The association of vertical and horizontal workplace social capital with employees’ job satisfaction, exhaustion and sleep disturbances: a prospective study

We analyzed the prospective association between vertical and horizontal WSC at baseline and job satisfaction, exhaustion and sleep disturbances at 24 months of follow-up. We used data from the Pioneer intervention study (Framke et al. 2016, 2018), a cluster-randomized participatory organization-level workplace intervention that aimed to improve well-being and reduce sickness absence by focusing on core job tasks. In September 2011 and September 2013, all employees within 78 municipal pre-schools received a questionnaire on WSC and well-being. Of the 78 workplaces, 7 were lost to follow-up. Therefore, this study is based on data from employees within 71 workplaces. Employees belonged to the following job groups: pedagogical leaders, nursery nurses, nursery nurse assistants, and other job groups. Other job groups were primarily kitchen and cleaning staff and school caretakers.

Figure 1 shows the flowchart toward the final study population. In total, 1560 employees received the baseline questionnaire, of which 1245 responded (79.8%). Of the 1245 baseline responders, 664 also responded to the follow-up questionnaire 24 months later. We excluded 58 pedagogical leaders, because their WSC may differ from the WSC of the employees, yielding a sample of 606 employees from 71 workplaces. Finally, we excluded employees with missing individual-level data on job satisfaction, exhaustion and sleep disturbances, yielding a final study sample of n = 581, n = 588 and n = 580 for the analyses on job satisfaction, exhaustion and sleep disturbances, respectively.

According to Danish law, research studies that use solely questionnaire and register data do not need approval from the National Committee on Health Research Ethics (Den Nationale Videnskabetiske Komité).

We measured WSC with self-administered questionnaires in September 2011. Of the nine WSC items, five were derived from the Danish Work Environment Cohort Study (DWECS) from the year 2010 (Det Nationale Forskningscenter for Arbejdsmiljø 2011), one was derived from DWECS from the year 2005 (Feveile et al. 2007), one was derived and slightly modified from a questionnaire on relational coordination (Hoffer et al. 2008), and two items were developed for the purpose of this study. A factor analysis (rotation method: varimax) showed two distinct factors with eigenvalues of 4.14 and 1.78, respectively. We named these factors “vertical WSC” (i.e., social capital in the relationship between employees and their superiors, five items) and “horizontal WSC” (i.e., social capital in the relationship between employees, four items) in accordance with theoretical considerations about different types of social capital in the literature (Oksanen et al. 2010). All rotated factor loadings were > 0.70 for vertical WSC and > 0.65 for horizontal WSC. Cronbach’s alpha was 0.87 and 0.80 for vertical and horizontal WSC, respectively.

A sample item for vertical WSC was “We have confidence in the management”, a sample item for horizontal WSC was “We help each other in achieving the best possible results”. Online Appendix 1 shows the full list of items and their response categories. Respondents were included if they responded to at least three of the five items of the vertical component of WSC and to at least two of the four items of the horizontal component of WSC. Response categories ranged from ‘To a very small extent’ (1) to ‘To a very large extent’ (5).

We calculated mean scores of vertical and horizontal WSC for each of the 71 workplaces. Then, we assigned the workplace-specific mean scores to all individuals working at the same workplace. Intra-class correlations were 0.35 and 0.14 for vertical and horizontal WSC, respectively, indicating that a considerable proportion of the variation in vertical and horizontal WSC was explained by the workplace and that the aggregation of the individual-level scores to the workplace level was justified.

We assessed employees’ well-being with three measures: job satisfaction, exhaustion and sleep disturbances. All three measures have been frequently used in studies on working conditions and employees’ well-being (Bond and Bunce 2001; Faragher et al. 2005; Lindeberg et al. 2011; Bültmann et al. 2013; Linton et al. 2015). We measured the three outcomes with self-administered questionnaires at baseline and at 24 months of follow-up. Job satisfaction, measured with the item ‘Regarding your work in general. How satisfied are you with your job as a whole, everything taken into consideration?’, was rated on a four-point scale (very satisfied, satisfied, dissatisfied, very dissatisfied) (Pejtersen et al. 2010). We measured exhaustion (‘Within the past 2 weeks, how much of the time have you felt lacking in energy and strength?’) and sleep disturbances (‘Within the past 2 weeks, how much of the time have you had trouble sleeping at night?’) with one item each, derived from the Major Depression Inventory (Bech et al. 2001). There were six response categories (all of the time, most of the time, slightly more than half of the time, slightly less than half of the time, some of the time, at no time). Higher scores indicate a higher level of job satisfaction, a higher level of exhaustion and a higher level of sleep disturbances.

