Employment status, working conditions and depressive symptoms among German employees born in 1959 and 1965

To analyse the association between current employment status and depressive symptoms, we used data from the first wave of the lidA study (‘German cohort study on work, age and health’). Within lidA, a sample of two birth cohorts of German employees subject to social security contributions [including minor employment (Eichhorst and Marx 2011)] as of 31 December 2009 was selected from Integrated Employment Biography (IEB) register data of the German Federal Employment Agency (Dorner et al. 2010).¹ In 2011, 6,585 personal interviews were conducted (i.e. the response rate was 27 %) (Hasselhorn et al. 2014). Selectivity analyses showed only minor differences between the lidA participants and the IEB sample, with a slightly lower probability of participation related to: (1) birth year (1965), (2) region (i.e. living in smaller cities or rural environments), (3) whether an individual was non-German and (4) whether a person had lower levels of qualification (Hasselhorn et al. 2014). At the time of the interview, 6,339 respondents were employed and 246 respondents were not. Of the employees, 4,921 respondents (78 %) gave written consent to link administrative records of their previous employment history to the survey data. Of those, 4,081 employed participants conducted a paper-and-pencil questionnaire on depressive symptoms [applied version of Beck Depression Inventory—BDI-V (Schmitt et al. 2003)] and answered questions on their current working conditions. Ultimately, 4,081 respondents are included in this paper’s analysis. Selected characteristics of our sample population can be found in Table 1.

The younger cohort, born in 1965, was oversampled in the first wave of lidA and represented approximately 57 % of the population studied. The reason for the oversampling was the expectation that in the following waves we would lose a considerable number of respondents of both cohorts for panel attrition reasons. In wave three (planned), the younger cohort is about the same age as the older cohort in wave one. To have approximately the same number of respondents of each cohort at each age, we included a higher amount of individuals of the younger cohort in the first wave. Due to gender-specific response rates, women made up 53 % of the study population. Both cohorts belonged to the German so-called baby boom generation and could be considered experienced by the time of this study (Tisch and Tophoven 2012). They showed similar employment trajectories and average levels of depressive symptoms. Intra-generational differences between these two cohorts were negligible.

A total of 22 % of the respondents held a university degree or an equivalent, approximately three-quarters of the respondents reported a medium level of secondary education, and fewer than 5 % of the respondents had never received a professional education. Almost nine in ten men and women were either married or cohabitating. With respect to current employment status, men predominantly held full-time employment (95 %), whereas women were more likely to work part-time (47 %) or in marginal employment (9 %) (cf. Table 1).

Depressive symptoms were the dependent variable in our analyses. Depressive symptoms were assessed using the applied version of BDI-V, which covered all 20 items of the original clinical questionnaire except for weight loss (Schmitt et al. 2003). During the personal interviews, to avoid both possible interviewer effects and social desirability effects, the respondents were asked to fill in a paper-and-pencil questionnaire containing the variables of the BDI-V. Answers were summed to a score with values from 0 to 100. The inter-item correlations ranged from 0.21 to 0.68; Cronbach’s alpha was 0.92.

Influence at work (i.e. decision authority) was measured by means of the following questions of the second version of the Copenhagen Psychosocial Questionnaire (COPSOQ II) (Pejtersen et al. 2010): ‘Do you have a say in choosing who you work with?’; ‘Can you influence the amount of work assigned to you?’; and ‘Do you have any influence on what you do at work?’. The response options (and values for the scale) were ‘Always’ (5); ‘Often’ (4); ‘Sometimes’ (3); ‘Seldom’ (2); and ‘Never/hardly ever’ (1). Mean values were calculated to represent the scale. Cronbach’s alpha was 0.67, and inter-item correlations ranged between 0.37 and 0.42.

Quality of leadership was measured by means of the following COPSOQ II questions (Pejtersen et al. 2010): ‘To what extent would you say that your immediate superior …’: (1) ‘… makes sure that the individual member of staff has good development opportunities?’; (2) ‘… gives high priority to job satisfaction?’; (3) ‘… is good at work planning?’; and (4) ‘… is good at solving conflicts?’. The response options (and values for the scale) were: ‘To a very large extent’ (5); ‘To a large extent’ (4); ‘Somewhat’ (3); ‘To a small extent’ (2); and ‘To a very small extent’ (1). Mean values were calculated to represent the scale. Cronbach’s alpha was 0.87; inter-item correlations ranged between 0.58 and 0.70.

Job insecurity was measured by a single item from the Effort Reward Questionnaire (Siegrist et al. 2009). Individuals were asked whether they consider their job at risk with the response options ‘Yes’ (1) and ‘No’ (0).

Employment history was generated based on administrative data from the German Federal Employment Agency that includes information on all individual periods of employment subject to social security contributions and unemployment. Based on previous employment history, we identified individuals’ main employment status for the previous 10 years (2001–2010). Accordingly, we distinguished between three categories: ‘Mostly full-time employment’ (majority of days in regular full-time employment), ‘Mostly part-time employment’ (majority of days in regular part-time employment) and ‘Mostly other’ (majority of days in minor employment, without employment or in employment not subject to social security contributions that is not included in the administrative data, e.g. self-employed or employed as a civil servant). Among men, ‘mostly other’ employment histories were dominated by days without social security contributions; among women, minor employment was the most dominant status (cf. Table 2). Men and women in current full-time employment were also mostly employed full-time in the past (88 %). Similarly, most marginally employed individuals had not been in regular full-time or part-time employment for the last 10 years. More than half of the currently part-time employed men and women also showed long periods of past part-time employment (cf. Table 3).

