A Longitudinal Study of the Relationship between Work Engagement and Symptoms of Anxiety and Depression


Siw Tone Innstrand, Department of Social Work and Health Sciences, Norwegian University of Science and Technology, 7491 Trondheim, Norway.

Email: siw.tone.innstrand@svt.ntnu.no


This longitudinal study examined the dynamic relationship between work engagement (vigour and dedication) and symptoms of anxiety and depression. A sample of 3475 respondents from eight different occupational groups (lawyers, physicians, nurses, teachers, church ministers, bus drivers, people working in advertising and people working in information technology) in Norway supplied data at two points in time with a 2-year time interval. The advantages of longitudinal design were utilized, including testing of reversed causation and controlling for unmeasured third variables. In general, the results showed that the hypothesized normal causal relationship was superior to a reversed causation model. In other words, this study supported the assumption that work engagement is more likely to be the antecedent for symptoms of depression and anxiety than the outcome. In particular, the vigour facet of work engagement provides lower levels of depression and anxiety 2 years later. However, additional analyses modelling unmeasured third variables indicate that unknown third variables may have created some spurious effects on the pattern of the observed relationship. Implications of the findings are discussed in the paper. Copyright © 2011 John Wiley & Sons, Ltd.


For several decades, occupational health researchers have been preoccupied with factors at work that cause stress and infirmity. However, recent findings indicate that individuals' sense of self and personal worth is increasingly dependent on work, which has become ‘a source of social integration, recreation and especially friendship, in addition to economic security’ (Roos, Trigg, & Hartman, 2006, p. 209). These benefits might have extensive effects on employee health. Work engagement, defined as ‘a positive, fulfilling, affective motivational state of work-related well-being’ (Bakker, Schaufeli, Leiter, & Taris, 2008, p. 187), reflects the recent trend toward ‘positive psychology’, where the concern is the positive aspects of employee health (Seligman & Csikszentmihalyi, 2000; Seligman, Steen, Park, & Peterson, 2005). This more positive vision implies a move beyond pathology to focus on understanding and promoting healthy functioning. Based on conservation of resources (COR) theory (Hobfoll, 1989, 1998, 2001), the present study explored the possible reciprocal relationship between work engagement and physiological symptoms of anxiety and depression. COR theory, a promising explanatory model of burnout (Lee & Ashforth, 1996), encompasses several theories of stress but extends these theories through a resource perspective.

Work engagement and symptoms of anxiety and depression

The basic tenet of COR theory is that people have a deeply rooted motivation to obtain, retain and protect what they value, labelled resources (Hobfoll, 1989). The innovative aspect of COR theory is that it describes not only what individuals do when confronted with stress but also how they behave in the absence of threats. Specifically, when confronted with stress, individuals are predicted by the model to strive to minimize the net loss of resources. Conversely, when not confronted with threats, people strive to develop resource surpluses to offset the possibility of future loss. Whereas the COR theory describes burnout as a state of extreme resource depletion (Hobfoll & Shirom, 2001; Neveu, 2007), work engagement might be regarded as a resource surplus. When people develop resource surpluses, they are likely to experience positive well-being and health.

In occupational health psychology research, work engagement is seen as an antithesis to the more familiar and investigated term, ‘burnout.’ In the present study, work engagement was operationalized and assessed by the Oldenburg Burnout Inventory (OLBI; Demerouti & Bakker, 2008), which agrees with the theoretical argumentation that burnout and engagement are two opposite poles of one continuum (Maslach & Leiter, 1997, 2008). The OLBI includes two dimensions: one ranging from exhaustion to vigour and a second ranging from disengagement to dedication (Demerouti & Bakker, 2008; Halbesleben & Demerouti, 2005). Together, they provide two bi-polar dimensions—energy and identification (Demerouti & Bakker, 2008). Recent research suggests that vigour and dedication constitute the core dimensions of engagement (e.g. González-Romá, Schaufeli, Bakker, & Lloreta, 2006). Vigour is characterized by high levels of energy and mental resilience while working, whereas dedication is characterized by a sense of significance, enthusiasm, inspiration and pride (Demerouti & Bakker, 2008). Research focusing on engagement is still in its infancy, and there are many questions that still need to be answered. According to Bakker et al. (2008), there is particularly a dearth of research on the relationship between work engagement and health.

