Elsevier

Social Science & Medicine

Volume 143, October 2015, Pages 171-178
Social Science & Medicine

Unemployment transitions and self-rated health in Europe: A longitudinal analysis of EU-SILC from 2008 to 2011

https://doi.org/10.1016/j.socscimed.2015.08.040Get rights and content

Highlights

  • Unemployed individuals report poorer self-rated health than employed individuals.

  • Self-rated health levels fall when people move from employment to unemployment.

  • This health fall is small compared to the health gap between employed and unemployed.

  • Self-rated health levels fall more among older workers.

  • The fall in levels of self-rated health differs between European countries.

Abstract

The Great Recession of 2008 has led to elevated unemployment in Europe and thereby revitalised the question of causal health effects of unemployment. This article applies fixed effects regression models to longitudinal panel data drawn from the European Union Statistics on Income and Living Conditions for 28 European countries from 2008 to 2011, in order to investigate changes in self-rated health around the event of becoming unemployed. The results show that the correlation between unemployment and health is partly due to a decrease in self-rated health as people enter unemployment. Such health changes vary by country of domicile, and by individual age; older workers have a steeper decline than younger workers. Health changes after the unemployment spell reveal no indication of adverse health effects of unemployment duration. Overall, this study indicates some adverse health effects of unemployment in Europe – predominantly among older workers.

Introduction

Following the Great Recession, unemployment rates in the European Union (EU-28) rose from 6.8 per cent in January 2008 to 10.0 per cent in January 2012 (OECD, 2014). Because it is well documented that unemployed people have poorer health than those who are employed (Bartley et al., 2005, Schmitz, 2011), this rise in unemployment has led to concern for the well-being and health of those affected (Catalano et al., 2011). Poorer health among the unemployed may be driven by various processes, including (1) causation – individuals becoming and remaining unemployed develop poorer health than those who continue working, and (2) health selection – individuals in poor health have elevated risks of becoming and staying unemployed. How far does self-rated health change when people move between employment and unemployment? This article investigates this issue using the panel of the European Union Statistics on Income and Living Conditions (EU-SILC) from 2008 to 2011.

Health selection means that people in poor health are more likely to become and to stay unemployed than people in good health. The reasons can be that poor health leads to unemployment or that various other factors affect both health and employment prospects, sometimes labelled direct and indirect health selection (Steele et al., 2013). Using various indicators of health, several studies have found that people in poor health are more likely to become unemployed than those who are healthier (Korpi, 2001, Virtanen et al., 2013). Indicators include self-rated health (Elstad and Krokstad, 2003, Van de Mheen et al., 1999, Virtanen et al., 2005), psychological distress (Mastekaasa, 1996), number of self-reported health symptoms (Korpi, 2001), and longstanding illness (Arrow, 1996). Both Virtanen et al. (2013) and Korpi (2001) found that poor self-rated health increases the risk of becoming and remaining unemployed in Sweden, and Schuring et al. (2007) drew similar findings from a more comprehensive panel from 12 European countries. A study from Great Britain (1973–2009) shows that over the last decades, people with limiting longstanding illness have had increasingly lower probability of employment compared to their counterparts in better health (Minton et al., 2012). In Europe Reeves et al. (2014) find that health selection processes are reinforced in the recent years.

Some of this selection might be due to indirect health selection into unemployment, i.e. through the effect of underlying causes on health and employment status. In Germany, Arrow (1996) found that immigrants, women, young adults, and previously unemployed people are at particularly high risk of health selection into unemployment. In their 12-country study, Schuring et al. (2007) found an elevated risk of health selection among unmarried women, parents of young children, elderly people, and low-income groups. Low education and poor health may also increase the risk of remaining unemployed (Bartley and Owen, 1996, Korpi, 2001, van der Wel et al., 2011). Nevertheless, disentangling such indirect health selection from direct health selection requires sophisticated methods because health and social position cannot (and should not) be randomised. Using dynamic panel models, which address the effect of previous health on current health, Steele et al. (2013) found limited evidence for direct selection but strong support for indirect selection; unmeasured individual factors were associated with higher risk of both unemployment and ill health.

Longitudinal data allow for investigations into changes in health as individuals become unemployed as well as temporal changes in health before and after becoming unemployed. Such methods come closer to causal effects than cross-sectional comparison because they can filter out all time-variant individual characteristic leading to both unemployment and poor health (Gunasekara et al., 2014).

However, there could be individual characteristics that change over time that might affect both health and the probability of unemployment. For example, alcoholism or marital dissolution could lead to both unemployment and poor health. These would be examples of time-varying confounding and health selection effects. Longitudinal data typically allow for investigating some – but not all – such effects.

