The longitudinal relationship between control over working hours and depressive symptoms: Results from SLOSH, a population-based cohort study

Author’s Accepted Manuscript The longitudinal relationship between control over working hours and depressive symptoms: Results from SLOSH, a population-based cohort study Sophie C Albrecht, Göran Kecklund, Kristiina Rajaleid, Constanze Leineweber www.elsevier.com/locate/jad

PII: DOI: Reference:

S0165-0327(16)31627-5 http://dx.doi.org/10.1016/j.jad.2017.03.010 JAD8809

To appear in: Journal of Affective Disorders Received date: 16 September 2016 Revised date: 1 February 2017 Accepted date: 5 March 2017 Cite this article as: Sophie C Albrecht, Göran Kecklund, Kristiina Rajaleid and Constanze Leineweber, The longitudinal relationship between control over working hours and depressive symptoms: Results from SLOSH, a populationbased cohort study, Journal of Affective Disorders, http://dx.doi.org/10.1016/j.jad.2017.03.010 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

The longitudinal relationship between control over working hours and depressive symptoms: Results from SLOSH, a population-based cohort study Sophie C Albrechta1, Göran Kecklunda,b, Kristiina Rajaleida, and Constanze Leinewebera a

Stress Research Institute, Stockholm University, Stockholm, Sweden

b

Behavioral Science Institute, Radboud University, Nijmegen, The Netherlands

[email protected]

Abstract Background: Psychosocial work factors can affect depressive moods, but research is inconclusive if flexibility to selfdetermine working hours (work-time control, WTC) is associated with depressive symptoms over time. We investigated if either sub-dimension of WTC, control over daily hours and control over time off, was related to depressive symptoms over time and examined causal, reversed-causal, and reciprocal pathways. Methods: The study was based on four waves of the Swedish Longitudinal Occupational Survey of Health which is a follow-up of representative samples of the Swedish working population. WTC was measured using a 5item index. Depressive symptoms were assessed with a brief subscale of the Symptom Checklist. Latent growth curve models and cross-lagged panel models were tested. Results:

1

Contact: Sophie C. Albrecht, Stress Research Institute, Stockholm University, 106 91 Stockholm, Sweden; Phone: +46 73 8925 717; Fax: +46 8 5537 8900

1

Best fit was found for a model with correlated intercepts (control over daily hours) and both correlated intercepts and slopes (control over time off) between WTC and depressive symptoms, with stronger associations for control over time off. Causal models estimating impacts from WTC to subsequent depressive symptoms were best fitting, with a standardised coefficient between -0.023 and -0.048. Limitations: Results were mainly based on self-report data and mean age in the study sample was relatively high. Conclusion: Higher WTC was related to fewer depressive symptoms over time albeit small effects. Giving workers control over working hours – especially over taking breaks and vacation – may improve working conditions and buffer against developing depression, potentially by enabling workers to recover more easily and promoting work-life balance. Abbreviations: WTC: work-time control; SLOSH: Swedish Longitudinal Occupational Survey of Health; SWES: Swedish Work Environment Survey; LGCM: Latent growth curve modelling; SEM: Structural equation modelling; RMSEA: root mean square error of approximation; CFI: comparative fit index; SRMR: standardised root mean squared residual; AIC: Akaike information criterion; BIC: Bayesian information criterion; Keywords: work-time control, flexible working hours, psychosocial work environment, prospective study, autonomy.

1.

Introduction Psychological conditions have become the most common reason for sickness absence accounting for

47% in women and 35% in men receiving sickness benefits in Sweden (Försäkringskassan, 2015) – with affective disorders like depressive, anxiety and stress-related disorders taking the lion’s share (Försäkringskassan, 2014). Depression not only affects labour capacity, but can also be the outcome of work environment factors such as social support, organisational justice, job control and psychological demands (Bonde, 2008; Magnusson Hanson et al., 2014b). Results from a meta-analysis found moder-

2

ately strong evidence for decision latitude (a dimension of job control), job strain (high psychological demands paired with low control), and bullying at work affecting depressive symptoms over time (Theorell et al., 2015). Preventing adverse work environments and work-related psychosocial stressors has received more attention as economic costs of mental health disorders make risk factors for depression at work a major public health concern (Berto et al., 2000). One aspect of working conditions that is understudied but potentially affects mental health and depression is control over working hours, or work-time control (WTC; Härmä, 2014). Some studies observed that high levels of WTC protect against mental health complaints like depressive symptoms while low WTC is associated with poor health (Ala-Mursula et al., 2005; Takahashi et al., 2012). But a systematic review concluded that the evidence of WTC affecting any health outcome is insufficient – at least partly because longitudinal studies are lacking (Nijp et al., 2012). We define WTC as the autonomy over duration and timing of work (Knauth, 1998) with two subdimensions regarding control over daily hours (self-determining starting and ending times of work and length of a workday) and control over time off (scheduling vacation, taking breaks and running private errands during work; Albrecht et al., 2016). Some studies observed that high levels of WTC are related to lower levels of depressive symptoms, fatigue and sleep disturbances (Ala-Mursula et al., 2005, 2002; Salo et al., 2014). Using a one-year follow-up design, Takahashi et al. (2012) observed that stable high or increasing levels of WTC were related to fewer depressive symptoms, longer sleep duration and less fatigue after one year compared to baseline. But several studies found no relation between WTC and depression/mental health (for a systematic review refer to Nijp et al., 2012). In contrast to Ala-Mursula et al. (2002), Jang et al. (2011) failed to find an effect of low WTC on psychological distress. In another study, low levels of WTC were unrelated to stress and burnout symptoms in physicians (Tucker et al., 2015). Moreover, the differential effects that

