© International Epidemiological Association 1999
International Journal of Epidemiology 1999;28:899–904
Printed in Great Britain
Bias related to the exclusion of the
economically inactive in studies on social
class differences in mortality
Pekka Martikainena,b and Tapani Valkonena
Background To assess how the exclusion of the economically inactive affects levels and trends
in social class differences in mortality among men and women at different durations of follow-up.
Methods
Records of the 1970, 1975, 1980 and 1985 censuses on Finnish men and women
aged 35–64 linked with records of all deaths during 1971–1990.
Results
Exclusion of the economically inactive population underestimates the class
differences in the total population by about 25% among men and 60% among
women. The bias does not disappear if the first 5 years of follow-up are excluded
and the bias can lead to erroneous conclusions about the trends in social class
differences in mortality.
Conclusions Analyses based on the economically active population may lead to significant
underestimation of social class differences in mortality, introduce biases in international comparison and may only partially capture the causal mechanisms underlying these mortality differences. Our results further show that although the bias
diminishes during the follow-up, it is by no means eliminated after the first
5 years. The underestimation of social class differences in mortality created by the
exclusion of the inactive should be more widely recognized and more accurate
data on previous occupations should be collected.
Keywords
Occupation-based social class, mortality differences, bias, economically inactive
Accepted
3 February 1999
Studies on socioeconomic differences in morbidity and mortality have employed a variety of indicators, for example level
of education, occupation-based social class, income, housing
tenure or car access. The strengths and weaknesses of all these
indicators have been discussed by several authors. 1–3 In
this study our focus is on one of these indicators, namely
occupation-based social class, which at least in Europe has been
the most commonly used indicator in studies on socioeconomic
mortality differentials.
A major problem in international comparative research on
social class mortality differentials is that different countries
have different occupational classifications and ways of grouping
occupations into broader social classes. Another major difficulty
lies in the absence of occupational information on economically
inactive people, for instance, disabled people, pensioners and
a Population Research Unit, Department of Sociology, University of Helsinki,
PO Box 33, 00014 University of Helsinki, Finland.
b International Centre for Health and Society, Department of Epidemiology
and Public Health, University College London, 1–19 Torrington Place,
London WC1E 6BT, UK.
homemakers.4 All or a significant proportion of these economically inactive people cannot usually be classified into social
classes and have often been excluded from the analyses.5–8 This
problem affects both studies of mortality and morbidity.
If the patterns and trends in social class mortality differences
are different for the economically active population compared
to the total population, any results based on the economically
active population will give a biased account of actual socioeconomic mortality differences. The extent of this bias cannot
be assumed to be constant, but it is likely to vary according to
sex, age, length of follow-up, country and other variables.9–13
In Finland the retrospective questions on former occupation
in the censuses of 1970 and 1975 allow a reliable study of mortality differences according to social class in the economically
active, inactive and the total population. In this study we use
these data to (1) assess how the exclusion of the economically
inactive affects social class mortality differences among men and
women at different durations of follow-up and (2) to examine
how the results on trends in social class differences in mortality
are influenced by the exclusion of the economically inactive
population.
899
900
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Methods
The data set compiled by Statistics Finland includes individuallevel census records for the years 1970, 1975, 1980 and 1985 for
the total population of Finland. These census records have been
linked to the records of all deaths in Finland from 1971 to
1990.14
In all censuses the economically active include people
who are working or are unemployed and seeking work. The
economically inactive include people receiving disability benefit
or old age pension, and family members (mostly housewives)
living on the income of other family members. In addition,
students and those whose source of income is not known have
been classified as economically inactive.
In the 1985 and 1980 censuses only current occupations were
asked. However, in the 1975 and 1970 censuses former occupations of both unemployed and the retired were also explicitly
asked. In order to use these data, occupational information from
all previous censuses have been linked to the records of the
censuses 1975, 1980 and 1985. With these data both the economically active and inactive can be classified into social classes
on the basis of one of the following: (1) own current occupation, (2) own former occupation, or (3) occupation of an associated person (e.g. head of household). The data do not allow us
to classify according to the longest held occupation or according
to lifetime ‘average occupation’.