Flint et al. (2013) found that unemployment transitions were associated with a decrease in self-reported mental distress, suggesting that unemployment generates psychological stress. In a review of longitudinal research on health and unemployment, Catalano et al. (2011) found that job losers are twice as likely as those who remain employed to have increased symptoms of depression and anxiety. On average, job losers tend to increase their report of symptoms by 15–30 per cent, suggesting a possible causal link between unemployment and health. Nevertheless, studies investigating how health changes around the time that unemployment occurs could be contaminated by direct health selection (when a sudden health decline precedes unemployment) and indirect selection (when a third factor affects both outcomes).

For such reasons, some analysts believe that plant closures or major layoffs are better indicators of true causal effects than instances of individual unemployment (Jin et al., 1995, Morris and Cook, 1991). Schmitz (2011) found a greater decline in health as measured by hospitalisation, mental health scores and satisfaction with health among people unemployed for individual reasons than among people becoming unemployed as a result of closures or mass layoffs. For those unemployed because of a closure, a similar finding was discovered for hospital visits, but not for satisfaction with health or mental health. Schmitz (2011) argues that the divergent results for the two groups are due to health selection. However, cases of downsizing and individual job terminations could be perceived as the result of selection based on the individuals' characteristics, unlike closures that affect the entire staff (Mastekaasa, 1996). Individuals who are laid off individually may relate their job loss to their inadequate job performance or other unattractive individual characteristics, and this interpretation may be more stressful than collective unemployment due to closure. As such, investigations of health effects of unemployment could benefit from a more direct investigation of health changes prior to unemployment.

We hypothesise (1) that changes in health when people become unemployed can explain some of the health difference between employed and unemployed individuals. We also hypothesise that these effects of unemployment will vary by individual characteristics. Because unemployment is more common among younger people and they are more likely than older workers to be reemployed (Skärlund et al., 2012, Wanberg et al., 2002), we hypothesise (2) that older workers will suffer more adverse health consequences than younger workers on becoming unemployed. Because it is probably easier for women than men to adopt social roles other than that of “breadwinner” (Kuhn et al., 2009), we expect (3) that the health consequences of unemployment to be more adverse for men than for women. We also expect (4) the health consequences of unemployment will be less severe for highly educated than for less educated individuals. One reason is that employers might prefer more highly educated workers, making those with more education more likely to gain reemployment than those with less (Carling et al., 1996). More educated individuals may also have resources that make it easier for them to engage in alternative activities during periods of unemployment – for example, pursuing further education or training opportunities.

Finally we hypothesise (5) that the relationship between unemployment and health may vary between European countries. The current analysis makes no assumptions about the countries or country in which various characteristics predict better or worse health effects following individual unemployment.

This analysis uses data from the 2008–2011 panel of the European Union Statistics on Income and Living Conditions (EU-SILC). It uses 404,843 yearly observations from 189,177 individuals who were in the labour force (working or unemployed) and living in 28 European countries (i.e. the EU-28, excluding Germany and Ireland and including Norway and Iceland). The data have been harmonised according to European Parliament and Council regulation 1177/2003, and they comprise an extraordinarily rich source of employment information. All variables – dependent and explanatory – can vary between the up-till four yearly observations of each individual (2008–2011).

The dependent variable is self-rated health, measured on a single item (“How is your health in general?”) and ranked on a 5-point scale (5 = “very good”, 4 = “good”, 3 = “fair”, 2 = “bad”, and 1 = “very bad”). This item has been shown empirically to be a powerful predictor of future morbidity and mortality (Burström and Fredlund, 2001, Eriksson et al., 2001, Idler et al., 2000). In EU-SILC, this question has an overall response rate of 85 per cent.

Data on unemployment versus employment, the main independent variable of interest, were collected retrospectively from the EU-SILC, which provides information on the main activity over the previous 12 months. Full-time, part-time and self-employment were given the value 1, unemployment was given the value 0, and all other activities (e.g. education/training, unpaid work experience, retirement, permanent disability/inability to work, compulsory military or community service, domestic responsibilities, etc.) were recorded as “missing”. If more than one type of activity occurred in the same month, priority was given to economic over non-economic activity or inactivity.

Unemployment (unemployed at t) is coded 1 if the respondent is unemployed at the time of the interview, 0 if employed. Unemployment transition (employed at t-1, t-2 or t-3) is coded 1 if the respondent is observed to be employed at previous interviews, but had a transition into unemployment between baseline and interview. Reemployment (employed at t, unemployment transition at t-1 or t-2) is coded 1 if the respondent re-entered employment after an unemployment transition.