3

both sub-dimensions of WTC – control over daily hours and control over time off – could have on depression are rarely assessed (Nijp et al., 2012). There are at least two hypothesised mechanisms through which WTC potentially affects mental health. The effort-recovery model provides an explanation for the influence of work stressors on workers’ well-being (Meijman and Mulder, 1998). Effort at work results in productivity gains but also costs in physio- and psychological outcomes. Workers need to recover from these expenditures during work breaks or after work. If time for recovery is insufficient, workers expend more effort which in turn builds up need for recovery. As high WTC allows to self-determine working hours to fit personal schedules, flexible recovery opportunities should increase and buffer against negative health outcomes associated with exhaustion and fatigue such as depression (Geurts and Sonnentag, 2006). Another hypothesis why higher levels of WTC should benefit health outcomes is that increased WTC enables workers to better align their working hours with demands stemming from outside of work. As such, WTC should improve work-life balance which in turn relates to favourable outcomes in workers’ health, energy, and satisfaction (Geurts and Demerouti, 2003; Leineweber et al., 2012). Likewise, WTC should benefit the psychosocial work environment and affect job and social climate, work demands, job satisfaction, and morale of workers (Joyce et al., 2010). Research is inconclusive if either of the sub-dimensions of WTC impacts mental health differently than the other one (Nijp et al., 2012). Either control over time off affects mental health more pronouncedly as it increases opportunities to recover – short- and long-term (Geurts and Sonnentag, 2006). Or control over daily hours is more effective as it allows to align daily working hours to acute needs and consistently promotes a psychological sense of control (Geurts and Demerouti, 2003). Although these proposed mechanisms can account for positive effects of WTC on depression, there are equally strong arguments for reversed causation. Hobfoll (1989) predicts in his Conservation of Resources (COR) model that ‘loss spirals’ follow initial depletion of resources. As individuals strive to main-

4

tain resources – which can be anything from material goods to personal, energy resources – losses are of greater salience than gains. With each incidence of resource loss, individuals are less equipped and less resourceful to face new situations that could mean depletion. At the same time, resources are connected to each other, meaning loss in one resource can result in losses in another, which results in a downward spiral. As such, resource loss is the main reason for stress (Hobfoll, 2001). Translated to the situation at work this means that healthy workers are more likely to perceive and develop resources at work such as WTC. Conversely, mental health complaints mean a loss in resource and would promote a less favourable perception of the work environment. Additionally, research on information processing showed that depressive individuals perceive their environment in a more negative way (Beck, 2002). These results support the notion that depressive symptoms could also adversely impact perceived WTC. If mental health complaints partly account for lower ratings of WTC, studies systematically overestimate the impact of WTC on health (Takahashi et al., 2012). This could explain some of the inconsistencies in results. A number of studies report reciprocal effects between psychosocial working conditions and mental health, for instance between job strain (including low job control which conceptually relates to WTC) and depression/distress (Ibrahim et al., 2009) and job demands and psychological distress (Dalgard et al., 2009). These results indicate that depressive symptoms likely affect ratings of job characteristics, and that reversed-causal as well as reciprocal pathways need to be investigated in this context.

Aim Due to a lack in longitudinal studies and inconsistencies in measurement, the strength and direction of the association between WTC and mental health remains unclear (Albrecht et al., 2016; Nijp et al., 2012). Differential effects of either sub-dimension – control over daily hours and control over time off – have rarely been investigated. The present study used Structural Equation Modelling (SEM) to assess the association of levels and change in WTC and depressive symptoms over a time span of six years in a 5

large, four-wave sample of Swedish workers. Based on evidence and theoretical mechanisms, we examined four hypotheses: (1) WTC is related to depressive symptoms over time, (2) WTC affects depressive symptoms, (3) depressive symptoms affect perceived WTC, and (4) WTC and depressive symptoms reciprocally affect each other. Each hypothesis was considered separately for control over daily hours and control over time off to explore any differential effects.

2.

Methods

Study Design and Population Data were obtained from the Swedish Longitudinal Occupational Survey of Health (SLOSH) which is an approximately representative sample of the Swedish working population aged between 16 and 64 years. In SLOSH, respondents of the Swedish Work Environment Survey (SWES; 2003-2011) were followed-up via self-administered questionnaires (with separate ones for those in and outside paid work) every other year since 2006. Gender as well as labour market sector distributions are approximately representative of the population. The current study was based on four SLOSH waves with response rates of 61.1%, (n=11441) in 2008 (Time 1), 56.4% (n=11525) in 2010 (Time 2), 56.8% (n=9880) in 2012 (Time 3), and 52.6% (n=20316) in 2014 (Time 4). Participants were included if they answered the questionnaire for those in paid work in at least two out of four waves and were not self-employed (i.e. entrepreneurs or farmers). Valid sample size was n=2791. Ethical approval for SLOSH and the present study was obtained from the Regional Research Ethics Board in Stockholm.

6

Measures Work time control WTC was measured using an adapted 5-item scale based on Ala-Mursula et al. (2005) which assessed the perceived control over length of working time, starting and ending times of work, taking breaks during work, running private errands at work, and scheduling vacation and other leave. Perceived influence was rated on a 5-point Likert scale from 1 (very little) to 5 (very much). The items cover two subdimensions of WTC: Control over daily work (two items: length of working time, starting and ending times of work) and control over time off (three items: taking breaks, running private errands, scheduling vacation and other leave; Ala-Mursula et al., 2005; Albrecht et al., 2016). Means were calculated for each sub-dimension. Cronbach’s alphas for control over daily hours in 2008, 2010, 2012, and 2014 were 0.91, 0.92, 0.93, and 0.93, and for control over time off 0.72, 0.75, 0.75, and 0.77.

Depressive Symptoms A brief, 6-item subscale of the Hopkins Symptom Checklist (SCL-90) was used to measure prevalence of core depressive symptoms regarding feeling blue, having no interest in things, feeling low in energy, worrying too much, blaming oneself, and perceiving everything as effortful (Lipman, 1986). Validity and unidimensionality were confirmed with the Major Depression Inventory as index (Magnusson Hanson et al., 2014c). Items were rated for the last week on how much symptoms troubled the responder from ‘not at all’ (0) to ‘extremely’ (4). Sum scores were used for analysis with scores ≥17 indicating major depression (Magnusson Hanson et al., 2014c). Cronbach’s alphas were 0.91, 0.92, 0.91, and 0.91 from 2008 to 2014.

7

Covariates Selection of covariates was based on theoretical and evidential associations with WTC (Ala-Mursula et al., 2005; Albrecht et al., 2016) and depressive symptoms (Bjelland et al., 2008; Harrington, 2001; Nolen-Hoeksema, 2001) as well as utilising directed acyclic graphs (Tu and Gilthorpe, 2012). Register data were used to access participants’ age (at end of year of the questionnaire), gender, highest level of education (five categories from 9 years compulsory school to doctoral education) and occupational status at baseline (manual, lower non-manual and medium to high non-manual workers). Working schedule types were assessed by self-report data. Shift workers were defined as those usually working shift or rostered working hours including or excluding night work or night hours only (between 6am to 6pm). Non-shift workers included those working daytime or evening hours (until 10pm). Weekly working hours were self-reported numerically in 2008 and in 7 categories from less than 10 hours to 60 hours or more per week from 2010 to 2014. Civil status (living alone/single versus married/cohabiting) and parental status (at least one child/no children living at home) were also assessed in questionnaires.

