International Journal of Epidemiology, 2017, 1147–1156

doi: 10.1093/ije/dyx016
Advance Access Publication Date: 27 February 2017
Original article

Social and Economic Determinants

Economic growth and mortality: do social

protection policies matter?
Usama Bilal,1,2* Richard Cooper,3 Francis Abreu,4 Claudia Nau,5

Downloaded from https://academic.oup.com/ije/article/46/4/1147/3057401 by guest on 20 January 2023

Manuel Franco1,2 and Thomas A Glass1
1
Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA,
2
Social and Cardiovascular Epidemiology Research Group, Universidad de Alcala, Madrid, Spain,
3
Department of Public Health Sciences, Loyola University Stritch School of Medicine, Chicago, IL,
USA, 4Department of Biostatistics and 5Department of Population, Family and Reproductive Health,
Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA
*Corresponding author. Johns Hopkins Bloomberg School of Public Health, Department of Epidemiology – Rm W6604, 615
N Wolfe St, Baltimore, MD, 21205, USA. E-mail: ubilal@jhmi.edu
Accepted 17 January 2017

Abstract
Background: In the 20th century, periods of macroeconomic growth have been associated
with increases in population mortality. Factors that cause or mitigate this association are
not well understood. Evidence suggests that social policy may buffer the deleterious
impact of economic growth. We sought to explore associations between changing un-
employment (as a proxy for economic change) and trends in mortality over 30 years in the
context of varying social protection expenditures.
Methods: We model change in all-cause mortality in 21 OECD (Organization for Economic
Cooperation and Development) countries from 1980 to 2010. Data from the Comparative
Welfare States Data Set and the WHO Mortality Database were used. A decrease in the un-
employment rate was used as a proxy for economic growth and age-adjusted mortality
rates as the outcome. Social protection expenditure was measured as percentage of gross
domestic product expended.
Results: A 1% decrease in unemployment (i.e. the proxy for economic growth) was asso-
ciated with a 0.24% increase in the overall mortality rate (95% confidence interval:
0.07;0.42) in countries with no changes in social protection. Reductions in social protec-
tion expenditure strengthened this association between unemployment and mortality.
The magnitude of the association was diminished over time.
Conclusions: Our results are consistent with the hypothesis that social protection policies
that accompany economic growth can mitigate its potential deleterious effects on health.
Further research should identify specific policies that are most effective.

Key words: Business cycles, macroeconomic conditions, mortality, unemployment, social policies

C The Author 2017; all rights reserved. Published by Oxford University Press on behalf of the International Epidemiological Association
V 1147
1148
Key Messages
• Economic change is an important macrosocial determinant of health.
• Economic growth has been associated with increases in mortality, whereas economic recessions seem to lower
mortality.
• We found that these associations are only present in countries with decreasing social protection expenditure.
• The nature of this association has changed over time and may be inverting from 2006 onwards.
Introduction
The relationship between short-term economic growth and
mortality patterns has been found to be consistently posi-
tive, but counterintuitive findings have been reported.1
Two key ideas by Geoffrey Rose2 may help interpret these
findings. First, economic change represents what Rose
called a ‘mass influence acting on populations as a whole’
that shifts the distribution of risk factors. Second, the
causes of individual risk (i.e. the effects of individual ex-
perience of unemployment) and population rates (e.g. the
effects of an increase in the unemployment rate) need not
be the same and may act in opposite directions.
The study of economic change and health has been a
topic of interest in the past century. Despite varying meth-
ods and across economic cycles, a consistent procyclical
pattern has been found (mortality increasing with eco-
nomic growth); suicide remains the main exception.3,4
This association would appear to be counterintuitive, given
that individual unemployment, under-employment and
loss of job security are consistently associated with higher
risk of illness and death.1
Rose argued2 that the causes of individual cases need not
be the same as the mass influences determining population
rates over time. Tapia-Granados5 and Noelke6 recently
demonstrated that whereas individual unemployment was
associated with increased mortality for the individual, at the
societal level unemployment was associated with decreased
mortality rates. Economic slowdown, on average, generates
health benefits for those who remain employed, potentially
due to reduced air pollution, traffic accidents, alcohol con-
sumption and other risk factors that decline with economic
slowdown.6
This report focuses on two aspects of the association be-
tween economic change and mortality: (i) changes over
time in the strength and direction of this association; and
(ii) the potential of effect modification through policy.
Ruhm recently reported that procyclical mortality associ-
ations have shifted toward the null in the late 1990s and
2000s (and potentially toward a more countercyclical pat-
tern).7 He attributes these changes to an intensification of
the counter-cyclicality of external mortality causes, espe-
cially drug overdoses and cancer. Case and Deaton recently
International Journal of Epidemiology, 2017, Vol. 46, No. 4
highlighted increases in drug overdoses and cirrhosis mor-
tality in US adults, but did not discuss macroeconomic con-
ditions.8 A second possibility, suggested recently by Lam
Downloaded from https://academic.oup.com/ije/article/46/4/1147/3057401 by guest on 20 January 2023
and Pierard, is a decrease in the value of unemployment as
a proxy for macroeconomic conditions over time.9 A third
explanation, as yet inadequately examined, could be the
role of health policy as a modifier of this association.10
Understanding the character and magnitude of policies
that buffer or exacerbate the health impacts of economic
change provides an opportunity to pursue a consequential-
ist epidemiology.11,12 Social policies represent mass influ-
ences on entire regions and countries. If plausible links to
secular trends in rates of disease in populations can be
defined, this suggests population interventions. Three re-
ports, by Reeves13 and Cylus,14,15 have explored how in
European countries and US states, higher spending on la-
bour market policies mitigates the association between sui-
cides, self-rated health and recessions. Nonetheless, few
studies looking at overall mortality burden have evaluated
the hypothesis that social protection policies blunt or coun-
ter cyclical trends. Those investigations that have pursued
this question16,17 used crude estimates of social protection
policies with time-fixed classifications obtained from
single-year estimates that may not be stable or representa-
tive of entire study periods. The latest period of economic
turmoil has sparked a debate on the role of ‘austerity meas-
ures’ that tighten social expenditure to promote a return to
economic growth.18 Social protection expenditures vary
with economic changes, either by entitlements that auto-
matically trigger when, for example, the unemployment
rate goes up (i.e. expenditure on unemployment benefits),
or by specific patterns of policies related to social expend-
iture (i.e. austerity policies). Hence, it is necessary to ac-
count for the time-varying nature of this potential effect
modifier (see review by Stuckler19 for an overview on the
importance of these effect modifiers).
We explored the association between economic growth
and all-cause mortality across 21 OECD countries, ad-
dressing two key limitations of previous studies. First, we
used a robust longitudinal model to examine change across
a 30-year time frame. Second, we evaluated effect modifi-
cation of the business cycle-mortality association by social
International Journal of Epidemiology, 2017, Vol. 46, No. 4 1149