The European Commission's Green Paper on mental health [COM (2005) 484] states that anxiety and depression are two of the most prevalent mental ill health problems facing European citizens today. Although anxiety and depression are closely related, they are regarded as separate disorders because of differences in presenting characteristics (see Barlow, 2000). There is substantial literature documenting that stress and stressful life events constitute a risk for depression and anxiety disorders. For example, McLaughlin and Hatzenbuehler (2009) found that stressful life events were longitudinally related to anxiety symptoms in a diverse sample of adolescents. Equally, comparisons of depressed patients with non-depressed controls indicates that, besides reporting more stressful events than the controls, depressed people also report more life strain associated with health, family, housing and work (Billings, Cronkite, & Moos, 1983). Demerouti, Bakker, and Bulters (2004) show that a person who feels distressed at one point in time is likely to be distressed at a later point in time unless a significant event changes their emotional state.

Just as life strain associated with health, family, housing and work may lead to depression and anxiety in the long run (Billings et al., 1983; McLaughlin & Hatzenbuehler, 2009) we argue that joy and satisfaction associated with these same life areas might ease symptoms of depression and anxiety. Judge and Locke (1993) suggest that job satisfaction influences subjective well-being through the centrality of work to an individual's life. Since most individuals spend the majority of their waking hours at work, a positive evaluation will have extensive effect on judgments of happiness and well-being overall. For example, Fredrickson's (1998) ‘broaden-and-build model’ of positive emotions suggests that positive emotions broaden the individual's attentional focus and behavioural repertoire and, as a consequence, build social, intellectual and physical resources. Similarly, the COR theory anticipates that positive experiences or resources are likely to accumulate, creating a positive spiral of resources, which is likely to have positive health-promoting effects. Such resource caravans might provide health-promoting effects because they may do the following: (1) undo some of the negative physiological effects associated with negative emotions; (2) positively influence the neuroendocrine system; and/or (3) interrupt and short-circuit the rumination spiral of stressful circumstances and prevent the decline into clinical depression (see Folkman & Moskowitz, 2000). In a study among older adults, Glass, Mendes deLeon, Bassuk, and Berkman (2006) found social engagement not only to be independently associated with depressive symptoms when examined cross-sectionally but also to be associated with change in depressive symptoms among people who were not depressed at the baseline.

Previous research has found support for both reciprocal gain spirals (Llorens, Schaufeli, Bakker, & Salanova, 2007) and resource caravans (Hakanen, Perhoniemi, & Toppinen-Tanner, 2008) of work engagement. However, neither Llorens et al. nor Hakanen, Perhoniemi et al. have examined the possible health-promoting effect of work engagement on symptoms of anxiety and depression.

Most closely related is Hakanen, Schaufeli, and Ahola's (2008) study that tested the longitudinal association between burnout and depression and between work engagement and commitment, respectively. Unfortunately, they did not test the possible association between work engagement and depression directly even though their correlational data indicated a negative relationship between them. To the authors' knowledge, the possible analogue relationship between work engagement and anxiety has not been examined previously. The aim of the present study was to explore the longitudinal relationship between work engagement (vigour and dedication) and symptoms of anxiety and depression. Based on the assumption of the COR theory and previous studies, we formulated the following hypotheses:

Hypothesis 1. Vigour at Time 1 has negative cross-lagged effects on anxiety at Time 2.

Hypothesis 2. Vigour at Time 1 has negative cross-lagged effects on depression at Time 2.

Hypothesis 3. Dedication at Time 1 has negative cross-lagged effects on anxiety at Time 2.

Hypothesis 4. Dedication at Time 1 has negative cross-lagged effects on depression at Time 2.