Statistical Analysis Two complementary analytical approaches were utilised to investigate the relation between WTC and depressive symptoms. Latent growth curve modelling (LGCM) is ideal to investigate if an association between two variables exists over time and if growth in one variable covaries with growth in another. While LGCM allow to model individual trajectories, no causal inferences can be drawn since information on which variable temporally precedes the other one is not included. Cross-lagged panel models are preferable to investigate the direction of an effect using all available, separate time points (Ferrer and McArdle, 2003; Little et al., 2007; McArdle, 2009). Data preparation was executed using SPSS 22.0 (IBM Corp., 2013). Structural equation modelling (SEM) was employed using Mplus 7 (Muthén and Muthén, 2012). Robust maximum likelihood was used 8

as estimator as some variables violated the standard normal distribution assumption. Full information maximum likelihood estimation allowed to reduce bias due to missing data (Enders and Bandalos, 2001). Model fit was assessed by model fit indexes rather than the Chi2 statistic which is strongly influenced by large sample sizes leading to a high likelihood of rejecting a model (Jöreskog and Sörbom, 1993). Satisfactory model fit is suggested by values of <0.06 (good model fit) to 0.08 (acceptable model fit) on the root mean square error of approximation (RMSEA; an absolute fit index; Moen et al., 2011), and values >0.95 on the comparative fit index (CFI; an incremental fit index; Bentler, 1990). For the standardised root mean squared residual (SRMR) <0.08 has been suggested as cut-off criteria (Hu and Bentler, 1999). The Akaike information criterion (AIC) and the Bayesian information criterion (BIC) were used to assess relative model fit between models, with lower values indicating better fit (Akaike, 1974; Schwarz, 1978). Additionally, models were compared using log-likelihood ratio tests based on the Chi-square statistic for nested models with significant p-values indicating model fit improvement (Vuong, 1989). Satisfactory measurement model fit was confirmed for both WTC dimensions and depressive symptoms for each time point as indicated by model fit indexes. To achieve satisfactory fit regarding depressive symptoms, correlations had to be freed between the items on blaming oneself and worrying too much. Tests of longitudinal measurement invariance were performed for each variable and resulted in at least strong factorial invariance for all latent variables. Model restrictions were used for consecutive models accordingly. Multivariate LGCM (utilising SEM) were employed to examine the association between initial level and change (=growth) in WTC and depressive symptoms over four time points. Developmental trajectories were modelled for both WTC factors and depressive symptoms. Each variable was predicted from latent intercepts (initial value, loadings fixed to one) and slopes (linear rate of change, loadings fixed from 0 to 6 in steps of two from Time 1 to Time 4, one unit equalling one year, centred at Time 1). Latent variables were created for depression for each wave. Identification issues arose when using latent

9

variables for WTC, so both WTC factors were treated as observed variables using the score mean for each individual per time point. We compared five models for the association between each sub-dimension of WTC and depressive symptoms. M0 included random intercepts and no slopes (i.e. variables are allowed to differ individually but not to change over time), M1 added fixed slopes (variables with fixed growth over time across all individuals), M2 allowed for random slopes (growth could vary between individuals). Additionally to the restrictions in M2, we compared models M3 with correlated slopes (change in one variable was associated with change in the other), M4 with correlated intercepts (initial levels were associated), and M5 with both correlated slopes and intercepts. To investigate the direction of effects, cross-lagged panel models were employed with WTC factors and depression as latent variables. Four different models were compared for each combination of variables: (1) a null model with only autoregressive paths within the variables from Time 1 to Time 4, (2) a causal model estimating forward paths from WTC to depression (Time 1 WTC to Time 2 depression, Time 2 WTC to Time 3 depression, etc.), (3) a reversed causality model estimating reversed paths from depression to WTC, and (4) a reciprocal model estimating causal and reversed causal relationships. Constraints were made for autoregressive, causal and reversed causal paths to be equal over time (Little et al., 2007). Finally, gender differences were investigated using the best fitting model and comparing between constraining relevant coefficients (i.e. intercepts and slopes correlation, cross-lagged paths) to be the same or allowing them to differ between women and men.

10

3.

Results

Descriptives Characteristics of the study sample are presented in Table 1. At Time 1, participants were on average 47 years old (SD=8.29), respectively they were 53 at Time 4. The proportion of participants scoring at or above the suggested cut-off value of 17 (which may indicate clinical levels of depression) was 4.0% at Time 1, 4.2% at Time 2, 2.7% at Time 3, and 3.0% at Time 4. Women reported significantly more depressive symptoms and lower levels of control over daily hours and time off across all time points than men (p<0.001). Table 2 presents the matrix for estimated means, standard deviations and inter-correlations between control over daily hours, control over time off and depressive symptoms from Time 1 to Time 4. Associations of covariates with both WTC dimensions and depressive symptoms were found to be significant for age, gender, highest level of education, weekly working hours, occupation and shift work. These variables were adjusted for in all subsequent analyses.

11

Table 1: Descriptive statistics of the study sample presenting mean and standard deviation (SD) or number (n) and percentage.

Total n Age (years): Time 4 – 2014 Highest Educational Level Primary/compulsory school (≤ 9 years) Secondary school (≤ 13 years) University (< 3 years) University (≥ 3 years) Postgraduate/doctoral education Civil status: Time 4 – 2014 Living alone/single Married/cohabiting Children living at home: Time 4 – 2014 At least one child No children Working time: Time 4 – 2014 Any shift work/nights Daytime/evening Weekly working hours: Time 4 – 2014 < 10h/week 10-19h/week 20-29h/week 30-39h/week 40-49h/week 50-59h/week >59h/week Occupation: Time 1 – 2008 Manual workers Lower non-manual workers Medium to high non-manual workers

