Content uploaded by Gabor Scheiring
Author content
All content in this area was uploaded by Gabor Scheiring on Mar 21, 2023
Content may be subject to copyright.
© The Author(s) 2023. Published by Oxford University Press on behalf of the Cambridge Political Economy Society.
All rights reserved.
Deindustrialisation and the post-socialist
mortality crisis
GáborScheiring , AytalinaAzarova , DarjaIrdam ,
KatarzynaDoniec, MartinMcKee , DavidStuckler and LawrenceKing*,
An unprecedented mortality crisis struck Eastern Europe during the 1990s, causing
around seven million excess deaths. We enter the debate about the causes of this
crisis by performing the rst quantitative analysis of the association between de-
industrialisation and mortality in Eastern Europe. We develop a theoretical frame-
work identifying deindustrialisation as a process of social disintegration rooted in
the lived experience of shock therapy. We test this theory relying on a novel multi-
level dataset, tting survival and panel models covering 52 towns and 42,800 people
in 1989–95 in Hungary and 514 towns in European Russia in 1991–99. The results
show that deindustrialisation was directly associated with male mortality and indir-
ectly mediated by hazardous drinking as a stress-coping strategy. The association
is not a spurious result of a legacy of dysfunctional working-class health culture
aggravated by low alcohol prices during the early years of the transition. Both coun-
tries experienced deindustrialisation, but social and economic policies have offset
Hungary’s more immense industrial employment loss. The results are relevant to
health crises in other regions, including the deaths of despair plaguing the American
Rust Belt. Policies addressing the underlying causes of stress and despair are vital to
save lives during painful economic transformations.
Key words: Deindustrialisation, Mortality, Stress, Eastern Europe, Multilevel
modelling
JEL classications: B520 Current Heterodox Approaches; I15 Health and
Economic Development; P3 Socialist Institutions and Their Transitions
1. Introduction
An unprecedented mortality crisis hit the former socialist countries in the early
1990s—a phenomenon that Ellman (1994) famously labelled ‘katastroika’. The
number of excess deaths could have been around 7.3 million in Eastern Europe in
1991–99 (Stuckler, 2009, p. 7). Male life expectancy in Russia declined by seven years
Manuscript received 14 August 2021; nal version received 25 August 2022
Address for correspondence: Department of Social and Political Sciences, Bocconi University, Via Guglielmo
Roentgen 1, 20132, Milan, Italy; email: gabor@gaborscheiring.com
* Department of Social and Political Sciences, Bocconi University, Milano, Italy (GS, DS); Department
of Public Health and Primary Care, University of Cambridge, Cambridge, UK (AA); Department of
Sociology, University of Cambridge, Cambridge, UK (DI); Department of Sociology and Nufeld College,
Leverhulme Centre for Demographic Science, Oxford University, Oxford, UK (KD); Department of Health
Services Research and Policy, London School of Hygiene and Tropical Medicine, London, UK (MM);
Department of Economics, University of Massachusetts Amherst, Amherst, MA, USA (LK)
Cambridge Journal of Economics 2023, 1 of 32
https://doi.org/10.1093/cje/beac072
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
between 1988 and 1995. As Popov (2012, p. 324) noted: ‘Never after the 1947 famine
had Russia had, in the post-war period, mortality rates as high as those in the 1990s.
Even in 1950–53, during the last years of Stalin’s regime, with the high death rates
in the labour camps and the [delayed] consequences of wartime malnutrition and
injuries, the mortality rate was only nine to ten per 1,000, compared with 14–16 in
1994’. Hungary also suffered a signicant though less dramatic mortality crisis, al-
though worse than its Visegrad neighbours (Chenet et al., 1996). Male life expectancy
in Hungary declined by 1.5 years between 1988 and 1994, and death rates reached
levels last observed during the Great Depression of the 1930s, that is, 14.5 per 1,000
in 1993 (Kopp et al., 2007, p. 326).
Life chances have improved since the second half of the 1990s, but wide health
inequalities still plague most countries. In addition to emigration and low fertility,
these health problems are the primary reasons why 15 out of the 20 fastest-shrinking
countries are located in Eastern Europe (United Nations, 2022). Figure 1 summarises
the post-socialist mortality crisis in Russia and Hungary, compared to the average of
Visegrad countries.1
Working-class men without a college degree suffered the most; therefore, we focus
on male life expectancy and death rates. In the 1990s, blue-collar male workers had
111% higher odds of dying than those with a college degree in Hungary, a 17%
Fig. 1. Post-socialist mortality crisis in Hungary and Russia.
Note: Visegrad countries’ average includes the Czech Republic, Poland and
Slovakia (excludes Hungary).
Source: WHO (2020) Health for All Europe Database.
1 Czech Republic, Poland and Slovakia, excluding Hungary.
Page 2 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
increase from the 1980s, and 50% higher odds of dying in Russia, a 14% increase from
the 1980s (Doniec et al., 2018). However, this does not mean that absolute material
deprivation drove the mortality crisis. Wealthier, more urbanised, and industrialised
regions in Russia suffered a more severe wave of excess deaths than impoverished re-
gions (Walberg et al., 1998). Poverty-related malnutrition played no meaningful role
(Ellman, 1994; Cornia and Paniccià, 2000).
In both countries, deaths considered sensitive to social disintegration and the re-
sulting psychosocial stress and associated behaviours increased the most, including
deaths due to mental disorders, homicides, alcohol (digestive system diseases, injury
and poisoning), and heart disease—with heavy alcohol consumption now known to
play a disproportionate role in cardiovascular deaths in Russia (Tomkins et al., 2012).
Appendix A of Supplementary Data (patterns of the post-socialist mortality crisis,
Supplementary Figure A1, Supplementary Tables A1 and A2) presents detailed data
on age- and cause-specic mortality trends, further underpinning the centrality of
stress-related mechanisms.
While there is widespread consensus on the proximal, or downstream, causes, es-
pecially easy access to alcohol, disagreement persists concerning the upstream, polit-
ical–economic factors—what Marmot (2018) called the ‘causes of causes’ of ill health.
Many argue that economic dislocation caused stress and despair, which led to ele-
vated mortality. Others hypothesise that adverse lifestyles—conditioned by the so-
cialist legacy of dysfunctional working-class health culture and aggravated by a marked
increase in the supply of alcohol, especially illicit or surrogate beverages (Gil et al.,
2009)—are the main culprits. Despite the intensity of this debate and the clear-cut
policy relevance, curiously, the role of industrial employment decline has not received
much attention. In a recent study, Scheiring and King (2022) qualitatively analysed
the health implication of deindustrialisation. However, a quantitative assessment of
this association is still lacking. This article aims to ll this gap.
Understanding the role of deindustrialisation in the post-socialist mortality crisis is
crucial. First, 30 years after the fall of the Soviet Union, there is a continued interest in
assessing the social costs of the transformation (Ellman, 2000; Ghodsee and Orenstein,
2021). Many connect the rise of populism in Eastern Europe to ‘the failure of liber-
alism to deliver’ (cf. Krastev, 2016; Scheiring, 2020). Workers’ physical and mental
suffering in left-behind areas is a critical correlate of anti-liberal, populist attitudes
(Koltai et al., 2020; Kavanagh et al., 2021). Therefore, the insights from analysing the
deindustrialisation–mortality association go beyond public health.
Second, many emerging economies are experiencing ‘premature deindustrialisation’
(Tregenna, 2008, 2016; Rodrik, 2016), with potentially severe health and well-being
costs. Third, global competition, technological change, and the imperative to mitigate
climate change will continue to put pressure on industrial employment, which could
harm workers’ health in rustbelt areas. Fourth, deindustrialisation appears to be a
critical factor in the ‘deaths of despair’ plaguing the USA (Case and Deaton, 2020;
Venkataramani et al., 2020). Better understanding the association of deindustrialisa-
tion with the wave of excess deaths hitting Eastern Europe in the 1990s promises im-
portant theoretical and policy insights for these elds (King et al., 2022).
We enter the debate about the causes of this crisis by performing the rst quanti-
tative analysis of the association between deindustrialisation and mortality in Eastern
Europe. We develop a theoretical framework identifying deindustrialisation as a process
of social disintegration rooted in the lived experience of shock therapy. Empirically, we
Deindustrialisation and mortality Page 3 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
ask whether deindustrialisation was directly and independently associated with the ex-
cess deaths and increased hazardous drinking driven by stress and despair. Second, we
ask whether the effect of deindustrialisation on mortality is a spurious consequence of
dysfunctional working-class health culture. We test whether hazardous drinking habits
could be considered a consequence of previous socialist industrialisation (as some
argue) and whether low alcohol prices acted on this dysfunctional working-class cul-
ture leading to elevated death rates. We use an innovative multilevel dataset and a
retrospective cohort study originating in the Privatisation and Mortality project, the
biggest exercise in individual data-gathering addressing the post-socialist mortality
crisis to date (see Irdam et al., 2016).
2. Competing perspectives on the lived experience of shock therapy
We rely on the heterodox literature critiquing shock therapy as the theoretical backdrop
for interpreting the political–economic determinants of the post-socialist mortality
crisis. Several problems had plagued socialist economies; there was a consensus that the
outdated industrial structure required reform. However, the modalities of this trans-
formation have been controversial (see the review by Ellman, 1997a). The transition
orthodoxy, inuenced by the Washington Consensus, advocated shock therapy. Its pro-
ponents hypothesised that there was an ‘enormous scope for increases in living stand-
ards in a few years, particularly as resources are shifted out of the military-industrial
complex into other sectors’ (Lipton et al., 1992, p. 214). Heterodox, institutionalist
economists and economic sociologists criticised this approach. They argued that shock
therapy unnecessarily destroyed companies and weakened institutions necessary for a
successful transition (Amsden et al., 1994; Kolodko, 2000; King, 2003; Popov, 2012).
Scholars tend to concentrate on the macroeconomic component of shock therapy,
that is, stabilisation. In some cases, macroeconomic stabilisation was necessary to
bring down ination, but this could also be achieved through heterodox stabilisation
measures without radical neoliberal monetary policy instruments, such as a currency
board that binds policy-makers to internal devaluation (austerity, real wage reduc-
tion) as the only option for macroeconomic adjustment. Scholars disagree whether
the macro-policies implemented in Russia in the early 1990s constituted an abortive
attempt at shock therapy (Aslund and Layard, 1993) or delivered an actual shock
therapy package (Murrell, 1993; King, 2002). Whichever is the case, it is clear that
these macroeconomic measures failed to achieve the stated goal of shock therapy,
which is to reduce ination to a modest level in Russia. Whether this was due to
a aw in the concept or its execution continues to be debated. We do not seek to
resolve this dispute here, as we concentrate on the microeconomic aspects of the
chosen policies.
In addition to the macro-stabilisation component, shock therapists also advocated
for a ‘big bang’ of liberalisation and privatisation. The orthodox approach contends
that the faster these microeconomic reforms are implemented, the less opportunity
there is for resistance to block them, and the earlier the benets will materialise. This
microeconomic component of shock therapy (i.e. rapid privatisation, trade liberalisa-
tion, strict bankruptcy laws, and the abandonment of activist industrial policy) was
particularly problematic. It neglected the inherent slowness of restructuring com-
panies, the need to create a dense domestic economic fabric, and to build state
capacity.
