Good jobs, good pay, better health? The effects of job quality on health among older European workers

This paper examines the effects of job quality on physical and mental health for a population of workers aged 50–65 between 2004 and 2013 in 15 European countries using data from the Survey of Ageing, Health and Retirement in Europe (SHARE).

Despite improving morbidity and mortality, an almost universal access to health care, health and safety regulations, and an overall decline in accidents at work, around a quarter of Europeans believe their jobs to be pathogenic [1]. The current cost of work-related morbidities is estimated to amount to 2.6–3.8% of annual EU GDP [2]. Moreover, multiple studies have confirmed that poor job quality influences retirement intentions [3], and contributes to early cessation of labour force participation [4, 5], which could partly be driven by work-related ill-health [e.g. 6, 7].

On the one hand, extended working lives, a largely unbroken trend to more intense work [8], and increasingly common precarious employment relations [9] have an impact on job quality and potentially add to work-related health hazards. On the other hand, improvements in work-life balance, greater skills utilisation, and reduced exposure to physical and environmental hazards have made workplaces safer and more flexible to the needs of employees. Within this context, the current study examines how job quality affects health and how these effects shape health over time in a general population of workers aged 50 and older in European countries. In doing so, this paper contributes to the growing economic literature on work and well-being.

Job quality is thought to affect health mainly through exposure to work stress. Situations in which job demands exceed the resources available to the individual—physical capacity, knowledge, psychosocial rewards or decision latitude—to cope with the demands are conjectured to be stressors with potentially adverse health effects [10‐12]. The negative consequences from continuous or repeated exposure to even minor work-related stressors can accumulate over time, cause dysregulation in the body and eventually lead to health problems [13]. Environmental hazards at work, psychosocial facets of job quality and material well-being have been found to correlate with a range of stress-related physical and mental health disorders [14, 15].

But while existing models of work stress have greatly helped to conceptualise a link between work and health, psychosocial job facets cover only a fraction of wider job quality. Building on interdisciplinary evidence, Green and Mostafa [16] and Green et al. [8] have proposed a model of objective job quality that combines a more comprehensive list of potentially health and well-being related job facets. Their model covers earnings, intrinsic job quality, job prospects and working time quality, which may affect physical and psychological well-being through stress or material positioning. Job prospects, in particular, may be of growing significance as non-standard employment situations are becoming more widespread in Europe [17, 18]. By focusing on objective job characteristics, Green et al.’s approach avoids the use of hard to compare expressions of job satisfaction that may change with individual aspirations, expectations or personalities. These objective job quality domains are thought to satisfy workers’ needs and endow individuals with the capability to “do and be things they value” [16]. The underlying principles have for example been used to map out the distribution of job quality in European countries [19, 20] or the gender gradient in work-related health [21]. Unlike psychosocial theories of work stress, Green et al.’s concept does not theorise on specific configurations of working conditions that create health risks (e.g. job strain or effort-reward imbalances), but argues that the satisfaction of needs will influence well-being. In the following, I will apply their job quality concept to examine job-facet specific health impacts. A clearer understanding of which job facet relates to which health outcomes will illuminate in greater detail the relation of work to well-being. In doing so, this study analyses the effects of everyday working conditions for the average older workers and not the consequences of specific work-related risk factors in narrowly defined fractions of the workforce.

A growing body of economic research has raised concerns about the causal nature of the association between job facets and health outcomes (see Barnay [1] for a recent review). Firstly, if workers with greater health risks sort into more stressful jobs, the estimated effects of poor job quality on health may be spurious. This is referred to as unobserved heterogeneity and arises from insufficient information on the determinants of job assignment. The concurrent labour market positioning, type of job and health trajectory will depend on events and circumstances throughout the life-course [22‐26]. In effect, people from better off socio-economic backgrounds tend to be in better health and employed in higher quality jobs [27, 28]. Secondly, work-related health impairments could lead to early labour force withdrawal and thus to an endogenous sample composition. Especially, workers who are eligible for early retirement pensions may opt for economic inactivity rather than further exposure to adverse working conditions and risk of health deteriorations [29]. Thirdly, health and work may be simultaneously determined in a bidirectional relationship. Work affects health, but health also influences labour supply and job characteristics such as wages [30, 31]. If this is true, a fraction of the cross-sectional association between job quality facets and health may reflect reverse causality.

Health is also highly persistent over the lifecycle [32, 33]. Health shocks may thus develop a multiplicative effect as transitional spells of ill-health evolve into chronic conditions. If work-related health deteriorations persist over time, even relatively modest short-term gains will help to improve individual health and well-being in the long-run. Moreover, work-related stressors may continue to affect health even after exposure has ended. Empirical evidence suggests, for example, that exposure to unfavourable working conditions during midlife may correlate with health during retirement in later life [34], which would indidcate that there is some persistence of work-related health impairments. But the drivers of this association are not fully understood. Apart from persistent health shocks, persistence in working conditions or unobserved heterogeneity that influences positioning in the job quality spectrum, as well as health outcomes, are further conceivable mechanisms.