Age, sex, job group, workplace size, workplace type, baseline scores of the three outcomes, intervention status and the interaction term intervention status × predictor variable were included as covariates. Age (continuous), sex, job group (nursery nurses, nursery nurse assistants, other job groups), workplace size (number of employees per workplace), workplace type (integrated, day care, kindergarten) and intervention status (intervention vs. control group) were register based.

All analyses were conducted using SAS 9.4.

Using the Genmod procedure we estimated the association of workplace aggregated measures of vertical and horizontal WSC at baseline with level of job satisfaction, exhaustion and sleep disturbances at 2 years of follow-up, while taking into account that employees were nested within workplaces. In supplementary analyses, we used individual-level measures of vertical and horizontal WSC instead of workplace aggregated measures of WSC.

We calculated estimates adjusted for sex and age, and further adjusted for job group, workplace type, workplace size and baseline scores of the outcomes. Because the data set was derived from an intervention study, we further adjusted for intervention status and for the interaction term intervention status × predictor variable.

Table 1 shows that the mean age of the study population was 43 years and that 87% were women. Of the total study population, 58% worked as nursery nurses, 32% worked as nursery nurse assistants and 10% worked in other job groups (primarily kitchen staff and school caretakers). Mean workplace size was about 23 employees per workplace. With regard to workplace type, the majority worked in integrated institutions (78%), with the remaining employees working in day care institutions (19%) and kindergartens (4%).

Table 1 further shows that younger employees and men were more likely to be lost to follow-up. Further, individuals working at workplaces with a high number of employees were more likely to be lost to follow-up. Job group and workplace type were not related to attrition. A higher level of vertical and horizontal WSC was associated with a lower risk of loss to follow-up. Regarding baseline values of the three outcome variables, individuals with a high level of exhaustion were more likely to be lost to follow-up than individuals with a low level of exhaustion. Level of job satisfaction and level of sleep disturbances were not related to risk of loss to follow-up.

Table 2 shows the associations between the scores of the predictor variables at baseline and the three outcomes after 2 years of follow-up. Higher levels of vertical and horizontal WSC were associated with a higher level of job satisfaction and a lower level of exhaustion at follow-up in all models. In the most adjusted model, estimates were 0.20, p = 0.01 (vertical WSC) and 0.24, p = 0.01 (horizontal WSC) for job satisfaction and − 0.33, p = 0.04 (vertical WSC) and − 0.43, p = 0.04 (horizontal WSC) for exhaustion at follow-up. Higher levels of vertical and horizontal WSC were also associated with higher levels of sleep disturbances in some of the models but not in the most adjusted model.

In the supplementary analyses (see Online Appendix 2), we repeated the analyses shown in Table 2 while using individual-level measures of vertical and horizontal WSC instead of workplace aggregated scores. Results from the supplementary analyses were similar to the results from the main analyses. As in the main analyses, higher levels of vertical and horizontal WSC were associated with a higher level of job satisfaction and a lower level of exhaustion at follow-up in all models. In contrast to the main analyses, a higher level of horizontal WSC was also associated with a lower level of sleep disturbances in the most adjusted model in the supplementary analysis. Vertical WSC showed a suggestive association (p = 0.08) with lower levels of sleep disturbances in the most adjusted model in the supplementary analysis.

The strengths of this study are the prospective design, the use of workplace aggregated measures of WSC and the availability of in total nine WSC items that allowed us to examine two types of WSC, namely vertical and horizontal WSC. An additional strength is that we have information about the workplace (i.e., the unit of aggregation) for all employees both at baseline and at follow-up. The psychometric properties of the two WSC scales were good, with factor analysis clearly showing two distinct factors where all rotated factor loadings were > 0.70 for vertical and > 0.65 for horizontal WSC. Further, both scales had a high internal reliability (Cronbach’s alpha of 0.87 and 0.80 for vertical and horizontal scales, respectively).

There are also weaknesses of the study. We used self-reported data on both exposure and outcome measures, leading to a notable proportion of employees for whom we have only baseline but not follow-up data. Compared to the study population used in the analyses, the population lost to follow-up were statistically significantly younger, more likely to be men, had lower levels of vertical and horizontal WSC and higher exhaustion scores. The size of the dropout proportion may partly be due to the rather long follow-up period of 24 months, which is another weakness of our study, since 2 years can be considered a rather long time lag when analyzing associations between psychosocial exposures and outcomes related to well-being. Ideally, we should have had at least one more measurement point between baseline and follow-up. That would have allowed us to monitor outcomes more closely and it would then have been possible to examine change in WSC (between first and second time point) and the effect of this change on the subsequent outcome (third time point). The lack of repeated measures of WSC will likely have resulted into exposure misclassification and an underestimation of the association between WSC and well-being. Finally, it is a weakness that all three end points have been operationalized from single items in questionnaires, which increases the risk of random measurement errors (Fayers and Machin 2000).