Content uploaded by Gabor Scheiring
Author content
All content in this area was uploaded by Gabor Scheiring on Jan 24, 2020
Content may be subject to copyright.
1
Analyzing the Health Impact of Economic Change:
Insights from a Multilevel Retrospective Cohort Study
Gábor Scheiring [1], Aytalina Azarova [2], Darja Irdam [2]
[1] Bocconi University, Italy - gabor@gaborscheiring.com
[2] University of Cambridge, UK
SAGE Research Methods Cases
Discipline: Public Health, International Health; Methods: Response rates, Cohort studies, Surveys
This is the pre-publication accepted manuscript, full citation: Scheiring, G., Azarova, A., & Irdam, D.
(2020). Analyzing the health impact of economic change: Insights from a multi-level retrospective
cohort study. SAGE Research Methods Cases. doi:10.4135/9781529711189
Abstract
An unprecedented mortality crisis befell the former socialist countries between 1989 and 1995,
representing one of the most significant demographic shocks of the post-Second World War period.
Academic research has identified economic transitions as a crucial factor behind the postsocialist
mortality crisis. However, most previous studies relied on either country-level or individual-level data,
which leaves the potential for modeling error, as they cannot assess both distal (economic) and
proximal (individual) causes of mortality simultaneously. We aimed to overcome these limitations and
investigate the role of economic transitions (rapid mass privatization, deindustrialization, and foreign
investment liberalization) and individual-level factors (e.g., alcohol consumption) in the mortality
crises in postsocialist countries. We identified towns with different privatization strategies and
collected administrative data on 539 towns in Russia, 96 towns in Belarus and 52 towns in Hungary.
In these towns, we identified the largest companies and collected data on their ownership structure.
We also conducted large-scale surveys using a retrospective cohort study approach. Respondents
provided information on themselves and their relatives, including socioeconomic characteristics,
health behavior, as well as the vital status of their relatives. In total, we collected data on 268,600
subjects in the three countries. Using this information, we created a complex multilevel database
linking towns’ industrial characteristics and individual health outcomes covering three decades from
1980 to 2010. We investigated how excess mortality of individuals is distributed across settlements
with different privatization strategies. The results confirmed that economic change and alcohol were
crucial determinants of mortality during the postsocialist transition.
Keywords: retrospective study, mortality, town, Russia, privatization, Belarus, settlement, population,
Hungary, surveying, crisis, propensity score
2
Dr. Gabor Scheiring is a research fellow at the Dondena Centre for Research on Social Dynamics
and Public Policy, at Bocconi University, Milan. After finishing his Ph.D. (University of Cambridge,
2017), he was a Democracy Fellow at the National Endowment for Democracy, Washington DC and
an ISRF Political Economy Research Fellow at the Department of Sociology, University of
Cambridge. Dr. Scheiring analyzes the human dimensions of economic change from the perspective
of political economy using both quantitative and qualitative methods, combining theoretical
innovation with empirical rigor. In his doctoral thesis and several related articles in the Lancet Global
Health and Sociology of Health and Illness, among others, he investigated the impact of
deindustrialization, privatization, and foreign investment on health. As part of his research on the
political economy of democratic backsliding in Central and Eastern Europe, Dr. Scheiring investigates
how the postsocialist economic reforms paved the way to the illiberal political backlash, with a book
on the dying of democracy in Hungary. As a founder of a progressive green political party, he served
as a member of the Hungarian Parliament between 2010 and 2014.
Dr. Aytalina Azarova is an Affiliated Researcher at the Department of Sociology, University of
Cambridge. Before coming to Cambridge in 2015, she studied political science and Russian studies at
the North Eastern Federal University (SVFU) and Central European University. She has worked in
the areas of health economics, comparative institutional economics, and Russian society. Her
publications have appeared in the Lancet, BMC, International Journal of Public Health, International
Journal of Cancer, and other flagship medical journals. During her time at the University of
Cambridge, she worked under the auspices of the PrivMort research group, where she combined
administrative and research tasks. Her research uses both qualitative and quantitative methods to study
institutional change and its effects on individual action and health. Relying primarily on an innovative
survey and statistical methods, she applies insights from political economy to the study of public health
and epidemiology. Her most recent work examined the impact of partners’ education on mortality in
Eastern Europe, as well as the consequences of between-country variation in the speed of economic
change for public health and longevity. She was previously a visiting scholar at the Harriman Centre
of Columbia University.
Dr. Darja Irdam is an affiliated researcher at the Department of Sociology, University of Cambridge
and Research Director in the Health and Social Care team at the National Centre for Social Research
(NatCen). Darja has completed her Ph.D. in the political economy of health at the University of
Cambridge (2017) to continue her postdoctoral research at the Department of Sociology, University
of Cambridge, where she was involved in the PrivMort project. In her current role as a Research
Director, Dr. Irdam leads a team of researchers on a variety of projects in the areas of public health
and social care. Dr. Irdam manages research projects for a range of government departments (including
the Department for Health and Social Care, Department for Exiting the European Union, the Ministry
of Defense, the Home Office) using both qualitative and quantitative research methods. Her research
interests are in the areas of economic transitions, gender, mortality, substance dependency, mental
health, pharmaceutical research, and palliative care. Her publications have appeared in such academic
journals as the Lancet, BMC, Sociology of Health and Illness, International Journal of Cancer.
