for male cohorts born about 1900 after 1970 when they
where 70 years (results not shown). A closer look at different causes of death, for example
suicide, CHD and Parkinson’s disease may have given better understanding of the
phenomenon studied, but was not carried out here.
The estimates from the refined and extended APC-model applied in this paper (see
expression (5)) are clearly sharper and trustworthier than the estimates from Horiuchi’s APC-
model (se expression (4)). The estimates are sharper because the number of observations is
six times higher (238 compared to 40), while the number of parameters estimated does not
differ much (33 compared to 23). This gives many more degrees of freedom (205 compared
to 17). The use of weights in the estimation of model (5) probably made the parameter
estimates trustworthier.
4. Discussion of some confounding factors
As discussed in section 3.3, one possible reason other than Spanish Influenza why the cohorts
born about 1900 have experienced relatively high later life mortality may be cigarette
smoking. In the following, three more hypotheses explaining the relatively high mortality
found for Norwegian cohorts born about 1900 are presented. It is discussed why each of the
three causes may have had an independent debilitating effects on the health of the cohorts
considered net of the effect of Spanish Influenza, whether they affected males and females
differently, and whether a specific cause, for example food shortage and rationing, during
WW I was severe or widespread enough to have deteriorating effects on later life health and
thereby the risk of premature death.

15
4.1. War accidents and post-traumatic stress disorders among WW I survivors in later
life
When considering that enormous number of soldiers from the belligerent nations in Europe
was wounded, exposed to toxic gas warfare, and experienced poor sanitary conditions and
nutritional deprivation, it is not surprising that it has been found the male cohorts directly
involved in combats experienced relatively high mortality in later life (Horiuchi 1983;
Boleslawski 1985; Caselli et al. 1987; Caselli and Capocaccia 1989; Caselli 1990; Wilmoth et
al. 1990). Of British cohorts born 1862-96 and of French cohorts born 1869-99, 60 per cent
were finally mobilized (Winter 1977; Wilmoth et al. 1990). Of all men in military age 80 per
cent were mobilized in Germany (Gregory 1997). Of the British soldiers, one in eight was
killed and more than one in four was wounded, and of the German soldiers 15 per cent were
killed. French cohorts born 1892-95 were reduced by more than a fourth due to battlefield
deaths.
In this section, evidence is presented that relatively few Norwegian men born about
1900 was injured or died in military accidents during and shortly after WW I. It is therefore
believed that the direct effects of WW I on male Norwegian cohort mortality must have been
small. The following discussion is not relevant for women as even fewer women were victims
of military accidents.
The Norwegian navy did not have to take part in combats to secure the Norwegian
neutrality or to protect the Norwegian merchant fleet. The only casualties registered in the
years 1914-20 are due to accidents. Four navy soldiers and six civilians, for instance, lost
their lives when navy soldiers were detonating mines (Marinens admiralstab 1940). There are
no reports of people wounded working on this dangerous task. The number of Norwegians
that volunteered for combats and were wounded or killed at the front is unknown to the
author, but the number is believed to be too small to have significant effect on the
phenomenon studied.
A substantial number of Norwegian seamen in the Norwegian merchant fleet and
Norwegian seamen on foreign ships were injured and died in military accidents, especially
after the onset of the all-out submarine war in April 1916. It is estimated that 300 or 0.1 per
cent of the seamen was physically disabled and that 1,892 died due to military accidents in the
years 1914-20 (Sjømennenes Minnehall, Stavern; Riksforsikringsanstalten 1923). Fear of
being dragged into the war, being recruited for frontier guard duty, and of being physically
injured or killed if torpedoed or mined may also have psychologically disabled sailors as well
as the navy conscripts in later life. However, the number of first-time hospitalized patients at
Norwegian asylum hospitals with mental disease linked to these reasons was only 17 over the
period 1914-20 (Medisinaldirektøren 1916, 1917, 1920a-b, 1921, 1923a-b). This figure is
small, but there are reasons to believe that many persons with equally strong or milder stress

16
disorders compared to those hospitalized did not see a psychiatrist. Askevold (1976), for
example, found that one third of the Norwegian sailors in the merchant navy, which survived
WW II, were psychologically disabled and on invalid pension in the late 1970s. Most suffered
from the “War Sailor Syndrome”, which consisted of two parts, the first being a non-neurotic
anxiety repeating the terrors of war, and the second being a brain damage caused by the
constant fear of death. Of the two-thirds that were not disabled, many were not able to take up
their work at sea and had to go ashore to low-status, low-wage jobs. If it is assumed that also
a third of the WW I veteran seamen developed the “War Sailor Syndrome” in later life,
10,000 Norwegian seamen were psychologically disabled after WW I. Only 3.4 and 2.5 per
cent of the male high mortality cohorts of 1901-05 and 1891-1900, however, were seamen by
occupation respectively in 1920 (Calculated from SSB 1924). This means that only about one
per cent of the cohort born 1891-1905 can have been debilitated due to psychological after-
effects due to service in the merchant fleet during WW I.