Health changes before and after the unemployment spell were investigated by utilising the time distance from the unemployment spell to the interview. To locate the exact month of unemployment transition, we created a job history file from the retrospective information on the main activity of each respondent for each month from 2007 through 2010. Transitions from employment to unemployment were recorded when at least three months of employment was followed by at least three months of unemployment. We then calculated the time from the month when a period of unemployment began to the time of the interview for all yearly observations. This variable was separated at zero to provide two variables, where health trend before unemployment spell denotes the temporal distance between interview and unemployment spell in the time before becoming unemployed while health trend after unemployment spell denotes the equivalent temporal distance in the time after becoming unemployed. On this variables, we recorded 7251 observations among 6156 individuals (mean = 1.18) before unemployment transition and 33,344 observations among 17,162 individuals (mean = 1.92) after unemployment transition. The unequal number of before and after unemployment observations is mainly attributable to the survey design. Respondents reported their monthly job history for the previous year at the time of the interview. Consequently, there will be more information on health after unemployment spells than before, providing stronger statistical power for health change after than the health trend before.

Time-varying covariates are current age (linear and squared), partnership (married or cohabiting) status and the number of dependent children (i.e. household members below 16 years) in the household. Disposable household income might mediate the effect of unemployment on health. This variable is recoded into logarithm because the impact of absolute changes may depend on the income level (Kawachi et al., 2010).

Gender and education level are time-invariant variables. Following Heggebø (2015) education is represented by two dummy-variables computed from the highest ISCED level attained. Pre-primary, primary and lower secondary is collapsed to primary education; (upper) secondary and post-secondary non-tertiary is collapsed to secondary education (reference category); and all higher educational qualifications are coded as higher education.

The data were analysed using linear regression models. Distributions in self-rated health were investigated using ordinary least squares (OLS) regression models, whereas changes in self-rated health were investigated using panel data models with individual fixed effects.

The OLS model estimates the mean self-rated health score among unemployed compared to the employed. Such estimates include both selection and causal effects. The fixed effects model estimates the within individual health change and thereby controls for all (measured and unmeasured) time-invariant confounding effects (Gunasekara et al., 2014). Health selection due to fixed factors is thereby eliminated.

Fixed effects estimates might be contaminated by health selection if there is a short time span between declining health and the onset of unemployment (Gunasekara et al., 2014). This possibility is tested by estimating health changes prior to entering unemployment; the data reveal no such tendencies. A lagged dependent variable is endogenous and cannot therefore be included in a regular fixed effects model. Thus, to control for path dependency – i.e. that previous health predicts current health changes – we employ Arellano–Bond dynamic fixed effects estimation (Arellano and Bond, 1991), which is a Generalised Method of Moments (GMM) estimator particularly appropriate for short panels with large number of observations (Arellano and Bond, 1991, Bond, 2002, Cameron and Trivedi, 2010). The Arellano-Bond estimator eliminates potential omitted variables bias by first-differencing, before estimating a system of year specific equations where first lag regressors constitute an instrument for the lagged dependent variable (Cameron and Trivedi, 2010, pp. 293–303).

Transitions from work to unemployment are associated with lower income. How far income mediates the relationship between unemployment and health is tested in a separate model.

Three models investigate how far the health effects of becoming unemployed are modified by three individual characteristics using interaction terms between unemployment and gender (female dummy), age (linearized) and education level (two dummy variables). Whether the results vary between the 28 European countries is investigated using interactions between unemployment and country dummies controlling for covariates and age interactions. The coefficients are estimated at age 40 and country-variation is tested by an associated (27 df) F-test.

Because national sample sizes do not correspond to the size of the national workforces, all OLS and regular fixed effects models apply population weights that provide estimates representative of the European population. Population weights were calculated as the function of pn, where p is the number of employees (aged 20–64) in the labour force, and n is the number of respondents in the analysis. Information on the number of employees (aged 20–64) in the labour force was extracted from Eurostat (2014). Test statistics are robust for heteroscedasticity and correct for the fact that repeated observations (2008, 2009, 2010 and 2011) for each individual are not statistically independent using the cluster option in Stata (2007). All tables present two-sided tests.

Section snippets

Descriptive statistics

Table 1 reports descriptive statistics of the data. At one interview or more, 37,413 (10.9 per cent) respondents were unemployed, and 9472 (4.0 per cent) moved from employment (three months or more) to unemployment (three months or more) during the time covered by the job history data.

Self-rated health (1–5) has a mean value of 4.056 (SD = 0.761). Employed Europeans reported better health (4.081) than unemployed individuals (3.851). Respondents were aged on average 42 years (SD = 11.6) and had

Discussion

The 2008 economic crisis has manifested itself in increased, and for several countries historically high, unemployment rates. Because the recession has been long-lasting and unemployment rates have remained high, there is good reason to be concerned about the welfare of those entering unemployment. Even a small individual health effect of unemployment could have substantial impact on health if accumulated at population level. This analysis investigates the association between a transition into

Conclusion

This study has investigated the individual health changes associated with unemployment transitions in Europe. Workers – especially older workers – who became unemployed during the Great Recession experienced a drop in self-rated health at the time of the transition. However, the potentially causal effect of unemployment on self-rated health appears to diminish after entering unemployment. The results indicate that workers in poor health face elevated risk of becoming unemployed. Taken together

Funding

This work was supported by the Research Council of Norway, grant number 221037.