Table 1. Description of the 21 OECD countries in the study sample for selected years in terms of unemployment rate (%), public
social protection expenditure (% of GDP) and all-cause mortality per 100 000. Countries are sorted by average level of social pro-
tection expenditure over the study period (1980–2010)

Unemployment rate (%) Public social protection Mortality per 100 000
expenditure (% of GDP)

Country 1980 1990 2000 2010 1980 1990 2000 2010 1980 1990 2000 2010

Low social Japan 2.0 2.1 4.7 5.0 10.4 11.3 16.2 22.5* 1441.1 1126.4 897.2 791.6
protection Australia 6.1 6.9 6.3 5.2 10.3 13.2 17.3 17.9 1542.1 1275.7 1036.9 864.6
expenditure USA 7.1 5.6 4.0 9.6 12.9 13.4 14.3 19.7 1477.1 1345.6 1251.7 1060.9
Portugal 8.1 4.8 4.0 11.0 9.9 12.5 18.9 25.6* 1986.9 1655.2 1335.7 1075.5
Switzerland 0.2 0.6 2.7 4.5 13.8 13.5 17.8 18.4* 1459.6 1273.2 1068.6 869.0

Downloaded from https://academic.oup.com/ije/article/46/4/1147/3057401 by guest on 20 January 2023

Ireland 7.3 13.4 4.2 13.9 16.5 17.3 13.4 23.7 1959.3 1673.9 1453.7 1085.7*
Canada 7.5 8.2 6.8 8.0 14.2 18.1 16.6 18.9 1419.8 1247.5 1083.8 893.3
Medium social Greece 2.8 7.1 11.2 12.6 10.3 16.6 19.3 23.9* 1502.3 1358.3 1329.3 1046.9
protection New Zealand 2.2 8.0 6.2 6.5 17.0 21.5 19.0 21.2 1791.2 1400.9 1115.8 923.1
expenditure UK 6.2 6.9 5.4 7.8 16.9 17.0 18.8 24.4* 1722.3 1441.9 1309.0* 1006.5
Spain 10.5 14.4 11.7 20.1 15.5 19.9 20.2 26.0* 1463.4 1319.0 1106.8 898.4
Italy 7.4 8.9 10.1 8.4 18.0 19.9 23.1 27.8* 1596.9 1326.1 1109.8 894.9
Norway 1.7 5.2 3.2 3.6 16.9 22.3 21.3 23.3* 1459.2 1395.1 1198.1 992.1
The Netherlands 6.1 5.1 3.1 4.5 24.8 25.6 19.8 23.2* 1437.1 1350.6 1271.3 1011.4
High social Germany 3.2 4.7 8.0 7.1 22.7 22.3 26.6 27.1 1693.6 1509.4 1213.7 1033.4
protection Finland 4.6 3.2 9.8 8.4 18.1 24.1 24.2 29.4* 1707.2 1527.8 1268.7 1034.1
expenditure Belgium 8.5 6.5 6.9 8.3 23.5 24.9 25.3 29.7* 1743.7 1406.1 1252.0 1045.5
Austria 1.7 3.0 3.6 4.4 22.4 23.8 26.6 29.1* 1784.6 1457.9 1212.8 1001.1
Denmark 6.1 7.2 4.3 7.5 24.8 25.1 26.4 30.2* 1599.2 1523.2 1339.7 1150.4
France 5.4 8.0 9.0 9.7 20.8 25.1 28.6 32.0* 1459.2 1197.2 1082.2 884.6
Sweden 2.0 1.7 5.6 8.6 27.1 30.2 28.4 29.8* 1490.8 1301.5 1125.7 972.7