The recruitment to the Norwegian navy was basically based on compulsory military
service of men 20-22 years old. The conscripts on duty in the navy during the years 1914-20
were thus born between 1893 and 1900. In the period 1914-20, 11,474 conscripts or less than
one per cent of the cohorts considered were recruited for active service in the navy (SSB
1920ab, 1921c, 1922a-c; Forsvarsdepartementet 1919). If one assumes that also one third of
the navy veterans experienced the “War Sailor Syndrome” in later life, only a little more than
one per cent of the cohorts born about 1900 could have been psychologically debilitated in
later life due to active service in the navy or in the merchant fleet during the war. This share is
so small that it cannot have caused the cohort effect identified in the analysis for men born
about 1900.
There are also studies that have reported that WW II debilitated (primarily males, but
also to a certain degree females) adolescent and young adults survivors in Japan, Germany,
the former Soviet Union, France, and Italy (Okubo 1981; Horiuchi 1983; Anderson and Silver
1989; Caselli et al. 1987; Caselli and Capocaccia 1989; Caselli 1990; Wilmoth et al. 1990).
Could WW II partly explain the debilitation of the male cohorts born about 1900 in Norway?
More than 10,000 war-deaths were reported in Norway 1940-45, of which nine out of ten
were men (Backer 1948). If one supposes that it were the healthiest and the fittest men that
were selected for duty and were injured or died in the war, and that men with the poorest
health were found unfit for service and therefore survived the war, this would on average lead
to a debilitation of the male cohorts considered. The soldiers that survived military combats,
and the soldiers and seamen that survived military accidents, probably also suffered poor
mental and physical health and thus higher mortality in later life (Askevold 1976). Those born
about 1900, however, were on average 40-45 years in the period 1940-45, far from the
average age of excess deaths due to the war (here assumed to be the average age of a mentally

17
or physically wounded soldier or seaman), which was 25 years (Gómez de León 1991). This
factor might therefore contribute to explain the relatively high cohort effects found for males
born 1910-14 and 1915-19 found in the APC-analysis (Figure 7). As it was mostly men that
were affected by military casualties, either by being a soldier or a sailor during WW II, it is
not surprising to find that the cohort effects for the corresponding female cohorts were not
similarly high. As suggested by Wilmoth et al. (1990) and Caselli and Capocaccia (1989) in the
case of the Italian and French population, an equally important explanation behind the high
later-life mortality might be that large proportion of the cohorts born 1918-1919 were exposed
to the Spanish Influenza virus in utero or as infants.
4.2. Shortage and rationing 1914-19
In this section it is discussed whether the shortage and rationing of food during WW I might
have affected immediate or later life mortality of Norwegians born about 1900 positively. The
evidence that is presented, however, gives little support to this hypothesis.
The daily calorie intake did not decline in Norway during the years 1914-17 as food
that was less expensive, more abundant and equally nutritious substituted consumption of
expensive food of short supply: Fish, whale oil margarine, whole milk, and whole meal bread
replaced meat, butter, eggs, skimmed milk and white bread (SSB 1917; SSB 1918). In
September 1917 (butter) and January 1918, however, milk, sugar, coffee, tea, bread,
macaroni, peas, beans, barley, and oat were rationed because of increasing shortage and
inflation. How did the rationing of food affect the weight and health of cohorts born about
1900? Pupils between 10 and 15 years (born 1903-1908) from two elementary schools in the
city of Bergen were found to suffer from malnutrition and weight loss during the spring of
1918, but this did not lead to an increase in absence from school due to sickness (Looft 1919).
In a study of height and weight among 12-18 years pupils (born 1900-06) in one secondary
school in Kristiania, Schreiner and Schreiner (1922) found that the body-mass index fell
during the spring of 1918, and that this might be explained by the rationing of food. The
emaciation, however, was fully compensated for most pupils during stays at summer camps or
visits to relatives in the countryside, where the supply of milk and other foodstuff was better
than in the cities.
Several socialist newspapers claimed that the Norwegian neutrality-keeping soldiers
were underfed due to the poor rations served in the army during the war, and for this reason
the exercises should be stopped and the recruits sent home (Sjefen for hærens sanitet 1919).
