The aim of our study was to analyse cross-sectional and longitudinal associations of ERI and WAI. After adjustment for age, gender, educational level, health-related behaviour and other factors of the work environment (physical demands, job control and psychological job demands), ERI had an effect on WAI in the cross-sectional and longitudinal analyses. These results are in line with recent cross-sectional [14
] and longitudinal findings [16
] and confirm that reciprocity and fairness at work, as operationalised by Siegrist’s ERI model, have a relevant impact on work ability independent of and above that of other known explanatory variables. However, it is notable that our findings also demonstrate that low work ability negatively affects the balance of effort and reward. This indicates that ERI might be a risk factor that mediates the transition from poor work ability to exit from paid employment and health-related early retirement [18
]. Furthermore, there seems to be a downward spiral of ERI affecting work ability that again might intensify the perception of an imbalanced relation of effort and reward.
Besides the adverse effect of ERI on WAI, our analyses showed that higher age, poorer health-related behaviour, higher psychological job demands and lower job control also predicted new cases of poor or moderate WAI. This is line with the review by van den Berg et al. [22
], which presented consistent evidence supporting the association of poor work ability primarily with older age, lack of leisure-time activity and obesity but also with high mental work demands and lack of autonomy. In contrast to the aforementioned review [22
], we could not establish a prospective association between physical demands and work ability, even though the cross-sectional association was relatively strong.
Despite a degree of concordance with the review of van den Berg [22
], our study suffers from several limitations.
Firstly, the response rate of the first survey was rather low. Other authors have described such a response rate as reasonable for an anonymous survey in the working population [38
]. However, we cannot rule out the possibility of bias from selective participation. Although our baseline responders and non-responders did not differ in terms of gender and age, we were unable to investigate the characteristics of the baseline non-responders in depth.
Secondly, our analyses of follow-up responders and non-responders indicated selective follow-up participation suggestive of a healthy worker effect.
Thirdly, as our results were based on an older, full-time employed white-collar sample, this certainly constrains the generalisation of our findings.
Fourthly, the gap between baseline and follow-up measurement was only one year, and baseline measurements were restricted to single-point measurements of the explanatory variables. As the time of exposure is relevant to the establishment of causal relations, a longer follow-up and measures of continuous exposure could provide a better understanding of the associations between ERI and work ability.
Fifthly, as both ERI and work ability were measured by self-report questionnaires, a tendency to respond negatively could have inflated the cross-sectional association. Some authors therefore propose to adjust regression models for negative affectivity when analysing associations between self-reported work stressors and measures of health [39
], whereas others strongly advise against adjusting for negative affectivity [42
]. We did not adjust for possible response bias due to negative affectivity. Indeed, a recent simulation study suggested that negative affectivity can affect associations of ERI and health-related outcomes even if mean scores are only slightly changed. However, substantial effects on the association are only plausible if a large proportion of participants and their questionnaire answers are influenced by negative affectivity [43
]. Moreover, we assume that negative affectivity did not affect our longitudinal analyses as these analyses were restricted to persons with good work ability and persons without effort-reward imbalance, respectively.
Sixthly, psychological job demands were strongly correlated with the binary ERI variable. This might result in overadjustment, with ERI as the dependent variable, and multicollinearity, with WAI as the dependent and ERI as the independent variable. We dropped psychological job demands from our final models and repeated the parameter estimations. Standard errors were slightly reduced. Estimates of the effect of ERI on work ability increased roughly 2-fold in the cross-sectional and longitudinal analyses. The estimate of the longitudinal effect of poor or moderate work ability on ERI was only slightly affected in the longitudinal analysis.
Seventhly, our study presents findings within the context of the German system of social security, employees’ rights and employers’ duties. However, as was recently shown for the associations between job insecurity and health-related outcomes, results may differ between welfare regimes [3
Nonetheless, these limitations are balanced by several strengths. Firstly, participants were recruited by random sampling. Secondly, we could refer to a relatively large sample for our analyses. Thirdly, our analyses were performed using a longitudinal design. Fourthly, the analyses were restricted to cases with good or excellent baseline work ability and cases without baseline ERI, respectively, in order to predict new cases of adverse events.
Our results concerning the effect of ERI on work ability indicate that an adequate effort-reward balance at work is a crucial dimension of healthy work. In this context, the ERI model offers options to promote work ability at the individual, interpersonal and organisational level [16
]. While individual-level interventions focus on coping with the existing stressors (e.g. reducing overcommitment in order to rebalance efforts and rewards), other interventions can be designed to modify stressors at the interpersonal or organisational level. For instance, Bourbonnais et al. [45
] described a participatory intervention approach in an acute care hospital in Canada. Following the concepts of German health circles, a multi-professional team of staff members and researchers identified 56 intervention targets and developed proposals for solutions. A controlled trial demonstrated that ERI decreased after one and three years in the intervention group compared to the controls and showed that most of the recommended solutions could be permanently implemented [46
Although there is a strong body of evidence concerning the impact of work ability on productivity loss at work, retirement intentions, long-term sickness absence, unemployment and early retirement [18
], there is less research regarding its more proximal consequences for work environment, work demands and quality of work. Cross-sectional studies that investigated associations between work characteristics and work ability were mostly interpreted unidirectionally in terms of work characteristics affecting work ability. Our longitudinal analyses show that causal relations also act conversely. This indicates that occupational health services for persons with poor work ability must not be restricted to the workers’ physical requirements. Employees also need support that addresses their psychosocial demands, especially esteem and security, in order to prevent effort-reward imbalance.