*Value from previous year.

protection policies, treated as a time-varying factor. We
hypothesized that increased public expenditures on social
protection policies may reduce the negative effect of eco-
nomic growth on all-cause mortality.
Methods
Data sources
This study is an ecological analysis of time trends in mor-
tality rates across 21 OECD (Organization for Economic
Cooperation and Development) countries from 1980 to
2010. These 21 countries were selected based on data
availability in the Comparative Welfare States Data Set
(CWSD: http://www.unc.edu/jdsteph/common/data-com
mon.html). We restricted our analysis to industrialized
high-income nations with functioning welfare systems, to
avoid comparability issues. Table 1 shows the included
countries.
The time period (1980 to 2010) was selected to ensure
enough coverage of a full range of experimental conditions
(i.e. several growth and recession periods and changes in
the social protection system).
Data on population estimates of all-cause mortality
rates, stratified by age and sex, were obtained from the
World Health Organization Mortality Database (updated
25 May 2015). We age-adjusted mortality rates to the
European Standard Population of 2013 using the direct
standardization method.
Economic data were obtained from the Comparative
Welfare States Data Set20 where data on welfare policies
from 22 countries have been compiled for five decades.
The main exposure of interest was the harmonized un-
employment rate (variable hunemr2 in the CWSD), our
surrogate measure of economic change. We relied on
unemployment since it has been extensively used in stud-
ies of the impact of macroeconomic change on health,
allowing us to better compare our findings with previous
research. Moreover, it represents a sensitive measure
of the effect of the economy on the people’s health.
Unemployment was defined as the number of unemployed
people as a percentage of the civilian labour force, using a
uniform definition of unemployment across all countries
(people of working age who, in the reference period, are
without work, are available to work and have taken spe-
cific steps to find work). This indicator pools several
1150
harmonized unemployment estimates from the OECD; a
more uniform indicator (variable hunemr in the CWSD)
uses a single definition but is not available for the entire
study period or for all countries. We performed a sensitiv-
ity analysis using this alternative, to determine how our
inferences were affected.
Social protection expenditure was defined as all public
social protection expenditures as a percentage of GDP
(gross domestic product) (available from 1980 to 2010;
variable socx_pub in the CWSD). These expenditures in-
clude old age and survivor policies, incapacity-related
benefits, health and family protection, active labour mar-
ket and unemployment expenditures, and housing policies.
Given that we used a measure of social protection expend-
iture that is relative to overall GDP, this measure may be
biased when GDP changes (i.e. if GDP goes down, social
protection expenditure may go up because the denomin-
ator becomes smaller). To test the robustness of our results
to this, we conducted a sensitivity analysis in which we fur-
ther adjusted our models with GDP growth [defined as
(GDP in current year – GDP in previous year)/GDP in pre-
vious year].
Statistical models
The goal of this analysis was to study the association
between economic change and mortality and interaction
by changes in social protection policies. Analyses were
conducted using linear models with the natural logarithm
of age-adjusted all-cause mortality rate as the dependent
variable. We used a mixed-effects model that decomposes
between- and within-group (country) variance compo-
nents21,22 by group-mean centring covariates and intro-
ducing these terms into the model along with a random
effect for the mean of each group. These models have the
advantages of removing time-fixed confounding on the
group-mean centred covariate (like fixed-effects models)
and still make it possible to study the effect of time-
fixed covariates, deal with variables missing at random
and increase the efficiency of the analysis.23 Recent
simulation studies have confirmed these aforementioned
advantages.24
The model building process included regressing mor-
tality on a series of year polynomials (including fixed and
random terms for each one) and adding a first-order auto-
correlation structure for the level-1 residuals. The model
with the lowest AIC (Akaike Information Criterion) was
selected. The final model was specified as follows (the j
index represents a country and the i index represents a year
of observation):
International Journal of Epidemiology, 2017, Vol. 46, No. 4
lnðMortalityÞji ¼b0 þu0j þðb1 þu1j ÞðYearij 1980Þij
þðb2 þu2j ÞðYearij 1980Þ2ij þb4 ðURij UR j Þ
þ b5 ðSPPij SPP j Þþb6 ðURij UR j ÞðSPPij SPP j Þ
þ b7 UR j þb8 SPP j þeij
2 3 0 2 31 0 2 31
l0j s00 1
6 7 B 6 7C B 6 7C
6 7 B 6 7C B 6 7C
6 l1j 7 NB0; 6 s10 s11 7C eji  N B0;r2  6 q1 1 7C
6 7 B 6 7C B 6 7C
4 5 @ 4 5A @ 4 5A
l2j s20 s12 s22 q2 q1 1
In order to check the robustness of our results to this
Downloaded from https://academic.oup.com/ije/article/46/4/1147/3057401 by guest on 20 January 2023
within-between-model specification, we also estimated a
two-way fixed-effects model (the typical specification in pre-
vious studies,25 including a fixed term for each country and
year) with a first-order autocorrelation structure of residuals.
In secondary analyses, we also assessed (i) whether the
effect of unemployment changed over time (as hypothesized
by Ruhm7) by introducing an interaction term of unemploy-
ment and social protection expenditure and time (linear
time, along with linear splines with knots every 10 years to
allow for changes in the associations to vary linearly every
decade); and (ii) if there were non-linear effects (as hypothe-
sized by Bonamore26), by adding a quadratic term for both
the unemployment rate and the social protection policies.
Main coefficients are presented as (1)*(eb1)*100,
and can be interpreted as the percentage change in mortal-
ity per 1% decrease in unemployment. Analyses were con-
ducted using STATA/SE 13.1 (STATACorp, College
Station, TX).
Results
A descriptive overview of the data is presented in Table 1.
The first four columns provide the unemployment rate for
a sample of time periods. Large differences can be seen
across countries and time. Structural unemployment (con-