1595 1595

Female Mean/ n 53.2

SD/% 7.9

Total n 1127 1127

Male Mean/ n 53.2

SD/% 8.8

66

4.1

103

9.1

635 84 782 28

39.8 5.3 49.0 1.8

497 123 378 26

44.1 10.9 33.5 2.3

367 1218

23.0 76.4

202 915

17.9 81.2

634 947

39.7 59.4

501 610

44.5 54.1

245 1350

15.4 84.6

132 995

11.7 88.3

74 27 107 457 849 69 12

4.7 1.7 6.7 28.7 53.2 4.3 0.8

31 13 33 184 768 85 13

2.8 1.2 2.9 16.3 68.1 7.5 1.2

360 266

22.6 16.7

395 79

35.0 7.0

1595

60.7

653

57.9

1585

1117

1581

1111

1595

1127

1595

1127

1595

1127

12

Table 2: Estimated means, standard deviations (SD) and inter-correlations between control over daily hours, control over time off and depressive symptoms between 2008 and 2014. Variables

Mea n

SD

1

2

3

4

5

6

7

8

9

10

11

1 2

1 Control over daily hours 2008

2.90 4

1.3 78

1

-

-

-

-

-

-

-

-

-

-

-

2 Control over daily hours 2010

2.83 6

1.3 08

0.72 4

1

-

-

-

-

-

-

-

-

-

-

3 Control over daily hours 2012

2.87 5

1.2 98

0.69 6

0.77 0

1

-

-

-

-

-

-

-

-

-

4 Control over daily hours 2014

2.89 5

1.3 50

0.67 3

0.72 6

0.79 4

1

-

-

-

-

-

-

-

-

5 Control over time off 2008

3.26 5

0.9 99

0.77 9

0.65 9

0.62 6

0.63 9

1

-

-

-

-

-

-

-

6 Control over time off 2010

3.14 6

1.0 02

0.66 0

0.80 4

0.69 5

0.68 6

0.81 9

1

-

-

-

-

-

-

7 Control over time off 2012

3.17 6

0.9 77

0.65 2

0.71 2

0.83 3

0.73 8

0.79 0

0.85 0

1

-

-

-

-

-

8 Control over time off 2014

3.18 4

1.0 12

-

-

-

0.6 20

1

-

-

11 Depressive symptoms 2012

10.4 86

4.6 70

0.5 67

0.6 59

1

-

12 Depressive symptoms 2014

10.9 56

4.7 31

0.86 9 0.14 3 0.15 5 0.14 5 0.17 3

1

4.9 26

0.82 3 0.14 5 0.16 6 0.12 1 0.16 7

-

11.1 65

0.76 7 0.20 5 0.19 0 0.15 5 0.18 5

-

10 Depressive symptoms 2010

0.86 3 0.02 2 0.03 3 0.02 3 0.07 0

-

5.0 97

0.72 6 0.03 0 0.06 2 0.03 6 0.04 9

-

11.3 57

0.68 0 0.02 9 0.06 0 0.02 5 0.05 2

1

9 Depressive symptoms 2008

0.64 7 0.06 3 0.07 0 0.04 9 0.07 8

0.5 76

0.6 38

0.6 80

1

0.11 5 0.13 8 0.11 8 0.18 7

Control over daily hours Association with depressive symptoms Results and fit indexes of latent growth curve models (LGCM) for control over daily hours and depressive symptoms can be inspected in Table 3. In LGCMs, model fit improved significantly with adding, random intercepts (M1), random slopes (M2), and correlated slopes and intercepts (M3-M5) compared to a no-change model (M0). Best model fit was achieved for M4 with intercepts correlating at -0.083 (p<0.001) but uncorrelated slopes, indicating lower control over daily hours was associated with more depressive symptoms at initial level. Intercepts were at 4.078 for control over daily hours and 3.184 for 13

depressive symptoms. Individuals differed in their initial levels of control over daily hours (residual variance=0.903, p<0.001) and depressive symptoms (residual variance=0.440, p<0.001). Control over daily hours was stable (slope mean=-0.001, p=0.773), while depressive symptoms increased over time (slope mean=0.055, p=0.013). Individuals differed in their rate of growth in control over daily hours (slope variance=0.011, p<0.001) and depressive symptoms (slope variance=0.004, p<0.001).

Direction of effect In cross-lagged panel models, differences between models were small (Table 3). Model fit improved with adding any cross-lagged path compared to the stability model. The reciprocal model improved model fit significantly compared to the reversed causality model, but not compared to the causal one. Since the causal model also showed the lowest AIC and BIC values, this model seemed to fit data best. Cross-lagged paths from control over daily hours to subsequent depressive symptoms were significant with a standardized coefficient of -0.023 (p=0.002, Figure 1).

14

Table 3: Latent growth curve models and cross-lagged panel models for depressive symptoms and control over daily hours adjusted for gender, age, education, weekly working hours and shift working status. Control over daily hours

2

n

Chi

df

2722

2947.630

741

M1: Random intercept; fixed slope M2: Random slope; random intercept M3: Correlated slopes

2972.010

743

2787.594

735

2782.562

734

M4: Correlated intercepts

2769.916

734

M5: Correlated slopes and intercepts

2768.363

733

4182.395

982

4145.000

980

4153.179

980

4143.881

979

M0: No Change; random intercept

Stability Model: Auto-regressive paths Causal Model: WTC → Depression Reversed Causality Model: Depression → WTC Reciprocal Model: WTC ↔ Depression

2722

RMSEA (90% CI)

0.033 (0.0320.034) 0.033 (0.0320.034) 0.032 (0.0310.033) 0.032 (0.0310.033) 0.032 (0.0310.033) 0.032 (0.0310.033)

0.035 (0.0340.036) 0.034 (0.0330.036) 0.034 (0.0330.036) 0.034 (0.0330.036)

CFI

SRMR

AIC

BIC

Log Likelihood Difference Test 2 Chi /df

Latent Growth Curve Models 0.954 0.033 167034 167371

0.953

0.033

167056

167381

Vs. M0: 27.897/2df (p<0.001)

0.957

0.032

166867

167239

Vs. M1: 190.941/8df (p<0.001)

0.957

0.031

166863

167241

Vs. M2: 5.0864/1df (p=0.024)

0.957

0.029

166849

167227

Vs. M2: 19.175/1df (p<0.001)

0.957

0.029

166850

167234

Vs. M2: 20.343/2df (p<0.001) Vs. M3: 15.738/1df (p<0.001) Vs. M4: 1.364/1df (p=0.243)

Cross-Lagged Panel Models 0.952 0.056 191175 191707

0.953

0.054

191141

191685

Vs. Stability: 36.065/2df (p<0.001)

0.953

0.054

191150

191693

Vs. Stability: 28.717/2df (p<0.001)

0.953

0.054

191142

191692

Vs. Stability: 37.978/3df (p<0.001) Vs. Causal: 1.032/1df (p=0.310) Vs. Reversed: 9.226/1df (p=0.002)

df = degrees of freedom, RMSEA = root mean square error of approximation, CFI = comparative fit index, SRMR = standardised root mean squared residual, BIC = Bayesian information criterion.

