Introduction

Research on the health effects of working hours mostly concentrates on the effects of weekly hours that deviate from the industrial norm (35–40 h per week). Specifically, studies either investigate part-time employment (Nylén et al. 2001; Bartoll et al. 2014) or long working hours (Virtanen et al. 2011; Artazcoz et al. 2013) and their links to different health outcomes. A more recent approach, though, uses the concept of ‘volition’ and compares the discrepancy between actual and preferred working hours (Maynard and Feldman 2011; Otterbach et al. 2016), instead of the number of working hours as such. From this perspective, a person can be underemployed (working less than preferred), overemployed (working more than preferred) or correctly matched (Pagan 2016). Data from the 6th wave of the European Working Conditions Survey show that, according to this definition, 14% of all workers in Europe are underemployed and 30% overemployed (Eurofound 2016).

Under- and overemployment can be considered psychosocial work stressors. Psychosocial work stressors are work characteristics that might cause chronic or repeated stress when there is no available response for the worker to cope with them (Mc Ewen 1998). In particular, under- and overemployment can hinder workers’ sense of control over the number of hours worked (Lyness et al. 2012). Cross-sectional studies have confirmed the link of under- and overemployment with poor mental health and poor mental well-being (Friedland and Price 2003; De Moortel et al. 2017), but longitudinal evidence is still rare (Angrave and Charlwood 2015; Otterbach et al. 2016). Because of the lack of previous longitudinal studies, the first aim of our study is to investigate whether both under- and overemployment are related to changes of mental health in a 2-year observation period.

Yet, any particular job can be conceived as a set of many different work characteristics, and therefore, each worker is exposed differently to stressors. Some workers are able to deal better with stressors and this can counteract the negative effect on mental health and may ultimately activate individuals, resulting in learning or motivation (Bosmans et al. 2015). According to the Effort-Reward Imbalance (ERI) model, effort at work leads only to poor mental health if appropriate gratifications or ‘rewards’, like money, esteem or appraisal and status control, are absent (Siegrist 1996).

If under- and overemployed workers are able to cope with their situation, through the availability of job rewards, we expect the association of under- and overemployment with poor mental health to be weaker. Therefore, our second aim is to investigate the moderating effect of job rewards on the change in mental health of under- and overemployed workers. Job rewards can, just like under- and overemployment, be conceived as employment conditions and relations (i.e. implicit and explicit conventions between the employer and the employee about the working conditions) (Eurofound 2013). We expect that poor conventions about working hours might be counteracted by better agreements about the other employment conditions and relations. Thus, the availability of job rewards (in the form of high earnings, job security, promotion prospects and occupational prestige) can act as buffers for the negative effects of under- and overemployment on mental health.

Methods

Study population

We used the German Socio-Economic Panel (GSOEP), a yearly repeated panel study among households in Germany that started in 1984 (Wagner et al. 2007). GSOEP is a representative longitudinal study of more than 20000 respondents from nearly 11000 households. From wave 19 (2002) onwards, a battery of questions on health was added biannually to the core questionnaire. We used information for the period 2006–2008, because in wave 23 (2006) a complementary battery of questions on job rewards was added to the core questionnaire. Then, the change in mental health was assessed during a 2-year follow-up period. Therefore, a baseline measurement, i.e. mental health in wave 23 (2006), is linked to a follow-up score 2 years later, i.e. in wave 25 (2008). Job change in waves 24 (2007) and 25 (2008) was used as an exclusion criterion, because job change could be used to solve under- or overemployment and may thus be associated with better mental health in the long term. The samples were restricted to 20–60 year olds. We also excluded those in (partial) retirement, in-service training, military service, voluntary service, and the self-employed—as their employment relations are different from those in standard salaried employment. Respondents working full-time, part-time and those who were marginally employed (mini-jobs) were included. This resulted in a sample with 4432 men and 4338 women. All variables but the dependent variable, are derived from wave 23.

Measurements

Mental health

Mental health was assessed using the Mental Component Summary (MCS), a subscale from the Short Form 12 Health Survey, Version 2. The MCS has four subscales: vitality, role limitation due to emotional problems, social functioning and general mental health (Andersen et al. 2007). The sum-score ranges from 0 to 100 (100 representing the highest level of health) with a mean of 50 and a standard deviation of 10, using the German population in 2004 as a reference (Andersen et al. 2007).