People who were working, were classified according to their
current occupation as reported in the census. Economically active,
but temporarily unemployed people as well as economically
inactive pensioners were classified on the basis of information
on former occupation in the earlier censuses.
Housewives and other family members not working outside
the home were classified according to the occupation of the
head of household. Among men this group was very small and
among women it was about 12% of all 35–64-year-old women
in the period 1981–1985.3 Housewives were classified on the
basis of head of household, because previous analyses suggest
that we can only find a former occupation for less than half of
these women. In general, there are both advantages and disadvantages from classifying people on the basis of head of household in comparison to using the participants’ own former or
current class.15,16 However, among married Finnish women—
for whom this method of allocation is most relevant—social
class differences in mortality according to own or spouse’s
occupation-based social class are very similar.16 This indicates
that the choice between household and individual methods of
class allocation may not be crucial.
The following social classes were used in this study: (1) upper
white-collar employees, (2) lower white-collar employees,
(3) skilled manual workers, (4) unskilled manual workers, (5)
farmers, (6) other self-employed, (7) others.17
The results presented here are based on data for men and
women aged 35–64 in the periods 1971–1975, 1976–1980,
1981–1985 and 1986–1990. The age-standardized relative death
rates are obtained from the Poisson regression analyses.18 These
relative death rates were then transformed into death rates by
multiplying them with the relevant crude death rate.
To summarize the size of the social class differences in mortality we have also calculated a variant of the index of dissimilarity (ID).19 The values of this index show the percentage
of all deaths that should be moved from the high mortality
social classes to the low mortality classes in order to achieve an
equal level of age-standardized mortality for all classes. This
index takes into account the sizes of social classes but it is not
based on assumptions of the hierarchic order of these classes.
Results
Mortality in 1981–1990 by social class and economic
activity in 1980
In 1981–1985 the age-standardized mortality of men who
were economically inactive in 1980 was about three times the
mortality of economically active men (Table 1). Among women
the corresponding excess was 2.4. Although 78% of men aged
35–64 were economically active in 1980, they accounted for
only 43% of the deaths. Among women only 34% of all deaths
occurred among those who were economically active in 1980.
The percentage of inactive people varies strongly by social class
among men (Table 2); 23.6% of skilled men are economically
inactive while only 11.5% of upper white collar men are inactive.
Among women class differences in the proportion of economically inactive are smaller. About 38% of both upper white-collar
women and skilled workers were economically inactive.
Relative differences in mortality between social classes are
greater among all men than among economically active men.
For example, the excess mortality of unskilled workers as compared to upper white-collar employees is 127% among all men
but only 59% among economically active men. Relative differences in mortality are smaller among the economically inactive
than the economically active.
Table 1 Number of person-years and number of deaths, age-standardized mortality (per 100 000), and relative age-standardized mortality
(economically active = 1.00) in 1981–1985 by economic activity in 1980, men and women aged 35–64 in Finland
Person-years
Deaths
Age-standardized mortality
Relative mortality
(1000s)
%
N
%
2990
77.9
18 415
43.2
749
1.00
849
22.1
24 165
56.8
2292
3.06
3839
100.0
42 580
100.0
1176
1.57
Economically active
2767
66.0
6397
33.8
273
1.00
Economically inactive
1426
34.0
12 555
66.2
656
2.40
All
4193
100.0
18 952
100.0
428
1.57
Men
Economically active
Economically inactive
All
Women
BIAS IN SOCIAL CLASS DIFFERENCES IN MORTALITY
901
Table 2 Person-years, age-standardized percentage of economically inactive people and age-standardized relative mortality (upper white-collar
employees = 1.00) in 1981–1985 by social class and economic activity in 1980, men and women aged 35–64 in Finland