15

Control over time off Association with depressive symptoms Model results and fit indexes for control over time off can be inspected in Table 4. In LGCMs, adding random slopes and correlated slopes and intercepts significantly improved model fit. Best model fit was found for model M5 with correlated intercepts and slopes. Intercepts were at 4.937 for control over time off and 3.201 for depressive symptoms. Individuals differed in their level of control over time off (residual variance=0.614, p<0.001) and depressive symptoms (residual variance=0.438, p<0.001). Control over time off slightly decreased over time (slope mean=-0.011, p<0.001), but individuals differed in that regard (slope variance=0.005, p<0.001). Depressive symptoms increased on average over time (slope mean=0.055, p=0.013) with participants differing in their rates of change (slope variance=0.004, p<0.001). The correlation between intercepts was -0.145 (p<0.001) indicating initial lower control over time off was related to more depressive symptoms. Slopes correlated at -0.217 (p=0.005) meaning a decrease in control over time off was associated with an increase in depressive symptoms over time.

Direction of effect In cross-lagged panel models (Table 4), although differences between models were small, the causal model was best-fitting as assessed by both model-fit indexes (lowest AIC and BIC value) and loglikelihood ratio tests. Cross-lagged paths from control over time off to subsequent depressive symptoms were significant (p<0.001) with a standardized coefficient of -0.043/-0.048 (Figure 2).

16

Table 4: Latent growth curve models (LGCM) and cross-lagged panel models for depressive symptoms and control over time off. 2

Control over time off

n

Chi

df

RMSEA (90% CI)

CFI

SRMR

AIC

M0: No Change; random intercept M1: Random intercept; fixed slope M2: Random slope; random intercept M3: Correlated slopes M4: Correlated intercepts M5: Correlated slopes and intercepts

2722

2957.278

741

2930.211

739

0.033 (0.0320.034)

0.954

0.040

160080

160429

Vs. M0: 26.724/2df (p<0.001)

2821.698

735

0.032 (0.0310.034)

0.956

0.039

159965

160337

Vs. M1: 106.513/4df (p<0.001)

2799.816

734

0.957

0.037

159942

160320

2772.993

734

0.957

0.030

159912

160290

2763.631

733

0.032 (0.0310.033) 0.032 (0.0310.033) 0.032 (0.0310.033)

0.957

0.030

159903

160287

Vs. M2: 20.949/1df (p<0.001) Vs. M2: 46.627/1df (p<0.001) Vs. M2: 57.195/2df (p<0.001) Vs. M3: 36.816/1df (p<0.001) Vs. M4: 9.440/1df (p=0.002)

Stability Model: Auto-regressive paths Causal Model: WTC → Depression Reversed Causality Model: Depression → WTC Reciprocal Model: WTC ↔ Depression

2722

4574.089

1191

Cross-Lagged Panel Models 0.032 (0.0312- 0.949 0.058 221971 0.033)

222485

4506.485

1189

0.032 (0.0310.033)

0.950

0.055

221906

222432

Vs. Stability: 61.968/2df (p<0.001)

4522.746

1189

0.032 (0.0310.033)

0.949

0.056

221922

222448

Vs. Stability: 45.582/2df (p<0.001)

4506.511

1188

0.032 (0.0310.033)

0.950

0.055

221908

222440

Vs. Stability: 62.409/3df (p<0.001) Vs. Causal: 0.319/1df (p=0.572) Vs. Reversed: 16.221/1df (p<0.001)

Latent Growth Curve Models 0.033 (0.0320.953 0.040 160107 0.034)

BIC

Log Likelihood 2 Chi /df

160443

df = degrees of freedom, RMSEA = root mean square error of approximation, CFI = comparative fit index, SRMR = standardised root mean squared residual, BIC = Bayesian information criterion.

17

Gender Differences We found no differences in LGCM and cross-lagged panel final models between men and women (results not reported). Models allowing for relevant coefficients to differ between genders failed to improve model fit.

4.

Discussion The present paper aimed at investigating, first, if there was an association between work-time con-

trol (WTC) and depressive symptoms over time, and second, which direction predominated the effect. Over a time period of six years in a large sample of Swedish workers we found a significant, genderindependent relation between WTC and depressive symptoms. Effects were stronger for control over time off which was highlighted by a significant correlation with changes in depressive symptoms for this sub-dimension only. When investigating the direction of effect, our results supported that causal models – estimating lagged impacts from WTC to subsequent depressive symptoms – were best fitting. Differences between models were small, indicating that reversed paths were only slightly weaker than causal ones. Few studies have investigated WTC and its effect on health in longitudinal samples. The majority of them used a dataset of Finnish public sector workers (Ala-Mursula et al., 2006, 2005; Salo et al., 2014) while some were based on one year of follow-up only (Ala-Mursula et al., 2002; Takahashi et al., 2012). The predominance of cross-sectional studies is in part responsible for the ambiguity regarding WTC affecting mental and physical health (Nijp et al., 2012). We tried to contribute to closing this gap in research by applying a longitudinal design to a sample of workers that is based on representative samples of the Swedish working population.