Page 4 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
Advocates of neoliberal shock therapy often cite Poland or former Czechoslovakia
as proof of its superiority. However, these countries steered far away from microeco-
nomic shock therapy, avoiding unnecessary bankruptcies and building strong state in-
stitutions (Orenstein, 2001; King and Sznajder, 2006). This contrasts with Hungary’s
and Russia’s more radical neoliberal approach to microeconomic restructuring.
Consequently, deindustrialisation in Poland, Slovakia and the Czech Republic was
much less severe than in Hungary or Russia (Scheiring, 2020, pp. 133–86).
Here, we add to the evidence on the lived experience of this radical industrial trans-
formation by examining the deindustrialisation–mortality association. A large body
of scholarship is devoted to the health consequences of the post-socialist transform-
ation (see the systematic reviews by Scheiring et al., 2018a, 2019). This scholarship
comprises two approaches: those arguing that economic stress associated with neo-
liberal reforms is the main factor of the excess deaths in the 1990s (from now on, the
‘dislocation-despair approach’), versus those questioning this link and emphasising the
dysfunctional health habits of working-class people aggravated by the affordability of
alcohol during the early transition (from now on, the ‘dysfunctional culture’ approach).
Followers of the dislocation-despair approach pointed to how microeconomic
shock therapy was implicated in a large part of the excess deaths of the early 1990s.
Unemployment and labour market turnover strongly correlated with the wave of post-
socialist mortality (Walberg et al., 1998; Perlman and Bobak, 2009). Even those who
kept their job but experienced fear of job loss, higher workload and decreased control
at work were at higher risk of dying (Lundberg et al., 2007). Deindustrialisation led to
social disintegration, status loss, the loss of communities and a cascade of infrastruc-
tural, social and health problems, depression and despair (Kideckel, 2008; Scheiring
and King, 2022). This despair and distress correlated with increased mortality (Kopp
et al., 2007). A related stream of studies showed that mass privatisation was a critical
economic policy factor behind the transformation-associated economic crisis (Hamm
et al., 2012), also driving the life expectancy decline (Stuckler et al., 2009; Azarova et
al., 2017), and alcohol-related deaths in Russia (King et al., 2009), and in Hungary
(Scheiring et al., 2018b).
However, proponents of the dysfunctional culture approach question the centrality
of socio-economic dislocation. Cockerham (1997, p. 127) argued that ‘evidence is
lacking that stress per se can account for the sharp rise in male deaths throughout the
region’ and proposed that unhealthy lifestyles are the main culprits. Regressing death
rates on a measure of socialist industrialisation, Carlson and Hoffmann (2011, p. 375)
concluded that state socialist development policies emphasising industrial employ-
ment ‘created anomic conditions leading to unhealthy lifestyles and self-destructive
behaviour among men,’ explaining the rise in mortality until the middle of the 1990s.
Other proponents of the dysfunctional culture approach suggested that the
Gorbachev anti-alcohol campaign ‘saved’ many male lives, and these men started to
die after the liberalisation reforms, as alcohol became cheaper and more accessible,
allowing working-class people to indulge in their unhealthy drinking habits, leading to
the wave of excess deaths observed in the early 1990s (Treisman, 2010; Bhattacharya
et al., 2013). However, as we have noted previously, this is inconsistent with detailed
demographic analysis (Stuckler et al., 2012). In sum, according to the dysfunctional
culture approach, self-destructive health behaviour—in the case of the post-socialist
mortality crisis primarily referring to hazardous drinking—is not a result of the radical
economic policies adopted during the transition from socialism to capitalism but of a
Deindustrialisation and mortality Page 5 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
dysfunctional, anomie-laden culture inherited from the socialist past whose effect was
made worse by low alcohol prices during the transition.
3. The political economy of deindustrialisation and mortality
In this section, we present a theoretical framework that captures the human dimen-
sion of neoliberal shock therapy through the lived experience of deindustrialisation.
According to Kaldor’s second growth law, industrialisation leads to economy-wide
productivity growth through dynamic economies of scale: manufacturing is the en-
gine of growth (Kaldor, 1967). Industry has more linkages than other sectors of the
economy; thus, it has a bigger multiplier effect (Hirschman, 2013). Heterodox econo-
mists also showed that industrialisation is tightly interwoven with the development of
social cohesion (Ramazzotti, 2009) and is necessary to ensure the provision of basic
needs (Singh, 1979). Therefore, a high rate of loss of industrial capacity could create
a cascade of economic and social problems. This deindustrialisation is driven not only
by the quest for productivity gains but also by external shocks facilitated by neoliberal
microeconomic policies in emerging and core capitalist countries (Rodrik, 2016;
Tregenna, 2016; Felipe et al., 2018).
Deindustrialisation can be captured in two dimensions, output and employment
decline. As Felipe et al. (2018) also show, compared with employment, output change
is a weak predictor of prosperity and is under less pressure in emerging economies.
Industrial employment decline is particularly important for health outcomes. A se-
vere drop in industrial employment is most likely to signify mass plant closures,
while production might decline for many reasons without necessarily affecting the
socio-economic fabric of the town in which the plants affected are located. If plants
survive, they can respond to a future recovery by adding hours worked or re-employing
those laid off. However, much more is lost when plants are closed, with large-scale job
losses, as the costs, in labour and capital, of restarting are often extremely high.
Deindustrialisation entails a loss of a complex set of socio-economic linkages that
are very difcult to re-establish. As capital escapes from deindustrialised areas, local
infrastructures collapse, with a loss of services, such as health, education, family sup-
port, or transport that were either provided directly by the large plants or by the
local authorities that they helped fund. This creates a downward spiral of social and
economic disintegration, leading to a regional lock-in of rustbelts (Hassink, 2010).
Deindustrialisation could lead to a cascade of social problems, such as increasing in-
come inequalities as it creates winners and losers (Morris and Western, 1999), growth
of precarious jobs and in-work poverty (Burchell, 2009), or the erosion of commu-
nities and communal identities (Ramazzotti, 2009; Rodríguez-Pose, 2018), which in
turn could lead to ill health. The growth of service sector jobs is no substitute for the
lost industrial capacity as ‘most skills acquired in manufacturing travel very poorly to
service occupations’ (Iversen and Cusack, 2011, p. 326).
Deindustrialisation in post-socialist Eastern Europe was a particularly painful social
process. Socialist industry played a crucial role in workers’ lives, providing stable, life-
time jobs and a comparatively high salary. Industrial workers enjoyed high social status
as the backbone of state socialist societies (Burawoy and Lukács, 1992). Companies
also provided many free services, including healthcare, housing, holiday homes, sports-
and cultural facilities. Russian enterprises spent around 3%–5% of GDP on social
provision, while East European rms spent about half this amount, which is still very
Page 6 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
important for the beneciaries (Cook, 2007, pp. 39–40). Industrial employment also
contributed to social integration, vibrant work- and neighbourhood communities
(Kideckel, 2008; Scheiring, 2020). Thus, although industrial working-class culture was
also associated with harmful behaviours, industrial plants contributed to health and
well-being in many ways. These company functions were lost with mass plant closures,
with signicant health implications.
Borrowing from Scheiring and King (2022), we conceptualise industry as a social
institution, allowing us to capture deindustrialisation’s multidimensional health impli-
cations. The collapse of the industry as an institution engenders social disintegration,
leading to ruptures in economic production and social reproduction. These ruptures
entail job and income loss, increased exploitation, social inequality and the disruption
of services previously provided by industrial companies. These ruptures in economic
production affect social reproduction, leading to adverse outcomes, such as material
deprivation, job strain, fatalism, increased domestic workload, anomie, community
disintegration and alienation. The framework shown in Figure 2 captures the multidi-
mensional and long-term socio-economic effects of deindustrialisation.2
These ruptures in social reproduction are sources of psychosocial stress, through
which deindustrialisation gets embodied as dysfunctional health behaviour and ill
health. Deindustrialisation is a short-term adverse life event (acute stress through eco-
nomic deprivation, job loss). It also increases long-term strain (chronic stress through
increased workload, loss of status, loss of communities, the stress of inequalities), re-
quiring a signicant behavioural change to adapt (Thoits, 2010, p. 45). The accu-
mulation of stressors eventually depletes individuals’ physical or psychological coping
resources, negatively affecting psychological health, immune and cardiovascular
Fig. 2. Theoretical framework: deindustrialisation, social disintegration and health.
Note: Based on Scheiring and King (2022, p. 12).
2 Scheiring and King (2022) combine insights from Marx, Durkheim and Polanyi, elaborating on their
theoretical framework in more detail and show its usefulness analysing life-history interviews of workers af-
fected by deindustrialisation in Hungary.
Deindustrialisation and mortality Page 7 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
systems (McEwen and Stellar, 1993), inducing harmful health behaviour, such as haz-
ardous drinking (Cooper et al., 1992; Tomkins et al., 2007).
Marmot and Bobak (2000) found that stressful situations cause a higher secretion
of cortisol, endorphins, platelets, brinogens, brinolysis and other substances. These
affect the level of plasma lipids, blood coagulability, blood pressure, cardiovascular re-
activity, central obesity, responses to inammation or infection, depression, coronary
artery atherogenesis and a weakening of the immune system, that is, changes affecting
cardiovascular mortality. Psychosocial stress has been shown to indirectly affect health
via the increased use of stress relievers such as alcohol, tobacco and drugs, which in-
uence health and social behaviours and the ability to maintain emotional balance
(Cooper et al., 1992). Through this stress mechanism, deindustrialisation can lead to
worse self-reported health (Mitchell et al., 2000), lower life expectancy (Nosrati et al.,
2018), and elevated drug- and alcohol-related deaths (Autor et al., 2019; Venkataramani
et al., 2020), especially when accompanied with a mix of neoliberal policies (Walsh et
al., 2009), as the extant literature on Western Europe and the USA has established.
Based on these considerations, we hypothesise that (a) deindustrialisation was as-
sociated with mortality directly during the post-socialist mortality crisis, (b) indirectly
mediated by increased drinking as a dysfunctional stress-coping mechanism, and (c)
its effects cannot be reduced to inherited working-class culture or alcohol price pol-
icies. We test the third hypothesis derived from the main alternative explanation by
examining whether hazardous drinking habits were inherited from the past due to
socialist industrialisation and whether low alcohol prices activated this dysfunctional
working-class culture leading to elevated death rates.
4. Data and methods
4.1 Data
The analysis relies on two datasets, one on Hungary and the other on Russia. For
Hungary, we assembled a novel multilevel dataset comprising town and individual-
level data as part of the Privatisation and Mortality (PrivMort) project (for a detailed
description of the study protocol, see Irdam et al., 2016). When analysing Hungary, we
focus on 1989–95. This was when deindustrialisation and the wave of excess deaths
were the most severe. There was no deindustrialisation in Hungary after 1995, and it
was less signicant before 1988.3
The rst pillar of the multilevel dataset (level 1) comprises a broad range of
individual-level data collected in a retrospective cohort study between January 2014
and December 2015. In each selected town, interviewers visited randomly selected ad-
dresses for face-to-face interviews. The primary screening criteria were having parents,
siblings or partners living in the same settlement between 1980 and 2010. Interviewers
only interviewed one respondent from each household, regardless of the family size.
Following the method of retrospective cohort studies, respondents provided informa-
tion on their relatives (parents, siblings and spouses). This indirect approach to col-
lecting mortality data from relatives (also known as the ‘Brass technique’) produces
3 See Figure 3. The Hungarian Central Statistical Ofce stopped reporting annual town-level industrial
employment from 1998. Since we are interested in the effect of deindustrialisation in the mortality crisis, we
do not extend the multilevel models to cover 1996–97 when the economy and life expectancy were already
growing. Robustness checks using town-level panel data including 1996 and 1997 conrm the signicant
associations (see Supplementary Table E4).