However, the study of the association between work engagement and depression and between work engagement and anxiety is complicated by the possibility of reversed causation and the effect of unmeasured third variables (see de Lange, Taris, Kompier, Houtman, & Bongers, 2004; Dormann, 2001; Zapf, Dormann, & Frese, 1996). People with depression or anxiety often lose interest in previously valued social roles, and the feeling of engagement in work might fade. Billings et al. (1983) suggest that depressed patients are doubly disadvantaged since not only do they suffer from more social isolation, but also their lower level of social resources illustrates less stress-buffering potential relative to that available to community controls. Therefore, initial losses beget further losses (Hobfoll, 1989, 1998, 2001). Judge and Locke (1993) supported a reciprocal relationship between job satisfaction and subjective well-being among a sample of clerical staff working at the university. Based on the cognitive theory of depression, which focuses on an individual's thought processes, they argued that subjective well-being affects job satisfaction through the way individuals collect and recall information about their jobs. For example, people suffering from anxiety or depression might store, evaluate or recall information about their jobs in a dysfunctional matter ensuing in less work engagement.

Previous studies on self-reported stressor–strain relations hypothesized that the results may be explained partly by a third variable such as negative affectivity (e.g. Burke, Brief, & George, 1993). Unmeasured third variables may threaten the internal validity of non-experimental studies. In most instances, relevant third variables are unknown. An unknown third variable can have a stable character such as time pressure or unstable character such as the weather. For instance, the impact of good weather on an individual's emotional state (i.e. occasion factors) may reduce the level of depressive symptoms or increase the feeling of work engagement, respectively. Dormann (2001) proposed a ‘less restrictive synchronous common factor’ (LRSCF) model to investigate the potential impact of unmeasured third variables as a source of spuriousness. The LRSCF models are described in more detail in the ‘Method’ section.

By utilizing a 2-year longitudinal panel design and advanced statistical analyses such as structural equation modelling (SEM), the present study also tested for reversed cross-lagged relationship between the study variables and the potential effect of unmeasured third variables. Hence, the following hypotheses were formulated:

Hypothesis 5. Anxiety at Time 1 has negative cross-lagged effects on vigour at Time 2.

Hypothesis 6. Anxiety at Time 1 has negative cross-lagged effects on dedication at Time 2.

Hypothesis 7. Depression at Time 1 has negative cross-lagged effects on vigour at Time 2.

Hypothesis 8. Depression at Time 1 has negative cross-lagged effects on dedication at Time 2.

Hypothesis 5. Hypotheses 1–4 are valid even if unmeasured third variables are taken into account.


Sample and procedure

This study was conducted as part of a two-wave panel study in Norway. Representative national samples of eight occupational groups (lawyers, physicians, nurses, teachers, church ministers, bus drivers, people working in advertising and people working in information technology) were drawn by Statistics Norway. For each occupation, a random sample of 1000 people was drawn from the central Norwegian registers of employees and employment. Equal numbers of men and women were drawn from all occupations except for the population of church ministers, which contained 599 men and 401 women.

In both study phases, the questionnaires were mailed to the employee's home address together with a one-page cover letter stating the goals of the survey and ensuring confidentiality. In the first phase (T1), 5017 participants returned the questionnaire, yielding a response rate of 63%. After 2 years, a follow-up questionnaire (T2) was sent to those who responded at T1 and were still alive, not hospitalized or emigrated (N = 4969). Seventy per cent returned the questionnaire. Therefore, the final panel consisted of 3475 respondents including 412 lawyers, 523 physicians, 496 nurses, 504 teachers, 500 church ministers, 381 bus drivers, 301 employees in advertising and 358 employees in information technology. The mean age of the sample was 42 years, and the average working hours were 41, as compared with the Norwegian norm of 37.5 h a week.



Engagement was assessed by recoding the negatively framed items of the Norwegian version of the 16-item OLBI (Demerouti & Bakker, 2008). Although originally developed to assess burnout, the inclusion of positively and negatively phrased items makes it suitable to assess work engagement as well (see González-Romá et al., 2006). The instrument was translated by one of the authors, back translated by a bilingual German psychiatrist and compared with the English and Swedish versions of the instruments. The construct and convergent validity of the measure has been confirmed in previous validation studies (Demerouti, Bakker, Nachreiner, & Schaufeli, 2001; Demerouti, Bakker, Vardakou, & Kantas, 2003; Halbesleben & Demerouti, 2005). In the present Norwegian version, one of the items in the disengagement scale was changed. The original item, ‘I always find new and interesting aspects in my work,’ is disconnected from the particular strain context and perceived mainly as a measure of a general personality trait (i.e. positive thinker). The new item, ‘I am less interested in my job now than in the beginning,’ makes such an interpretation less likely without changing the factor structure.