Person-years
Relative age-standardized mortality
(1000s)
Of which economically inactive (%)
Economically active
Economically inactive
All
Upper white-collar
499
11.5
1.00
1.00
1.00
Lower white-collar
627
21.4
1.19
1.15
1.38
1527
23.6
1.31
1.29
1.60
Unskilled workers
375
42.3
1.59
1.43
2.27
Farmers
528
21.8
1.18
1.18
1.39
Other self-employed
238
7.6
1.42
0.85
1.17
45
98.6
0.66
1.51
3.55
3839
23.3
1.26
1.26
1.53
Upper white-collar
390
38.4
1.00
1.00
1.00
Lower white-collar
1363
24.3
1.10
1.55
1.14
Skilled workers
1132
38.5
1.16
1.41
1.29
Unskilled workers
536
33.8
1.11
1.82
1.42
Farmers
551
32.1
1.14
1.61
1.25
Other self-employed
174
31.2
1.24
0.76
0.90
47
99.0
1.06
2.61
3.34
4193
32.8
1.13
1.48
1.24
Men
Skilled workers
Others
All
Women
Others
All
Table 3 Person-years, age-standardized percentage of economically inactive people and age-standardized relative mortality (upper white-collar
employees = 1.00) in 1986–1990 by social class and economic activity in 1980, men and women aged 35–64 in Finland
Person-years
Relative age-standardized mortality
(1000s)
Of which economically inactive (%)
Economically active
Economically inactive
All
Men
Upper white-collar
623
5.9
1.00
1.00
1.00
Lower white-collar
757
12.2
1.18
1.39
1.36
1806
15.1
1.48
1.59
1.71
Unskilled workers
393
32.8
1.89
1.80
2.53
Farmers
467
15.8
1.30
1.38
1.50
Other self-employed
264
6.7
1.51
1.00
1.40
44
98.2
2.71
1.89
4.18
4355
15.4
1.37
1.55
1.62
Upper white-collar
479
33.1
1.00
1.00
1.00
Lower white-collar
1675
16.3
1.07
1.48
1.10
Skilled workers
1234
32.9
1.17
1.46
1.30
Unskilled workers
495
26.9
1.22
2.00
1.49
Farmers
453
24.9
1.03
1.82
1.26
Other self-employed
194
31.5
1.21
1.03
1.11
38
98.6
1.41
3.12
3.97
4569
26.0
1.11
1.52
1.24
Skilled workers
Others
All
Women
Others
All
Among women the biggest class differences are found among
the economically inactive. The excess mortality of unskilled
workers as compared to upper white-collar employees is 82%
among the economically inactive but only 11% among the
economically active and 42% among all women.
Table 3 is based on the population aged 35–64 at the census
of 1985 and on the deaths in 1986–1990. However, the data on
social class and economic activity status are from the 1980
census. The results on mortality in 1986–1990 thus correspond
to the results that would have been obtained in the second
5-year period of a 10-year follow-up study.
A comparison of all men in Tables 2 and 3 shows that the
relative mortality differences between social classes were larger
in 1986–1990 than in 1981–1985. Also among economically
902
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Table 4 Values of the Index of Dissimilarity for social class by employment status in 1980, mortality in 1981–1985, 1986–1990 and 1981–1990,
men and women aged 35–64 in Finland
(1)
Economically active
(2)
Economically inactive
(3)
All
(4)
[(1)/(3)]*100
Men
5.63
4.04
8.51
66.2
Women
1.24
5.27
4.97
24.9
Men
7.42
4.83
9.52
77.9
Women
2.99
7.00
5.64
53.0
Men
6.65
4.34
8.97
74.1
Women
2.12
5.74
5.30
40.0
1981–1985
1986–1990
1981–1990
active men the relative differences in mortality were larger in
the second period than the first period. After an average followup time of 7.5 years (Table 3), the social class mortality differences
among those who were economically active at the beginning of
the follow-up (1980), were still markedly smaller than among
all men. For example, the excess mortality of unskilled workers
as compared to upper white-collar employees was 153% among
all men but 89% among economically active men.
The results for women are largely similar to those for men.
The relative mortality differences between social classes were
somewhat larger in the second than the first period. The social
class differences in mortality among women who were
economically active in 1980 were still in 1986–1990 markedly
smaller than among all women.
Index of dissimilarity (ID, see Methods for the interpretation)
values confirm the conclusions made on the basis of Tables 2
and 3 (Table 4). The increase in ID for the whole population
from the first to the second period was 12% for men and 13%
for women. The ID for the economically active population show
an increase of 32% for men and 140% for women. For the
period 1981–1990 the ID for the economically active population
as per cent of the ID for the total population is 74% among men
and 40% among women.