Page 8 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
results that are robustly consistent with mortality estimates from administrative sources
in Russia (Bobak et al., 2003).
As respondents could not have died, we only included respondents’ relatives in the
analysis who lived in the surveyed towns during the 1980s and the 1990s. We further
reduced the sample by excluding cases with missing information on gender. In total,
42,800 people met these criteria and were included in this analysis: 24,377 men and
18,423 women. The average number of subjects per town was 881. The average re-
sponse rate was 85%. Survey questions and the resulting variables used for the analysis
include the following: relationship status of the survey respondent and the subject, vital
status, year of birth and death (if applicable), age in 1989, gender, education, smoking,
alcohol consumption and marital status. The subjects’ mean age in 1989 was 51 years.
Supplementary Table B1 presents descriptive statistics of the individual-level data for
Hungary. Supplementary Table B2 gives a detailed overview of the survey items and
variable coding.
The second pillar of the multilevel dataset (level 2) covers town-level data. We ran-
domly selected towns with inhabitants between 5,000 and 100,000 and industrial em-
ployment (as a share of total employment) in 1989 exceeding 30%. TÁRKI Social
Research Institute, Hungary’s leading polling agency, conducted individual surveys in
these towns as described above. Because our primary concern is the health impact of
the collapse of industrial plants, we concentrated on towns with signicant industrial
capacities. Predominantly agricultural small towns and villages experienced a different
type of economic shock.
Our sample includes only medium-sized towns in both Hungary and Russia.4 The
main reason for this is that we rely on the PrivMort project database, whose primary
aim was to analyse the association between privatisation and mortality. There are many
potential privatised companies in large cities, making data collection and linkage of
individuals to companies daunting. In contrast, in medium-sized towns, the ve lar-
gest companies capture a large enough share of the population to detect the effect of
privatisation. This dataset is not ideal, but it is still the best dataset available to analyse
the impact of deindustrialisation on mortality in Eastern Europe.
Regarding economic geography, Hungary is a highly polarised country, dominated
by Budapest and its agglomeration, which could skew the results; thus, the Budapest
agglomeration was excluded from the sample. This way, we generated a set of 52 towns
covering the entire geographical area of Hungary outside the capital. The sample rep-
resents the types of mid-sized towns where most Hungarians live and can be used to as-
sess the impact of deindustrialisation in non-metropolitan urban areas. Supplementary
Tables B3–B5 present an overview of town-level data for Hungary.
We collected data from the Hungarian Central Statistical Ofce on annual industrial
employment in each town and calculated industrial employment as a percentage of the
population. While the decline in manufacturing employment in the share of total em-
ployment might be a better measure, such data are not available at the level of towns
in Hungary. The Hungarian Central Statistical Agency only published the number of
persons employed in industry (manufacturing, mining, construction) at the level of
towns. Total employment is only available from the censuses conducted every 10 years.
It would be misleading to calculate the industrial employment share by relying on total
employment in 1990. Therefore, we concentrate on industrial employment decline
4 Hungary only has nine cities with more than 100,000 inhabitants.
Deindustrialisation and mortality Page 9 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
as a percentage of each town’s population. Additionally, we gathered data on towns’
unemployment, dependency ratio, death rates, income per capita, number of general
practitioners, outmigration and immigration as control variables.
The second dataset covers towns in the European part of Russia, also collected
under the PrivMort project’s auspices. Because Russia experienced two economic
crises during the transformation in the 1990s and more prolonged deindustrialisation
continuing at least until 1998, we analyse the 1991–99 period. The dataset includes
514 medium-sized towns in the European part of the country where the population
exceeded 3,000 but was smaller than 200,000 in 1989. These towns are in the 49 most
populous regions of Russia. The total population in the sampled towns was 20.9 mil-
lion, that is, 14.1% of Russia’s total population.
We assessed the quality of mortality data in the sample of 514 towns by calculating
the mean crude death rates across the towns each year between 1991 and 1999 and
compared them with the respective national-level statistics, as shown in Supplementary
Figure B1. While at lower levels, the line for the sample of towns generally tracks the
national-level trend. Therefore, we conclude that the selected towns provide a robust
sample to assess the association between deindustrialisation and mortality in European
Russia.
For Russia, gender-specic town-level death rates were unavailable, so we use the
overall death rates covering men and women. In addition to the income, age structure,
number of inhabitants, net migration, healthcare provision and housing conditions,
we were also able to collect data on regional-level average alcohol consumption (per
capita) and the regional-level average alcohol price.5 The consumption variable does
not include consumption of illicitly distilled alcohol, as the ofcial statistics did not
report these. The price of alcohol is expressed in roubles, corresponding to the average
for each year in the regional capital where the town is located, deated to 1991 by
the CPI to avoid bias arising from the hyperination characterising the early 1990s in
Russia. Even though 1991 was the last year of the Soviet command economy, deating
wages and prices to the price level in that year is a reasonable strategy, given the hyper-
ination in the following periods. For example, in January 1992, the rate was as high
as 245%, totalling 2509% yearly in 1992.
We use these two alcohol-related variables to check the robustness of the deindus-
trialisation hypothesis against the alcohol policy hypothesis, which suggested that the
excess deaths in the 1990s resulted from the relative drop in alcohol price after the
Gorbachev anti-alcohol campaign ended. We obtained information on industrial em-
ployment, death rates, income, age structure, number of inhabitants, migration, health
care provision, housing conditions, alcohol price and total alcohol consumption from
the Federal Statistics Service (Rosstat) yearbooks. Supplementary Table B6 presents
summary statistics, while Supplementary Table B7 shows the variable denitions for
the town-level data used in analyses covering Russia.
5 We did not include the unemployment variable due to the unreliability of this measure in the Russian
Federation during 1990s. Many formally employed workers were often sent on unpaid leave for indenite
durations or did not have gainful employment while being de facto employed. It has been widely recognised
that these records vastly underestimated the extent of unemployment in Russia during the transition years
(Standing, 1996; Grogan and van den Berg, 2001). The Hungarian data on registered unemployment is
more reliable, and the practice of sending employed workers on unpaid leave was uncommon in the country
in the 1990s.
Page 10 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
4.2 Estimation strategy
For Hungary, we t two types of models. First, we examine the relationship between
deindustrialisation and mortality using multilevel discrete-time survival analysis, with
individuals (level 1) nested in towns (level 2).6 The dependent variable in the multilevel
models in Hungary reects the odds of dying in the 52 towns during the 1989–1995
period, adding dummy variables for each year (xed effects).7
The primary independent variable is deindustrialisation. To capture the con-
textual, non-contemporaneous effect of deindustrialisation in the multilevel models,
we measure deindustrialisation as a change in industrial employment (manufacturing,
mining and construction) from 1989 to 1995, expressed as a percentage of each town’s
total population. The mean value of deindustrialisation in the 52 towns was 41%, and
the median was 40%. We group towns into two categories: those where deindustrialisa-
tion equals or exceeds 50% (n = 16) and those below 50% (n = 36). This approach to
dening deindustrialisation as a contextual factor is widely used in the extant literature
on deindustrialisation and health (Mitchell et al., 2000; Rind et al., 2014).
We use a series of covariates to control for town-level heterogeneity. First, the towns’
population size could also inuence mortality and the effect of modernisation and
industrialisation; therefore, we include a variable in our models measuring the mean
number of inhabitants. The age composition of towns could also inuence the pri-
mary association. To account for any effect, we use the towns’ dependency ratio (ratio
of those 0–14 and those over 64 to those of working age) to proxy for age structure.
We also control for the number of unemployed people and use the number of general
practitioners as a proxy for the towns’ health infrastructure. We control for the towns’
death rate in 1989 to reduce the potential for selection bias. We also assess the robust-
ness of the deindustrialisation–mortality association against the initial level of indus-
trial employment. We control for subjects’ age, gender, education, smoking, alcohol
consumption and marital status at the individual level. All models include additional
controls for the subjects’ relationship with the respondent to account for potential bias
due to the survey design.
Complementing the multilevel models, we also t models using only town-level ad-
ministrative time-series data, with dummies for towns and years (two-way xed effects)
and standard errors clustered on towns, estimated through Ordinary Least Squares.
In these panel models, the independent variable is continuous, measuring the annual
deindustrialisation compared to 1989, using the same data as constructing the con-
textual deindustrialisation variable in the multilevel models. The dependent variable is
the annual town-level all-age male death rate (number of male deaths/100,000). The
panel covers the same 52 towns with 310 observations (town-years) and is strongly
balanced, with only two missing data points. We use a similar set of control variables
as in the case of the multilevel models. The models lter out towns’ time-invariant
characteristics by design, and we also included year dummies to lter out unobserved
time-variant heterogeneity.
6 We specify random intercepts and random coefcients, allowing for the coefcient of deindustrialisation
to vary across towns. We rely on pseudo maximum-likelihood estimation with robust standard errors clus-
tered at the level of towns.
7 This is the established practice in discrete-time survival analysis (Rabe-Hesketh and Skrondal, 2012).
Survival analysis uses a ‘censored’ outcome variable. In our case, individuals who died after 1995 or were
still alive at the date of the questionnaire (2014 or 2015) were censored in 1995.
Deindustrialisation and mortality Page 11 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
We use only ecological (regional- and town-level) data for Russia, tting two-way
xed-effects models with clustered standard errors. The dependent variable meas-
ures town-level annual all-age deaths per 100,000 population (both genders). The
primary independent variable is the cumulative change in the town-level industrial
employment-to-population ratio (1991 as baseline). In our sample of 514 middle-sized
towns, the median decrease in industrial employment between 1991–99 was 39%.
Only 13% of towns suffered no reduction in industrial employment.
We t models covering the period from 1991 to 1999. We control for annual real wage
change measured in 1991 roubles, the number of inhabitants in 1,000s, dependency
ratio, net of out and in-migration per person, per-person oor space, the number of phys-
icians per 100,000 and hospital beds per 10,000. We conduct additional analyses to assess
whether the association between industrial employment and mortality is robust to alcohol
policy. First, we check the correlation between industrial employment and alcohol price
with pure alcohol consumption, using the same dataset and control variables as above.
We carried out all statistical analyses using STATA 16.0. Despite the multiple robust-
ness checks against potential selection bias and unobserved heterogeneity, establishing
causality is beyond the scope of our work. We hope the correlational patterns identied
will inspire future work designed to assess causal connections.
5. Results
5.1 An overview of post-socialist deindustrialisation
Figure 3 summarises the key trends of labour market transformation in Hungary and
Russia. While overall employment declined by 24% in Hungary between 1986 and 1995,
industrial employment fell by 43% in the same period, the years of the country’s most
pronounced liberalisation measures. This extreme deindustrialisation was very severe
and rapid by international standards. The most deindustrialised American metropolises,
such as Philadelphia, Cleveland and Chicago, lost around 30% of their manufacturing
labour force between 1972 and 1987 (Wallace et al., 1999, p. 115). Representing one
of the worst cases globally, it took 30 years for deindustrialisation to reach 60% in West
Central Scotland between 1971 and 2005 (Walsh et al., 2009). In comparison, within
less than a decade, almost every second person employed in manufacturing in Hungary
lost their job. This represents a massive shock to the social fabric, whose wide-ranging
implications have not received much attention until recently.