Both the vigour and the dedication subscales were described by eight items each. Sample items are ‘I feel emotionally depleted by work’ (recoded positively; vigour) and ‘With time I have lost the deep interest in my job’ (recoded positively; dedication). The items were scored on a five-point scale ranging from totally disagree to totally agree. The internal consistencies of the variables (see Table 1) were satisfactory (α ≥ 0.86).

Table 1. Means, standard deviations (SD), Chronbach's α (on the diagonal in italics) and correlations for the study variables
  1. Note: T1: Time 1; T2: Time 2; all significant p < 0.01.
1. Vigour T13.280.770.87       
2. Vigour T23.310.770.610.88      
3. Dedication T13.710.800.540.330.86     
4. Dedication T23.720.800.360.580.580.88    
5. Anxiety T11.340.34−0.53−0.37−0.37−0.280.79   
6. Anxiety T21.330.35−0.37−0.52−0.24−0.380.630.81  
7. Depression T11.380.41−0.61−0.43−0.46−0.340.730.510.90 
8. Depression T21.350.41−0.42−0.59−0.30−0.460.510.730.640.90

Anxiety and depression

Anxiety and depression were measured with the Norwegian version of the Hopkins Symptom Checklist (SCL-25) derived from the 90-item Symptom Checklist (Derogatis, 1994). The SCL-25 is a 25-item self-report scale that includes symptoms of anxiety (10 items) and depression (15 items) accompanied by four-point scales ranging from 1 = not bothered at all to 4 = extremely bothered. For example, participants are asked to indicate how bothered they have been by shivering, headache, sudden fear without reason, loneliness, loss of appetite and sleep disturbance during the preceding month. The internal consistencies of the variables (see Table 1) were satisfactory (α ≥ 0.79). The SCL-25 has proven to have good psychometric properties as a measure of psychological distress (Strand, Dalgard, Tambs, & Rognerud, 2003).

Statistical analyses

Cross-lagged structural equation analyses, by means of lisrel 8.7 (Jöreskog & Sörbom, 2004), were used to test and compare various competing models of the relationship between work engagement and mental health across time. SEM has the advantage of determining causal priority and causal predominance when finding reciprocal relationships (de Lange et al., 2004). Cross-lagged techniques are designed to test causal structures in particular where measurements of the same variables have been made at two different times in the same sample (Edwards, Guppy, & Cockerton, 2007). Preliminary descriptive analyses indicated some deviations from normality in the anxiety and depression scale; hence, the models tests were based on the asymptotic covariance matrix (Jöreskog, Sörbom, du Toit, & du Toit, 2000; West, Finch, & Curran, 1995). Missing values were handled using the listwise procedure. Since the standard errors are estimated under non-normality, the Satorra–Bentler scaled chi-square statistic was applied as a goodness-of-fit statistic, which is the correct asymptotic mean even under non-normality (Jöreskog et al., 2000). However, due to the sensitivity of sample size in chi-square statistics (Diamantopoulos & Siguaw, 2000; Hair, Anderson, Tatham, & Black, 1998; Hu & Bentler, 1995; Sharma, 1996), the root mean square error of approximation (RMSEA), the non-normed fit index (NNFI) and the comparative fit index (CFI) were used as additional measures of fit. By convention, the model fit is considered to be excellent if the RMSEA is less than or equal to 0.05. A RMSEA less than or equal to 0.08 is considered to indicate a sufficient model fit to the observed data. The NNFI and CFI should be equal to or greater than 0.90 for the model to be accepted (Diamantopoulos & Siguaw, 2000).

