Data source and country coverage
We analyse the third and seventh waves of the Survey of Health, Ageing and Retirement in Europe (SHARE)
1 [
28], ,which is a cross-national panel database of micro data on health and socio-economic status of individuals aged 50 or older covering 27 European countries and Israel [
29,
30] . The third and seventh waves include retrospective questions about respondents’ life history, such as employment history, periods of stress and financial difficulties, and health at younger ages. Data were collected in 2009 and 2017 respectively, thus even the youngest cohorts of the sample were already of active age during the times of transition. The seventh wave of SHARE questionnaire contains a retrospective questionnaire for all respondents who did not participate in the third wave, as well as a regular panel questionnaire for all respondents who already answered the retrospective questions in the third wave. Thus, each SHARE respondent who participated in the seventh wave answered the retrospective questions exactly once (either in the third or the seventh wave). We group the countries into post-socialist CEE countries (Bulgaria, Croatia, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, Slovak Republic, Slovenia) and the rest, labelled as ‘West’ (Austria, Belgium, Cyprus, Denmark, Finland, France, Germany, Greece, Ireland, Israel, Italy, Luxembourg, Malta, Netherlands, Portugal, Spain, Sweden, Switzerland). We split the German sample according to the place of residence on 1st November 1989 (i.e. before the Berlin wall came down).
We split the post-socialist CEE countries (except for East Germany) further into three groups: the Visegrad countries (V4: Czech Republic, Hungary, Poland, Slovak Republic), Baltic countries (Estonia, Latvia, Lithuania) and Southern countries (Bulgaria, Croatia, Romania, Slovenia).
Measures
We assess the current health (as measured in 2017, the 7th wave of the SHARE data) of the 50+ population with several indicators. Self-rated general health is measured on a 5-point Likert scale from excellent to poor, which is a strong predictor of morbidity and mortality [
31]. From this variable, following the standard approach in the literature [
32‐
35], we generate a binary indicator of poor health which equals 1 if the self-rated health is fair or poor, 0 otherwise. Other binary outcome variables indicate whether the respondent suffers from chronic or long-term health problems (long-term illness, henceforth), has a health problem that limits paid work, has certain conditions, such as heart problems, hypertension, diabetes, ulcer, cancer, and chronic lung disease (each condition is assessed by a separate dichotomous variable). Besides reported health conditions, dependent variables include obesity (Body Mass Index 30 or greater) and an indicator of grip strength, which was shown to explain old age disability [
36]. Since grip strength, on average, varies by age, gender and the build of the individuals, we create a binary indicator of weak grip strength, which equals one if the grip strength is below the gender, 10-year age group and country specific median of grip strength. For the sake of brevity, we focus on the binary indicators of poor health and long-term illness in the main analysis, as two composite health indicators, with the indicator of long-term illness being more objective. We relegate the results on the other health measures to the Appendix.
To identify shocks around the transition, we look at retrospectively reported periods of stress and financial hardship that started between 1987 and 1993 and at reported end of jobs between 1987 and 1993 with at least 6 months of gap without employment or immediate retirement afterwards. The latter two indicators measure whether the respondent suffered from economic difficulties, while the stress variable may capture the general burden of uncertainties experienced during the system change as well. These binary measures of hardship are set to zero for those who do not report the analysed hardship ever (i.e. no stress, no hardship and no end of job with 6 months gap afterwards, except for retirement, respectively). Descriptive statistics are provided in Table
1.