sistent high levels) is evident in countries like Spain, where
rates never go below 10%, but also in Italy, Greece and
Ireland. Period effects are also evident as indicated by in-
creases in unemployment in 2010 in 18 of the 21 countries,
as a result of the Great Recession of 2009. Cross-country
variations are large also for social protection. Some coun-
tries, like Australia, Canada, New Zealand and the USA,
have low levels of social protection expenditure, never
reaching 20% in any given year. Other countries, mostly in
Scandinavia and Central Europe, consistently have expend-
itures of 23% and above. Mortality rates have declined
monotonically in all countries during the study period. In
1980, mortality rates varied from a high of 1987 in
Portugal to a low of 1437 in The Netherlands. Rates of
International Journal of Epidemiology, 2017, Vol. 46, No. 4
decline varied widely; Japan declined by 45% whereas the
USA saw a 28% decline.
The primary results are based on a series of multilevel
mixed-effects models that capture change in mortality rates
as a function of unemployment, social protection expend-
iture and time (Table 2). In model 1 without an interaction
term between unemployment and social protection ex-
penditure, a decrease in unemployment (a proxy for eco-
nomic growth) is associated on average with an increase in
mortality. Model 2 includes an interaction term between
unemployment and social protection expenditure; this
model indicates that the association between unemploy-
ment and mortality is attenuated in countries with higher
expenditures.
A more complete characterization of these findings
can be seen in Figure 1: a 1% decrease in unemployment
(a proxy for economic growth) is associated with a
0.24% increase in mortality [95% confidence interval
(CI) ¼ 0.07;0.42]. This is the estimated association (Figure 1,
centre) for countries that do not change their social protec-
tion expenditure levels in the same time period. In countries
with a 5% decline in social protection expenditures, the as-
sociation between economic growth and mortality is almost
double in magnitude (0.45% increase in mortality, 95%
CI ¼ 0.17;0.73; left side of Figure 1). In contrast, in countries
where social expenditures increase, changes in unemploy-
ment are associated with smaller or null changes in mortality
(Figure 1, right side).
We conducted two secondary analyses to test hypothe-
ses suggested in previous studies. Figure 2 shows the asso-
ciation between economic growth and mortality and
interaction with social protection expenditure over time.
Two trends are apparent. Economic growth (as proxied by
a decrease in unemployment) is associated with increased
mortality from the 1980s through the early 2000s; this as-
sociation is especially strong in countries with a decrease in
social protection expenditure (left panel, Figure 2).
Moreover, the size of these changes is stronger in countries
with decreasing social protection expenditure, whereas
countries that increase expenditure only see variations of
this association around the null (right panel, Figure 2). The
strength of the unemployment-mortality association is di-
minished in the 2000s.
Table 2 also shows the results of an additional second-
ary analysis (model 3), adding a quadratic term for un-
employment to test the hypothesis that the association
between unemployment and mortality was not linear.
Model fit was not improved suggesting a linear associ-
ation. The sensitivity analysis using the alternative out-
come (completely harmonized unemployment rates, model
4 in Table 2) demonstrates consistent inferences. The sensi-
tivity analysis using a regular fixed-effects model also
1151
showed analogous inferences to our main analysis (model
5 in Table 2). The sensitivity analysis adjusting for GDP
growth (%) to control for potential changes in the de-
nominator of social protection expenditure showed no dif-
ferences compared with the main model (see model 6 in
Table 2).
Discussion
The primary finding of this study was that the association
between economic change and mortality is weakened in
countries that concurrently increase social protection ex-
Downloaded from https://academic.oup.com/ije/article/46/4/1147/3057401 by guest on 20 January 2023
penditure. These results add support to previous reports
that economic growth is associated with an increase in
mortality rates. A key contribution of the present study
is that this association is only apparent in countries that
concurrently decrease social protection expenditures.
Secondary analyses provide evidence that the negative con-
sequences of growth on mortality may have diminished or
reversed since 2006.
Previous research has consistently identified increases in
mortality accompanying economic growth.4,16,27–34 Two
previous studies have identified country-level effect modi-
fiers using percentage of GDP spent on social protection
services;16,17 one study measured expenditures on services
in 1995 and the other averaged over 10 years. Both studies
found similar results to ours: the effects of macroeconomic
change are more pronounced in countries with weaker so-
cial protection systems. Nonetheless no study, to our
knowledge, has assessed the effect of changes in expend-
itures and their interaction with trends in economic activ-
ity. Our results provide evidence to support pro-active
policies to strengthen welfare systems as a means of ensur-
ing optimal public health35.
The effects of economic growth on health can be medi-
ated through changes on the supply side or the demand
side. Supply side changes include rising working hours and
increased demand for labour as a function of increasing in-
dustrial production. These changes may lead, inter alia, to
increased stress resulting from rising competition and work
demands1,36–40 or to an increase in the cost of time, reduc-
ing the probability of engaging in healthy (and time-costly)
recreational or health-affirming behaviours.4 Changes in
the demand side41 could include increases in consumption
of products which erode health, primarily tobacco, alcohol
and a diet high in animal products and calories,41–45 to-
gether with exposure to worsening air pollution and
increased risk of accidents associated with greater motor-
ized transport and longer working hours.46 A recent report
by Cutler28 highlights the role of pollution, accidents, alco-
hol and tobacco consumption in this association, and
stresses the importance of considering age-specific effects
 1152