3
Selected publications related to the project:
Azarova, A., Irdam, D., Gugushvili, A., Fazekas, M., Sheiring, G., Horvat, P.,
… King, L. (2017). The effect of rapid privatization on mortality in mono-industrial towns in
post-Soviet Russia: A retrospective cohort study. The Lancet Public Health, 2, e231–e238.
Irdam, D., Scheiring, G., & King, L. (2015). Mass privatization.
In J.Hölscher and H. Tomann (Eds.), Palgrave dictionary of emerging markets and
transition economics (pp. 488–507). Basingstoke, UK: Palgrave Macmillan.
Irdam, D., King, L., Gugushvili, A., Azarova, A., Fazekas, M., Scheiring, G.,
… Bobak, M. (2016). Mortality in transition: A multilevel indirect demographic cohort study
PrivMort. BMC Public Health, 16, 1–8.
Scheiring, G., Irdam, D., & King, L. (2018). The wounds of post-socialism: A systematic review
of the social determinants of mortality in Hungary. Journal of Contemporary Central and
Eastern Europe, 26, 1–31.
Scheiring, G., Irdam, D., & 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, 41, 673–691.
Scheiring, G., Doniec, K., Irdam, D., Azarova, A., Stuckler, D., Murphy, M.,
… King, L. (2019). Deindustrialization, foreign investment and social development: The
impact of industrial transformations on mortality inequalities. Cambridge, UK: Department
of Sociology, University of Cambridge.
Scheiring, G., Stefler, D., Irdamet, D., Fazekas, M., Azarova, A., Kolesnikova, I.,
… King, L. (2018). The gendered effects of foreign investment and prolonged state
ownership on mortality in Hungary: An indirect demographic, retrospective cohort study. The
Lancet Global Health, 6, 95–102.
Stefler, D., Azarova, A., Irdam, D., Scheiring, G., Murphy, M., McKee, M.,
… Bobak, M. (2018). Smoking, alcohol and cancer mortality in Eastern European men:
Findings from the PrivMort retrospective cohort study. International Journal of
Cancer, 143, 1128–1133.
4
Learning Outcomes
By the end of this case, students should be able to achieve both methodological and conceptual
understanding of several topics and should:
• Have a clear understanding of methodological challenges of working with large-scale
surveys and how such challenges can be overcome.
• Have an overview of the retrospective cohort study method of data collection that can be
used to construct retrospective cohort studies.
• Understand the application of collecting administrative town-level data and company
information and some of the strategies to tackle these difficulties.
• Understand how to fit multilevel models to analyze the health impact of economic change.
• Have an insight into the postsocialist mortality crisis induced by rapid economic change.
Project Overview and Context
An unprecedented mortality crisis befell the former socialist countries between 1989 and 1995 with
the transition from socialism to capitalism. The leading direct cause of death during the postsocialist
mortality crisis was acute heart disease and alcohol poisoning. Russia and some other former members
of the Soviet Union were affected the most (Billingsley 2011). As a result, while in the 1970s Russia
lagged behind western European countries by only 2-3 years in life expectancy, this gap increased to
15–17 years for men by the early 2000s (Shkolnikov et al. 2004), with a total excess mortality reaching
3.26 million in 1990–1999 (UNICEF 2001). The loss of healthy life expectancy had severe indirect
costs on the economy and the population as well. Researchers estimated the national income loss due
to illness in Russia at 1.8–4.7 percent of one year’s GDP during the first half of the 1990s (Bloom and
Malaney 1998).
Epidemiological research has uncovered various proximal (individual) causes of these deaths,
identifying alcohol, social isolation, strain, and psychological stress as the key reasons for increased
mortality (Scheiring et al. 2019a). However, distal (socio-economic) factors are less clear. The
magnitude of the mortality crisis suggests that these social determinants go beyond poverty,
encompassing broader segments of society through insecurity, stress, unemployment, and inequality
(Brainerd 1998). Economic recessions and crises have also been convincingly linked to the loss of
healthy lives (Stuckler et al. 2009b). However, there is no direct link between the level of economic
production and mortality; other large-scale factors mediate the health impact of macroeconomic
cycles, like unemployment, inequality, the welfare state, or economic uncertainty. In turn, these
macro-level factors are influenced by economic policies, such as liberalization and privatization.
Earlier work in Russia pointed out the role of rapid transition, as reflected by high labor market
turnover and privatization (Brainerd 2001; Stuckler et al. 2009a). However, these studies have mostly
relied on either country-level or individual-level data, which leaves the potential for modeling error,
as they cannot assess both distal (economic) and proximal (individual) causes of mortality
5
simultaneously. Therefore, these approaches are unsuitable for investigating the pathways linking
economic factors with mortality.