18

To our knowledge this study is the first one to examine the direction of effect between WTC and any health outcome. Disentangling causal from reversed causal pathways is crucial – especially regarding mental health outcomes such as depression. De Lange et al. (2005) investigated several mechanisms underlying reversed causal and reciprocal paths in a number of longitudinal samples between job demands, control and support and depression. Results showed support for depressive mood indeed affecting job strain ratings with no difference between initially healthy or unhealthy respondents. Such evidence highlights not only the necessity to investigate reciprocal relations between work-related factors and depression, but also the importance to prevent workers from developing depressive moods in the first place as they may enter a vicious cycle of perceiving their work environment as more strenuous which accelerates development of depression (Bonde, 2008; Theorell et al., 2015). To investigate reciprocity, structural equation modelling using panel data is a useful approach to compare causal, reversed causal and reciprocal cross-lagged paths (Farrell, 1994) – and an even more robust technique when using more than two time points (Cole and Maxwell, 2003). In contrast to latent growth curve modelling – which we used in the present study to investigate if WTC is related to depressive symptoms over time – cross-lagged panel models allow to estimate lagged effects in different directions between variables and time points (Ferrer and McArdle, 2003). Our results add support for a predictive relation between WTC and depressive symptoms, but also suggest that effects are rather small. We found causal models to fit the data best for both subdimensions of WTC. Results on control over daily hours have to be interpreted carefully. The relation is weaker and only with initial levels of depressive symptoms when using a time lag of two years. Higher levels of control over time off were related to fewer depressive symptoms over time with a standardized coefficient of -0.05. In comparison, the influence of other psychosocial work factors on subsequent depressive symptoms (using a similar statistical technique) seems to be only slightly stronger with standardized coefficients of 0.06 for job demands and -0.07 to -0.08 for workplace support

19

(Magnusson Hanson et al., 2014a). WTC is potentially easier to implement by companies which may be an advantage over other factors of psychosocial work environment. A different way to investigate reciprocity would be to use objective instead of self-rated WTC. Objective control would not be influenced by depressive moods compared to the individual’s perception of WTC. However, objective WTC is difficult to assess in practice – for instance employer-reported levels or within working contracts. Only one study compared objective and subjective control over daily hours and their joint effect with work stress on sickness absence rates. Interestingly, subjective results were replicable using objective control in women, but not in men (Ala-Mursula et al., 2005). However, using objective levels of control instead of subjective ones comes with other shortcomings. The mere perception of control appears to be more relevant for effects on health than objective ratings, which has been confirmed for general autonomy (Lefcourt, 1973; Perlmuter and Monty, 1977) and specifically for WTC (Nijp et al., 2015). These results are in line with theories on the psychological sense of control (Deci and Ryan, 1985; Skinner, 1996). Perceived control even entails a subjective rating over feelings of control in contrast to objectively available control (Skinner, 1996). Future studies should investigate the mechanisms underlying the predictive relation between WTC and depressive symptoms as well as important mediators. Few studies have investigated differential effects of the sub-dimensions of WTC on health. We found larger effects for control over time off compared to control over daily hours. Likewise, a cross-sectional study found more favourable and consistent effects on depressive symptoms, sleep and fatigue with high control over time off than with control over daily hours (Takahashi et al., 2011). The underlying mechanism remains unclear. Speculatively, control over time off may promote opportunities to recover to a greater deal (by taking breaks and vacation when necessary) and facilitate aligning work and private life from a more long-term perspective (by running private errands during working hours and scheduling vacation in accordance with family needs). In contrast, workers may utilise control over daily hours sub-

20

optimally and choose working hours that are unideal in terms of health (e.g. by working long hours). This is underlined by an intervention study on self-rostering showing that workers did not entirely adhere to recommendations on how to self-determine shift-work schedules best (Garde et al., 2012). Control over daily hours could contribute to the issue of boundaryless work and prevent workers from optimal mental recovery and hence advance development of burnout and depression. This notion is supported by evidence for more favourable effects on work-life balance and sleep when high control over time off coincided with low variability of working hours, but not with high control over daily hours (Kubo et al., 2013). Control over daily hours may be used by workers to increase variability of working time which facilitates aligning work and private life up to a point where it blurs boundaries too much and reverses any positive effects on health. Nevertheless, studies consistently show moderate to high correlations between the sub-dimensions of WTC (Ala-Mursula et al., 2005; Albrecht et al., 2016). Even though levels of control over time off seem to be higher in general than control over daily hours (Albrecht et al., 2016), workers reported the greatest need for control over when to take vacation and other leave in a Dutch sample (Nijp et al., 2015). Future studies should investigate if high levels of control over time off paired with moderate levels of control over daily hours are more beneficial for health than high control on both dimensions.

Limitations Results of this study are based on mainly self-report data and all known limitations for these kind of data apply to our study as well as common-method bias. Objective measures such a sickness absence based on diagnoses of depression should be included in future research. Missing data remain a potential issue even though full information maximum likelihood estimation allowed us to reduce the likelihood of any bias. Due to the time lag of two years used in this study, we were unable to investigate more short-term, immediate associations between WTC and depressive symptoms. 21

The scale used to assess depressive core symptoms is not widely used and may have failed to capture aspects of especially clinical levels of depression. However, a validation study confirmed satisfactory psychometric values in comparison to longer, more established measures of depression (Magnusson Hanson et al., 2014c). The present study controlled for a number of covariates, but excluded factors of psychosocial work environments. This could mean that effects are overestimated or that WTC does not explain additional variance in depressive symptoms than other psychosocial variables. However, using directed acyclic graphs (DAGs; Tu and Gilthorpe, 2012) showed that WTC precedes and may impact many work-related factors (results not shown). For instance, if workers can self-determine working hours it may be easier to cope with higher job demands and prevent conflict with family and private life. Support at work on the other hand may be reduced if workers have less overlap at work due to flexible working hours. Job control shares important variance with WTC regarding autonomy and as a covariate would have overcontrolled effects. For that reason, it is advised to exclude such variables (Tu and Gilthorpe, 2012). All analyses were however controlled for occupational status at baseline. Manual work (e.g. goods/service production) and lower-level non-manual work (e.g. secretary/assistant) are related to low WTC. In contrast, high WTC is more common (though not consistently) in high-level occupations (e.g. managers/knowledge workers (Ala-Mursula et al., 2005). At the same time, occupation similarly captures aspects of psychosocial work factors such as job demands and control (Rahkonen et al., 2006). Although occupational status was a significant covariate, it did not attenuate effects between WTC and depressive symptoms. The mean age of the present sample is relatively high meaning that participants may have already developed depressive symptoms or even experienced clinical levels of depression. Our results could be biased if WTC fails to reverse manifested depressive symptoms. We repeated all analyses excluding individuals above an identified cut-off value indicating major depression (Magnusson Hanson et al., 2013)

22

and found no difference in results. Still, an older sample could mean that manifest depressive symptoms allow for little variance and hence effects of WTC could be underestimated – even though symptoms did vary over time within individuals to some degree.

Conclusion Results from this study suggest that higher levels of WTC, in specific control over time off, are associated with fewer depressive symptoms over time albeit small effects. The direction of this effect predominantly seems to be from WTC to subsequent depressive symptoms. Given that sickness absence from work due to depression diagnoses is increasing (Försäkringskassan, 2015), employers and policy-makers need to consider increasing WTC by introducing for instance flextime work contracts or self-scheduling software in shift-work occupations. Greater autonomy over working hours may help buffer against builtup need for recovery, burnout and depression.