Accounting for unobserved heterogeneity and reverse causality has been shown to generally reduce the association between work and health [1], whilst addressing endogenous selection into economic inactivity tends to increase the estimated health impacts of work [29]. Disentangling the contribution of individual characteristics from work-related effects among older workers will help to gauge the potential health benefit that better working conditions may deliver during the final career stages.

By analysing the effects of job quality on multiple health outcomes in a cross-national representative panel of older workers, this paper contributes to the wider literature on work and well-being. In combining earnings, psychosocial job characteristics and job insecurity, it analyses the different pathways through which work may affect physical health, mental well-being as well as functional disabilities. Longitudinal data from SHARE allows accounting for time-invariant unobserved heterogeneity, model potential reverse causality and examine the persistence of work-related health effects over time. In conjunction with tests and correction for endogenous sample selection, this paper examines to what extent good jobs and good pay determine good health among older workers in European countries now and over time.

The remainder of the paper is organised as follows: Section two introduces the data and variables for the analysis. The empirical model is described in Section three. Section four reports the key findings, before Section five concludes.

Let \(h_{it}^{*j}\) be latent health in dimension \(j\) of individual \(i\) at time \(t\). Concurrent health in dimension \(j\) is modelled to depend on time-variant and time-invariant individual background characteristics \(X_{it}\), job quality \({\text{JQ}}_{it}\) and a random component capturing unobserved individual frailty \(\vartheta_{i}^{j}\) and idiosyncratic health shocks \(\varepsilon_{it}^{j}\)

The error term is split into a random normally distributed time constant individual effect to measure unobserved frailty differences between individuals (\(\vartheta_{i}^{j}\)), and an idiosyncratic error \(\left( {\varepsilon_{it} } \right)\). Random effects will pick up unobserved, time-constant differences in the risk of developing certain health conditions. Both error components are assumed to neither correlate with the independent variables nor with each other.

This baseline specification may be insufficient to identify \(\beta^{j}\) if job quality is correlated with the time-constant frailty term or the idiosyncratic error due to unobserved heterogeneity, sample selection or reverse causality.

Firstly, to enhance the specification of unobserved heterogeneity, within-individual means of job quality variables are added as covariates. The frailty component changes to \(\vartheta_{i}^{j} = \psi^{j} + \pi^{j} \overline{\text{JQ}}_{i} + \eta_{i}^{j}\), where \(\eta_{i}^{j}\) represents random effects that are independent of the observed covariates. Including within-individual means of job quality items allows for non-zero correlation between job facets and time-constant unobserved individual factors. In other words, this specification allows for characteristics that influence positioning in the job quality spectrum as well as health. In the case of linear panel models correlated random effects are equivalent to a fixed-effects approach. For non-linear models the approach provides a more flexible alternative to conditional maximum likelihood estimation at the expense of additional parametric assumption about the distribution of unobserved heterogeneity [52]. A test of joint significance of the added within-individual means helps to decide between random effects and correlated random effects models [53]. To allow for systematic differences between the sub-panels the empirical implementation will include intercepts for each sub-panel besides a full set of period effects. Doing so accounts for potentially unobserved sample selection mechanisms that may correlate with the covariates. Retrospective information on healthiness and socio-economic positioning during childhood are added in the sensitivity analysis to test whether these background measures suffice to make random effects specifications consistent.

Secondly, endogenous sample selection is modelled through a two-step Heckman style selection approach [54, 55]. In the first step, the propensity to work is estimated conditional on a set of instruments \(Z_{it}\), i.e. early retirement eligibility (age ≥ early retirement age) and the proportion of retired household members, and the other model covariates. Observing an individual in work is conditional on the propensity to work \(w_{it}^{*}\) which is captured by the indicator \(w_{it}\). The indicator will equal one if respondent \(i\) is in work at time \(t\). Selection may for instance pick up a healthy worker effect, where healthier and more resilient workers are more likely to work beyond the early retirement threshold. The selection model includes within-individual means of the instrument variables \(\overline{{Z_{i} }}\) and individual constant effects \(\tau_{\text{i}}\) to account for time-invariant unobserved heterogeneity that influences the timing of retirement

Endogenous sample selection implies that selection processes correlate with the idiosyncratic error components in Eq. (1). This can be easily tested within the extended approach.

Equation (2) is estimated by a set of probit regressions separately for each period in the pooled sample. The estimates are used to derive the inverse Mills ratio \(m_{it} \left( \cdot \right)\). In the second step, the inverse Mills ratio is added to the list of covariates in Eq. (1). The inverse Mills ratios represent the nonselection hazard and will capture associations between the decision to cease work and the work-health nexus over and above the time-invariant unobservables that are already included in the main models. Significance tests can be used to decide whether sample selection biases the unadjusted estimates.