The Privatization and Mortality in Eastern Europe (PrivMort) is a multi-disciplinary project in the
subfield of the Political Economy of Public Health, funded by the European Research Council that ran
between 2012 and 2017. Primary Investigator, Professor Lawrence King at the Department of
Sociology, University of Cambridge, led the project. The senior project partners were Sir Michael
Marmot (University College London), Professor Martin McKee (London School of Hygiene and
Tropical Medicine), Professor Martin Bobak (University College London) and Professor Mike
Murphy (London School of Economics and Political Science). In addition, the project involved several
postdoctoral researchers and fieldworkers, as well as external contractors.
The project was designed to address the methodological drawbacks of existing approaches through
multilevel modeling linking individual- and town-level data and determining the hierarchy of causes
of mortality in three postsocialist countries (Russia, Belarus, and Hungary). The project aimed to
investigate the role of economic change (rapid mass privatization, deindustrialization, and foreign
investment liberalization) and individual-level factors (e.g., alcohol consumption) in the mortality
changes simultaneously.
Section summary
• An unprecedented mortality crisis befell the former socialist countries between 1989 and
1995 with the transition from socialism to capitalism.
• Earlier studies mostly relied on either country-level or individual-level data, which leaves
the potential for modeling error, as they cannot assess both distal (economic) and
proximal (individual) causes of mortality simultaneously.
• The goal of the PrivMort project was to address these challenges and construct multilevel
dataset (combining individual- and town-level data) covering multiple countries to
measure the health impact of economic change in Belarus, Hungary, and Russia.
Research Design
The PrivMort project created and analyzed two large-scale datasets. The first is a set of retrospective
cohort study surveys (where respondents provide information about themselves and their relatives
post-hoc, looking back in time, covering many years) conducted in 30 settlements in Russia, 20 in
Belarus and 52 in Hungary. The second is a set of annual time series (administrative data retrieved
from official sources at the level of settlements) covering the period of 1990–2010 for 539 settlements
in Russia, 96 settlements in Belarus, and 52 settlements in Hungary, which include the settlements in
which we conducted the surveys. As part of the settlement-level data, we also collected data on the
ownership structure of the largest companies in these towns. To link changes in companies’
ownership-structure to individual health, we aggregated company-level information to the settlement
level.
6
Settlement-level data
In Russia 10 mono-industrial towns with fast privatization were matched with 10 mono-industrial
towns with slow privatization, (fast privatized municipalities are towns, where 90 or more percent of
state shares were privatized within two consecutive years, and slow privatized towns are towns in
which less than 50 % of state shares were privatized within two straight years.) After that, we matched
a control group of five multi-industrial towns with fast privatization to five multi-industrial towns with
slow privatization. In Russia, we matched towns using standard propensity score matching
1
based on
the pre-transition demographic and socio-economic conditions in the settlements. We used eight
potential predictors of mortality levels, all measured for the pre-transition year, 1991, except wage in
USD, which was available from 1992 onwards.
We obtained the settlement-level data in Russia from the Economy of Russian Cities database
provided by the Main Interregional Centre for Processing and Dissemination of Statistical Information
of the Federal State Statistics Service (GMC Rosstat). In Belarus, we acquired the settlement-level
data through the National Statistical Committee of the Republic of Belarus (Belstat); from various
regional bodies of the National Statistical Committee; from the Ministry of Health of the Republic of
Belarus; and the Ministry of Internal Affairs. Figure 1 presents a map of the towns in Russia.
In Hungary, we selected settlements between 5,000 and 100,000 inhabitants with industrial
employment exceeding 30 percent in the first round (N=110). We eliminated towns close to Budapest,
together with towns where we were not able to obtain data on the most significant companies from the
Ministry of Justice responsible for the management of the central company registry system. This was
necessary to eliminate bias related to the unprecedentedly high concentration of economic activity in
Budapest, not reflected in other parts of the country. This has resulted in a sample of 83 settlements,
out of which we randomly selected 52 for practical reasons. The geographical, size, and age-structure
distribution of the settlements are representative of medium-sized towns outside the proximate vicinity
of Budapest.
We collected the settlement-level data from the Hungarian Central Statistical Office and a variety of
other state institutions. Information is available for several indicators, including population size,
unemployment rate, income per capita, dependency ratio, number of general practitioners and number
of deaths for people aged 15-64, for each year between 1990 and 2006. Figure 2 presents a map of the
towns in Hungary.
1
Propensity Score Matching is a statistical analysis used to reduce selection bias and thus filter out the effect of potential confounding
variables. The purpose of matching is to make the two groups that are compared (“treatment” and “control groups”) more alike based
on characteristics that could be influencing the main association between “treatment” and “effect”. In our case, treatment refers to
rapid privatization, the control group comprises towns with slow privatization, and effect refers to mortality. As a result of propensity
score matching, the treatment and control groups only differ on the treatment variable but not on other pre-transition demographic and
socio-economic conditions.