Table 2. Results from the main model and secondary and sensitivity analyses

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

(without Main model Secondary Sensitivity Sensitivity Sensitivity
interaction) (incl. interaction) analysis: analysis: analysis: analysis:
quadratic UE harmonized UE two-way FE control for GDP
b*100 (95% CI) b*100 (95% CI) b*100 (95% CI) b*100 (95% CI) b*100 (95% CI)

Year 1.14 (1.44;0.83) 1.11 (1.42;0.80) 1.10 (1.41;0.79) 1.11 (1.46;0.75) 1.04 (1.36;0.72)
Year2 0.02 (0.02;0.01) 0.02 (0.02;0.01) 0.02 (0.02;0.01) 0.02 (0.03;0.01) 0.02 (0.03;0.01)
1% increase in 0.21 (0.39;0.04) 0.24 (0.42;0.07) 0.17 (0.41;0.07) 0.27 (0.48;0.07) 0.27 (0.51;0.04) 0.23 (0.41;0.05)
unemployment rate
Unemployment x 0.04 (0.00;0.08) 0.07 (0.01;0.13) 0.04 (0.01;0.09) 0.04 (0.01;0.08) 0.05 (0.01;0.09)
social protection
expenditure
1% increase in 0.22 (0.01;0.43) 0.12 (0.10;0.35) 0.10 (0.14;0.33) 0.18 (0.05;0.41) 0.05 (0.23;0.32) 0.01 (0.26;0.24)
social protection
expenditure
1% increase in 0.01 (0.05;0.02)
unemployment rate2
Unemployment2 x 0.00 (0.01;0.00)
social protection
expenditure
AIC 2751.9215 2754.558 2749.9507 2257.447 2880.3987 2758.966
AIC change – 2.63 þ 1.97 N/A N/A 7.04
No. parameters 15 16 19 16 34 17
No. observations 613 613 613 504 592 613

All coefficients are shown untransformed to facilitate comparison, but scaled times 100 to ease visualization. Model 2 is the one shown in the methods section. Model 1 is analogous to model 2 without the unemployment
x social protection expenditure interaction coefficient. Model 3 is analogous to model 2, with a quadratic coefficient for unemployment (and its interaction with social protection expenditure). Model 4 uses the completely
harmonized unemployment rate as the main exposure. Models 1–4 further include a term for mean unemployment rate and mean social protection expenditure for each country. Model 5 is a two-way fixed-effects model
with a fixed effect for year and a fixed effect for time, with a first-order autoregressive residual structure.
UE: unemployment rates; FE, fixed effects; incl, including; N/A, not available; no., number.
International Journal of Epidemiology, 2017, Vol. 46, No. 4

Downloaded from https://academic.oup.com/ije/article/46/4/1147/3057401 by guest on 20 January 2023

International Journal of Epidemiology, 2017, Vol. 46, No. 4 1153

Downloaded from https://academic.oup.com/ije/article/46/4/1147/3057401 by guest on 20 January 2023

Figure 1. Percentage change in mortality (and 95% CI) per 1% decrease in unemployment across levels of change of social protection expenditure.
The histogram represents the observed distribution of social protection expenditure changes

Figure 2. Change in the unemployment/mortality association over time by levels of social protection expenditure. The left panel represents countries
with decreasing social protection expenditure, the right panel represents countries with increasing social protection expenditure and in the middle
one are countries with no changes in social protection expenditure. Coefficients come from a random-effects model that includes linear and quadratic
time (both fixed and random), unemployment (within-country change and mean over the study period), social protection expenditure (within-country
change and mean over the study period), form and linear spline terms with knots every 10 years, along with the interactions between unemployment,
social protection expenditure and the linear time splines. The grey band represents the 95% confidence interval of the unemployment/mortality
association