5.

References

Akaike, H., 1974. A New Look at the Statistical Model Identification. IEEE Trans Automat Contr 19, 716– 723. Ala-Mursula, L., Vahtera, J., Kivimäki, M., Kevin, M. V, Pentti, J., 2002. Employee control over working times: Associations with subjective health and sickness absences. J Epidemiol Community Heal 56, 272–8. Ala-Mursula, L., Vahtera, J., Kouvonen, A., Väänänen, A., Linna, A., Pentti, J., Kivimäki, M., 2006. Long hours in paid and domestic work and subsequent sickness absence: Does control over daily working hours matter? Occup Env Med 63, 608–16. Ala-Mursula, L., Vahtera, J., Linna, A., Pentti, J., Kivimäki, M., 2005. Employee worktime control moderates the effects of job strain and effort-reward imbalance on sickness absence: The 10-town study. J Epidemiol Community Health 59, 851–7. Albrecht, S.C., Kecklund, G., Tucker, P., Leineweber, C., 2016. Investigating the factorial structure and availability of work time control in a representative sample of the Swedish working population. Scand J Public Health 44, 320–328. Beck, A.T., 2002. Cognitive Models of Depression, in: Leahy, R. (Ed.), Clinical Advances in Cognitive 23

Psychotherapy: Theory and Application. Springer Publishing Company, New York, pp. 29–61. Bentler, P.M., 1990. Comparative fit indexes in structural models. Psychol Bull 107, 238–246. Berto, P., D’ilario, D., Ruffo, P., Di Virgilio, R., Rizzo, F., 2000. Depression: Cost-of-illness studies in the international literature, a review. J Ment Health Policy Econ 3, 3–10. Bjelland, I., Krokstad, S., Mykletun, A., Dahl, A.A., Tell, G.S., Tambs, K., 2008. Does a higher educational level protect against anxiety and depression? The HUNT study. Soc Sci Med 66, 1334–1345. Bonde, J.P.E., 2008. Psychosocial factors at work and risk of depression: A systematic review of the epidemiological evidence. Occup Environ Med 65, 438–45. Cole, D.A., Maxwell, S.E., 2003. Testing mediational models with longitudinal data: questions and tips in the use of structural equation modeling. J Abnorm Psychol 112, 558–577. Dalgard, O.S., Sorensen, T., Sandanger, I., Nygård, J.F., Svensson, E., Reas, D.L., 2009. Job demands, job control, and mental health in an 11-year follow-up study: Normal and reversed relationships. Work Stress 23, 284–296. de Lange, A.H., Taris, T.W., Kompier, M.A., Houtman, I.L., Bongers, P.M., 2005. Different mechanisms to explain the reversed effects of mental health on work characteristics. Scand J Work Environ Health 31, 3–14. Deci, E.L., Ryan, R.M., 1985. Intrinsic Motivation and Self-Determination in Human Behavior. Plenum Press, New York. Enders, C.K., Bandalos, D.L., 2001. The relative performance of full information maximum likelihood estimation for missing mata in structural equation models. Struct Equ Model A Multidiscip J 8, 430–457. Farrell, A.D., 1994. Structural equation modeling with longitudinal data: Strategies for examining group differences with reciprocal relationships. J Consult Clin Psychol 62, 477–487. Ferrer, E., McArdle, J., 2003. Alternative structural models for multivariate longitudinal data analysis. Struct Equ Model A Multidiscip J 10, 493–524. Försäkringskassan, 2015. Social Insurance in Figures 2015. Försäkringskassan (Swedish Social Insurance Agency), Stockholm. Försäkringskassan, 2014. Sjukfrånvaro i psykiska diagnoser (Sick leave in psychiatric diagnoses). Försäkringskassan (Swedish Social Insurance Agency), Stockholm. Garde, A.H., Albertsen, K., Nabe-Nielsen, K., Carneiro, I.G., Skotte, J., Hansen, S.M., Lund, H., Hvid, H., Hansen, Å.M., 2012. Implementation of self-rostering (the PRIO-project): Effects on working hours, recovery, and health. Scand J Work Environ Health 38, 314–26. Geurts, S.A.E., Demerouti, E., 2003. Work/Non-Work Interface: A Review of Theories and Findings, in: Schabracq, M.J., Winnubst, J.A.M., Cooper, C.L. (Eds.), The Handbook of Work and Health Psychology. John Wiley & Sons, Chichester, pp. 279–312. Geurts, S.A.E., Sonnentag, S., 2006. Recovery as an explanatory mechanism in the relation between acute stress reactions and chronic health impairment. Scand J Work Environ Health 32, 482–492. Harrington, J., 2001. Health effects of shift work and extended hours of work. Occup Environ Med 58, 24