Russia experienced a steep decline in output and wages, but industrial employment
declined over a more extended period. Total employment fell by 15% between 1987
and 1999, while industrial employment dropped by 38% in the same period. Economic
adjustment to the new competitive capitalist environment took place more through the
income channel in Russia, leading to a more pronounced decline in aggregate income
and wages, while income in Hungary fell less, and adjustment took place more through
the employment channel (Boeri and Terrell, 2002; Gimpelson and Kapelyushnikov,
2011). Russian policy-makers allowed companies to keep their workers while reducing
and delaying their salaries.
5.2 Deindustrialisation and mortality in Hungary
Table 1 presents the results of multilevel survival models for Hungary, showing that de-
industrialisation is robustly and statistically signicantly associated with male mortality
Page 12 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
in each specication. The results for women were not signicant; see Supplementary
Table C1. Therefore, we concentrate on men in the subsequent models. The constant
in model 1 shows the unadjusted odds of dying in the 1989–1995 period (0.025), and
the variance of the constant at the town level is 0.017, suggesting large differences
across towns, underpinning the multilevel modelling strategy.8 Next, we included the
Fig. 3. The deindustrialisation of Hungary and Russia. Panel A: Hungary. Panel B: Russia.
Note: Hungary: industrial employment includes mining and excludes con-
struction. Russia: industrial employment includes mining and construction.
Source: Hungary: Population and employment: Feenstra et al. (2019),
manufacturing employment: Brada et al. (1994), and Laky (2000). Russia:
Estimated from: Obzor zanyatosti Rossii (1991-2000), Issue 1, 2002.
Moscow: Teis; Goskomstat SSSR. 1980–1989. Narodnoe khoziaistvo RSFSR:
statisticheskii ezhegodnik. Moscow: Goskomstat.
8 In multilevel modelling, this is called an ‘empty model’ or ‘variance components model’, regressing the
odds of dying on the intercept, which is simply used to partition out the variance across the levels.
Deindustrialisation and mortality Page 13 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
Table 1. Deindustrialisation and male mortality in Hungary, 1989–95, multilevel survival models
Dependent variable Subject dying between 1989 and 1995
(1) (2) (3) (4) (5) (6) (7)
Deindustrialisation 89–95 (ref.: Moderate (<50%))
Severe (≥50%) 1.146*1.182*1.182*1.141*** 1.188*1.149**
(0.066) (0.078) (0.078) (0.043) (0.091) (0.052)
Individual-level control variables
Age in 1989 (centred) 1.066*** 1.066***
(0.002) (0.002)
Education (ref.: Primary)
Secondary 0.855*** 0.853***
(0.040) (0.040)
Tertiary 0.678*** 0.675***
(0.060) (0.061)
Smoking (ref.: Quit or never)
Regularly 1.510*** 1.517***
(0.059) (0.060)
Alcohol (ref.: Max 1–4 times a month)
Daily or several times a week 1.229*** 1.227***
(0.039) (0.038)
Town-level control variables
Average population 89–95 10,000 persons
(centred)
0.989 0.995
(0.010) (0.014)
Average unemployment 89–95 (centred) 0.985 0.972**
(0.009) (0.010)
Average dependency ratio 89–95 (centred) 1.008 1.011
(0.008) (0.007)
Death rate 89 per 1,000 (centred) 1.003 0.992
(0.013) (0.011)
Constant 0.025*** 0.024*** 0.024*** 0.024*** 0.017*** 0.024*** 0.016***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Random effects
Town: var(Constant) 0.017*0.013*0.006 0.006 0.015 0.004 0.008
(0.008) (0.006) (0.005) (0.005) (0.009) (0.005) (0.005)
Page 14 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
Dependent variable Subject dying between 1989 and 1995
(1) (2) (3) (4) (5) (6) (7)
Town: var(Deindustrialisation) 0.031*0.031*0.017*0.028** 0.006
(0.014) (0.014) (0.008) (0.010) (0.004)
Prob > Chi2. 0.018 0.012 0.000 0.000 0.000 0.000
Year xed effects No No No Ye s Ye s Ye s Ye s
Type of relative No No No No Ye s No Ye s
No. of observations (person-years) 139,211 139,211 139,211 139,211 139,211 139,211 139,211
Notes: Multilevel survival models with random coefcient for deindustrialisation. The reference category for the dependent variable is the subject not dying between
1989 and 1995. Coefcients reported as odds ratios, with cluster-robust SEs in parentheses.
* p < 0.05, ** p < 0.01, *** p < 0.001.
Table 1. Continued
Deindustrialisation and mortality Page 15 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
primary independent variable, deindustrialisation. In this unadjusted setting, men in
severely deindustrialised towns have 14.6% higher odds of dying than men in moder-
ately deindustrialised towns. In model 3, we also specify a random coefcient, that is,
allow the effect of deindustrialisation to vary across towns. The effect size increases,
and the town-level variance of the constant decreases signicantly, suggesting that the
random coefcient for deindustrialisation improves the models. The chi-square test
also shows an improving model t. In model 4, we add year xed effects.
In model 5, we add individual-level controls for age, education, smoking, alcohol
consumption and relationship status. In model 6, we add town-level control variables
for population size, age structure, initial mortality rates and the unemployment vari-
able.9 These controls result in a signicant reduction of town-level variance. Finally,
model 7 includes each control variable simultaneously. In this fully adjusted model,
men in severely deindustrialised towns have 14.9% higher odds of dying in 1989–95
than men in moderately deindustrialised towns.
It is also worth mentioning that working-class men with primary education (the ref-
erence category) have 32.5% higher odds of dying than those with a college degree,
net of the effect of other factors. The only town-level control variable that reaches stat-
istical signicance is registered unemployment. A 1% increase in registered unemploy-
ment (thus the number of people receiving unemployment benets) is associated with
2.8% lower odds of dying among men in 1989–95.10
Figure 4 shows the predicted marginal probability of dying in the 1989–95 period
by the extent of deindustrialisation (using industrial employment change deciles) ad-
justed according to model 7. The association between deindustrialisation and mortality
is non-linear. People living in the most severely deindustrialised decile of towns have
a 28% higher probability of dying than those living in the least deindustrialised decile.
Next, we carry out a ‘placebo test’ to rule out the potential that pre-existing mor-
tality differentials may have played a role. The models with town-level controls al-
ready included death rates in 1989, showing no association with mortality in 1989–95.
However, our placebo test offers a more robust guarantee against selection bias. We
construct a separate dataset using subjects who lived in 52 towns during the 1980s be-
fore deindustrialisation began in 1989. We use this dataset to investigate the town and
individual-level determinants of subjects’ death between 1985 and 1988, including the
deindustrialisation level after 1989.
Supplementary Table C2 shows that we found no statistically signicant pre-existing
mortality differences among men living in the towns that later underwent different in-
dustrial transformations. Comparing the coefcient for deindustrialisation reported in
Table 1 (model 7) and the coefcient obtained through the placebo test, we can ascer-
tain that the ‘treatment effect’ is signicantly different from the ‘placebo effect’.11 The
Z-score for the difference between the two coefcients is 2.529, corresponding to p =
0.006 (one-tailed test). This means that the effect of deindustrialisation is signicantly
different in the ‘treatment group’ (1989–95) than in the ‘placebo group’ (1985–88).
9 Outmigration, the number of GPs and town-level income were not signicantly associated with the de-
pendent variable, so we removed them from the analysis to avoid multicollinearity. Supplementary Table D3
presents the association between these additional town-level control variables and mortality.
10 We offer an interpretation of this nding in the discussion section.
11 We can use a Z-test to compare coefcients obtained through maximum-likelihood estimation (Brame
et al., 1998).
Page 16 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
Proponents of the dysfunctional working-class culture thesis argue that hazardous
drinking is inherited from the past due to socialist industrialisation that led to nor-
mative disorientation and unhealthy lifestyles among socialist workers (Cockerham,
1997; Carlson and Hoffmann, 2011). In Hungary, there was no anti-alcohol cam-
paign, so the increased affordability of alcohol in the post-campaign period should
not play a major role. Therefore, to assess the explanatory power of the dysfunctional
working-class culture thesis, we analyse how industrial employment in 1989 inuences
the association. The correlation between initial industrial employment and deindus-
trialisation is very low (r = −0.04). Thus the initial level of industrial employment does
not seem to explain subsequent deindustrialisation.
Next, we analyse whether industrial employment in 1989 inuences mortality differ-
entials in multilevel regression models. We split the sample of 52 towns in half, above
and below the median industrial employment in 1989. According to the dysfunctional
working-class culture hypothesis, a higher share of industrial employment should cor-
relate with higher mortality in 1989–95. As we report in model 7, Supplementary Table
C3, we nd the opposite. Men in severely deindustrialised towns with below-median
Fig. 4. Predicted probability of men’s dying by deindustrialisation in 52 towns of Hungary, 1989–95.
Note: Predicted probabilities from logistic regression with clustered standard
errors. The dependent variable is subject dying between 1989 and 1995 (ref.:
subject not dying between 1989 and 1995), the primary independent vari-
able is deindustrialisation (in deciles), adjusted for town-level (total death rate
in 1989, average unemployment in 1990–95, and average dependency ratio
1989–95), and individual-level control variables (smoking, alcohol consump-
tion and education). Number of observations at (person-years): 139,211,
number of groups (towns): 52.
Deindustrialisation and mortality Page 17 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
industrial employment in 1989 have higher odds of dying between 1989 and 1995 than
men in towns with above-median industrial employment in 1989, where the effect of
deindustrialisation is not signicant. This means that the positive association of dein-
dustrialisation with mortality is not a spurious effect of initial industrial employment.
These results contradict the dysfunctional industrial culture hypothesis.
Figure 5 shows the unadjusted correlation between town-level mortality and dein-
dustrialisation in Hungary. This visual inspection of the raw data reveals a potential
association at the town level, with a bivariate correlation of r = 0.314.
Table 2 shows that the town-level two-way xed-effects panel regressions conrm the
multilevel model results. Deindustrialisation is robustly associated with male mortality
in every model. Model 4—controlling for towns’ size, age structure, unemployment,
per capita income, outmigration and the number of GPs—shows a highly signicant
association (p < 0.01); 1% deindustrialisation correlates with 1.74 additional male
deaths per 100,000 inhabitants.
5.3 Deindustrialisation and mortality in Russia
As described above, we investigate the correlation between deindustrialisation and
health in Russia using town-level xed effects modelling covering mid-sized towns in
European Russia. Table 3 presents the regression results.
The association between deindustrialisation and mortality is positive and highly sig-
nicant in all four models. We start by tting regressions with year dummies and town
Fig. 5. The town-level association between mortality and deindustrialisation in Hungary.
Note: Unadjusted, bivariate correlation. Deindustrialisation is measured as
the percentage decline in industrial employment ratio from 1989 to 1995.
Number of observations (towns): 52.
Page 18 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
xed effects. Given declining trends in industrial employment, we can interpret the
coefcients in model 1 as follows: a decrease in industrial employment of 1% was asso-
ciated with 0.407 (p < 0.01) more deaths per year per 100,000 population in the 1991–
99 period. This estimate is not attenuated by accounting for per capita income and
square metres of accommodation in model 2. The association becomes even stronger
after controlling for the number of inhabitants, dependency ratio and per capita net
migration in model 3. Finally, in the fully adjusted model 4, a 1% deindustrialisation is
associated with 0.516 additional deaths per year per 100,000 population (p < 0.001).
Figure 6 shows the predicted town-level male deaths by deindustrialisation in Russia,
suggesting a strong association.