Under- and overemployment

Employees reported how many hours they normally work in a week (overtime included) and how many hours they would choose to work in a week, bearing in mind that earnings would increase or decrease depending on the chosen number of working hours. We created a three-category variable: (1) ‘matched’: desired = actual hours; (2) ‘underemployed’: desired > actual hours and (3) ‘overemployed’: desired < actual hours (Pagan 2016).

Previous studies have shown that more than 60% of the German work force are overemployed (Wunder and Heineck 2013; Pagan 2016). Therefore, one can question whether there are a lot of false-positive cases in the overemployment-category. As a consequence, sensitivity analyses were conducted with two alternative definitions of overemployment. First, we distinguished those overemployed (i.e. desired < actual hours) with and without overtime work (defined as working more than the number of hours included in the work contract). It can be argued that workers with overtime work will benefit more from job rewards, as these job rewards counteract the imbalanced reciprocity between the employee and the employer caused by the disregard of the work hours described in the work contract. This variable distinguished between four categories: (1) ‘matched’; (2) ‘underemployed’; (3) ‘overemployed: overtime’; and (4) ‘overemployed: no overtime’. Second, only those workers with more than or equal to 4 h deviation between desired and actual working hours are defined as being under- or overemployed (Bell et al. 2011).

Job rewards

A short version of the ERI questionnaire was included in the GSOEP wave 23 (Siegrist 1996). The internal consistency of the scales was validated in a previous study using GSOEP-data (Siegrist et al. 2009). The short version included seven items for rewards, including salary, esteem, job security and career opportunities: (1) I receive the respect I deserve from my superior or a respective relevant person; (2) My job promotion prospects are poor; (3) I have experienced or I expect to experience an undesirable change in my work situation; (4) My job security is poor; (5) Considering my efforts and achievements, I receive the respect and prestige I deserve at work; (6) Considering all my efforts and achievements, my job promotion prospects are adequate; and (7) Considering all my efforts and achievements, my salary/income is adequate. This was asked in a two-stage procedure. First, for each item, respondents indicated whether they were confronted with it (yes or no). If so, they also indicated to what extend this bothered them (not at all, somewhat, heavily and very heavily). Seven dummies were created (one for each item): 0 = those responding no or yes, but not at all bothered and 1 = those confronted with the item and somewhat, heavily and very heavily bothered by the respective item. We created a sum scale (range 0 to 7) of all items and divided the sample into three equal parts using tertiles (low, medium and high job rewards).

Additional variables

We included five sociodemographic variables (partnership status, number of young children, income, education and age groups), the total number of weekly working hours and physical health in our models as control variables. Younger and older age, lower income, lower education, lower overall health status and not living with a partner are risk factors for mental health problems (Silva et al. 2016). Parenthood has also been associated with behaviors that are not beneficial to health (Umberson and Montez 2010). The number of working hours is a predictor of being over- or underemployed (Reynolds and Aletraris 2006). The presence of a steady partner was measured using a dummy variable with value 1 if respondents live with a partner (irrespective of the marital status) and 0 if not. The number of children in the household aged 14 or younger was grouped into ‘none’, ‘one’ and ‘more than one’. A variable for household income was constructed using tertiles (low, medium, high). Household income was based on the monthly household income, that was adjusted for household size in accordance with the OECD equivalence-scale (Hagenaars et al. 1994). The variable for educational level distinguished ‘no vocational training’, ‘vocational training’ and ‘higher education’. Age was grouped into four categories ‘job starters’ (20–29), ‘early midlife’ (30–39), ‘late midlife’ (40–49) and ‘older working life’ (50–60). The number of working hours was measured using the actual number of working hours per week (overtime included). Physical health was assessed using the question: ‘Does your health limit you in doing demanding everyday activities, such as heavy lifting’. Answer categories were: ‘greatly’, ‘somewhat’ and ‘not at all’.

Data analysis

Because of the average lower working hours of women, all analyses were done for men and women separately. First, descriptive analyses were performed. We presented percentages, means and standard deviations of all included variables. Then, we showed the mean MCS scores of 2006 and 2008 across the main independent variables. The significance of the differences between the means across years was tested using a paired t test. Differences between means across under-/overemployment and level of job rewards were compared using one-way ANOVA. Throughout the descriptive analyses, data were weighted to correct for chances of unequal selection probability.