Because of this critique, the 20-22 year old recruits (born 1897-99) were weighted at the
beginning of the recruit-school and every second week in 1918. The analysis of the weights
showed that in every battalion, average weight increased by 0.5 to 3.0 kilogram per recruit per
year. The average daily calorie-intake was 3,700 gram a recruit compared to 3,000 gram for

18
the average worker and low paid public clerical officer in the two cities of Kristiania and
Bergen. The rationing of food was gradually lifted in the five month period from March to
July 1919 and during the summer of 1919 most food articles were in normal pre-war supply
(SSB 1920c; SSB 1921). The average calorie consumption and consumption of earlier
rationed food articles increased markedly after this, but as people learned that demand was
fulfilled, as normal, consumption went down (SSB 1921).
From the above presentation, it seems reasonable to conclude that although the period
of heavy rationing in 1918-19 affected food intake of cohorts born about 1900, it is not likely
that the rationing was serious and long lived enough to leave lasting marks which can give a
significant rise in later-life mortality. Also, it has been reported elsewhere that the mortality
during the first half of 1918, which was a period of relatively heavy rationing of food, was
lower than the average of the first halves of the three years 1915-17 which was characterized
by increasing shortage of food, but no rationing (Mamelund 2003). A comparison with
Germany also shows how relatively well-off Norway was when it comes to nutritional
problems during and after WW I. Grebler and Winkler (1940) have estimated that the German
population in 1920 only had 50 per cent of the supply of the most need food items compared
to 1913. The average Frankfurt resident subsisted on roughly 1,500 calories a day in the
summer of 1918, with 2,500 calories daily considered normal, and the average weight loss a
person approached 23 per cent of pre-war pounds (Fritz 1992).
4.3. Mass emigration and selection
Gómez de León (1991) was tempted to explain the high relative mortality in later life of the
Norwegians that were adolescents in 1918 by debilitation effects of the Spanish Influenza
Pandemic. But he also assumed that a late wave of predominantly male overseas emigration,
which peaked in the mid 1920s, might have introduced additional selection if the most
selective of the survivors left. The question of selection effects imposed by emigration is
surprisingly not discussed in the previous papers on this topic (Horiuchi 1983; Boleslawski
1985; Caselli et al. 1987; Caselli and Capocaccia 1989; Caselli 1990; Wilmoth et al. 1990).
Gómez de León’s hypothesis is appealing. Bævre (2001), for example, has found that the
massive overseas emigration 1846-1939 raised mortality of Norwegian cohorts, male
mortality more than female mortality because men migrated more than women, suggesting a
strong selection mechanism. Moreover, Bævre et al. (2001) have documented that by age 50
male and female cohorts born 1865-85 were reduced by 30 and 20 per cent due to emigration
respectively, while male and female cohorts born 1890-1900 were reduced by only 13-19 and
7-13 per cent due to emigration respectively. If those with the best health were selected for
emigration, this would on average leave more persons with poor health among those born
1865-85 as compared to the average health of those born 1890-1900. The 1865-85 cohort is

19
thus assumed to have higher mortality in later life than the cohort born 1890-1900 due to
emigration.
5. Conclusion
The analysis showed that male and female cohorts born about 1900 in Norway, which was
neutral during WW I, experienced significantly higher all-cause mortality in middle and old
ages than preceding and succeeding cohorts. This finding demonstrates that Horiuchi (1983)
who has insisted that this cohort effect is exclusive for males in the countries that were
belligerent during WW I, for instance Germany, Poland, Austria, Hungary, Italy, and France,
seems to be wrong. Clearly at least, the search for a possible explanation in neutral countries
must go beyond the direct (soldiers wounded physically and mentally) and indirect effects
(shortage and rationing of food) of WW I on later life mortality by sex. Hourichi (1983)
explained the high relative mortality of German men born about 1900 by high mortality of
cardiovascular diseases in middle and old age caused by malnutrition experienced in
adolescence during WWI. The finding in this paper demonstrates that Horiuchi’s explanation
may be incomplete. This paper has argued that Spanish Influenza may be the most important
candidate explaining the debilitation of the male and female cohorts considered in neutral
Norway. The cohorts born 1890-1910 experienced the highest incidence, but the lowest
immediate death rates of Spanish Influenza 1918-19. The disease did thus kill relatively few
but marked relatively many of the persons belonging to these cohorts. Mental distractions,
hepatitis, ear illnesses, deafness, blindness, and baldness are examples of the peculiar after-
effects that the survivors faced in later life. The survivors seem to have had high risks of
committing suicide immediately or shortly after 1918-19. The influenza survivors may also
have had higher risk of dying from the Sleeping Sickness, Parkinson’s disease, and coronary
heart disease in later life.