The proponents of the dysfunctional culture approach questioned the idea that
stress caused by economic dislocation would be an essential determinant of haz-
ardous drinking, arguing for the centrality of dysfunctional culture amplied by the
high affordability of alcohol in Russia. If this hypothesis were valid, we would expect
(a) a positive association between industrial employment and alcohol consumption
and (b) that controlling for alcohol price eliminates the association between dein-
dustrialisation and alcohol consumption. Table 4 shows the coefcients and standard
errors for a percentage share of industrial employment predicting alcohol consump-
tion for this period in models 1 and 3 (full model). To facilitate cross-model com-
parison, we also show the effect of alcohol price on its consumption in a bivariate
Table 2. Deindustrialisation and male mortality in Hungary, 1989–95, town-level xed-effects panel
models
Dependent variable Male deaths per 100,000 population
(1) (2) (3) (4)
Cumulative deindustrialisation from 1989 (%) 1.193*1.526*1.697** 1.737**
(0.580) (0.581) (0.578) (0.589)
Income per capita 10,000 HUF (centred) 8.720 1.663
(6.339) (7.010)
Unemployment (centred) −0.344 −0.256
(2.480) (2.618)
Population 10,000 persons (centred) −1.138 −3.322
(101.033) (101.330)
Dependency ratio (centred) −13.201** −12.444*
(4.592) (5.343)
Outmigration % of population (centred) 1.381 1.796
(11.701) (11.949)
No of GPs per 10,000 inhabitants (centred) 8.496 7.559
(11.314) (11.896)
Constant 644.139*** 878.826*** 748.686*** 785.867***
(13.109) (180.021) (34.478) (182.541)
Year xed effects Yes Ye s Ye s Ye s
R20.037 0.049 0.071 0.071
No. of observations (town-years) 310 310 310 310
No. of groups (towns) 52 52 52 52
Note: Coefcient estimates from town-level two-way xed-effects panel models, cluster-robust SEs in
parentheses.
* p < 0.05, ** p < 0.01, *** p < 0.001.
Deindustrialisation and mortality Page 19 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
model 2 and in a model with industrial employment share, price of alcohol and other
covariates, namely wage in 1991 roubles, number of inhabitants in 1000s and per-
person oor space.
Contrary to the expectations derived from the dysfunctional culture hypothesis, the
association between industrial employment and alcohol consumption is signicant and
negative in both specications, indicating an increase of 0.013 l per person per year
with every 1% decrease in the industrial employment-per-population ratio. Thus, de-
industrialisation seems to be associated with increased alcohol consumption, ltering
out the effect of alcohol price. This suggests that hazardous alcohol consumption was
a stress-coping strategy, above and beyond the effects of alcohol price. The correlation
of price with alcohol consumption is also signicant in both models; however, the co-
efcient has the opposite sign, as suggested by the dysfunctional culture argument. In
full model 3, each ruble of increase in alcohol price correlates with 0.017 l (p < 0.001)
more pure alcohol consumption.
Table 3 Deindustrialisation and mortality in Russia, 1991–99, town-level xed-effects
panel models.
Dependent variable Deaths per 100,000 population
(1) (2) (3) (4)
Deindustrialisation (%) 0.407*** 0.394*** 0.520*** 0.516***
(0.105) (0.106) (0.095) (0.096)
Income per capita (in
1991 RUB) (centred) 0.100** 0.147***
(0.034) (0.037)
Floor area per person
(centred) 1.103 0.646
(1.425) (1.478)
Population 10,000
persons (centred) −37.807** −35.995**
(11.826) (11.399)
Dependency ratio
(centred) 1.583*** 1.420**
(0.389) (0.409)
Migration % of
population (centred) −37.355 −38.062
(41.380) (41.216)
Physicians per 100,000
(centred) −0.551 −0.410
(0.461) (0.462)
Constant 1,193.056*** 1,198.453*** 1,205.301*** 1,210.978***
(6.359) (7.009) (7.334) (8.102)
R20.189 0.180 0.239 0.230
No. of observations
[town-years] 4,527 4,450 4,02S9 3,968
No. of groups [towns] 514 513 493 492
Notes: Deindustrialisation is dened as the cumulative change in the industrial employment-to-population
ratio (1991 as baseline). All models include controls for year xed effects (year dummies)—Cluster-robust
SEs in parentheses.
* p < 0.05, ** p < 0.01, *** p < 0.001.
Page 20 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
6. Robustness checks
First, we assess the stability of multilevel modelling results using a different estimation
strategy. Logistic regression is sensitive to model specication and can yield biased re-
sults compared to linear models (Mood, 2009). Therefore, we re-estimate regressions
from Table 1 using multilevel linear predicted probability modelling. As Supplementary
Table D1 shows, the positive association between deindustrialisation and male mor-
tality remains robust and signicant in each model.
We also re-estimate the multilevel association between mortality and deindustrialisa-
tion, operationalising the latter as a categorical variable with three and four categories.
These results shown in Supplementary Table D2 again conrm the main ndings.
Supplementary Table D3 shows that the fully adjusted association also holds when
we measure deindustrialisation as a continuous annual industrial employment ratio
(mean = 17.5, standard deviation [SD] = 8.5). In this specication, the effect is even
bigger; a 50% higher industrial employment ratio implies 40% lower odds of dying.
Supplementary Table D4 shows multilevel modelling estimates with additional
town-level control variables (income per capita, no of GPs, industrial employment
1989). With these new covariates, the effect of deindustrialisation is even bigger (OR
= 1.193, p < 0.001). Industrial employment in 1989 was measured as a continuous
Fig. 6. Predicted town-level male deaths by deindustrialisation in 490 towns of Russia, 1991–99.
Note: Linear predictions from town-level xed-effects panel models adjusted
for income per capita, oor area per person, population, dependency ratio,
migration, number of GPs and unobserved time-varying heterogeneity (year
xed effects), and the dependent variable lagged by 1 year. Number of obser-
vations (town-years): 3,589, number of groups (towns): 490.
Deindustrialisation and mortality Page 21 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
variable (min.: 0.1, max.: 0.63), expressing the share of the total population employed
in industry in percentages. The association is negative but not signicant, again sug-
gesting that our main results cannot be explained by the dysfunctional culture hypoth-
esis, according to which this association should be signicantly positive.
Next, we check against another potential source of selection bias stemming from mi-
gration. Towns that lose younger or healthier people would exhibit a higher mortality
rate. We test for in- and outmigration during the analytical period and following the
analytical period until the individual surveys were conducted. As Supplementary Table
D5 shows, none of the migration-related variables is signicant, and their inclusion
does not inuence the signicance of the effect of deindustrialisation, which is even
bigger in this setting (1.143 ≤ OR ≤ 1.215).
We also test the dysfunctional culture hypothesis by splitting the town-level sample
into half (above and below-median industrial employment in 1989) end estimating
the panel models as specied above this way. The results in Supplementary Table D6
again are inconsistent with the dysfunctional culture thesis by showing that the effect
of deindustrialisation on mortality is particularly severe in towns with lower industrial
employment at the end of the 1980s. Supplementary Figure D1 conrms the same
for Russia; the association between the change in death rate and initial industrial em-
ployment is at. The superimposed regression lines demonstrate that the initial level of
industrialisation does not correlate with subsequent changes in death rates.
In Supplementary Table D7, we re-estimate the panel models with a lagged de-
pendent variable. In the fully adjusted model 6 with lagged male death rates, the
Table 4. Industrial employment and alcohol consumption in Russia, 1991–999, town-level xed-
effects panel models
Dependent variable Regional-level annual alcohol consumption
(litres of pure alcohol per capita)
(1) (2) (3)
Industrial employment (%) −0.013*** −0.013**
(0.004) (0.004)
Alcohol price, deated (centred) 0.023*** 0.017**
(0.006) (0.006)
Income per capita (in 1991 RUB) (centred) 0.0005*
(0.0002)
Population 10,000 persons (centred) −0.039
(0.070)
Floor area per person (centred) 0.010
(0.08)
Constant 5.178*** 4.948*** 5.259***
(0.081) (0.037) (0.086)
R20.410 0.417 0.421
No. of observations[town-years] 4,532 4,495 4,417
No. of groups[towns] 514 514 513
Note: Deindustrialisation is dened as the annual industrial employment-to-population ratio in percent.
All models include controls for year xed effects (year dummies)—Cluster-robust SEs in parentheses.
* p < 0.05, ** p < 0.01, *** p < 0.001.
Page 22 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
association between deindustrialisation and death rates is even stronger (b = 2.342,
p < 0.001). The same holds for results using data on Russian towns. Supplementary
Table D8 shows that the deindustrialisation–mortality association remains signicant
in Russia when including a lagged dependent variable (b = 0.499, p < 0.001 in the
fully adjusted model 4). Next, we measure deindustrialisation as a continuous annual
industrial employment ratio (mean = 16.2, SD = 8.2). These panel models imply rst
differences in this case and not cumulative change. Supplementary Table D9 shows
that the association between the industrial employment ratio and death rates is ro-
bustly and strongly negative in Hungary, conrming the positive association between
deindustrialisation and mortality.
Finally, we test the multilevel and panel results against potentially inuential
outliers. First, we identify outliers in Supplementary Figure E1 for Hungary, and
Supplementary Figure E2 for Russia, by checking the deviation of town-level death
rates from the overall sample average. Supplementary Table E1 shows that multilevel
model estimates for Hungary are robust when excluding the three towns with the
highest and lowest mortalities. Supplementary Table E2 shows the same for town-level
panel models. Supplementary Table E3 excludes 20 towns with the lowest mean death
rates and 20 towns with the highest mean death rates in 1991–99 in Russia. A signi-
cant mortality effect remains for deindustrialisation. Supplementary Table E4 shows
that Hungarian panel models remain signicant and robust when including data from
1996 and 1997, while Supplementary Table E5 shows the same when excluding 1989
and 1990. In Supplementary Table E6, we retain a signicant mortality effect for dein-
dustrialisation in Russia, excluding 1998 and 1999. Finally, Supplementary Table E7
shows that the same holds when excluding 1991 and 1992.
7. Discussion and conclusions
In this article, we utilised an innovative multilevel dataset on the economic determin-
ants of mortality in Eastern Europe. We showed that deindustrialisation was signi-
cantly associated with male death rates in Russia and Hungary in the 1990s, but social
and economic policies have offset Hungary’s more immense industrial employment
loss. Furthermore, we showed that the association between deindustrialisation and
the post-socialist mortality crisis is not a spurious result of a legacy of dysfunctional
working-class health culture aggravated by low alcohol prices during the early years
of the transition. Even if dysfunctional working-class culture played a role, our re-
sults suggest that researchers should not pit culture against political–economic dis-
locations. Hazardous drinking was, in part, a dysfunctional coping strategy in response
to a highly stressful experience. To understand this process, we proposed a political–
economic theory of deindustrialisation and mortality based on heterodox economics
and economic sociology. This conceptual framework identies deindustrialisation as
a process of social disintegration rooted in the lived experience of neoliberal shock
therapy. Deindustrialisation acts as a source of stress and despair through individual-
level (income loss, job loss, precarity) and community-level channels (loss of sense of
community, increased inequality, loss of company-related services), leading to reduced
coping ability and ill health.