Afterwards, conditional change models (Aickin 2009) were estimated using Ordinary Least Squares (OLS) regressions with the change in MCS score after 2 years as dependent variable. These models allowed to examine whether change over time in mental health was related to under- and overemployment in 2006. The first model included the baseline MCS score of 2006. In Model 2 all confounders were added. This model was extended by the categorical variable on under- and overemployment in 2006 (model 3). Afterwards, model 3 was extended by job rewards (model 4). To formally test if reward modified the effect of under- and overemployment on mental health change, we included interaction terms between under-/overemployment and each level of reward in model 5. By comparing models without and with interactions on the basis of a likelihood-ratio (LR) test, we tested for significant interactions. At all steps, parameter effects of the covariates in relation with change in mental health were presented as unstandardized regression coefficients (B), with their related confidence intervals (CI). We applied complete case analyses, reducing the final sample to 3266 men and 3139 women. All calculations were done using Stata v14.2.

Results

The average MCS scores were slightly higher for men than for women, but comparable over the two time points. Overemployed workers more often were men, while underemployed workers more often were women. Most respondents belonged to the age group of 40–49 years old, had a partner and had vocational training. More women, than men reported to have no children (See Table 1).

Table 1 Description (in %) of the population studied (Population in salaried employment, 20–60 years old, 3108 men and 2999 women, weighted, GSOEP wave 23–25)
Full size table

In Table 2, the mean MCS scores of 2006 and 2008 are shown. Under- and overemployed workers differed with respect to MCS scores, with significantly worse mean scores for under- and overemployed workers compared with correctly matched workers. Underemployed workers reported significantly higher mean scores than overemployed workers. No significant changes in mean scores between 2006 and 2008 were found, except for overemployed women. For overemployed women, the mean scores increased significantly, indicating better mental health in 2008 than in 2006. Workers with low job rewards had significantly lower mean scores in mental health compared with workers with higher job rewards. Significant changes in mean scores were found for male workers with high job rewards and for female workers with low rewards. For female workers with low job rewards, the mean scores increased significantly, indicating better mental health in 2008, compared with 2006. For male workers with high job rewards the mean scores were lower in 2008 than in 2006. In all categories, the mean MCS scores for women were lower, compared to those of men.

Table 2 Mean MCS scores in 2006 and 2008 by under- and overemployment and by level of job rewards (MCS scores range from 0 to 100, higher scores reflect better mental health, population in salaried employment, 20–60 years old, 3108 men and 2999 women, weighted, GSOEP wave 23–25)
Full size table

Table 3 presents results of our regression models for men and women. Models 1 show that, for men and women, an increase of 1 point in baseline MCS score resulted in a significant decrease in mental health change (B = − 0.52; CI: − 0.55—− 0.49 and B = − 0.53; CI: − 0.56—− 0.50, respectively). When additionally controlling for all confounders (Models 2), the explained variance of the models rose with 1.6 and 1.7 percentage points for men and women, respectively. Models 3 show that overemployment (for men and women) was significantly related to a negative change in mental health, compared with workers who’s actual and preferred hours match. Underemployment was not related to a negative change in mental health. Adding under- and overemployment to the models does not increase the explained variance of the models. When job rewards are added in models 4, for men and women, the significant negative association between overemployment and mental health change remained. From models 4 it is clear that, for both men and women, medium and high rewards were significantly related to a positive change in mental health, compared to low rewards. Adding job rewards very slightly increased the explained variance, compared to models 3. The interaction models (Models 5) show that there were no significant interactions between job rewards and under- and overemployment. For men, due to the added interaction, the significant negative association between overemployment and mental health disappeared. For women, due to the added interaction, the significant negative association between job rewards and mental health disappeared.

Table 3 Conditional change models for change in MCS score (t2–t1) (population in salaried employment, 20–60 years old, 3159 men and 3038 women, GSOEP wave 23–25)
Full size table

Sensitivity analysis

The first alternative definition of overemployment distinguished those overemployed who worked more than agreed in their contracts (doing overtime), and those reporting to be overemployed, but did not work more hours than agreed upon in their contract (no overtime). Most overemployed workers worked more hours than agreed upon in their contract (see Appendix Table S1). These workers had significantly lower mean MCS scores, compared to overemployed workers without overtime (except for women in 2008). In Models 3 (Appendix Table S2), we see that, overemployment when doing overtime (for men and women) and overemployment without doing overtime (for men) led to a decrease in mental health. Secondly, we defined under- and overemployment as a deviation of four hours or more between the actual and the preferred working hours. Using this stricter definition, the group of under- and overemployed workers decreased considerably: nearly half of the workers now belonged to the correctly matched group (see Appendix Table S1). Moreover, no statistically significant effects of under- and overemployment on mental health were found (See Appendix Table S3). Similar to the previous interaction models, the interaction models using the alternative definitions showed no significant effects.