7
Figure 1. Map of the towns in Russia
Note: 1. Zhukovka; 2. Karabash; 3. Kohma; 4. Navoloki; 5. Privolzhsk; 6. Yuzha; 7. Nieman; 8. Kirov-Chepetsk; 9.
Kulebaki; 10. Mtsensk; 11. Otradny; 12. Yasnogorsk; 13. Lakinsk; 14. Nikolsk; 15. Semiluki; 16. Seltso; 17. Starodub;
18. Bahcall; 19. Sim; 20. Dalmatovo; 21. Belinsky; 22. Nikolsk; 23. Plavsk; 24. Boguchar; 25. Danilov; 26. Alekseevka;
27. Svetlyj; 28. Buturlinovka; 29. Moorom; 30. Pechora.
8
Figure 2. Map of the in Hungary
Note: 1 Ajka, 2 Alsózsolca, 3 Baja, 4 Bátonyterenye, 5 Békéscsaba, 6 Berhida, 7 Celldömölk, 8 Csongrád, 9 Dorog, 10
Dunaújváros, 11 Edelény, 12 Eger, 13 Hajdúböszörmény, 14 Hajdúhadház, 15 Jászberény, 16 Kaba, 17 Kalocsa, 18
Kaposvár, 19 Kapuvár, 20 Karcag, 21 Kisbér, 22 Kiskunhalas, 23 Kisújszállás, 24 Kisvárda, 25 Komló, 26 Kőszeg, 27
Kunhegyes, 28 Kunszentmárton, 29 Lábatlan, 30 Lenti, 31 Martfű, 32 Mosonmagyaróvár, 33 Nagyatád, 34
Nagykanizsa, 35 Nyírbátor, 36 Oroszlány, 37 Paks, 38 Salgótarján, 39 Sárbogárd, 40 Sárvár, 41 Sátoraljaújhely, 42
Sopron, 43 Szeghalom, 44 Szekszárd, 45 Szerencs, 46 Szolnok, 47 Szombathely, 48 Tab, 49 Tiszafüred, 50
Törökszentmiklós, 51 Vásárosnamény, 52 Zalaegerszeg
Company-level data
In Russia, we collected the enterprise-level data from a variety of state sources. First, we eliminated
enterprises with less than 500 employees from the sample. Then, we matched the samples based on
the OKPO (National Classification of Enterprises and Organizations by the State Statistics Service)
registration code for enterprises, while for those firms that changed their code after being privatized;
we based the matching on the address, enterprise name and the approximate size of the enterprise
within a settlement in some cases. Were collected the enterprise-level data for Belarus from the
following agencies: the Bureau van Dijk, the National Statistical Committee of the Republic of Belarus
(Belstat), the Unified State Registry of Legal Entities including the State Property Committee of the
Republic of Belarus, and JSC Belarusian Currency and Stock exchange.
9
In Hungary, we identified the largest companies in the selected settlements based on registered capital
in 1990 through the Company Information Service of the Ministry of Justice. In addition, we extend
this information with data from the Hungarian Privatization Agency. After selecting the three largest
companies in each settlement, we collected data on the number of employees, ownership structure and
profitability of the enterprises from the archives of the local registry courts of and various private
digital company information archives.
Most of the companies that existed in 1990 changed their names or their legal form of operation. To
ensure continuity, we identified the successors of the original parent companies and obtained data on
them as well. Overall, we collected data on 383 Hungarian companies, 550, including parent
companies and successor companies. For analytic purposes, we later treated original parent companies
and successors as one company. When there were more than one successor companies, we selected
the most significant company by registered capital or the one closest to the original company by type
of activity.
Individual-level data
The collection of the survey data used the Brass method named after William Brass and colleagues
who developed this technique initially in the framework of UN-sponsored demographic analyses. This
demographic method was initially designed to make it possible to estimate mortality in countries that
have no vital registration system, or whose official registry data are unreliable (Graham et al. 1989;
Hill 1977; Merdad et al. 2013). In the absence of official mortality statistics, the Brass method collects
information from respondents about the survival status of their relatives post-hoc, looking back in
time. This way, it is possible to estimate mortality retrospectively. We may use the information
collected on subjects’ relatives to produce standard indicators of mortality. Given the fact that these
studies collect mortality information retrospectively and not directly, researchers also refer to this
method as “indirect estimation.” In countries with higher rates of literacy, we can also ask complex
questions, collecting information on topics such as individuals’ employment histories, smoking, and
drinking.