Trends in mortality by economic activity and
social class 1971–1990
Figure 1 shows the mortality ratios of all manual workers as
compared to upper white-collar employees for the periods
1971–1975, 1976–1980, 1981–1985 and 1986–1990. In 1971–
1975 the excess mortality of all manual workers was 76%
among all men and 50% among economically active men. A
decrease in the difference between all manual workers and
upper white-collar employees occurred among economically
active men from 1971–1975 to 1976–1980, whereas the data for
all men show an increase in the difference. After 1980 the
mortality differences have increased both in the economically
active and the total male population.
Among economically active women mortality differentials
decreased from the period 1971–1975 to 1976–1980 and 1981–
1985. Among all women the differentials first increased and
then decreased from 1976–1980 to 1981–1985. After 1985 class
differences in mortality increased in the economically active
and the total population.
Figure 1 The ratio of age-standardized mortality of all manual workers
to that of upper white-collar employees among all and economically
active men and women aged 35–64, Finland 1971–1990
Discussion
Our findings show that among middle-aged men the exclusion
of the economically inactive population from the analyses of
social class differences in 10-year mortality underestimates the
class differences in the total population by about 25%. Among
women the corresponding bias is about 60%. For women, the
underestimation of social class differences in mortality is so
large that the analyses for the economically active seem practically useless. The large underestimation is consistent with
earlier results which show very small differentials among economically active women in Finland.3
The ‘healthy worker effect’—a tendency of those working
being selected for their good health and mental and physical
ability to work—wears off with increasing duration of follow-up
and by discarding the first years of follow-up more reliable
estimates of class differences in mortality can be achieved.10,13
BIAS IN SOCIAL CLASS DIFFERENCES IN MORTALITY
Our results show that although the bias introduced by analysing
the economically active population diminishes during the
follow-up, it is by no means eliminated. In the second 5-year
follow-up period the underestimation of the class differences in
mortality in the total population is still about 22% among men
and 47% among women.
We have also shown that using data on the economically
active population leads to wrong conclusions about trends in
socioeconomic mortality differences in Finland. In the period
from 1971–1975 to 1976–1980 relative class differences in
mortality declined in the economically active male and female
populations, although it increased in the total population.
It is not possible to give a general estimate of the size of the
bias that applies to all age groups, follow-up periods, as well as
regions and countries. Thus, international comparisons of social
class differences in male mortality on the basis of information
concerning the economically active population are difficult. For
women such comparison will be even more problematic as the
labour force participation rates vary widely between countries.
Correction factors to account for the bias have been
developed,20,21 and the factor presented by Kunst—based on
information on the proportion of inactive persons and the
mortality ratio of inactive versus active people in the total population, and survey estimates of the proportion of the inactive in
a particular class divided by the same proportion for all classes—
is most likely to perform quite well.20 Use of such correction
factors should be encouraged in international comparative studies
when information on occupations is not available. However,
applying such adjustments is not always very simple, because
they are specific for each age group and class scheme and need
external information that is not always available or easy to process. Furthermore, the applicability and accuracy of these correction factors have not been tested among women, for whom
the bias is most severe. In addition, correction factors can only
be used to calibrate observed social class differences in mortality
at the aggregate level (e.g. in international comparisons) and
are not of any use in analyses of underlying pathways and
determinants of social class differences in individual data.
In summary, we have shown that excluding the economically
inactive from the analyses of social class differences in mortality
underestimates the differences observed in the total population
markedly. Furthermore, although the bias diminishes during
the follow-up, it is by no means eliminated after the first 5
years. Withdrawal from the labour market, especially when
related to early retirement is often associated with a process of
declining health that may ultimately lead to premature mortality. In a working age population a large part of mortality and
ill health occur among those economically inactive. Thus, possible explanations for social class differences in health based on
the economically active population may give a partial view
of the relevant causal factors and processes. The biases created
by the exclusion of economically inactive from the analyses of
social class differences in mortality and morbidity should be
better recognized and more accurate data on previous occupations should be collected.
Acknowledgements
We are grateful to Statistics Finland for permission (TK 531783–96) to use the data, and Marika Jalovaara and Jere Sillanpää
903
for their assistance in the preparation of the manuscript. The
study is part of a research project funded by the Social Science
Research Council of the Academy of Finland (grant 37631) and
the Signe and Ane Gyllenberg Foundation.