(i.e. most work-related or risk factors apply only to
adults).
It should be emphasized that though unemployment de-
creases during growth and increases during recessions,
only a fraction of the population is affected by unemploy-
ment and the majority of the population remains in stable
employment. Thus, whereas the experience of unemploy-
ment is negative for the health of an individual,47 the over-
all effect we measure here reflects the population-wide
influences of changing unemployment rates (including indi-
viduals whose status did not change).48 As hypothesized by
Cooper,41 economic growth (prosperity) may be associated
with increased consumption; in contemporary consumer
culture, a proportion of that consumption may be detri-
mental to health.
A key contribution of this paper is our finding suggest-
ing that social protection policies may buffer the conse-
quences of economic growth on mortality. Two
mechanisms may be at play. First, well-developed social
protection systems may push fewer people with health vul-
nerabilities into the work force. A recent report by Regidor
et al.27 supports this hypothesis, as it found that the in-
crease in mortality during periods of growth (or the de-
crease during recessions) was stronger in people of lower
1154
socioeconomic position, who are most prone to be pushed
into the work force while sick.49 Second, buffering may
arise from a reduction of either the negative contextual ef-
fect of economic growth or the positive contextual effect of
economic recessions. Although the latter seems likely, a hy-
pothesis regarding the former can be found in the downsiz-
ing/job insecurity literature, where previous research has
shown the negative effects of these practices.50 Moreover,
as hypothesized by Ruhm, social protection expenditure
may be a marker for ‘broader patterns of government in-
volvement’ that may reduce the overall effects of economic
changes.10 As mentioned in the introduction to this paper,
social protection expenditure changes vary with the econ-
omy, as certain sources of expenditure are entitlements
that trigger when the economy shrinks (like unemployment
benefits), whereas other changes are contemporaneous,
like austerity policies. Our measure of social protection ex-
penditure combines both, but our model implicitly adjusts
for the first (entitlements triggered during recessions) by
including unemployment changes in the model.
The strengths of this study include the use of data on a
large number of countries (N ¼ 21) over a long time period
(30 years) encompassing several economic cycles. This en-
hances the generalizability and robustness of our results.
Our time frame included (but was not limited to) the recent
economic turmoil beginning in 2008 which has been the
subject of research and policy scrutiny. Lastly, this large
temporal scope allowed us to test for heterogeneity over
time by policy-sensitive factors, a mechanism that may
underlie these changes7.
A potential limitation of this study is the possibility of
confounding. Using both random (primary analysis) and
fixed-effects models24 (sensitivity), we found little evidence
that a diverse set of country-specific time-fixed confound-
ers might explain this association. We are unable to rule
out additional unmeasured time-varying confounders. A
second limitation is that the use of a general measure of so-
cial protection expenditure does not allow testing of the
buffering effects of particular policies. This limits our abil-
ity to provide guidance about whether, for example, in-
come transfers, job training or active labour policies are
most effective. Future studies should consider investigating
which specific policies may be driving the effect modifica-
tion. Third, we focus our inferences on the changes in so-
cial protection expenditure, instead of focusing on the
consequences of (cross-sectional) levels of social ex-
penditure. These two variables are independent in our data
set (See Supplementary Figure 2, available in the
Supplementary data at IJE online); we therefore believe
that the effect-modifying nature of social protection ex-
penditure changes is not due to baseline or average levels
of social protection. However, this limits the scope of our
International Journal of Epidemiology, 2017, Vol. 46, No. 4
study to the impact of specific concurrent social expend-
iture policies, but at the same time reduces the possibility
of time-fixed confounding by unmeasured country-level
characteristics (e.g. history of egalitarianism) that may af-
fect levels of social protection expenditure. Finally, we
relied on unemployment changes as our surrogate of eco-
nomic growth. Although not perfect, this proxy has been
extensively used in studies of the impact of macroeconomic
change on health. Other alternatives, such as change in
GDP, may not be sensitive measures of the effect of eco-
nomic change on people’s lives. Sensitivity analyses
showed that the use of harmonized or non-harmonized un-
Downloaded from https://academic.oup.com/ije/article/46/4/1147/3057401 by guest on 20 January 2023