It has been shown that neutral Norway fits well as a test case to study effects of Spanish
Influenza on later life mortality as it is partly counterbalancing effects of WWI. One of the
indirect effects of the war that may have caused an increase in later life mortality is shortage
and rationing of food. Although some of the Norwegian cohorts considered experienced a
drop in weight during the relatively heavy rationing of food spring of 1918, there was little or
no immediate increase in morbidity or mortality. In the years 1915-17 the daily calorie intake
was maintained as food that was less expensive, more abundant and equally nutritious
substituted consumption of expensive food of short supply. It is therefore believed that the
shortage and rationing of food is confounding Spanish Influenza relatively little as the major
factor debilitating the cohorts considered. Although the Norwegian navy did not have to take
part in combats to secure the Norwegian neutrality or to protect the Norwegian merchant
fleet, the paper has documented that post-traumatic stress disorders was present among the

20
neutrality-keeping soldiers and the seamen in the merchant fleet, who survived WW I. Thus,
direct effects of WW I are a possible confounder of Spanish Influenza. However, as it is
estimated that only a little more than one per cent of the men born about 1900 can have
experienced poorer health and thereby higher mortality due to this event, it is believed to be
of relatively little importance compared to Spanish Influenza.
Another possible confounding factor, surprisingly not dealt with in previous studies
on this topic, is selection effects imposed by periods of mass emigration. Did the late wave of
predominantly male overseas mass emigration, which peaked in 1923 and 1927, select those
with the best health among those born 1900, leaving more persons with poor health? The
answer might be yes. The reason why the cohort effects in neutral Norway were found to be
more pronounced for men than for women may be that more men than women emigrated. The
sex-difference in the cohort effects, however, may also be explained by the sex difference in
mortality of coronary heart disease; Men have always had higher CHD mortality than women,
but what was the reason for this? Two possible answers stands out; higher incidence of
Spanish Influenza among men than women in 1918-19 and/or that more men than women are
or have been cigarette smokers.
The APC-analysis applied in the paper, which utilizes no other covariates than age,
period and cohort to explain the variance in the death rates, cannot contribute to the explanation
of why cohorts born about 1900 were debilitated. It is possible though, to create and to discuss
plausible hypotheses that caused the cohort differentiation identified in the analysis. The answer
whether or why the cohorts considered were debilitated in later life due to Spanish Influenza
1918-19 can most likely only be found by linking the data on incidence by occupation, housing
standard, and socioeconomic status from the individual “influenza-censuses” carried out in the
years 1918-19 in Bergen, Norway, and in a number of cities in the United States and England
with modern individual death registers.
Acknowledgements
I am most grateful to Alberto Palloni, Nico Keilman, Eivind Bernhardsen, and especially
Shiro Horiuchi for helpful comments and suggestions to the paper. My thanks also goes to
The Center for Demography and Ecology (CDE) at the University of Wisconsin-Madison
which provided excellent working conditions while I wrote a preliminary version of the paper
fall 2001. Earlier versions of the paper have been presented at the Young Scholar’s Historical
Demography Conference, Leuven, Belgium, April 25-27
th
2002 and the annual meeting of the
Population Association of America, Minneapolis, Minnesota, 1-3 May 2003. The paper is part
of the research project Spanish Influenza and beyond: the case of Norway, which is

21
financially supported by the Norwegian Research Council and Department of Economics,
University of Oslo, to whom I extend my thanks.
6. References
Anderson, B.A. and B.D. Silver. 1989. Patterns of Cohort Mortality in the Soviet Population,
Population and Development Review 15(3): 471-501
Askevold, F. 1976. War Sailor Syndrome, Psychother Psychosom 27(3-6): 133-8.
Azambuja, M.I.R., and B.B. Duncan. 2002. Similarities in mortality from influenza in the first
half of the 20th century and the rise and fall of ischemic heart disease in the United States: a
new hypothesis concerning the coronary heart disease, Cad. Saúde Pública 18(3): 557-77
[Rio de Janeiro].
Backer, J. 1948. Statistiske meddelelser, statistisk oversikt over krigsdødsfallene 1940-1945,
NOS, Statistisk sentralbyrå.
Basler, F.B., Reid, A.H., Dybing, J.K., Janczewski, T.A., Fanning, T.G., Zheng, H., Salvatore,
M., Perdue, M.L., Swayne, D.E, García-Sastre, A., Palese, P., Tauenberger, J.K. 2001.