Table 1
Descriptive statistics
Health in 2017 |
poor health | 0.393 | 0.488 | 0.452 | 0.498 | 0.462 | 0.499 | 0.642 | 0.479 |
long-term illness | 0.493 | 0.500 | 0.615 | 0.487 | 0.494 | 0.500 | 0.597 | 0.490 |
health limits work | 0.213 | 0.410 | 0.324 | 0.468 | 0.182 | 0.386 | 0.318 | 0.466 |
any chronic disease | 0.529 | 0.499 | 0.593 | 0.491 | 0.574 | 0.495 | 0.606 | 0.489 |
hypertension | 0.393 | 0.488 | 0.480 | 0.500 | 0.472 | 0.499 | 0.474 | 0.499 |
heart problem | 0.106 | 0.307 | 0.145 | 0.353 | 0.103 | 0.303 | 0.163 | 0.370 |
diabetes | 0.130 | 0.336 | 0.161 | 0.368 | 0.117 | 0.321 | 0.100 | 0.300 |
ulcer | 0.028 | 0.165 | 0.052 | 0.222 | 0.045 | 0.208 | 0.088 | 0.283 |
cancer | 0.046 | 0.210 | 0.044 | 0.205 | 0.024 | 0.153 | 0.053 | 0.225 |
lung disease | 0.058 | 0.234 | 0.050 | 0.217 | 0.045 | 0.207 | 0.056 | 0.231 |
obese | 0.199 | 0.399 | 0.299 | 0.458 | 0.256 | 0.437 | 0.308 | 0.462 |
weak grip strength | 0.531 | 0.499 | 0.517 | 0.500 | 0.540 | 0.498 | 0.534 | 0.499 |
Hardship around transition |
(1 = yes; 0 = no hardship ever) |
stressful period | 0.137 | 0.344 | 0.081 | 0.272 | 0.104 | 0.305 | 0.123 | 0.328 |
financial difficulties | 0.056 | 0.229 | 0.062 | 0.241 | 0.072 | 0.259 | 0.120 | 0.325 |
job ends with gap after | 0.085 | 0.279 | 0.109 | 0.312 | 0.125 | 0.330 | 0.148 | 0.355 |
Start of hardship conditional on hardship ever |
(1 = around transition; 0 = before/after transition) |
stressful period | 0.151 | 0.358 | 0.130 | 0.336 | 0.161 | 0.368 | 0.178 | 0.383 |
financial difficulties | 0.129 | 0.336 | 0.158 | 0.365 | 0.155 | 0.362 | 0.271 | 0.445 |
job ends with gap after | 0.242 | 0.428 | 0.355 | 0.479 | 0.451 | 0.498 | 0.314 | 0.464 |
Individual characteristics |
age in 2017 | 67.396 | 10.629 | 65.901 | 10.069 | 66.529 | 10.122 | 66.624 | 10.585 |
female | 0.536 | 0.499 | 0.559 | 0.497 | 0.555 | 0.497 | 0.604 | 0.489 |
education (0 = primary, 2 = secondary, 3 = primary) | 0.937 | 0.685 | 1.012 | 0.468 | 0.940 | 0.506 | 1.210 | 0.553 |
childhood health (1 = excellent to 5-poor) | 2.192 | 1.036 | 2.200 | 0.981 | 1.970 | 0.964 | 2.616 | 1.044 |
hospitalisation during childhood | 0.053 | 0.224 | 0.063 | 0.243 | 0.034 | 0.183 | 0.092 | 0.289 |
Industry of last job prior 1987 |
agriculture, hunting, forestry, fishing | 0.078 | 0.268 | 0.186 | 0.389 | 0.154 | 0.361 | 0.242 | 0.428 |
mining and quarrying | 0.013 | 0.114 | 0.032 | 0.177 | 0.041 | 0.198 | 0.010 | 0.099 |
manufacturing | 0.181 | 0.385 | 0.252 | 0.434 | 0.305 | 0.460 | 0.211 | 0.408 |
electricity, gas and water supply | 0.019 | 0.137 | 0.022 | 0.146 | 0.023 | 0.150 | 0.024 | 0.153 |
construction | 0.087 | 0.282 | 0.077 | 0.266 | 0.083 | 0.276 | 0.077 | 0.267 |
wholesale and retail trade | 0.117 | 0.321 | 0.081 | 0.273 | 0.067 | 0.251 | 0.056 | 0.230 |
hotels and restaurants | 0.033 | 0.180 | 0.021 | 0.143 | 0.020 | 0.141 | 0.022 | 0.147 |
transport, storage and communication | 0.049 | 0.216 | 0.069 | 0.253 | 0.080 | 0.271 | 0.082 | 0.274 |