employment rates did not affect our inferences.
Ecological studies are known to have additional limita-
tions. We address this concern in part by limiting our infer-
ences to the population level, making no claims about the
effects of changing economic conditions on the risk of
death for individuals. We were interested in identifying
macrosocial processes that affect population-level mortal-
ity rates. These factors may differ from individual-level
risk factors.2,51 In a recent review, Stuckler19 argued for a
better conceptualization of causal frameworks to study
this problem. We believe that population-level inferences
are relevant to a consequentialist epidemiology12 by pro-
viding guidance about how national policies can buffer the
effects of important mass influences on health. Given the
interest in studying how austerity affects population
health,18 this study can inform future policy making. If
economic growth is associated with mortality only in coun-
tries that concomitantly decrease social protection expend-
itures, then interventions that increase or maintain such
expenditures may promote growth with fewer negative
consequences.
Conclusions
Economic growth over the period 1980–2010 was associ-
ated with a slower decline in mortality rates over time in a
sample of 21 OECD countries. This association is observed
only in countries that decrease social protection expend-
itures. This suggests that maintaining or increasing the
strength of the social safety net can diminish the associ-
ation between economic change and population health.
Replication of these results in other industrialized coun-
tries may provide more information on the mechanisms
and specific policies behind this phenomenon and may in-
form future interventions to avoid increases in mortality
with economic growth.
Supplementary Data
Supplementary data are available at IJE online.
International Journal of Epidemiology, 2017, Vol. 46, No. 4
Funding
U.B. was supported by a Johns Hopkins Center for a Livable Future
– Lerner Fellowship, and a Postgraduate Fellowship from the Obra
Social La Caixa.
Conflicts of interest: F.A. currently works at Genentech and con-
ducts research sponsored by this company. This study is not part of
any research supported or sponsored by Genentech.
References
1. Catalano R, Goldman-Mellor S, Saxton K et al. The health effects
of economic decline. Annu Rev Public Health 2011;32:431–50.
2. Rose G. Sick individuals and sick populations. Int J Epidemiol
1985;14:32–38.
3. Ogburn WF, Thomas DS. The influence of the business cycle on
certain social conditions. J Am Stat Assoc 1922;18:324–40.
4. Ruhm CJ. Are recessions good for your health? Q J Econ
2000;115:617–50.
5. Tapia Granados JA, House JS, Ionides EL, Burgard S, Schoeni
RS. Individual joblessness, contextual unemployment, and mor-
tality risk. Am J Epidemiol 2014;180:280–87.
6. Noelke C, Avendano M. Who suffers during recessions?
Economic downturns, job loss, and cardiovascular disease in
older Americans. Am J Epidemiol 2015;182:873–82.
7. Ruhm CJ. Recessions, healthy no more? J Health Econ
2015;42:17–28.
8. Case A, Deaton A. Rising morbidity and mortality in midlife
among white non-Hispanic Americans in the 21st century. Proc
Natl Acad Sci U S A 2015;112:15078–83.
9. Lam J-P, Pierard E. The time-varying relationship between mortal-
ity and business cycles in the US. Health Econ 2017;26:164–83.
10. Ruhm CJ. Macroeconomic conditions, health, and government
policy. In: Schoeni RF (ed). Making Americans Healthier:Social
and Economic Policy as Health Policy. New York, NY: Russell
Sage Foundation, 2008.
11. Galea S, Link BG. Six paths for the future of social epidemiology.
Am J Epidemiol 2013;178:843–49.
12. Galea S. An argument for a consequentialist epidemiology. Am J
Epidemiol 2013;178:1185–91.
13. Reeves A, McKee M, Gunnell D et al. Economic shocks, resilience,
and male suicides in the Great Recession: cross-national analysis
of 20 EU countries. Eur J Public Health 2015;25:404–09.
14. Cylus J, Glymour MM, Avendano M. Do generous unemploy-
ment benefit programs reduce suicide rates? A State fixed-effect
analysis covering 1968–2008. Am J Epidemiol 2014;180:45–52.
15. Cylus J, Glymour MM, Avendano M. Health effects of un-
employment benefit program generosity. Am J Public Health
2015;105:317–23.
16. Toffolutti V, Suhrcke M. Assessing the short term health impact
of the Great Recession in the European Union: A cross-country
panel analysis. Prev Med 2014;64:54–62.
17. Gerdtham UG, Ruhm CJ. Deaths rise in good economic times:
evidence from the OECD. Econ Hum Biol 2006;4:298–316.
18. Stuckler D, Basu S. The Body Economic: Why Austerity Kills.
New York, NY: Basic Books, 2013.
19. Stuckler D, Reeves A, Karanikolos M, McKee M. The health ef-
fects of the global financial crisis: can we reconcile the differing
1155
views? A network analysis of literature across disciplines. Health
Econ Policy Law 2015;10:83–99.
20. Brady D, Huber E, Stephens JD. Comparative Welfare States