Sequence of the 1918 pandemic influenza virus non-structural gene (NS) segment and
characterization of the recombinant viruses bearing the 1918 NS genes,
Microbiology 98(5): 2746-51.
Bævre, K. 2001. Two wrongs do not make a right: sex-differences in mortality in Norway 1846-
1939, Paper presented at IUSSP’s XXIVth General Population Conference, Salvador-Bahia,
Brazil, 18-26 August 2001.
Bævre, K., Riis, Christians, and T. Thonstad. 2001. Norwegian Cohort Emigration, J Popul
Econ 14:473-489.
Ben-Shlomo, Y., Finnan, F., Allwright S., and G. Davey Smith. 1993. The Epidemiology of
Parkinson's disease in the Republic of Ireland: observations from routine data sources,
Ir Med J 86:190-192.
Boleslawski, K. 1985. Roznice w umieranosci miedzy generacjami jako skutek wojen
swiatowych, Studia demograficzne, 4:51-7.
Britten, R.H., 1932. The incidence of epidemic influenza, 1918-19, Public Health Reports 47(6):
304-339.
Casals, J., T.S. Elizan and M.D. Yahr. 1998. Postencephalitic parkinsonism – a review,
J Neural Transm 105: 645-676.
Caselli, G. 1990. The Influence of Cohort Effects on Differentials and Trends in Mortality, In
Vallin, J, D’Souza, S., and A. Palloni (eds.), Measurement and Analysis of Mortality. New
Approaches. Oxford: Clarendon Press, pp. 229-249.
Caselli, G. and R. Capocaccia. 1989. Age, Period Cohort and Early Mortality: An Analysis of
Adult Mortality in Italy, Population Studies 43:133-153.
Caselli, G., J. Vallin, J.W. Vaupel, and A. Yashin. 1987. Age-specific mortality trends in France
and Italy since 1900: period and cohort effects, European Journal of Population 3(1):33-60.
Collier, R. 1974. The Plague of the Spanish Lady. The influenza Pandemic of 1918-1919.
London: Macmillan.
Collins, S.D., 1931. Age and sex incidence of influenza and pneumonia morbidity and mortality in
the epidemic of 1928-29 with comparative data for the epidemic of 1918-19, Public Health Reports
46(33): 1909-1937.
Crosby, A. 1989. America’s Forgotten Pandemic. The Influenza of 1918. Cambridge:
Cambridge University Press.
Davis, J.L., Heginbottom, J.A., Annan, P.A., Daniels, R.D., Berdal B.P., Bergan, T., Duncan,
K.E., Lewin P.K., Oxford J.S., Roberts, N., Skehel, J.J., and C.R. Smith. 1999. Ground
Penetrating Radar Surveys to Locate 1918 Spanish Flu victims in Permafrost, Journal of
Forsenic Sciences 45(1): 68-76.
Det civile medisinalvesen. 1921-1925. Sundhetstilstanden og medisinalforholdene 1917-1921,
NOS VII.3, NOS VII.58, NOS VII.108, NOS VII.138, NOS VII.152. Kristiania: Aschehoug
& Co.
Forsvarsdepartementet. 1919. Rekruttering for den norske hær. NOS VI. 146. Kristiania:
Aschehoug & Co.
Fritz, G. 1992. Frankfurt, In van Hartesveldt, F.R. (ed), The 1918-1919 Pandemic of Influenza.
The Urban Impact in the Western World. New York: Edwin Mellen Press, pp. 13-32.

22
Gibbs, M.J., Armstrong, J.S., and A.J. Gibbs. 2001. The Haemagglutinin Gene, but not the
Neuraminidase Gene, of Spanish Flu Was a Recombinant, Phil. Trans. R. Soc. Lond. B.,
356: 1845-1855
Gómez de León, J. 1991. Empirical Models to Fit and Project. Time Series of Age-specific
Mortality Rates, Report from a Nordic seminar on prognoses. Drammen, Norway 24-26
April 1990. Tekniske Rapporter no. 53. Nordisk Statistisk Sekretariat.
Great Britain Minsitry of Health. 1920. Report on the 1918-19 Pandemic of Influenza, Reports
on Public Health and Medical Subjects, No. 4, London, Great Britain Ministry of Health.
Grebler, L. and W. Winkler. 1940. The Cost of the World War to Germany and to Austria-
Hungary. New Haven: Yale University Press.
Gregory, A. 1997. Lost Generations: the impact of military casualties on Paris, London, and
Berlin, in Winter, J.M. and J-L. Robert (eds.) Capital Cities at War, Paris, London, Berlin
1914-1919, Cambridge: Cambridge University Press.