References
1 Kitagawa EM, Hauser PM. Differential Mortality in the United States.
A Study in Socioeconomic Epidemiology. Cambridge, Mass: Harvard
University Press, 1973.
2 Kunst AE, Mackenbach JP. Measuring Socio-economic Inequalities in
Health. Copenhagen: World Health Organization, Regional Office for
Europe, 1994.
3 Martikainen, P. Mortality and socio-economic status among Finnish
women. Popul Stud 1995;49:71–90.
4 Valkonen T. Problems in the measurement and international
comparisons of socio-economic differences in mortality. Soc Sci Med
1993;36:409–18.
5 Vågerö D, Lundberg O. Health inequalities in Britain and Sweden.
Lancet 1989;2(8653):35–36.
6 Leclerc A, Lert F, Fabien C. Differential mortality: some comparisons
between England and Wales, Finland and France, based on inequality
measures. Int J Epidemiol 1990;19:1–10.
7 Kunst AE, Mackenbach JP. International variation in the size of
mortality differences associated with occupational status. Int J
Epidemiol 1994;23:742–49.
8 Mackenbach JP, Kunst AE, Cavelaars AE, Groenhof F, Geurts JJ.
Socioeconomic inequalities in morbidity and mortality in western
Europe. The EU Working Group on Socioeconomic Inequalities in
Health. Lancet 1997;349:1655–59.
9 Fox AJ, Goldblatt PO, Jones DR. Social class mortality differentials:
artefact, selection or life circumstances. J Epidemiol Community Health
1985;38:1–8.
10 Goldblatt P, Fox J, Leon D. Mortality of employed men and women.
In: Goldblatt P (ed.). Longitudinal Study. Mortality and Social Organisation. London: HMSO, OPCS Series LS No. 6, 1990.
11 Valkonen T, Martelin T, Rimpelä A. Eriarvoisuus Kuoleman Edessä.
Sosioekonomiset Kuolleisuuserot Suomessa 1971–85. Helsinki: Central
Statistical Office of Finland, Studies 172, 1990.
12 Valkonen T, Martikainen P. Problems of comparability resulting from
the exclusion of economically inactive persons in studies on social
class differences in mortality. In: Wunsch G, Hancioglu A (eds).
Morbidity and Morbidity Data: Problems of Comparability. Proceedings
of the European Association for Population Studies and the
Hacettepe Institute of Population Studies. Hacettepe University,
Institute of Population Studies, Publication No. IPS-HU.97–01, 1997,
pp.95–114.
13 Drever F, Whitehead M (eds). Health Inequalities. Decennial Supplement.
Office for National Statistics Series DS No.15. London: The Stationery
Office, 1997.
14 Valkonen T, Martelin T, Rimpelä A, Notkola V, Savela S. Socio-Economic
Mortality Differences in Finland 1981–90. Helsinki: Statistics Finland,
Population 1, 1993.
15 Moser K, Pugh H, Goldblatt P. Mortality and the social classification of
women. In: Goldblatt P (ed.). Longitudinal Study. Mortality and Social
Organisation. London: HMSO, OPCS Series LS No. 6, 1990.
16 Martikainen P. Socioeconomic mortality differentials in men and
women according to own and spouse’s characteristics in Finland.
Sociol Health Illness 1995:17:353–75.
17 Central Statistical Office of Finland. Classification of Socio-Economic
Status, Classification of Occupational Status. Helsinki: Central Statistical
Office of Finland, Handbooks No. 17, 1983.
904
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
18 Aitkin M, Clayton D. The fitting of exponential, Weibull, and extreme
20 Kunst AE. Cross-national Comparisons of Socio-economic Differences in
value distributions to complex censored survival data using GLIM.
Appl Stat 1980;29:156–63.
Mortality. Department of Public Health, Erasmus University, Rotterdam, 1997.
19 Koskinen S, Martelin T. Why are socioeconomic mortality differences
21 Diderichsen F, Hallqvist J. Trends in occupational mortality among
smaller among women than among men? Soc Sci Med 1994;38:
1385–96.
middle-aged men in Sweden 1961–1990. Int J Epidemiol 1997;28:
782–87.