financial intermediation | 0.031 | 0.172 | 0.010 | 0.097 | 0.006 | 0.079 | 0.008 | 0.091 |
real estate, renting and business activity | 0.013 | 0.114 | 0.004 | 0.059 | 0.001 | 0.032 | 0.003 | 0.057 |
public administration and defence | 0.093 | 0.290 | 0.051 | 0.220 | 0.033 | 0.178 | 0.033 | 0.180 |
education | 0.081 | 0.272 | 0.071 | 0.257 | 0.050 | 0.217 | 0.102 | 0.303 |
health and social work | 0.075 | 0.264 | 0.053 | 0.224 | 0.034 | 0.182 | 0.054 | 0.227 |
other community | 0.130 | 0.336 | 0.072 | 0.259 | 0.102 | 0.303 | 0.075 | 0.263 |
Total number of individuals | | 43,424 | | 12,310 | | 10,025 | | 8739 |
Statistical models
We estimated multivariate logistic regressions of current health indicators, with binary measures of hardship during the transition as explanatory variables. We added the following confounding variables to the multivariate models that are likely to influence both health outcomes and our explanatory variable (shock indicator): age in 2017, gender, education (categorised as primary, secondary and tertiary, based on the international classification, ISCED-97), the industry code of the last job before the transition, and measures of childhood health (self-evaluated overall childhood health and a dummy for hospitalisation during childhood). To account for country specific differences, we included country dummies. We calculated cluster-robust standard errors, clustering on the country level, using the
vce (cluster clustvar) option of Stata, as explained by Cameron and Trivedi (2009), section 3.3.5 [
37]. All our results are based on weighted data, using calibrated individual weights. Hence, in the weighted sample, smaller countries have smaller weights. Also, with using the calibrated weights, we avoid bias due to unit nonresponse and panel attrition (see Malter and Börsch-Supan, 2015 [
38] for details).
In the first group of models, we estimated the effects of the shocks around the transition on the subsample of CEE countries for each health outcome and type of hardship, allowing the effects to differ by subgroups of the CEE countries:
$$ \Pr \left({\mathrm{h}}_{\mathrm{igc}}=1\right)=\Lambda \left({\mathrm{s}}_{\mathrm{igc}}{\mathrm{D}}_{\mathrm{g}}{\boldsymbol{\upalpha}}_{\mathbf{1}}+{\mathrm{x}}_{\mathrm{igc}}{\boldsymbol{\upbeta}}_{\mathbf{1}}+{\gamma}_{\mathrm{c}}\right) $$
(1)
where Λ is the logistic function,
higc is the binary indicator of current health problem of individual
i living in country-group
g and country
c,
Dg is a binary indicator of living in country group
g,
s is the indicator of hardship during transition,
x is the set of confounding variables listed above, and γ
c captures the country effects. Our focus is on the exponential of the coefficient vector
α1 (reported in Table
2), showing how the odds of a health problem in 2017 relates to having had hardships during transition in a specific country-group. Individuals who never had such hardships (according to the retrospective survey) serve as the comparison group.