The specific choice of estimator for Eq. (1) will change with health outcomes and follows common practice in the literature. Acute health conditions (AHC), functional disability (FD) and self-assessed health (SAH) which are either dummies or ordinal variables will be estimated using logit or ordered logit models. Cardiovascular risk factors (CVR), musculoskeletal disorders (MSD) and mental health (MH) are counts of reported symptoms and will be estimated using Poisson regressions. All models include period dummies, a set of dummies for each sub-panel, and are estimated with robust standard errors that allow for heteroscedasticity and serial correlation. In the following, job quality effects on concurrent health are discussed first, before examining the persistence of job quality related health effects over time.

This study examines the effects of job quality on a range of physical and mental health outcomes in an unbalanced panel of workers aged 50–65 years across 15 European countries. By applying a multi-facetted concept of job quality, the analyses shed light on different pathways through which work can affect physical and mental health in the short term and beyond. Unobserved heterogeneity, endogenous sample selection and reverse causality may bias the cross-sectional relation of job quality with health and well-being. Addressing these issues, I examine to what extent good pay and good jobs lead to good health.

Three themes emerge from the estimates. Firstly, inequities in job quality correlate with inequities in health. Musculoskeletal disorders, mental well-being, functional disabilities and general health worsen as we move down the job quality ladder. There is even evidence to suggest that joint job quality and especially pay are associated with the occurrence of potentially fatal acute conditions. Except for a few cases, job quality facets correlate with health outcomes in tandem. The health gradient by job quality is quantitatively moderate on the individual level, but may sum up to a substantial number of avoidable conditions at population level. Health differences by level of income and job insecurity further widen work-related health inequities. Modelling job quality by a single facet would have underestimated these differences.

Secondly, a substantial proportion, but not all, of the cross-sectional correlation between job quality and health outcomes is accounted for by unobserved heterogeneity. Especially health inequities by intrinsic job quality may reflect more inter-individual differences than work-related consequences, whereas pay and job insecurity effects change less when allowing for correlated random effects. Though the effect sizes are overall modest, better jobs can help to improve individual mental well-being and reduce musculoskeletal complaints—two key work-related health dimensions. Despite evidence for a healthy worker effect on sample composition, i.e. economically inactive individuals tended to be in worse health, there was no indication that this affected the job quality-health nexus.

Thirdly, job quality related health changes appear to be transitional for the average older worker. There is no clear indication that effects last and accumulate beyond a single period. Workers who were in low quality jobs in consecutive data sweeps were not in worse health than workers who transitioned into a low quality job between sweeps. Thus, differences in job quality do not seem to contribute to a further widening of work-related health inequities among older workers in Europe. Existing mechanisms of social protection, health care and prevention may be sufficient to support full recovery from work-related ailments for the average worker within relative short periods of time. Findings from the selection model confirm that differences in the propensity to retire do not affect the relationship of job quality with health. This dovetails with the transitional nature of work-related health impairments. If workers can recover from ailments and disorders within employment, retirement is a less essential mechanism to recuperate health.

However, some caveats have to be borne in mind when interpreting the findings. The job quality model by Green et al. conceptualises working time quality as an additional facet of job quality, but for which SHARE does not provide information. The ability to flexibly change, for example, the number of working hours or the work schedule may be of special importance for older workers who have to accommodate health impairments or family care responsibilities. Job-quality related effects on health and well-being may thus not be captured fully. It is also conceivable that by focussing on job quality and not specific work stressors, the approach is conceptually too far removed from well-being and introduces confounding influences. However, the measured dimensions of job quality in the analysis share largely the same set of information that would have been used to measure psychosocial work stressors (control, effort and rewards). Furthermore, the panel is relatively short and unevenly spaced since wave 3 includes only limited contemporary information. More data would have allowed estimation of determinants of rarer conditions with greater confidence and would have enriched the study of health dynamics. Though selection into retirement did not emerge as a critical confounder, there are other sources of endogeneity that could still bias the estimated relations. For example, workers may have sufficient discretion to mould their job to their needs after, for example, a health shock. The flexibility to recover in work from health shock would be a key dimension of job quality among older workers and warrants further investigation.

In all, the results support the importance of recent efforts by the OECD, the European Foundation for the Improvement of Working Conditions and the European Commission to bring intrinsic job quality to the forefront of the policy debate [59‐61]. Bad jobs may create negative externalities in the form of greater health care expenses for mental health problems and musculoskeletal disorders, increase sickness absence and induce productivity losses. Better jobs would directly help to limit the incidence of avoidable musculoskeletal disorders and deteriorations of mental well-being, if taken to the population level. But the findings also highlight that job quality-related health impairments are transitional. Existing health inequities along the job quality ladder appear to have largely formed during earlier stages of working life. For the wider policy agenda to extend working lives, the main obstacle may rather be the sorting of people with lower health into jobs of lower quality. This range of jobs appears ill-suited to cater for further health deteriorations and thus may create additional early retirement incentives.