Hanssen, O. 1923. Undersøkelser over influenzaens optræden specielt i Bergen 1918-1922.
Arbeider fra Den medicinske Avdeling av Haukeland sykehus. Skrifter utgit ved Klaus
Hanssens Fond. Nr. III. Bergen: A.S. John Griegs Boktrykkeri og N. Nilssen & søn.
Horiuchi, S. 1983. The long-term Impact of War on Mortality of World War I Survivors in the
Federal Republic of Germany, Population Bulletin of the United Nations 15:80-92.
Huserbråten, K. 1993. “Helse”, in Sosialt Utsyn, Statistiske Analyser, Statistisk Sentralbyrå.
Johnson, N.P.A.S. and Mueller, J. 2002. Updating the Accounts: Global Mortality of the 1918-
1920 Spanish Influenza Pandemic, Bulletin of the History of Medicine 76: 105-115.
Keilman, N. and R. Gill 1986. On the Estimation of Multidimensional Demographic Models
with Population Registration Data. NIDI Working Paper no. 68. The Hague: Netherlands
Interuniversity Demographic Institute.
Laidlaw, P.P. 1935. Epidemic Influenza: a virus disease, Lancet i: 1118-1124.
Looft, C. 1919. Kostrationering og barnehelse, Tidsskrift for den norske legeforening 39(2-3): 57-65,
95-103.
Lund, C. 1997. Encephalitis lethargica – Sykdommen som forsvant. Hoveoppgave i
Mikrobiologi. Universitetet i Trondheim.
Mamelund, S-E. 1996. Redusert kjønnsforskjell i forventet levealder de siste årene,
Samfunnsspeielt 10(3): 13-17.
Mamelund, S-E. 1998. Spanskesyken i Norge 1918-1920: Diffusjon og demografiske
konsekvenser. Hovedoppgave i Samfunnsgeografi høsten 1998. Oslo: Institutt for Sosiologi
og Samfunnsgeografi, Universitetet i Oslo [Master thesis]
Mamelund, S-E. 1999. Spanskesyken i Norge: Kostnadene og konsekvensene, Samfunnsspeilet
13(6): 22-31. Oslo: Statistisk sentralbyrå.
Mamelund, S-E. 2003. Can the Spanish Influenza pandemic of 1918 explain the baby-boom of
1920 in neutral Norway? Memorandum no. 01, Department of Economics, University of
Oslo.
Mamelund, S-E. and J.K. Borgan. 1996. Kohort- og periodedødelighet i Norge 1846-1994,
Reports 96/9. Statistics Norway. Oslo-Kongsvinger.
Mamelund, S-E. og B.G. Iversen. 2000. Sykelighet og dødelighet ved pandemisk influensa i
Norge, Tidsskr nor lægeforen 130(3): 53-56.
Marinens admiralstab. 1940. Marinen. Nøytralitetsvernet 1914-1918, samt nøytralitetsvernets
avvikling. Oslo: Industritrykkeriet A/S.
Martyn, C.N. 1997. Infection in childhood and neurological diseases in adult life,
Br Med Bull 53(1): 24-39.
Matheson Commission. 1929. Epidemic encephalitis. Etiology, epidemiology, treatment. New
York: Columbia University Press.
Okubo, M. 1981. Increase in Mortality of Middle-Aged Males in Japan. Tokyo: Nihon
University Population Research Institute.
Oxford, J.S., Sefton, A., Jackson, R., Johnson, N.P.A.S., and R.S. Daniels. 1999. Who’s that
lady?, Nature Medicine 5(12):1351-1352.
Phillips, H. 1990. Black October: The Impact of the Spanish Influenza Epidemic of 1918 on
South Africa. D.Phil. Thesis, Pretoria: University of Cape Town.
Pampel, F. C. 2002. Cigarette Use and the Narrowing Sex Differential in Mortality, Population
and Development Review, 28(1): 77-104.
Poskanzer, D.C. and R.S. Schwab. 1963. Cohort analysis of Parkinson’s syndrome: evidence for
a single etiology related to subclinical infection about 1920, J Chron Dis 16: 961-973.
Ravenholt, R.T. and W.H. Foege. 1982. Before our time. 1918 influenza, Encephalities
Lethargica, Parkinsonism, The Lancet (16 Oct.): 860-864.

23
Reid, A.H., Fanning T.G, Hultin J.V., and J.K. Tauenberger. 1999. Origin and evolution of the
1918 Spanish influenza virus hemagglutinin gene, Proc Natl. Acad. Sci, USA, 96: 1651-
1656.