Table 2
First and second groups of models - Health measures regressed on difficulties occurring between 1987 and 1993 in CEE country groups and in CEE and West
Stress x V4 | 1.728*** | 2.610*** | Fin. difficulties x V4 | 1.923*** | 2.112*** | Job ends x V4 | 1.502*** | 1.419*** |
| [1.502–1.988] | [2.243–3.037] | | [1.211–3.054] | [1.676–2.663] | | [1.355–1.665] | [1.331–1.514] |
Stress x South | 2.042*** | 2.236*** | Fin. difficulties x South | 1.771*** | 1.549*** | Job ends x South | 1.599*** | 1.343*** |
| [1.607–2.596] | [2.052–2.438] | | [1.263–2.484] | [1.111–2.159] | | [1.430–1.788] | [1.222–1.476] |
Stress x Baltic | 1.592** | 1.724*** | Fin. difficulties x Baltic | 1.175*** | 1.522*** | Job ends x Baltic | 1.967*** | 1.707*** |
| [1.112–2.280] | [1.222–2.434] | | [1.111–1.243] | [1.208–1.917] | | [1.585–2.442] | [1.502–1.940] |
Observations | 17,452 | 17,452 | Observations | 20,503 | 20,503 | Observations | 20,524 | 20,525 |
p-value of Wald test | 0.425 | 0.034 | p-value of Wald test | 0.007 | 0.121 | p-value of Wald test | 0.081 | 0.008 |
Second group of models |
| Poor health | Long-term illness | | Poor health | Long-term illness | | Poor health | Long-term illness |
CEE x stress | 1.563*** | 1.966*** | CEE x fin. difficulties | 1.773*** | 1.588*** | CEE x job ends | 1.502*** | 1.238* |
| [1.233–1.980] | [1.473–2.623] | | [1.408–2.232] | [1.243–2.029] | | [1.368–1.650] | [0.979–1.564] |
West x stress | 1.526*** | 1.506*** | West x fin. difficulties | 1.670*** | 1.780*** | West x job ends | 1.270 | 1.282*** |
| [1.114–2.088] | [1.343–1.689] | | [1.343–2.077] | [1.540–2.057] | | [0.931–1.733] | [1.094–1.501] |
Observations | 35,273 | 35,276 | Observations | 43,539 | 43,541 | Observations | 43,344 | 43,344 |
p-value of Wald test | 0.826 | 0.075 | p-value of Wald test | 0.731 | 0.399 | p-value of Wald test | 0.177 | 0.707 |
To analyse to what extent do the associations between transition related shocks and later health vary within the country groups, we estimate a modified version of Eq. (
1). Here, we replace
Dg with the binary indicators of living in the specific country in the CEE region. Also, we replace
s with the binary indicator of experiencing any of the analysed three shocks around the transition, with individuals who never had such hardships serving as the comparison group. We analyse the three shocks jointly in the country-specific analysis to ensure that we have a sufficient number of observations of transition related shocks in each country.
In the second group of models, we included the Western countries and analysed a possible interaction between the shocks and the region (CEE versus West) based on the following equation:
$$ \Pr \left({\mathrm{h}}_{\mathrm{irc}}=1\right)=\Lambda \left({\mathrm{s}}_{\mathrm{irc}}{\mathrm{D}}_{\mathrm{r}}{\boldsymbol{\upalpha}}_{\mathbf{2}}+{\mathrm{x}}_{\mathrm{irc}}{\boldsymbol{\upbeta}}_{\mathbf{2}}+{\varepsilon}_{\mathrm{c}}\right) $$
(2)
where the notation is the same as in Eq. (
1), with country-group specific coefficients replaced with region (
r) specific coefficients. The exponential of
α2 (reported in Table
2) shows how the odds of a health problem in 2017 relates to having had hardships during transition in CEE or in the West, with individuals who never had such hardships serving as the comparison group.