68–72. Hobfoll, S.E., 2001. The influence of culture, community, and the nested-self in the stress process: Advancing Conservation of Resources theory. Appl Psychol an Int Rev 50, 337–421. Hobfoll, S.E., 1989. Conservation of resources: A new attempt at conceptualizing stress. Am Psychol 44, 513–524. Hu, L., Bentler, P.M., 1999. Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Struct Equ Model A Multidiscip J 6, 1–55. Härmä, M., 2014. Workhours in relation to work stress, recovery and health. Scand J Work Environ Health 32, 502–514. IBM Corp., 2013. IBM SPSS Statistics for Windows, Version 22.0. IBM Corp, Armonk, NY. Ibrahim, S., Smith, P., Muntaner, C., 2009. A multi-group cross-lagged analyses of work stressors and health using Canadian National sample. Soc Sci Med 68, 49–59. Jang, S.J., Park, R., Zippay, A., 2011. The interaction effects of scheduling control and work-life balance programs on job satisfaction and mental health. Int J Soc Welf 20, 135–143. Joyce, K., Pabayo, R., Critchley, J.A., Bambra, C., 2010. Flexible working conditions and their effects on employee health and wellbeing (review). Cochrane Database Syst Rev 2010 CD008009. Jöreskog, K.G., Sörbom, D., 1993. LISREL 8: Structural Equation Modeling with the SIMPLIS command language. Scientific Software International, Chicago. Knauth, P., 1998. Innovative worktime arrangements. Scand J Work Env Heal 24, 13–17. Kubo, T., Takahashi, M., Togo, F., Liu, X., Shimazu, A., Tanaka, K., Takaya, M., 2013. Effects on employees of controlling working hours and working schedules. Occup Med (Chic Ill) 63, 148–151. Lefcourt, H.M., 1973. The function of the illusions of control and freedom. Am Psychol 28, 417–425. Leineweber, C., Baltzer, M., Magnusson Hanson, L.L., Westerlund, H., 2012. Work-family conflict and health in Swedish working women and men: A 2-year prospective analysis (the SLOSH study). Eur J Public Health 23, 710–716. Lipman, R.S., 1986. Depression Scales Derived from the Hopkins Symptom Checklist, in: Sartorius, N., Ban, T.A. (Eds.), Assessment of Depression. Springer, Berlin, Heidelberg, pp. 232–248. Little, T.D., Preacher, K.J., Selig, J.P., Card, N.A., 2007. New developments in latent variable panel analyses of longitudinal data. Int J Behav Dev 31, 357–365. Magnusson Hanson, L.L., Chungkham, H.S., Åkerstedt, T., Westerlund, H., 2014a. The role of sleep disturbances in the longitudinal relationship between psychosocial working conditions, measured by work demands and support, and depression. Sleep 37, 1977–85. Magnusson Hanson, L.L., Leineweber, C., Chungkham, H.S., Westerlund, H., 2014b. Work-home interference and its prospective relation to major depression and treatment with antidepressants. Scand J Work Environ Heal 40, 66–73. Magnusson Hanson, L.L., Westerlund, H., Leineweber, C., Rugulies, R., Osika, W., Theorell, T., Bech, P., 2014c. The Symptom Checklist-core depression (SCL-CD6) scale: psychometric properties of a brief 25

six item scale for the assessment of depression. Scand J Public Health 42, 82–88. Magnusson Hanson, L.L., Westerlund, H., Leineweber, C., Rugulies, R., Osika, W., Theorell, T., Bech, P., 2013. The Symptom Checklist-core depression (SCL-CD6) scale: psychometric properties of a brief six item scale for the assessment of depression. Scand J Public Health 42, 82–8. McArdle, J.J., 2009. Latent variable modeling of differences and changes with longitudinal data. Annu Rev Psychol 60, 577–605. Meijman, T.F., Mulder, G., 1998. Psychological aspects of workload, in: Drenth, P.J.D., Thierry, H., de Wolff, C. (Eds.), A Handbook of Work and Organizational Psychology: Volume 2: Work Psychology. Psychology Press, Hove (UK), pp. 5–33. Moen, P., Kelly, E.L., Tranby, E., Huang, Q., 2011. Changing work, changing health: Can real work-time flexibility promote health behaviors and well-being? J Health Soc Behav 52, 404–429. Muthén, L.K., Muthén, B.O., 2012. Mplus User’s Guide, Seventh Ed. ed. Muthén & Muthén, Los Angeles, CA. Nijp, H.H., Beckers, D.G.J., Geurts, S.A.E., Tucker, P., Kompier, M.A.J., 2012. Systematic review on the association between employee worktime control and work-non-work balance, health and wellbeing, and job-related outcomes. Scand J Work Environ Health 38, 299–313. Nijp, H.H., Beckers, D.G.J., Kompier, M.A., Bossche, S.N. van den, Geurts, S.A., 2015. Worktime control access, need and use in relation to work-home interference, fatigue, and job motivation. Scand J Work Environ Health 41, 327–336. Nolen-Hoeksema, S., 2001. Gender differences in depression. Curr Dir Psychol Sci 10, 173–176. Perlmuter, L.C., Monty, R.A., 1977. The Importance of Perceived Control: Fact or Fantasy? Am Sci 65, 759–765. Rahkonen, O., Laaksonen, M., Martikainen, P., Roos, E., Lahelma, E., 2006. Job control, job demands, or social class? The impact of working conditions on the relation between social class and health. J Epidemiol Community Health 60, 50–54. Salo, P., Ala-Mursula, L., Rod, N.H., Tucker, P., Pentti, J., Kivimäki, M., Vahtera, J., 2014. Work Time Control and sleep disturbances: Prospective cohort study of Finnish public sector employees. Sleep 37, 1217–1225. Schwarz, G., 1978. Estimating the dimension of a model. Ann Stat 6, 461–464. Skinner, E.A., 1996. A guide to constructs of control. J Pers Soc Psychol 71, 549–570. Takahashi, M., Iwasaki, K., Sasaki, T., Kubo, T., Mori, I., Otsuka, Y., 2012. Sleep, fatigue, recovery, and depression after change in work time control. J Occup Environ Med 54, 1078–1085. Takahashi, M., Iwasaki, K., Sasaki, T., Kubo, T., Mori, I., Otsuka, Y., 2011. Worktime control-dependent reductions in fatigue, sleep problems, and depression. Appl Ergon 42, 244–50. Theorell, T., Hammarström, A., Aronsson, G., Träskman Bendz, L., Grape, T., Hogstedt, C., Marteinsdottir, I., Skoog, I., Hall, C., 2015. A systematic review including meta-analysis of work environment and depressive symptoms. BMC Public Health 15, 738. Tu, Y.-K., Gilthorpe, M.S., 2012. Path Diagrams and Directed Acyclic Graphs, in: Statistical Thinking in 26

Epidemiology. Chapman and Hall/CRC, Boca Raton, pp. 17–26. Tucker, P., Bejerot, E., Kecklund, G., Aronsson, G., Åkerstedt, T., 2015. The impact of work time control on physicians’ sleep and well-being. Appl Ergon 47, 109–116. Vuong, Q.H., 1989. Likelihood ratio tests for model selection and non-nested hypotheses. Econometrica 57, 307–333.

Figure 1: Best fitting cross-lagged panel model with standardised coefficients between control over daily hours and depressive symptoms. Observed variables, covariates and non-significant paths are not displayed. (*** p<.001)

27

Figure 2: Best fitting cross-lagged panel model with standardised coefficients between control over time off and depressive symptoms. Observed variables, covariates and non-significant paths are not displayed. (*** p<.001)

HIGHLIGHTS 

The longitudinal association between work-time control and depressive symptoms was assessed



Causal, reversed-causal and reciprocal pathways were examined



Control over time off was inversely associated with depressive symptoms over time



Control over working hours seems beneficial for workers’ mental health

28