Data Set. 2014 http://www.unc.edu/jdsteph/common/data-
common.html (21 January 2016, date last accessed).
21. Allison PD. Fixed Effects Regression Models. Thousand Oaks,
CA: SAGE Publications, 2009.
22. Fitzmaurice GM, Laird NM, Ware JH. Applied Longitudinal
Analysis. Hoboken, NJ: Wiley, 2011.
23. Clark TS, Linzer DA. Should I use fixed or random effects? Polit
Sci Res Methods 2015;3:399–408.
24. Bell A, Jones K. Explaining fixed effects:random effects modeling
of time-series cross-sectional and panel data. Polit Sci Res
Methods 2015;3:133––53.
Downloaded from https://academic.oup.com/ije/article/46/4/1147/3057401 by guest on 20 January 2023
25. Ruhm CJ. Commentary: Macroeconomic conditions and social out-
comes through a 90-year lens. Int J Epidemiol 2015;44:1490–92.
26. Bonamore G, Carmignani F, Colombo E. Addressing the
unemployment-mortality conundrum: non-linearity is the an-
swer. Soc Sci Med 2015;126:67–72.
27. Regidor E, Vallejo F, Granados JAT, Viciana-Fern andez FJ, de la
Fuente L, Barrio G. Mortality decrease according to socioeco-
nomic groups during the economic crisis in Spain: a cohort study
of 36 million people. Lancet 2016;388:2642–52.
28. Cutler DM, Huang W, Lleras-Muney A. Economic conditions
and mortality: evidence from 200 years of data. National Bureau
of Economic Research Working Paper Series No. 22690.
Cambridge, MA: NBER, 2016.
29. Ruhm CJ. Economic conditions and health behaviors: are reces-
sions good for your health? N C Med J 2009;70:328–29.
30. Tapia Granados JA. Increasing mortality during the expansions of
the US economy, 1900–1996. Int J Epidemiol 2005;34:1194–202.
31. Tapia Granados JA. Economic growth and health progress in
England and Wales: 160 years of a changing relation. Soc Sci
Med 2012;74:688–95.
32. Tapia Granados JA, Ionides EL. The reversal of the relation be-
tween economic growth and health progress: Sweden in the 19th
and 20th centuries. J Health Econ 2008;27:544–63.
33. Tapia-Granados JA. Macroeconomic fluctuations and mortality
in postwar Japan. Demography 2008;45:323–43.
34. Thomas D. Social Aspects of the Business Cycle. London:
George Routledge and Sons, 1925.
35. Reeves A, McKee M, Stuckler D. The attack on universal health
coverage in Europe: recession, austerity and unmet needs. Eur J
Public Health 2015;25:364–65.
36. Cooper M, Lynne MR, Frone MR. Work stress and alcohol ef-
fects: A test of stress-induced drinking. J Health Soc Behav
1990;31:260–76.
37. Hlatky MA, Lam LC, Lee KL et al. Job strain and the preva-
lence and outcome of coronary artery disease. Circulation
1995;92:327–33.
38. James SA, Keenan NL, Strogatz DS, Browning SR, Garrett JM.
Socioeconomic status, John Henryism, and blood pressure in black
adults. The Pitt County Study. Am J Epidemiol 1992;135:59–67.
39. Karasek R. The stress-disequilibrium theory: chronic disease de-
velopment, low social control, and physiological de-regulation.
Med Lav 2006;97:258–71.
40. Karasek R, Baker D, Marxer F, Ahlbom A, Theorell T. Job deci-
sion latitude, job demands, and cardiovascular disease: a
1156
prospective study of Swedish men. Am J Public Health 1981;
71:694–705.
41. Cooper RS. Prosperity, of the capitalist variety, as a cause of
death. Int J Health Serv 1977;9:155–59.
42. Colell E, Sanchez-Niub o A, Delclos GL, Benavides FG,
Domingo-Salvany A. Economic crisis and changes in drug use
in the Spanish economically active population. Addiction
2015;110:1129–37.
43. Courtemanche CJ, Pinkston JC, Ruhm CJ, Wehby G. Can chang-
ing economic factors explain the rise in obesity? National
Bureau of Economic Research Working Paper Series No. 20892.
Cambridge, MA: NBER, 2015.
44. Harhay MO, Bor J, Basu S et al. Differential impact of the eco-
nomic recession on alcohol use among white British adults,
2004–2010. Eur J Public Health 2014;24:410–15.
45. Nandi A, Charters TJ, Strumpf EC, Heymann J, Harper S.
Economic conditions and health behaviours during the ‘Great
Recession’. J Epidemiol Community Health 2013;67:1038–46.
International Journal of Epidemiology, 2017, Vol. 46, No. 4
46. Chen G. Association between economic fluctuations and road
mortality in OECD –countries. Eur J Public Health
2014;24:I612–14.
47. Halliday TJ. Unemployment and mortality: Evidence from the
PSID. Soc Sci Med 2014;113:15–22.
48. Tapia Granados JA. Response: on economic growth, business
fluctuations, and health progress. Int J Epidemiol 2005;
34:1226–33.
49. Catalina-Romero C, Sainz JC, Pastrana-Jiménez JI et al. The im-
pact of poor psychosocial work environment on non-work-
related sickness absence. Soc Sci Med 2015;138:210–16.
50. Quinlan M, Bohle P. Overstretched and unreciprocated commit-
ment: reviewing research on the occupational health and safety
effects of downsizing and job insecurity. Int J Health Serv
Downloaded from https://academic.oup.com/ije/article/46/4/1147/3057401 by guest on 20 January 2023
2009;39:144.
51. Rockhill B. Theorizing about causes at the individual level while
estimating effects at the population level: implications for pre-
vention. Epidemiology 2005;16:124–29.