In the third group of models (Table
3), we extended the time period and assessed the impact of shocks in CEE between 1984 and 1996, to see whether difficulties around and probably due to the transition are specific or not:
$$ \Pr \left({\mathrm{h}}_{\mathrm{ic}}=1\right)=\Lambda \left({\mathrm{s}}_{\mathrm{ic}}{\mathrm{T}}_{\mathrm{ic}}{\boldsymbol{\upalpha}}_{\mathbf{3}}+{\mathrm{x}}_{\mathrm{ic}}{\boldsymbol{\upbeta}}_{\mathbf{3}}+{\omega}_{\mathrm{c}}\right) $$
(3)
Table 3
Third group of models – Health measures regressed on difficulties occurring between 1987 and 1993 versus 1984–1986 and 1994–1996 in CEE
transition x stress | 1.695*** | 2.045*** | transition x fin. difficulties | 1.923*** | 1.993*** | transition x job ends | 1.379*** | 1.441*** |
| [1.429–2.010] | [1.812–2.306] | | [1.515–2.441] | [1.581–2.512] | | [1.174–1.620] | [1.189–1.745] |
before/after transition x stress | 1.562*** | 1.940*** | before/after transition x fin. difficulties | 1.780*** | 1.589*** | before/after transition x job ends | 1.403*** | 1.156 |
| [1.203–2.028] | [1.378–2.731] | | [1.428–2.219] | [1.187–2.127] | | [1.147–1.715] | [0.804–1.660] |
observations | 19,174 | 19,174 | observations | 21,837 | 21,837 | observations | 21,773 | 21,774 |
where the notation is the same as in Eq. (
1), but instead of estimating country-group specific coefficients, we allow the health implications of hardships to vary with the time period when the difficulties occurred, denoted by
Tic (during the transition period versus before or after the transition period).
Finally, we estimated a modified version of Eq. (
3), where we allow the health implications of hardships to vary with gender, education and age category, restricting the sample again to CEE and considering shocks occurring between 1987 and 1993 (Table
4).
Table 4
Heterogeneity analysis results - Health measures regressed on difficulties occurring between 1987 and 1993 in CEE
stress | 2.206*** | 2.650*** | fin. difficulties | 1.910*** | 1.874*** | job ends | 1.880*** | 1.625*** |
| [1.728–2.817] | [1.742–4.032] | | [1.494–2.441] | [1.315–2.669] | | [1.466–2.410] | [1.223–2.159] |
stress x female | 0.701** | 0.814 | fin. difficulties x female | 0.874 | 0.942 | job ends x female | 0.736 | 0.787 |
| [0.512–0.960] | [0.475–1.394] | | [0.679–1.125] | [0.686–1.293] | | [0.465–1.166] | [0.500–1.237] |
observations | 17,452 | 17,452 | observations | 20,503 | 20,503 | observations | 20,524 | 20,525 |
stress | 2.707*** | 3.097*** | fin. difficulties | 1.629 | 1.496*** | job ends | 1.019 | 1.082 |
| [1.652–4.434] | [1.573–6.098] | | [0.868–3.060] | [1.295–1.729] | | [0.881–1.178] | [0.866–1.351] |
stress x secondary education | 0.655* | 0.787 | fin. difficulties x secondary edu. | 1.090 | 1.259 | job ends x secondary edu. | 1.672*** | 1.383** |
| [0.416–1.030] | [0.415–1.493] | | [0.599–1.983] | [0.902–1.757] | | [1.407–1.988] | [1.070–1.787] |
stress x tertiary education | 0.645 | 0.615 | fin. difficulties x tertiary edu. | 1.157 | 1.060 | job ends x tertiary edu. | 1.148 | 1.017 |
| [0.321–1.297] | [0.269–1.407] | | [0.544–2.461] | [0.649–1.730] | | [0.861–1.532] | [0.740–1.398] |
observations | 17,452 | 17,452 | observations | 20,503 | 20,503 | observations | 20,524 | 20,525 |
stress | 1.743*** | 2.135*** | fin. difficulties | 1.939*** | 1.744*** | job ends | 1.415*** | 1.342** |
| [1.374–2.212] | [1.873–2.433] | | [1.185–3.170] | [1.310–2.321] | | [1.257–1.593] | [1.030–1.749] |
aged< 36 in 1990 x stress | 1.092 | 1.219* | aged< 36 in 1990 x fin. difficulties | 0.853 | 1.066 | aged< 36 in 1990 x job ends | 1.211** | 1.090 |
| [0.756–1.577] | [0.965–1.540] | | [0.517–1.407] | [0.753–1.507] | | [1.011–1.451] | [0.694–1.714] |
observations | 17,452 | 17,452 | observations | 20,503 | 20,503 | observations | 20,524 | 20,525 |