As further control variables, we first included level of education, which was classified according to the CASMIN scheme using the two questions ‘What is the highest school qualification you have obtained?’ and ‘What is the highest vocational qualification you have obtained?’ (König et al. 1988). We distinguished between the three categories of ‘low’, ‘middle’ and ‘high’. Other control variables were gender, cohort affiliation and partnership status, which were based on information that the respondents gave during the interview.

The analyses were conducted with the statistics software Stata 12.0. Using survey set commands, we considered the sample point structure of the data set. To test our assumptions, we used a multiple linear regression analysis on depressive symptoms (BDI-V). First, we estimated bivariate regression models to see the uncontrolled contribution of each independent variable. In multivariate analyses, we first considered the subjects’ current employment contracts as an independent variable. In four additional regression analyses, we additionally controlled for current working conditions to see whether these mediated the relationship between current employment and depressive symptoms. First, we adjusted for job insecurity. Second, we considered leadership quality. Third, we considered influence in the workplace. Fourth, we considered the influence of the subjects’ previous employment histories in order to see whether they mediated the relationship between current employment and depressive symptoms. Finally, we included all tested variables and concepts in one analysis. At every step of the multivariate analyses, we further controlled for birth cohort and occupational education (CASMIN).

The assumption of normality was examined by the inspection of residual plots. A deviation was not detected. We also inspected possible multicollinearity for the intended regressions. Although we found that 88 % of men and women in regular full-time employment in 2011 had been full-time employed throughout the previous 10 years (2001–2010), the assumption of multicollinearity could be rejected due to low variance inflation factors (all values for VIF were below 1.7).

To identify gender differences in the influence of job insecurity, quality of leadership, influence at work/decision authority and the main type of current employment status during the previous 10 years on depressive symptoms and on the association between current employment status and depressive symptoms, we calculated bivariate and multiple linear regression models separately for women and men.

The strengths of the present study are that it (a) relied on a representative sample of two cohorts (1959 and 1965) of employees in Germany, (b) had retrospective information on employment biographies, (c) used validated measures of three psychosocial factors [quality of leadership (Pejtersen et al. 2010), influence at work (Pejtersen et al. 2010) and job insecurity (Siegrist et al. 2012)] and (d) controlled for relevant potential confounders, namely vocational education and partnership status. Regarding point (a), the study analysed a heterogeneous population of employees, where the effect of age was fixed by choosing two birth cohorts (1959 and 1965) of the German baby boom generation. Regarding point (b), employment register information was used to study the relationship between employment history and health. Using register information is beneficial to avoid recall or reporting problems.

However, some limitations of the study remain to be addressed. (a) The study is—apart from the retrospective information on employment histories—cross-sectional in terms of health and working conditions. The study does not have information on previous depressive symptoms; therefore, we were not able to determine to what extent changes in employment status resulted from depressive symptoms or to what extent changes in depressive symptoms resulted from employment status (Fig. 1b). The employment biography data do not contain information on employment contracts (fixed term vs. open-ended) or periods of sickness and are limited to employment subject to social security contributions.

The present study confirms earlier German findings (Thielen and Kroll 2013) showing that employees in minor employment (i.e. with comparatively low income and reduced social security contributions) are more likely to show depressive symptoms than people in regular employment. The study also confirmed that employment history is more strongly related to depressive symptoms among men than among women. The study confirms the Canadian findings that some of the effects of employment on health can be mediated by specific working conditions, such as the experience of job insecurity (Scott-Marshall and Tompa 2011). The present study showed, however, that sometimes mediation can suppress the effect of employment, such as in the example of quality of leadership and marginal employment among women. Furthermore, the present study has confirmed a Danish study, finding that—among a range of psychosocial factors—job insecurity, low influence and poor quality of leadership were associated with the subsequent development of depressive symptoms (Rugulies et al. 2008).

To determine why depressive symptoms are associated with male part-time work and female marginal employment, two research activities should be encouraged. First, analyses should investigate the relative importance of selection of people into and out of various types of employment depending on their level of depressive symptoms and of the possible impact of employment status on the development of depressive symptoms (causation). Second, longitudinal analyses considering life courses should be carried out, to study effects at the beginning of the employment career (including the period of vocational education) and through adulthood as well as in the years before and after retirement age.

Our results suggest that among German women, current marginal employment is positively related to the prevalence of depressive symptoms; employment histories with mostly non-standard employment, however, are negatively related to depressive symptoms. This relationship should be investigated further, and a better description of those women is needed with regard to family situation. It might be anticipated that factors concerning private life histories are more relevant for mental well-being among women than employment trajectory.

Finally, further aspects of part-time employment and marginal employment—and subjective indicators hereof—should be examined, which might explain elevated depressive symptoms. However, this effect is not unidirectional: reduced working time can result in negative effects in some populations and positive effects in other populations (Barnett and Gareis 2000; Gareis and Barnett 2002; Herold and Waldron 1985).