Before comparing competing causal models and examining the hypothesized relationships in the cross-lagged structural equation analyses, the measurement model at T1 and T2 was tested by confirmatory factor analyses (CFA). The CFA provided a good fit to the observed data at T1 [χ2 (773) = 15,851.76, p < 0.001, RMSEA = 0.068, NNFI/CFI = 0.95/0.95] and T2 [χ2 (773) = 7144.63, p < 0.001, RMSEA = 0.052, NNFI/CFI = 0.97/0.97]. All parameter estimates were significant (p < 0.05) and loaded positively and clearly on their intended latent variable with factor loadings (standardized solution) ranging from 0.36 to 0.81 in T1 and from 0.40 to 0.86 in T2. To ensure measure invariance across time, two multigroup CFAs (T1 and T2) were performed and compared with each other, one with constrained parameters (invariant factor loadings) and the other with unconstrained parameters. Given the limitation of the χ2 statistics with large sample sizes and consistent with recommendations set forth in the literature (Chen, 2007; Cheung & Rensvold, 2002; Steenkamp & Baumgartner, 1998), we examined the change in CFI and two other practical fit measures, including the RMSEA and NNFI. With adequate sample size (N > 300), changes in CFI values of 0.01 or less and a change of ≥0.015 in RMSEA indicate factor invariance (Chen, 2007; Cheung & Rensvold, 2002). Metric invariance (i.e. equal factor loadings) across time was demonstrated, since the CFI (0.96) and NNFI (0.96) was the same in these two models, whereas the RMSEA improved slightly in the test of metric invariance (0.062 versus 0.061).

Because of a high correlation between anxiety and depression (r = 0.73), multicollinarity was considered a potential problem. Multicollinarity may occur when highly correlated factors are included in the same regression model and may provide unstable coefficients that are difficult to interpret (Cohen, Cohen, West, & Aiken, 2003). Although highly correlated factors might be due to spuriousity and at least in part accounted for by means of a LRSCF model, identification of such large models is complex (Dormann, 2001). Therefore, cross-lagged analyses were performed on work engagement and depression and work engagement and anxiety separately.

Stability models and causality models

The Stability Model (Mstabil) is a constrained model without cross-lagged effects but with temporal stabilities for all the latent variables at Time 1. The stability models and causality models are illustrated in Figure 1. The structural paths between the variables across time, e.g. vigour Time 1 to vigour Time 2, illustrate the assumed and estimated stability across time. As is conventional in SEM for longitudinal models, the measurement errors of indicators measuring the same factors on both occasions were allowed to correlate in all of the estimated models. The Causality Model (Mcausal) is identical to the Stability Model (Mstabil) except for the structural paths from the two work engagement variables, vigour and dedication at Time 1 to anxiety and depression at Time 2. Nested within the stability model, the Reversed Causality Model (Mrevers) has structural paths from anxiety and depression at Time 1 to the two work engagement variables vigour and dedication at Time 2.

Figure 1.

Stability and causal models for (A) work engagement (vigour and dedication) and depression and (B) work engagement (vigour and dedication) and anxiety. Double-headed arrows indicate correlation between Time 1 variables. One-headed arrows indicate stability Time 1–Time 2 variables and the structural paths in the causal and reversed causality models (error autocorrelations of observed indicators omitted for clarity). Note: T1, Time 1; T2, Time 2

Unmeasured third variables

Unmeasured third variables were ruled out by performing a series of LRSCF models as suggested by Dormann (2001). In these models, unmeasured third variables are expressed in terms of phantom constructs with their variance fixed at 1.0 for identification reasons. The first type of unmeasured variables tested were occasion factors, which are assumed to be completely unstable. According to Dormann (2001), occasion factors can be considered a special case of LRSCF in which the stability of the common factor has to be fixed at 0.0. In the second test, denoted the LRSCF model, it is assumed that the unmeasured variables are stable to some degree and their stability estimated (see Dormann, 2001; Dormann & Zapf, 2002). The occasion factor model and the LRSCF model are nested and can be compared statistically using a χ2 difference test.