We collected Individual-level data between January 2014 and December 2015. In each selected town,
the polling agency identified 20-45 starting points, by the help of the grid method. The map of the
town was divided into squares of equal size, and each was numbered. The numbers were randomly
selected, and the starting points were at the center of the selected squares. Interviewers visited
randomly selected addresses for face-to-face interviews, starting from the center of the square. The
starting points in each settlement were distributed among the interviewers of the polling agency, and
each of them visited up to 25 households on the route. Interviewers only interviewed one respondent
from each household regardless of the size of the family. If more than one person in the home matched
the screening criteria, interviewers selected the person whose birthday was closer to the date of the
survey. These procedures ensured randomization. If the subject of the interview was temporarily
unavailable, interviewers made four attempts at interviewing the person. Sample size depended on the
number of citizens over 40 y.o. in each town. The total sample size of respondents for Russia and
10
Hungary was 24,000 for each and 16,000 for Belarus. Together with respondents and their relatives,
the resulting survey dataset comprises information on 94,006 subjects in Russia, 102,698 subjects in
Hungary and 71,976 subjects in Russia.
The primary screening criteria for the interviews was having parents, siblings, or male partners (for
females) living in the same settlement between 1980 and 2010. All respondents were at least 42 years
old or born before 1972 to ensure that they and their relatives were of working age in 1991 and hence
could potentially be affected by the transition. Respondents provided information on their socio-
economic characteristics, employment history, lifestyle, and health, as portrayed in Table 1.
Table 1. Overview of the survey items
Domain
Measures
Demographic information
Date of birth; gender; marital status; religion
Residential history
Residence places for the last three decades; reasons for moving
Socio-economic status
Position; occupation; number of subordinates
Labor market history
Employment history for the last three decades; ISCO
Education
Highest level of education obtained
Health Behaviors
Smoking; drinking (frequency; binge drinking; hazardous
drinking)
Material deprivation
Absolute poverty proxies
Ownership
Ownership of material resources
Social capital
Communication with relatives
Source: Irdam et al. (2016: 4)
In addition to reporting on themselves, respondents also answered questions about three types of
relatives (parents, siblings, and partners of female respondents), including their vital status, year of
birth and death (if not alive). Finally, we only gathered information on partners of female respondents
as previous studies have shown that male respondents in surveys provide less reliable data on non-
residential female partners. To avoid the exclusion of men who died early during the transition, thus
biasing the results, interviewers collected data on the first partners of the respondents. The University
of Cambridge Department of Sociology ethics committee approved the study. We anonymized all data
to prevent any potential identification of individual respondents.
Section summary
We conducted retrospective cohort surveys in 30 settlements in Russia, 20 in Belarus and 52
in Hungary.
The surveys resulted in information on 94,006 subjects in Russia, 102,698 subjects in
Hungary, and 71,976 subjects in Russia.
We collected a set of annual time series covering the period of 1990–2010 for 539 settlements
in Russia, 96 settlements in Belarus, and 52 settlements in Hungary.
11
We also collected data on the ownership structure of the largest companies in these towns. To
link changes in companies’ ownership-structure to individual health, we aggregated
company-level information to the settlement level.
Research Practicalities
First, during the study preparation phase, complying with public procurement regulations, we issued
a tender for fieldwork. It was essential to ensure that all relevant polling agencies received invitations.
TARKI, a Budapest-based private social research-institute with a long tradition in survey research,
took the lead in developing the sampling procedure, which was later adopted in Russia and Belarus.
TARKI’s team developed all necessary survey design and accompanying technical documentation.
During the actual fieldwork, it was crucial to ensure respondent consent prior to each interview – each
respondent had to confirm explicitly his or her permission to participate. Because the questionnaire
included sensitive questions, the design of the questionnaire allowed interviewees to decline to answer
any questions apart from a set of basic demographic questions. Three large teams of fieldwork
specialists in three countries were involved at all stages of the project, design of the study, development
of the survey protocols, and carrying out the surveys. To ensure seamless coordination, the project
appointed a special liaison officer at the Cambridge team, who traveled to all three counties several
times. Overall, the Cambridge team coordinated the entire process of international fieldwork, and
quality assured the data.
In parallel with conducting the surveys, the fieldwork team also collected information on towns. For
this, first, the project team identified the cities as described above. Then we contacted the central
statistical agency of each country respectively to collect administrative statistical information on the
cities. We did this by email and over the telephone. As we requested large amounts of data in each
country, it took several months for the agencies to reply. As this administrative data came from
multiple sources, they were in different formats, which meant we needed to invest significant energy
into cleaning and harmonizing the settlement-level data to have one large unified dataset for each
country.
Finally, the fieldwork team collected information on the biggest companies in these towns, as
described above. The information on companies came from two types of sources: existing digital
registries and the non-digital archives of registry courts. When digital information was not available,
the fieldwork team visited the archives of regional registry courts, copied all the relevant information
from the archival files, and finally entered this data into a digital dataset. Collecting settlement and
company-level data proved to be challenging in Hungary, encompassing more than three years. There
is no reliable digital company information in Hungary covering the 1980s and 1990s, so, for this
reason, we hired additional fieldworkers who then visited the 19 regional registry courts in the country.
This needed extensive travel and coordination among the fieldwork team and the central research team.
12
Section summary
The project involved a large number of senior and junior researchers and several external
contractors.
To ensure seamless coordination, the project appointed a special liaison officer, and the
Cambridge team traveled to all three counties several times throughout the project.