Our study has some important limitations. The individual-level sample used for the
multilevel models in Hungary is constructed from the close kin of those who were
still living in the specic town in 2014–15. This data collection procedure excludes
Deindustrialisation and mortality Page 23 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
those families that were in the town from 1989 to 1995 but not when the interviews
were conducted. To address this weakness, our models controlled for outmigration and
in-migration. Furthermore, the United Nations has validated and recommended this
indirect demographic method for censuses (United Nations, 1983). Previous studies
conducted in Russia using a similar method also showed results consistent with the
ofcial data and the extant literature (Bobak et al., 2003; Azarova et al., 2017).
In response to the much-debated study by Ruhm (2000), which found that mor-
tality decreases during economic recessions, Arthi et al. (2017) showed that mortality
increased during a recession that they analysed after adjusting for migration. The fact
that adjusting for migration turns the recession–mortality association positive implies
that if migration biases such associations, it does so because less-healthy people mi-
grate away from affected areas. It is plausible to assume that people negatively affected
by deindustrialisation migrated in search of better conditions, while those who could
keep their jobs stayed in the town. While this runs contrary to the oft-quoted healthy
migrant effect, it is consistent with some populations moving away from adverse eco-
nomic conditions, such as the Irish in the 19th and 20th centuries (Delaney et al.,
2013). Although we cannot robustly rule out such a bias, we nd no evidence for it,
and even if it exists, it makes our estimates more conservative than they really are.
Second, the retrospective cohort method might also be prone to recall bias. People
might err a few years when recollecting when their relative died. We also found subjects’
answers about the occupation and place of work of their relatives unreliable. Thus, we
excluded these variables from the analysis. Because of recall bias, we could not gen-
erate an annual death variable for ne-tuned year-on-year mortality risk analysis and
had to rely on an indicator variable encompassing the 1989–95 period. Therefore,
we could not generate a panel dataset using individual data. However, our town-level
xed-effects panel models using administratively collected death rates from two coun-
tries conrmed the multilevel models.
Third, the towns involved in the analysis might differ in unobservable characteris-
tics in both countries. We tackled this limitation by controlling for pre-existing health
differentials, suggesting that the towns did not differ in terms of health before dein-
dustrialisation. Furthermore, the xed-effects models, by denition, lter out unob-
served time-invariant heterogeneity. We also controlled for time-variant global factors
by including year xed and tested our models against the inclusion of a lagged de-
pendent variable. We could also control for the most critical characteristics identied
in the literature. Nevertheless, unobserved heterogeneity might still be an issue, and
the associations cannot be interpreted as causal links.
Theoretically, it is possible to handle selection and heterogeneity bias statistically.
However, designing such a quasi-experiment for our study faces insurmountable chal-
lenges. There is either no administrative data available for the pre-treatment period
or what is available is not reliable. For example, neither the Eurostat nor the IMF
publishes comparative national-level data on social spending before 1995 for most
countries in post-socialist Eastern Europe. The same data availability limitation is even
more severe at the town level. Fine-grained digitalised statistics from the socialist era
are a rarity, and if they exist, collecting them from non-digital archives is costly. While
mortality data from the socialist period are reliable, the quality of information on town-
level social and economic characteristics used in the present paper is hard to ascertain.
All in all, our research question, and the data available to answer it severely limit
the viability of credible quasi-experimental designs. However, we do not think that
Page 24 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
quasi-experimental methods are magic bullets, as they have some important limitations
(see Cook et al., 2002). First, what quasi-experiments gain in internal validity, they
frequently lose in external validity. The more they resemble an experiment, the more
they can identify causality, but the less generalisable they become outside the study
population. Second, quasi-experiments rest on assumptions about some exogenous
source of ‘quasi-random’ treatment variation that cannot be proven statistically. Third,
researchers’ pre-analytical visions and ‘meta-scientic’ commitments irreconcilably
inuence the questions they ask, the variables they dene and the associations they
seek. Fourth, quasi-experiments, even in the most highly ranked journals, are at risk
of p-hacking12 that distorts the overall picture emerging from the research, leaving
much room for researchers’ preferences to inuence the results (Brodeur et al., 2020).
As Kuhn (2012) famously argued, these pre-analytical visions are so important that
science typically does not change by gradually accumulating knowledge but through
scientic revolutions as incommensurable paradigms succeed each other.
Altogether, while quasi-experimental methods bring signicant added value to re-
search, they should not devalue studies of phenomena for which it is very difcult, if
not impossible, to design a credible quasi-experiment. As Angus Deaton (2020, p. 1)
emphasises in his criticism of randomised control trials: ‘It is a mistake to put method
ahead of substance.... Methodological prejudice can only tie our hands. Context is al-
ways important, and we must adapt our methods to the problem at hand’.
The fourth limitation is that our samples are restricted to medium-sized towns in
both countries, curtailing the generalisability of our results. For example, the detri-
mental effect of deindustrialisation could be buffered in big cities due to the broader
availability of jobs in non-industrial branches of the economy. Smaller towns and vil-
lages experienced a different shock related to the collapse of agriculture, which would
require a different sample. Also, working-class culture likely differs across cities of
various sizes. If people living in large cities, who are omitted from our sample, are more
prone to hazardous drinking, then our models underestimate its effect. However, if
hazardous drinking is more severe in small- and medium-sized industrial towns, then
our models magnify its effect compared to the whole population of the two countries.
Even though such differences across the towns might exist, we showed that the mor-
tality trend in our sampled towns in Russia mirrors the national mortality trend. This
parallel trend suggests that large cities omitted from our sample did not experience
signicantly different mortality shocks. We also took account of the towns’ population
size in every regression model to reduce the bias arising from the size of the towns.
Nevertheless, our results cannot be directly generalised to small villages and the largest
cities. Notwithstanding the limited generalisability from our sample of medium-sized
towns, we nd it remarkable that the models behave consistently in the Hungarian and
Russian samples.
Finally, during the transformation from socialism to capitalism, many policies and in-
stitutions changed simultaneously. This makes it difcult to isolate deindustrialisation,
even if this was desirable, given the implausibility of a single causal factor. However,
several of these policies, such as political–institutional change, did not vary between
towns, only between countries, and thus should not bias our results. Following the ex-
tant literature, we controlled for health service variables, which were not signicant and
12 P-hacking ‘occurs when researchers try out several statistical analyses and/or data eligibility specica-
tions and then selectively report those that produce signicant results’ (Head et al., 2015, p. 1).
Deindustrialisation and mortality Page 25 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
did not inuence the primary association (Brainerd and Cutler, 2005). Yet, this should
not be interpreted as suggesting that the post-socialist health system was functioning
well, which it was not (Andreev et al., 2003). Other elements of shock therapy—such
as bankruptcy policies, rapid import opening, capital account liberalisation and rapid
mass privatisation—could have played a role, and deindustrialisation likely picks up
some of their effect (Amsden et al., 1994; Tregenna, 2008).
Notwithstanding these limitations, our results align with several bodies of research.
First, we found that deindustrialisation was correlated only with men’s health in the
short run, echoing the existing literature on the health effect of privatisation in Russia
(Azarova et al., 2017). The gendered expectations of the male breadwinner model
could have made men more vulnerable to labour market upheaval, especially because
men were overrepresented in industrial jobs. However, men might benet more from
the new jobs replacing old industrial ones in the long run. Women in towns with more
privatisation had higher odds of dying compared to women in towns with higher state
ownership in the 1995–2004 period in Hungary (Scheiring et al., 2018b).
Second, concerning the weakness of the alcohol affordability argument, others also
found that the increase in the price of vodka led to a rise in the consumption of il-
licit and surrogate alcohol (Goryakin et al., 2015). A recent study by Azarova et al.
(2021) also showed that the correlation between the Gorbachev anti-alcohol campaign
in the 1980s, alcohol prices in the 1990s, and mortality reported by Bhattacharya et al.
(2013) and Treisman (2010) is likely spurious. Furthermore, death rates improved in
the second half of the 1980s and worsened in the 1990s in several countries, such as
Hungary, that did not have a Soviet-type alcohol campaign that led to a large alcohol
price increase in the 1980s and a subsequent fall in the 1990s (Ellman, 1997b, p. 358).
Third, the negative association between registered unemployment and mortality
likely reects the cushioning impact of unemployment benets. The maximum un-
employment benet duration in Hungary between 1989 and 1993 was 24 months,
with the unemployed receiving 70% of their previous gross earnings during the rst
6 months of unemployment (Vodopivec et al., 2005). Hungary’s privatisation strategy
was also more gradual and orderly than Russia’s rapid mass privatisation and the at-
tendant disintegration of bureaucratic capacities. These social and economic policy
differences likely played an important role in offsetting deindustrialisation’s health
effect in Hungary.
The negative association between registered unemployment and mortality also sup-
ports our hypothesis that the adverse health effect of deindustrialisation goes beyond
job loss. There were many people in Hungary (and in other countries in the region)
who lost their job and did not register as unemployed. Registered unemployment only
reects the net balance of those who are actively looking for a job through formal state
institutions, thus overlooking the issue of high labour market turnover, high inactivity,
as well as the community-level implications of deindustrialisation. Industrial employ-
ment loss captures these multidimensional ripple effects, while registered unemploy-
ment does not.
In contrast to Hungary, the Russian welfare state was less developed, and its bur-
eaucratic capacity was especially severely affected by mass privatisation and transform-
ation (Hamm et al., 2012; Irdam et al., 2015). The unemployment insurance system
was practically non-existent (Clarke, 1998, pp. 50–52). The average unemployment
benet in 1993 was 12% of average pay, 18% of average pay in 1994, and the average
length of benet was 5 months. As a result, only 12% of unemployed people received
Page 26 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
any help in Russia in 1992. The welfare state’s weakness, the chaotic mass privatisation
programme, the precarity of employment, and the more severe income loss could ex-
plain why Russia experienced a more severe wave of excess deaths.
Of course, these two countries were already different before the transition in terms
of initial conditions, such as the extent of industrial distortions, the size of the economy
and non-state sector, and the structure of safety nets, among others. Future research
should look in more detail at how these pre-existing differences conditioned policy
choices that inuence how deindustrialisation affects health. As King et al. (2022)
argue, comparative historical research has much to add to our understanding of the
causes of mortality crises. We hope our results will inspire further work to compare
countries within the region and explore the similarities and differences between the
post-socialist mortality crisis and deaths of despair plaguing the North American
rustbelt.
The insights emerging from our results are relevant for other countries beyond post-
socialist Eastern Europe. As Case and Deaton (2020, p. 108) noted, ‘it is no exagger-
ation to compare the long-standing misery of these Eastern Europeans with the wave
of despair that is driving suicides, alcohol, and drug abuse among less-educated white
Americans’. Our study adds an important element to such comparisons. Robust wel-
fare provisions, active labour market policies and community regeneration programmes
are necessary to offset the adverse health effect of deindustrialisation (Venkataramani
et al., 2021). Well-designed strategies, such as regionally targeted industrial policies,
can create jobs and contribute to the health of vulnerable populations (Rodrik, 2004).
A green new deal and just transition policies (Newell and Mulvaney, 2013) can steer
society towards a lower carbon future underpinned by equity, justice and workers’
health in rustbelt areas.
Supplementary data
Supplementary data are available at Cambridge Journal of Economics online.
Acknowledgements
The authors acknowledge the nancial support of The European Research Council
(ERC) [grant number 269036]; as well as the support of the Cariplo Foundation and
the Lombardy Region as part of the POTES project (‘Enhancing the Attractiveness
of Lombardy through the Excellence of Health Economics Research’—Ref. 2017-
2077). The funding bodies had no role in study design, data collection, data analysis,
and reporting of this study. The authors also express their gratitude for the superb
research assistance provided by Péter Mátyás, András Kövesdi, as well as Dr. Irina
Kolesnikova at The Institute of Economics of the National Academy of Sciences of
Belarus (NASB).