Discussion

This study has produced three main findings, for both men and women: (1) Being overemployed was related to a reduction in mental health after 2 years. (2) Underemployment was not related to a reduction in mental health after 2 years. (3) Higher job rewards did not protect the mental health of under- and overemployed workers.

In line with previous, mostly cross-sectional research (Friedland and Price 2003; De Moortel et al. 2017), we found that overemployment is related with a reduction in mental health after 2 years. Underemployment was not related to a reduction in mental health after 2 years. However, when fitting models 3 without all confounders, there was a significant negative relation between underemployment and change in mental health (results not shown). Using stepwise inclusion of the control variables, showed that when controlling the model for household income, the effect of underemployment disappears (results not shown). This is in line with previous research indicating that underemployed workers are overexposed to a lack of (financial) stability and low-skilled routine jobs (Stier and Lewin-Epstein 2003).

The second aim of this study was to investigate whether the mental health of under- and overemployed workers is protected by the availability of job rewards. We hypothesized that poor conventions about working hours might be counteracted with better agreements about the other employment conditions and relations (in the form of high job rewards). The overall interaction effects were not significant; thus, we could not confirm our second hypothesis. For women, the positive relation between job rewards and change in mental health disappeared when adding the interaction terms. This suggests that correct hours are more important for women, than job rewards. The lower importance of job rewards for female workers’ mental health can be related to their lower average working hours. It can be assumed that the lower exposure time of these women leads to less strong effects of rewards on their mental health. However, difficulties in combining household and paid work responsibilities might also offer part of the explanation (Artazcoz et al. 2001). For men, in contrast, the negative relation between overemployment and change in mental health disappeared when adding the interaction terms. This could indicate a higher importance of job rewards for men, compared to correctly matched working hours.

The sensitivity analysis did not lead to different results concerning the second aim of our study. Moreover, irrespective of doing overwork, a negative relation between overemployment and change in mental health was found. In contrast, overemployment defined as more than or equal to 4 h deviation between actual and preferred working hours was not related to mental health changes. This might indicate that small deviations between actual and preferred hours do matter. However, the group of under- and overemployed workers decreased considerably using this alternative definition, therefore, the non-significant relations could also be due to a lack of statistical power.

This study has several limitations. The reference group (correctly matched workers) are a rather small group, which might explain the low statistical power of our models. To discover strong health effects of under- and overemployment, the delay between the two measurement points might also be too long. A recent study showed that the mental health penalty of under- and overemployment on mental well-being became manifest after a relatively short time (Angrave and Charlwood 2015). That study also showed the subjective well-being of the under- and overemployed quickly returning to pre-mismatch levels when mismatch came to an end (Angrave and Charlwood 2015). Only overemployment during more than 2 years had a long-lasting negative effect (Angrave and Charlwood 2015). Unfortunately, as mental health status was only asked every 2 years we were unable to reduce the time lag. Although, information on job rewards was also available in GSOEP wave 28 (2011), we did not reproduce our results using this wave and wave 29 (2012), because wave 28 did not contain information on (baseline) mental health. Another limitation of this study is that our sample only included workers that stayed in the same job for 2 years. Yet, this does not mean that the level or presence of under- and overemployment remained stable during this period. The exclusion of respondents who changed jobs during the observational period might have (disproportionally) selected workers less likely to change job (such as those with no work hours mismatch). This could lead to an underestimation of the mental health effects of under- and overemployment. However, comparing the models with and without the exclusion of those who changed jobs did not lead to different conclusions.

Despite its limitations, this study has some clear strengths as well. The use of conditional change models is a first strength. Under- and overemployed workers have lower mental health at baseline, compared to workers whose actual and preferred hours match. The conditional change model is seen as an attempt to remove baseline differences between different groups of workers (Aickin 2009). The use of the GSOEP is also a clear strength. The GSOEP provides a large longitudinal dataset representative of the German population.

In sum, this study is one of the first to shed light on the underlying mechanisms explaining the relation between under-/overemployment and mental health. The findings demonstrate that overemployment was related to a negative mental health change after 2 years, and that this relationship held true both for people receiving high and low reward at work. However, more research is needed to confirm our results in other countries.