Method in Action
As we deployed the same survey instrument in three different countries, with different cultural
underpinnings, we paid particular attention to achieving consistency of the surveying procedures and
instruments across all three countries. Some questions, which were easily understandable in one
language, did not make sense in the other. Careful and meticulous survey instrument piloting followed
by cognitive testing of the instrument
2
in all three locations ensured perfect calibration of the
instrument to the specific cultural contexts of the countries in question.
A multidisciplinary team of researchers, followed by cognitive testing, developed the questionnaire.
We carried out the cognitive tests in a controlled environment to identify problematic wording and
sensitive questions. Based on the results of the cognitive tests, we modified the survey questions to
ensure that participants can easily follow the interview, feel comfortable and confident about their
responses. Cognitive tests during the early phase of the PrivMort project have also demonstrated that
the sensitivity of health-related questions is less of a problem. This careful approach ensures that
responses to the questions in the PrivMort questionnaire are reliable and accurately measure the
intended items.
The most challenging part of the individual-level data collection procedure was to ensure the response
rate of 60%: due to the unfortunate timing – we originally planned the fieldwork to overlap with the
summer period – the survey was at risk of very low response rate, as most of the potential respondents
were away from their homes. We made special arrangements to postpone the peak of the survey to the
fall-winter period. Relying on intensive pretesting, careful design, and relying on highly skilled
research teams at the survey institutes, we were able to assemble a novel retrospective cohort dataset.
In Russia, we were also able to use an extensive set of town-level data to match towns before
conducting the interviews using propensity score matching. This method made it possible to rule out
selection bias. Without prior matching, the effect of privatization would have been harder to identify.
This study design overcame the constraints of all previous analysis of the health effects of
privatization. In general, the availability of administrative information in Russia is excellent, with
high-quality time-series data on the towns. We faced no problems collecting this information from the
Russian statistical agency.
2
Cognitive testing is a set of procedures used with potential survey respondents in lab settings to understand how respondents
understand and interpret questions asked in the survey instrument and the instructions provided in the instrument.
13
The availability of company information was more limited in Hungary than we expected, especially
compared to Russia, where we could find everything we needed for the project in digital format.
Though some digital data exists in Hungary for the early years of the transition as well, we needed to
significantly extend these and check their reliability by visiting the local registry courts. This required
much more time than we originally envisioned. We first attempted to find digital information by
contacting every known company and institution that distributes company ownership information, but
this turned out to be a dead end. However, emailing and liaising with these companies took a very
long time, which caused delays. Finally, we decided that the only way forward is to visit the non-
digital archives. For this, we needed to find, vet, hire, and train new fieldworkers.
Despite the significant logistical challenges, we managed to collect all the individual, town, and
company-level data that we planned. Using this complex multi-level dataset, we were able to test the
health impact of postsocialist economic change in a robust way, ruling out individual and town-level
confounders and significantly reducing the potential for selection bias. This allowed us to test the
privatization – health, deindustrialization – health and foreign economic liberalization – health
mechanisms more robustly than previous, existing studies. This way, our project contributed to the
literature substantially both with new data, with a new methodology and with new findings.
Altogether, the project was thus a success.
Section summary
We paid particular attention to achieving consistency of the surveying procedures and
instruments across all three countries.
The most challenging part of the individual-level data collection procedure was to ensure the
response rate of 60%.
The availability of company information was more limited in Hungary than we expected.
Practical Lessons Learned
During the PrivMort project, we assembled a unique multi-country database that encompasses
information at multiple levels allowing multilevel modeling (modeling information on individuals
nested within towns) to investigate the association between micro-level data on individuals’ health
with meso-level variables on economic change. This methodological strategy allowed us to overcome
the limitations of cross-country studies that are prone to ecological fallacy. At the same time, we could
also avoid the individualist bias of studies relying solely on survey-based data to estimate the social
determinants of health. To our knowledge, this research project is the first one that investigates
a) the association between rapid privatization and mortality using matched towns;
b) the health impact of foreign direct investment using a mixed-method multi-level analysis;
c) the health impact of prolonged state ownership using quantitative company information;
d) the health effect of deindustrialization during the postsocialist mortality crisis in Eastern
Europe.
14
The project has shown the benefits of using a retrospective cohort approach to analyze the health
impact of economic change. William Brass and his colleagues originally developed this strategy to
assess mortality and demographic trends in countries with a missing or unreliable vital registry. The
lesson from this study is that application of this method in countries with high literacy rates not only
confirms the underlying mortality rates and produces reliable mortality information, but also ties them
to a wide range of demographic and socio-economic characteristics of individuals that would not be
available via the official vital registration system. This allowed us to measure the health impact of
economic change robustly, combining individual and ecological-level data.
A challenge specifically related to the retrospective cohort study approach that we followed is that the
resulting sample is not directly representative. The chance of inclusion is not uniformly distributed
and might be influenced by the size of the family and the type of relatives that were included.