Reid, A.H., Fanning T.G., Janczewski, T.A., and J.K. Tauenberger 2000. Characterization of the
1918 Spanish influenza virus neuraminidase gene, Proc Natl. Acad. Sci, USA. 97: 6785-
6790.
Reid, A.H., Tauenberger, J.K., and T.G. Fanning 2001. The 1918 Spanish influenza: integrating
history and biology, Microbes and Infection 3: 81-87.
Reid, A.H., Fanning T.G., Janczewski, T.A., Shermann, M., Tauenberger, J.K., and J.K.
Tauenberger 2002. Characterization of the 1918 “Spanish” Influenza Virus Matrix Gene
Segment, Journal of Virology 76(21): 10717-23
Rice, G. 1988. Black November. The 1918 Influenza Epidemic in New Zealand. Wellington:
Allen &Unwin/Historical Branch.
Riksforsikringsanstalten. 1923. Sjømannsforsikringen for året 1920. Ulykkesforsikring for
sjømenn / Fiskeriforsikringen for året 1921 (1/1-31/3). Ulykkesforsikringen for fiskere m.v.
NOS VII. 75. Kristiania: Aschehoug & Co.
Sacks, O. 1999. Awakenings, New York: Vintage Books.
Shortridge, KF. 1999. The 1918 Spanish Flu: Pearls from swine?, Nature Medicine 5(4): 384-
385.
Sjefen for hærens sanitet. 1919. Kostholdet under vaabenøvelserne 1918. Hygieniske
meddeleser og social medicin. Tidsskr nor lægeforen 39(7): 293-294.
Sjømennenes Minnehall, Stavern (http://www.minnehallen.no/historie.htm
).
Statistisk sentralbyrå. 1917. Dyrtidens virkninger på levevilkaarene. 1ste del. NOS VI 105. Kristiania:
Aschehoug
Statistisk sentralbyrå. 1918. Dyrtidens virkninger på levevilkaarene. 2den del. NOS VI 124. Kristiania:
Aschehoug
Statistisk sentralbyrå. 1919. Husholdningsforbruket før og under krigen, Meddelelser fra Det
Statistiske Centralbyrå 1918, 36(12): 129-144. NOS. Kristiania.
Statistisk sentralbyrå 1920a-b, 1921c, 1922a-c. Rekruttering for den norske hær 1914-1920. NOS VI.
168, NOS VI. 193, NOS VII. 27, NOS VII. 33, NOS VII. 42, NOS VII. 69. Kristiania: Aschehough
& Co.
Statistisk sentralbyrå. 1920c. Undersøkelser over husholdningsforbruket i perioden 27 Januar til
20 april 1919, Meddelelser fra Det Statistiske Centralbyrå, 37(11): 164-173. NOS.
Kristiania.
Statistisk sentralbyrå. 1921. Undersøkelser over husholdningsforbruket i tidsrummet 9 sept.
1918-7 sept. 1919, Meddelelser fra Det Statistiske Centralbyrå, 38(2-4): 41-50. NOS.
Kristiania.
Statistisk sentralbyrå. 1924. Folketellingen i Norge 1 desember 1920. Folkemengden fordelt
etter livsstilling, alder og ekteskapelig stilling. Tiende hefte, NOS VII. 111. Kristiania:
Aschehoug & Co.
Sydenstricker, E. 1931. The Incidence of Influenza among Persons of Different Economic
Status During the Epidemic of 1918, Public Health Reports 46(4): 154-170.
Tauenberger, J.K., Reid, A.H., Janczewski, T.A., and T.G. Fanning. 2001. Integrating historical,
clinical and molecular genetic data in order to explain the origin and virulence of the 1918
Spanish influenza virus, Phil. Trans. R. Soc. Lond. B, 356: 1829-1839.
Tauenberger, J.K., Reid, A.H., Kraft, A.E., Bijwaard K.E., and T.G. Fanning. 1997. Intial
genetic characterization of the 1918 Spanish influenza virus. Science, 275:1793-1796.
Tumpey, T.M., García-Sastre, A., Mikulasova, A., Tauenberger, J.K, Swayne, D.E., Palese, P.,
and C.F. Basler. 2002. Existing antivirals are effective against influenza viruses with genes
from the 1918 pandemic virus, Microbiology 99(21): 13849-54
Vaughan, W.T. 1921. Influenza. An Epidemiological Study. The American Journal of Hygiene.
Monopgraph Series. No. 1. Baltimore: The American Journal of Hygiene.
Wasserman, I.M. 1992. The Impact of Epidemic, War, Prohibition and Media on Suicide:
United States, 1910-1920, Suicide and Life-threatening Behavior 22(2): 240-254.