Acknowledgements

This paper was written as a part of the Health Inequalities, Economic Crisis, and the Welfare State project. We thank Espen Dahl and Ruth Bell for valuable comments on earlier drafts. We would like to thank the reviewers for the detailed suggestions for improving the manuscript.

References (45)

  • K.A. van der Wel et al.

    Social inequalities in ‘sickness’: European welfare states and non-employment among the chronically ill

    Soc. Sci. Med.

    (2011)
  • P. Virtanen et al.

    Health status and health behaviour as predictors of the occurrence of unemployment and prolonged unemployment

    Public Health

    (2013)
  • M. Arellano et al.

    Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations

    Rev. Econ. Stud.

    (1991)
  • C. Bambra

    Work, Worklessness, and the Political Economy of Health

    (2011)
  • M. Bartley

    Unemployment and ill health: understanding the relationship

    J. Epidemiol. Community Health

    (1994)
  • M. Bartley et al.

    Health and labour market disadvantage: unemployment, non-employment and job insecurity

  • M. Bartley et al.

    Relation between socioeconomic status, employment, and health during economic change, 1973–93

    BMJ

    (1996)
  • S. Bond

    Dynamic panel data models: a guide to micro data methods and practice

    Port. Econ. J.

    (2002)
  • P. Brickman et al.

    Hedonic relativism and planning the good society

  • B. Burström et al.

    Self rated health: is it as good a predictor of subsequent mortality among adults in lower as well as in higher social classes?

    J. Epidemiol. Community Health

    (2001)
  • A.C. Cameron et al.

    Microeconometrics Using Stata

    (2010)
  • R. Catalano et al.

    The health effects of economic decline

    Annu. Rev. Public Health

    (2011)
  • Cited by (59)

    • Do changes in network structure explain why unemployment damages health? Evidence from German panel data

      2022, Social Science and Medicine
      Citation Excerpt :

      As a stressful life event (Pearlin, 1989), unemployment has severe negative consequences on individuals’ health and health behavior. Studies have documented effects on self-rated health (Tøge and Blekesaune, 2015), health satisfaction (Krug and Eberl, 2018), mental health and/or physical health (Gebel and Voβemer, 2014; Stauder, 2019), and health behaviors such as smoking (Marcus, 2014) and physician contacts (Åhs and Westerling, 2006); however, some authors have also reported null effects for one or more of these health measures (e.g., Böckerman and Ilmakunnas, 2009; Salm, 2009; Schmitz, 2011). Such expertly performed studies offer what Shadish et al. (2002) call causal description.

    • Does perceived social mobility affect health? Evidence from a fixed effects approach

      2022, Social Science and Medicine
      Citation Excerpt :

      Considering that there is a large educational gradient in health in Poland (Korzeniowska and Puchalski, 2015; Szklarska and Anita Jankowska, 2003), we account for participants’ educational attainment (coded as for fathers). Although some studies in the field treat education as time-invariant (e.g. Tøge and Blekesaune, 2015), in POLPAN the chances of respondents moving from one educational category to another from wave to wave are significantly higher than zero. Table S3 in the supplementary materials shows a matrix of transition probabilities of educational attainment between POLPAN waves.

    • Public health in restructuring coastal communities: Generational trends in self-rated health following the decline in small-scale fishing. The HUNT study, Norway

      2021, Journal of Rural Studies
      Citation Excerpt :

      The decline in small-scale fishing undoubtably resulted in an extensive reduction in coastal-based livelihoods, potentially leaving people unemployed. The correlation between unemployment and poor health has been widely studied and confirmed (Abebe et al., 2016; Tøge and Blekesaune, 2015), also in countries with long-standing welfare services (Bambra and Eikemo, 2008). Unemployment has been found to negatively affect well-being, possibly resulting in poor physical health over time (Bartley et al., 2006; Korpi, 2001; Ytterdahl and Fugelli, 2000), in addition to long-term consequences for the displaced workers, with earning losses and worsened labour-market position exceeding the transitory period of adjustment (Eliason and Storrie, 2006).

    • Regional employment and individual worklessness during the Great Recession and the health of the working-age population: Cross-national analysis of 16 European countries

      2020, Social Science and Medicine
      Citation Excerpt :

      Our study is unfortunately limited by the use of self-reported data which may be subject to reporting bias. However, our research builds on existing cross-national studies in the area of employment and health in which self-rated health has often been the sole outcome variable (Huijts et al., 2015; Shaw et al., 2014; Tøge and Blekesaune, 2015). Our study is also cross-sectional which precludes the inference of causal relationships.

    View all citing articles on Scopus
    View full text