Migration could also introduce a potential bias to the indirect estimates on mortality. However, we
relied on multiple strategies to offset this potential bias. Due to the large number of interviews
conducted in each town, the project collected information from a range of different types of
individuals. We also used various survey techniques to increase the randomness of the sample, such
as the grid method and the random walk procedure. We collected town-level data, and this could
control for migration, either including this data into the regression models or using this information as
part of the matching procedure.
Recall bias is another challenge that we needed to address. The long time-span covered by the
questionnaire makes it difficult for respondents to remember accurately relevant information,
especially regarding their relatives. We were not able to link some of the information collected to
specific years explicitly, only to decades, which limits the accuracy of the individual-level data. To
cope with this challenge, we used cognitive tests before launching the final surveys and used proxy
questions to minimize the difficulties with the questions. For example, we first asked interviewees
about political events that most people remember easily, thus helping them to recall other events that
also happened in the distant past.
Nevertheless, these techniques cannot fully ensure the complete accuracy of individual-level data,
especially regarding retrospective health behavior. As compared to currently available survey and
registry data, the PrivMort database is the only information source that allows researchers to connect
individual social and health information to town level data. Thus, this approach currently represents
the best available method to analyze the economic determinants of mortality.
As part of the research project, we measured several individual-level determinants of health, such as
smoking and partners’ education, contributing to the already robust literature by analyzing new
multilevel data. However, the real substantive contribution of the PrivMort project lies in its approach
to link economic change to health. Using multilevel survival models, we tested whether
deindustrialization increased the relative chance of mortality, whether foreign investment
liberalization affected death rates, and whether privatization to domestic or international capitalists
made a difference to health, compared to prolonged state ownership in the long term. To maximize
15
the contrast in the speed of privatization between the towns, in Russia, we selected one-company cities
that were privatized rapidly and matched them with cities that were privatized gradually using
propensity score matching. Using data from the 1980s, as well as a substantive amount of town-level
control variables, we were also able to reduce the potential for selection bias in Hungary as well
significantly.
Section summary
Our methodological strategy allowed us to overcome the limitations of cross-country studies
that are prone to ecological fallacy.
We were also able to avoid the individualist bias of studies relying solely on survey-based
data to estimate the social determinants of health.
We employed multiple strategies to address the challenges specifically arising from our
methodology, devising approaches to reduce the potential for recall bias, and to address the
non-uniform distribution of respondents.
Conclusion
The postsocialist mortality crisis represents the most significant demographic catastrophe seen outside
China since the Second World War (Eberstadt 2010). The magnitude of this human crisis stands in
contrast to our limited understanding of the complexities of how postsocialist economic
transformations influenced this mortality shock. Our project sought to tackle this gap.
Methodologically, we aimed to create a new multi-country dataset that we can use to analyze the health
impact of economic change robustly, combining information on individuals, companies, and towns.
We fitted random intercept multilevel survival models to investigate the association between micro-
level data with meso-level variables. Collecting a wide range of town-level data, we could rule out
selection bias by matching towns in Russia, and by analyzing pre-privatization mortality trends in
Hungary. The retrospective cohort study approach also allowed us to filter out the influence of the
essential individual-level determinants of health. Thus, it is unlikely that the differences between the
towns with different trajectories of economic change would be the results of differences in individual
health behavior, age, education, or other relevant factors.
We faced several difficulties during the project. We had to carefully design and test the surveys, we
had to modify the timing of the survey fieldwork, we had to overcome the lack of reliable digital
company ownership information in Hungary, and we had to harmonize administrative data from
multiple sources. This also necessitated intensive coordination across a large research team, liaising
with numerous external contractors and field workers to ensure the timely progress of the project. The
retrospective approach also has certain limitations that we had to take into account. We could address
some of these limitations during data collection, while others during the analysis phase and the
statistical modeling.
16
We have found the following associations between economic change and individual health. Severe
deindustrialization is associated with significantly larger odds of mortality for men between 1989 and
1995 in Hungary (Scheiring et al. 2019b). On the other hand, analyzing Hungary, we also found that
prolonged state ownership is related to significantly lower odds of dying among women, compared to
towns dominated by domestic private property and to cities affected by foreign investment
liberalization (Scheiring et al. 2018). Analyzing matched towns in Russia, we confirmed previous
cross-country analyses that rapid privatization is indeed statistically significantly associated with male
mortality in Russia (Azarova et al. 2017).
This project could form the basis of further research on the impact of economic change on individual
mortality and beyond. One of the exciting findings of our project is the gendered difference in the
effects of deindustrialization, foreign investment, and privatization. Future research could explore this
gender-related difference in coping in more detail. There is no official data available on settlement-
level income inequalities, and the individual data did not allow us to calculate the distribution of
personal income within towns. Multi-level data collection techniques in the future could thus focus
more on town-level inequalities and social cohesion that could be important determinants of mortality
differentials.