Wilmoth, J., J. Vallin and G. Caselli 1990. When Does a Cohort's Mortality Differ From What
We Might Expect?, Population, English Selection, 2: 93-126.
Winter, J.M. 1977. Britain’s Lost Generation of the First World War, Population Studies, 31(3):
449-466.
Aase, A., and I. Storm-Furru 1996. Nasjonalatlas for Norge, Hønefoss: Statens Kartverk.

24
Figure 1. Age-sex specific incidence rates of influenza in the city of Bergen, July-
September 1918 and October-December 1918 (male solid and female dotted)
Source: Hanssen 1923.
0
1
2
3
4
5
6
7
8
0-1 30-39 40-49 50-59 60-69 70-79 80-89
Deaths per 1 000
5-9
2-4
10-14
15-19
20-24 25-29
1918-1919
1917+1921
Figure 2. Age-sex specific death rates of influenza and pneumonia, Norway 1918-19
and 1917 and 1921 (male solid and female dotted).
Source: DCM (1921, 1922, 1923, and 1925).
0
50
100
150
200
250
300
350
400
0-9 10-19 20-29 30-39 40-49 50-59 60-69 70+
Age
Influenza cases per 1 000
July-September 1918
October-December 1918

25
Figure 3. Lethality of influenza and pneumonia in Norway 1918-19 (males solid and females
dotted).
1
Source: Calculated from DCM (1921, 1922) and Hanssen (1923).
1
The number of influenza and pneumonia cases by age and sex are estimated by multiplying age-sex specific incidence rates
from Bergen July 1918-March 1919 by the population in Norway 31.12 1918 by age and sex.
Figure 4. Rate of mortality change with age, K (x), for Norwegian males 1959, 1964, 1969
and 1974 using five-year age groups
Note: The downward-pointing
and
upward-pointing arrows mark the last
and the first of high mortality cohorts respectively.
0
2
4
6
8
10
12
0-9 10-19 20-29 30-39 40-49 50-59 60-69 70+
Age
Per cent of infected who died
0,020
0,025
0,030
0,035
0,040
0,045
0,050
0,055
40 45 50 55 60 65 70 75 80
Age
k(x)
1959
1964
1969
1974

26
Figure 5.
Rate of mortality change with age, K (x), for Norwegian females 1959, 1964, 1969
and 1974 using five-year age groups.
Figure 6.
Estimates of cohort effects (
δ
’s) for Norwegian males and females in model (4)
proposed by Horiuchi 1983, pp. 81-84.
0,020
0,025
0,030
0,035
0,040
0,045
0,050
0,055
40 45 50 55 60 65 70 75 80
Age
k(x)
1959
1964
1969
1974
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
0,3
0,4
0,5
0,6
1875-1879
1880-1884
1885-1889
1890-1894
1895-1899
1900-1904
1905-1909
1910-1914
1915-1919
1920-1924
1925-1929
1930-1934
1935-1939
Cohorts
Estimated cohort-effects
Males
Females Note that all effects are relative to that for the cohort
1935-39, which is the reference cohort

27
Figure 7. Estimates of the cohort effects (δ’s) for Norwegian males and females in model (5)
Figure 8. Estimates of the age effects (
β
1
i + β
2
i
2
) for Norwegian males and females in model
(5)
0,0
0,5
1,0
1,5
2,0
2,5
3,0
3,5
4,0
4,5
35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 85-89 90-94 95-99 100-104
Age
Estimated age effects
Males
Females
-0,15
-0,10
-0,05
0,00
0,05
0,10
0,15
0,20
1865-1869
1870-1874
1875-1879
1880-1884
1885-1889
1890-1894
1895-1899
1900-1904
1905-1909
1910-1914
1915-1919
1920-1924
1925-1929
1930-1934
1935-1939
Cohorts
Female cohort effects
0,00
0,02
0,04
0,06
0,08
0,10
0,12
0,14
0,16
Male cohort effects
Males (right axis)
Females (left axis)
Note that all effects are relative to that for cohort 1865-69,
which is the reference cohort

28
Figure 9. Estimates of the period effects (γ’s) for Norwegian males and females in model (5)
-0,45
-0,40
-0,35
-0,30
-0,25
-0,20
-0,15
-0,10
-0,05
0,00
1917 1922 1927 1932 1937 1942 1947 1952 1957 1962 1967 1972 1977 1982 1987 1992 1997
Calendar years
Estimated period effects
Males
Females
Note that all effects are relative to that for
1917 which is the reference period