Bibliography
Amsden, A. H., Kochanowicz, J. and Taylor, L. 1994. The Market Meets Its Match: Restructuring
the Economies of Eastern Europe, Cambridge, Harvard University Press.
Andreev, E. M., Nolte, E., Shkolnikov, V. M., Varavikova, E. and McKee M. 2003. The evolving
pattern of avoidable mortality in Russia, International Journal of Epidemiology, vol. 32, no. 3,
437–46.
Deindustrialisation and mortality Page 27 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
Arthi, V., Beach, B. and Hanlon, W. W. 2017. ‘Estimating the Recession-Mortality Relationship When
Migration Matters’, Working Paper 23507, Cambridge, MA, National Bureau of Economic
Research.
Aslund, A. and Layard, R. 1993. Changing the Economic System in Russia, New York, St. Martin’s
Press.
Autor, D., Dorn, D. and Hanson, G. 2019. When work disappears: manufacturing decline and
the falling marriage market value of young men, American Economic Review: Insights, vol. 1,
no. 2, 161–78.
Azarova, A., Irdam, D., Gugushvili, A., Fazekas, M., Scheiring, G., Horvat, P., Steer, D.,
Kolesnikova, I., Popov, V., Szelényi, I., Stuckler, D., Marmot, M., Murphy, M., McKee,
M., Bobak, M. and King, L. 2017. The effect of rapid privatisation on mortality in mono-
industrial towns in post-Soviet Russia: a retrospective cohort study, The Lancet Public Health,
vol. 2, no. 5, e231–8.
Azarova, A., Scheiring, G., Ash, M. and King, L. 2021. ‘Did Alcohol Policy Really Cause the
Postsocialist Mortality Crisis? Revisiting the Rebound and Affordability Hypotheses’, Working Paper
Series, Number 540, Amherst, MA, Political Economy Research Institute (PERI), University
of Massachusetts,.
Bhattacharya, J., Gathmann, C. and Miller, G. 2013. The Gorbachev anti-alcohol campaign and
Russia’s mortality crisis, American Economic Journal: Applied Economics, vol. 5, no. 2, 232–60.
Bobak, M., Murphy, M., Rose, R. and Marmot, M. 2003. Determinants of adult mortality in
Russia: estimates from sibling data, Epidemiology, vol. 14, no. 5, 603–11.
Boeri, T. and Terrell, K. 2002. Institutional determinants of labor reallocation in transition,
Journal of Economic Perspectives, vol. 16, no. 1, 51–76.
Brada, J. C., Singh, I. and Török A. 1994. Firms aoat and rms adrift: Hungarian industry and
the economic transition, Eastern European Economics, vol. 32, no. 1, 1–104.
Brainerd, E. and Cutler, D. M. 2005. Autopsy on an empire: understanding mortality in Russia
and the former Soviet Union, The Journal of Economic Perspectives, vol. 19, no. 1, 107–30.
Brame, R., Paternoster, R., Mazerolle, P. and Piquero, A. 1998. Testing for the equality of
maximum-likelihood regression coefcients between two independent equations, Journal of
Quantitative Criminology, vol. 14, no. 3, 245–61.
Brodeur, A., Cook, N. and Heyes, A. 2020. Methods matter: p-hacking and publication bias in
causal analysis in economics, American Economic Review, vol. 110, no. 11, 3634–60.
Burawoy, M. and Lukács, J. 1992. The Radiant Past: Ideology and Reality in Hungar y’s Road to
Capitalism, Cambridge, Cambridge University Press.
Burchell, B. 2009. Flexicurity as a moderator of the relationship between job insecurity and psy-
chological well-being, Cambridge Journal of Regions, Economy and Society, vol. 2, no. 3, 365–78.
Carlson, E. and Hoffmann, R. 2011. The state socialist mortality syndrome, Population Research
Policy Review, vol. 30, no. 3, 355–79.
Case, A. and Deaton, A. 2020. Deaths of Despair and the Future of Capitalism, Princeton, NJ,
Princeton University Press.
Chenet, L., McKee, M., Fulop, N., Bojan, F., Brand, H., Hort, A. and Kalbarczyk, P. 1996.
Changing life expectancy in central Europe: is there a single reason? Journal of Public Health,
vol. 18, no. 3, 329–36.
Clarke, S. (ed.) 1998. Str uctural Adjustment without Mass Unemployment? Lessons from Russia,
Cheltenham, Edward Elgar.
Cockerham, W. C. 1997. The social determinants of the decline of life expectancy in Russia and
eastern Europe: a lifestyle explanation, Journal of Health and Social Behavior, vol. 38, no. 2,
117–30.
Cook, L. J. 2007. Postcommunist Welfare States: Reform Politics in Russia and Eastern Europe, Ithaca,
NY and London, Cornell University Press.
Cook, T. D., Campbell, D. T. Shadish, W. 2002. Experimental and Quasi-experimental Designs for
Generalised Causal Inference, Boston, MA, Houghton Mifin.
Cooper, M. L., Russell, M., Skinner, J. B., Frone, M. R. and Mudar, P. 1992. Stress and al-
cohol use: moderating effects of gender, coping, and alcohol expectancies, Journal of Abnormal
Psychology, vol. 101, no. 1, 139–52.
Cornia, G. A. and Paniccià, R. 2000. The transition mortality crisis: evidence, interpretation
and policy responses, pp. 3–37 in Cornia, G. A. and Paniccià, R. (eds.), The Mortality Crisis in
Transitional Economies, Oxford, Oxford University Press.
Page 28 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
Deaton, A. 2020. ‘Randomisation in the Tropics Revisited: A Theme and Eleven Variations’, Working
Paper Series, No. 27600, National Bureau of Economic Research.
Delaney, L., Fernihough, A. and Smith, J. P. 2013. Exporting poor health: the Irish in England,
Demography, vol. 50, no. 6, 2013–35.
Doniec, K., Steer, D., Murphy, M., Gugushvili, A., McKee, M., Marmot, M., Bobak, M. and
King, L. 2018. Education and mortality in three eastern European populations: ndings from
the PrivMort retrospective cohort study, European Journal of Public Health, vol. 29, no. 3,
549–54.
Ellman, M. 1994. The increase in death and disease under ‘Katastroika’, Cambridge Journal of
Economics, vol. 18, no. 4, 329–55.
Ellman, M. 1997a. The political economy of transformation, Oxford Review of Economic Policy,
vol. 13, no. 2, 23–32.
Ellman, M. 2000. The social costs and consequences of the transformation process, Economic
Survey of Europe, vol. 2000, no. 2–3, 125–40.
Ellman, M. 1997b. Transformation as a demographic crisis, pp. 351–71 in Zecchini, S. (ed.),
Lessons from the Economic Transition: Central and Eastern Europe in the 1990s, New York, Springer
Science+Business Media.
Feenstra, R. C., Inklaar, R. and Timmer, M. P. 2019. Penn World Table Version 9.1 [April 2019].
https://www.rug.nl/ggdc/productivity/pwt/ (last accessed 12 September 2019): Groningen
Growth and Development Centre, Faculty of Economics and Business, University of
Groningen. doi:10.15141/S50T0R.
Felipe, J., Mehta, A. and Rhee, C. 2018. Manufacturing matters…but it’s the jobs that count,
Cambridge Journal of Economics, vol. 43, no. 1, 139–68.
Ghodsee, K. and Orenstein, M. 2021. Taking Stock of Shock: Social Consequences of the 1989
Revolutions, Oxford, Oxford University Press.
Gil, A., Polikina, O., Koroleva, N., McKee, M., Tomkins, S. and Leon, D. A. 2009. Availability
and characteristics of nonbeverage alcohols sold in 17 Russian cities in 2007, Alcoholism
Clinical and Experimental Research, vol. 33, no. 1, 79–85.
Gimpelson, V. and Kapelyushnikov, R. 2011. ‘Labor Market Adjustment: Is Russia Different?’ IZA
Discussion Paper No. 5588, March 2011, Bonn, Germany, Forschungsinstitut zur Zukunft
der Arbeit Institute for the Study of Labor.
Goryakin, Y., Roberts, B. and McKee, M. 2015. Price elasticities of alcohol demand: evidence
from Russia, The European Journal of Health Economics, vol. 16, no. 2, 185–99.
Grogan, L. and van den Berg, G. J. 2001. The duration of unemployment in Russia, Journal of
Population Economics, vol. 14, no. 3, 549–68.
Hamm, P., King, L. and Stuckler, D. 2012. Mass privatisation, state capacity, and economic
growth in post-communist countries, American Sociological Review, vol. 77, no. 2, 295–324.
Hassink, R. 2010. Locked in decline? On the role of regional lock-ins in old industrial areas,
pp. 450–68 in Boschma, R. and Martin, R. (eds.), The Handbook of Evolutionary Economic
Geography, Cheltenham, Edward Elgar Publishing.
Head, M. L., Holman, L., Lanfear, R., Kahn, A. T. and Jennions, M. D. 2015. The extent and
consequences of P-hacking in science, PLoS Biology, vol. 13, no. 3, e1002106.
Hirschman, A. O. 2013. A generalised linkage approach to development, with special reference
to staples, pp. 155–94 in Hirschman, A. O. and Adelman J. (eds.), The Essential Hirschman,
Princeton, NJ, Princeton University Press.
Irdam, D., King, L., Gugushvili, A., Azarova, A., Fazekas, M., Scheiring, G., Steer, D., Doniec,
K., Horvat, P., Kolesnikova, I., Popov, V., Szelenyi, I., Marmot, M., Murphy, M., McKee, M.
and Bobak, M. 2016. Mortality in transition: study protocol of the PrivMort project, a multi-
level convenience cohort study, BMC Public Health, vol. 16, no. 672, 1–8.
Irdam, D., Scheiring, G. and King, L. 2015. Mass privatisation, pp. 488–507 in Hölscher, J.
and Tomann, H. (eds.), Palgrave Dictionary of Emerging Markets and Transition Economics,
Houndmills, Basingstoke, Palgrave Macmillan.
Iversen, T. and Cusack, T. R. 2011. The causes of welfare state expansion: deindustrialisation or
globalisation? World Politics, vol. 52, no. 3, 313–49.
Kaldor, N. 1967. Strategic Factors in Economic Development, Ithaca, NY, Cornell University Press.
Kavanagh, N. M., Menon, A. and Heinze, J. E. 2021. Does health vulnerability predict voting
for right-wing populist parties in Europe? American Political Science Review, vol. 115, no. 3,
1104–9.
Deindustrialisation and mortality Page 29 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
Kideckel, D. A. 2008. Getting by in Postsocialist Romania: Labor, the Body, and Working-Class
Culture, Bloomington, IN, Indiana University Press.
King, L. 2002. Postcommunist divergence: a comparative analysis of the transition to capitalism
in Poland and Russia, Studies in Comparative International Development, vol. 37, no. 3, 3–34.
King, L. 2003. Shock privatisation: the effects of rapid large-scale privatisation on enterprise
restructuring, Politics & Society, vol. 31, no. 1, 3–30.
King, L., Hamm, P. and Stuckler, D. 2009. Rapid large-scale privatisation and death rates
in ex-communist countries: an analysis of stress-related and health system mechanisms,
International Journal of Health Services, vol. 39, no. 3, 461–89.