Future research could also look at the health impact of trade liberalization and the collapse of the
socialist states. The literature suggests that beyond the availability of resources and the problems of
productivity, the destruction of socialist markets could have been a critical factor behind company
failures and terminations in Eastern Europe during the 1990s. Finally, following Amartya Sen’s
assertion that mortality is a good measure of development, the results presented open up a research
potential on the relationship between health, mortality, and democratic consolidation in post-socialist
countries. The recent backsliding in democratic quality in post-socialist countries is not understood
enough, linking health and democracy could be a new way of looking at the social roots of current
authoritarian tendencies. Future research could use individual- or settlement-level measures of
mortality, as we did in the PrivMort project, to analyze the social foundations of the rise in support for
authoritarian parties in the region. The questions related to individual health in today’s increasingly
chaotic world contribute to the need for sophisticated multilevel approaches.
Section summary
We created a new multi-country dataset that we used to analyze the health impact of economic
change, robustly combining information on individuals, companies, and towns.
Our project confirmed that deindustrialization, foreign economic liberalization, and
privatization were significant determinants of health during the postsocialist change.
Our approach could form the basis of further research on the impact of economic change on
individual mortality and beyond.
17
Teaching resources
Classroom Discussion Questions
1. What types of data were collected as part of the PrivMort project?
2. How does a multilevel approach help to analyze the association between economic change
and mortality compared to other approaches?
3. What do you think is the main advantage of the retrospective cohort study approach?
4. What are the main limitations of the method described in this case study?
Multiple Choice Quiz Questions
One of the most significant benefits of a multi-level analysis on the economic determinants of health
is that it can:
1. Filter out individual confounders while avoiding ecological fallacy (correct answer)
2. Account for the health implications of economic change cross-sectionally
3. Reduce the potential of recall bias
The retrospective cohort study approach was initially developed to collect data in…:
a. Populations affected by conflict
b. Populations with high infant mortality
c. Countries with no reliable registry data (correct answer)
In order to select settlements, which statistical technique was applied before the commencement of
the fieldwork in Russia?
a. Propensity score matching (correct answer)
b. The birthday method
c. The random numbers method
18
References
Azarova, Aytalina, et al. (2017), 'The effect of rapid privatization on mortality in mono-industrial
towns in post-Soviet Russia: a retrospective cohort study', The Lancet Public Health, 2 (5), e231-e38.
Billingsley, Sunnee (2011), 'Exploring the Conditions for a Mortality Crisis: Bringing Context Back
into the Debate', Population, Space and Place, 17 (3), 267-89.
Bloom, David E. and Malaney, Pia N. (1998), 'Macroeconomic consequences of the Russian mortality
crisis', World Development, 26, 2073-85.
Brainerd, Elizabeth (1998), 'Market reform and mortality in transition economies', World
Development, 26 (11), 2013-27.
--- (2001), 'Economic reform and mortality in the former Soviet Union: A study of the suicide epidemic
in the 1990s', European Economic Review, 45 (4–6), 1007-19.
Eberstadt, Nicholas (2010), 'The Enigma of Russian Mortality', Current History, 109 (729), 288.
Graham, Wendy, Brass, William, and Snow, Robert W. (1989), 'Estimating Maternal Mortality: The
Sisterhood Method', Studies in Family Planning, 20 (3), 125-35.
Hill, K. (1977), 'Estimating adult mortality levels from information on widowhood', Population
Studies, 31 (1), 75-84.
Irdam, D., et al. (2016), 'Mortality in Transition: A Multilevel Indirect Demographic Cohort Study
PrivMort', BMC Public Health, 16 (672), 1-8.
Merdad, Leena, Hill, Kenneth, and Graham, Wendy (2013), 'Improving the measurement of maternal
mortality: the sisterhood method revisited', PLoS One, 8 (4), e59834.
Scheiring, Gábor, Irdam, Darja, and King, Lawrence (2019a), '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, 41 (4), 673-91.
Scheiring, Gábor, et al. (2019b), 'Deindustrialisation, Foreign Investment and Social Development:
The Impact of Industrial Transformations on Mortality Inequalities', (Cambridge, UK: Manuscript,
Department Of Sociology, University of Cambridge).
Scheiring, Gábor, et al. (2018), 'The gendered effects of foreign investment and prolonged state
ownership on mortality in Hungary: an indirect demographic, retrospective cohort study', The Lancet
Global Health, 6 (1), 95-102.
Shkolnikov, Vladimir M, et al. (2004), 'Mortality Reversal in Russia: The Story So Far', Hygiea
Internationalis, 4 (1), 29-80.
19
Stuckler, David, King, Lawrence, and McKee, Martin (2009a), 'Mass privatization and the post-
communist mortality crisis: a cross-national analysis', The Lancet, 373 (9661), 399-407.
Stuckler, David, et al. (2009b), 'The public health effect of economic crises and alternative policy
responses in Europe: an empirical analysis', The Lancet, 374 (9686), 315-23.
UNICEF (2001), A Decade of Transition: Regional Monitoring Report 8 (Florence: UNICEF).