King, L., Scheiring, G. and Nosrati, E. 2022. Deaths of despair in comparative perspective,
Annual Review of Sociology, vol. 48, no. July 2022, 299–317.
King, L. and Sznajder, A. 2006. The state-led transition to liberal capitalism: neoliberal, or-
ganizational, world-systems, and social structural explanations of Poland’s economic success,
American Journal of Sociology, vol. 112, no. 3, 751–801.
Kolodko, G. W. 2000. From Shock to Therapy: The Political Economy of Postsocialist Transformation,
Oxford, Oxford University Press.
Koltai, J., Varchetta, F. M., McKee, M. and Stuckler, D. 2020. Deaths of despair and Brexit
votes: cross-local authority statistical analysis in England and Wales, American Journal of Public
Health, vol. 110, no. 3, 401–6.
Kopp, M., Skrabski, A., Székely, S., Stauder, A. and Williams, R. 2007. Chronic stress and social
changes: socioeconomic determination of chronic stress, Annals of the New York Academy of
Sciences, vol. 1113, no. 1, 325–38.
Krastev, I. 2016. What’s wrong with east-central Europe? Liberalism’s failure to deliver, Journal
of Democracy, vol. 27, no. 1, 35–8.
Kuhn, T. S. 2012. The Structure of Scientic Revolutions, Chicago, University of Chicago Press.
Laky, T. 2000. Labour market in Hungary 1999, 304–24 in Fazekas, K. (ed.), Munkaerőpiaci
Tükör 2000, Budapest, Hungarian Academy of Sciences, Institute of Economics.
Lipton, D., Sachs, J. D., Mau, V. and Phelps, E. S. 1992. Prospects for Russia’s economic re-
forms, Brookings Papers on Economic Activity, vol. 1992, no. 2, 213–83.
Lundberg, J., Bobak, M., Malyutina, S., Kristenson, M. and Pikhart, H. 2007. Adverse health
effects of low levels of perceived control in Swedish and Russian community samples, BMC
Public Health, vol. 7, no. 1, 1–11.
Marmot, M. 2018. Inclusion health: addressing the causes of the causes, The Lancet, vol. 391,
no. 10117, 186–8.
Marmot, M. and Bobak, M. 2000. Psychosocial and biological mechanisms behind the recent
mortality crisis in central and eastern Europe, pp. 127–50 in Cornia, G. A. and Paniccia, R.
(eds.), The Mortality Crisis in Transitional Economies, Oxford, Oxford University Press.
McEwen, B. S. and Stellar, E. 1993. Stress and the individual: mechanisms leading to disease,
Archives of Internal Medicine, vol. 153, no. 18, 2093–101.
Mitchell, R., Gleave, S., Bartley, M., Wiggins, D. and Joshi, H. 2000. Do attitude and area inu-
ence health?, Health & Place, vol. 6, no. 2, 67–79.
Mood, C. 2009. Logistic regression: Why we cannot do what we think we can do, and what we
can do about it, European Sociological Review, vol. 26, no. 1, 67–82.
Morris, M. and Western, B. 1999. Inequality in earnings at the close of the twentieth century,
Annual Review of Sociology, vol. 25, no. 1, 623–57.
Murrell, P. 1993. What is shock therapy? What did it do in Poland and Russia? Post-Soviet Affairs,
vol. 9, no. 2, 111–40.
Newell, P. and Mulvaney, D. 2013. The political economy of the ‘just transition’, The Geographical
Journal, vol. 179, no. 2, 132–40.
Nosrati, E., Ash, M., Marmot, M., McKee, M. and King, L. P. 2018. The association between
income and life expectancy revisited: deindustrialisation, incarceration and the widening
health gap, International Journal of Epidemiology, vol. 47, no. 3, 720–30.
Orenstein, M. A. 2001. Out of the Red: Building Capitalism and Democracy in Postcommunist
Europe, Ann Arbor, MI, The University of Michigan Press.
Perlman, F. and Bobak, M. 2009. Assessing the contribution of unstable employment to mor-
tality in posttransition Russia: prospective individual-level analyses from the Russian longitu-
dinal monitoring survey, American Journal of Public Health, vol. 99, no. 10, 1818–25.
Page 30 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
Popov, V. 2012. Russia: austerity and decit reduction in historical and comparative perspective,
Cambridge Journal of Economics, vol. 36, no. 1, 313–34.
Rabe-Hesketh, S. and Skrondal, A. 2012. Discrete-time survival, pp. 749–96 in Multilevel and
Longitudinal Modelling Using Stata, Vol II: Categorical Responses, Counts and Survival, 3rd
Edition, College Station, TX, Stata Press.
Ramazzotti, P. 2009. Industrial districts, social cohesion and economic decline in Italy, Cambridge
Journal of Economics, vol. 34, no. 6, 955–74.
Rind, E., Jones, A. and Southall, H. 2014. How is post-industrial decline associated with the
geography of physical activity? Evidence from the health survey for England, Social Science &
Medicine, vol. 104, 88–97.
Rodríguez-Pose, A. 2018. The revenge of the places that don’t matter (and what to do about it),
Cambridge Journal of Regions, Economy and Society, vol. 11, no. 1, 189–209.
Rodrik, D. 2004. Industrial Policy for the Twenty-First Century, , Discussion Paper No. 4767,
Centre for Economic Policy Research. Available at SSRN: https://ssrn.com/abstract=617544
or http://dx.doi.org/10.2139/ssrn.617544.
Rodrik, D. 2016. Premature deindustrialisation, Jour nal of Economic Growth, vol. 21, no. 1, 1–33.
Ruhm, C. J. 2000. Are recessions good for your health? The Quarterly Journal of Economics, vol.
115, no. 2, 617–50.
Scheiring, G. 2020. The Retreat of Liberal Democracy: Authoritarian Capitalism and the Accumulative
State in Hungary, New York, Palgrave Macmillan.
Scheiring, G., Irdam, D. and King, L. 2019. Cross-country evidence on the social determinants
of the post-socialist mortality crisis in Europe: a review and performance-based hierarchy of
variables, Sociology of Health & Illness, vol. 41, no. 4, 673–91.
Scheiring, G., Irdam, D. and King, L. 2018a. The wounds of post-socialism: a systematic re-
view of the social determinants of mortality in Hungary, Journal of Contemporary Central and
Eastern Europe, vol. 26, no. 1, 1–31.
Scheiring, G. and King, L. 2022. Deindustrialisation, social disintegration, and health: a neo-
classical sociological approach, Theory and Society, no. Advance access, published on Mar 14
2022.
Scheiring, G., Steer, D., Irdam, D., Fazekas, M., Azarova, A., Kolesnikova, I., Köllő, J., Popov,
V., Szelényi, I., Marmot, M., Murphy, M., McKee, M., Bobak, M. and King, L. 2018b.
The gendered effects of foreign investment and prolonged state ownership on mortality in
Hungary: an indirect demographic, retrospective cohort study, The Lancet Global Health, vol.
6, no. 1, e9595–e102.
Singh, A. 1979. The ‘basic needs’ approach to development vs the new international eco-
nomic order: the signicance of third world industrialization, World Development, vol. 7, no.
6, 585–606.
Standing, G. 1996. Russian Unemployment and Enterprise Restructuring: Reviving Dead Souls,
Houndmills, Basingstoke, Palgrave Macmillan.
Stuckler, D. 2009. ‘Social Causes of Post-Communist Mortality: A Macro-sociological Analysis’,
Ph.D. dissertation, Cambridge, Department of Sociology, University of Cambridge.
Stuckler, D., King, L. and McKee, M. 2012. The disappearing health effects of rapid privatisa-
tion: a case of statistical obscurantism? Social Science & Medicine, vol. 75, no. 1, 23–31.
Stuckler, D., King, L. and McKee, M. 2009. Mass privatisation and the post-communist mor-
tality crisis: a cross-national analysis, The Lancet, vol. 373, no. 9661, 399–407.
Thoits, P. A. 2010. Stress and health: major ndings and policy implications, Journal of Health
and Social Behavior, vol. 51, no. 1_suppl, S41–53.
Tomkins, S., Collier, T., Oralov, A., Saburova, L., McKee, M., Shkolnikov, V., Kiryanov, N. and
Leon, D. 2012. Hazardous alcohol consumption is a major factor in male premature mortality
in a typical Russian city: prospective cohort study 2003-2009, PLoS One, vol. 7, no. 2, e30274.
Tomkins, S., Saburova, L., Kiryanov, N., Andreev, E., McKee, M., Shkolnikov, V. and Leon,
D. A. 2007. Prevalence and socio-economic distribution of hazardous patterns of alcohol
drinking: study of alcohol consumption in men aged 25-54 years in Izhevsk, Russia, Addiction,
vol. 102, no. 4, 544–53.
Tregenna, F. 2008. Characterising deindustrialisation: an analysis of changes in manufacturing
employment and output internationally, Cambridge Journal of Economics, vol. 33, no. 3,
433–66.
Deindustrialisation and mortality Page 31 of 32
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
Tregenna, F. 2016. Deindustrialisation and premature deindustrialization, 710–28 in Reinert, E.
S., Ghosh, J. and Kattel, R. (eds.), Handbook of Alternative Theories of Economic Development,
Cheltenham, Edward Elgar Publishing.
Treisman, D. 2010. Death and prices: the political economy of Russia’s alcohol crisis, Economics
of Transition, vol. 18, no. 2, 281–331.
United Nations. 1983. ‘Manual X: Indirect Techniques for Demographic Estimation’, Population
Studies, No. 81, New York, United Nations, Department of International Economic and
Social Affairs.
United Nations. 2022. World Population Prospects 2022, Department of Economic and Social
Affairs, Population Division, UN. https://population.un.org/wpp/ (last accessed 21 August
2022).
Venkataramani, A. S., Bair, E. F., O’Brien, R. L. and Tsai, A. C. 2020. Association between
automotive assembly plant closures and opioid overdose mortality in the United States: a
difference-in-differences analysis, JAMA Internal Medicine, vol. 180, no. 2, 254–62.
Venkataramani, A. S., O’Brien, R. L. and Tsai, A. C. 2021. Declining life expectancy in the
United States: the need for social policy as health policy, JAMA, vol. 325, no. 7, 621–2.
Vodopivec, M., Wörgötter, A. and Raju, D. 2005. unemployment benet systems in central and
eastern Europe: a review of the 1990s, Comparative Economic Studies, vol. 47, no. 4, 615–51.
Walberg, P., McKee, M., Shkolnikov, V., Chenet, L. and Leon, D. 1998. Economic change,
crime, and mortality crisis in Russia: regional analysis, BMJ, vol. 317, no. 7154, 312–8.
Wallace, R., Wallace, D., Ullmann, J. E. and Andrews, H. 1999. Deindustrialisation, inner-city
decay, and the hierarchical diffusion of aids in the USA: how neoliberal and cold war policies
magnied the ecological niche for emerging infections and created a national security crisis,
Environment and Planning A, vol. 31, no. 1, 113–39.
Walsh, D., Taulbut, M. and Hanlon, P. 2009. The aftershock of deindustrialisation—trends in
mortality in Scotland and other parts of post-industrial Europe, European Journal of Public
Health, vol. 20, no. 1, 58–64.
WHO. 2020. European Health for All Family of Databases [HFA-DB], World Health Organisation.
Online Database (last accessed 26 October 2020).
Page 32 of 32 G. Scheiring et al.
Downloaded from https://academic.oup.com/cje/advance-article/doi/10.1093/cje/beac072/7081084 by guest on 21 March 2023