Descriptive analysis

Table 1 displays the means, standard deviations, Cronbach's alpha and correlations for the two dimensions of engagement (vigour and dedication) and symptoms of anxiety and depression for Time 1 and Time 2. The alpha levels for the various measures indicate an acceptable level of inter-item consistency in the measures (Nunnally & Bernstein, 1994) with Chronbach's alpha coefficients of 0.79 or higher. Correlations between the measures were in the expected direction. The moderately high correlation between vigour, dedication, depression and anxiety are in line with previous studies (i.e. Toker, Shirom, Shapira, Berliner, & Melamed, 2005) and can be explained in part by the synchronism between these affective states. Nevertheless, the high correlation between anxiety and depression (r = 0.73) was considered a potential problem, and causal analyses were performed for work engagement (vigour and dedication) and depression and for work engagement (vigour and dedication) and anxiety separately. The across-time stability of these variables was high, with test–retest (Time 1–Time 2) correlations ranging from 0.58 (for dedication) to 0.64 (for depression).

Causal relationships

In order to test causal relationships, the results of the three competing structural models for the depression model and anxiety model were compared. Table 2 presents the fit indices of the competing causal models under the heading ‘baseline model’. The fit of the causal models were all satisfactory (RMSEA ≤ 0.05 and NNFI, CFI ≥ 0.90). The chi-square differences test indicates that the causality model fit the data significantly better than the stability model for depression (Mstabil versus Mcausal: Δχ2 (2) = 9.86, p < 0.01) and anxiety (Mstabil versus Mcausal: Δχ2 (2) = 8.57, p < 0.025). In contrast, the reversed causality model increased the chi-square significantly as compared with the model without reversed effects for depression (Mstabil versus Mrevers: Δχ2 (2) = −26.04, p < 0.001) and anxiety (Mstabil versus Mrevers: Δχ2 (2) = −6.45, p < 0.05). Therefore, the causal model is superior to the reversed causal model. In the causal model testing the cross-lagged relationship between the two dimensions of work engagement and depression, vigour explained 43.7% of the variance, dedication explained 37.4% of the variance and depression explained 46.1% of the variance. The equivalent numbers for the causal anxiety model were 43.4% (vigour), 37.2% (dedication) and 44.6% (anxiety).

Table 2. Goodness-of-fit indices for the baseline model, the occasion factor model and the LRSCF model
Model descriptionχ2dfRMSEANNFICFI
  1. Note: All chi-square values significant at p < 0.001; coefficients and numbers refer to model fit indices: χ2: Satorra–Bentler chi-square; df: degrees of freedom; RMSEA: root mean square error of approximation; NNFI: non-normed fit index; CFI: comparative fit index; LRSCF: less restrictive synchronous common factor.
Cross-lagged relationships between work engagement (vigour and dedication) and depression
Baseline model (third variables not accounted for)
MstabilStability model9929.4017920.0410.980.98
McausalCausality model (work engagement → depression)9919.5417900.0410.980.98
MreversReversed causality model (depression → work engagement)9955.4417900.0410.980.98
Occasion factor model (non-stable third variables accounted for)
MOcausalCausality model (work engagement → depression)8501.4417870.0370.980.98
MOreversReversed causality model (depression → work engagement)8517.0617870.0370.980.98
LRSCF model (stable third variables accounted for)
MLcausalCausality model (work engagement → depression)8501.8617860.0370.980.98
MLreversReversed causality model (depression → work engagement)8513.7517860.0370.980.98
Cross-lagged relationships between work engagement (vigour and dedication) and anxiety
Baseline model (third variables not accounted for)
MstabilStability model9376.0112420.0490.970.97
McausalCausality model (work engagement → anxiety)9367.4412400.0490.970.97
MreversReversed causality model (anxiety → work engagement)9382.4612400.0490.970.97
Occasion factor model (non-stable third variables accounted for)
MOcausalCausality model (work engagement → anxiety)8235.5712370.0450.970.97
MOreversReversed causality model (anxiety → work engagement)8236.2512370.0450.970.97
LRSCF model (stable third variables accounted for)
MLcausalCausality model (work engagement → anxiety)8232.3312360.0450.970.97
MLreversReversed causality model (anxiety → work engagement)8233.6812360.0450.970.97

The standardized path coefficients linking vigour and dedication to depression and anxiety are shown in Table 3. Vigour at T1 had a negative cross-lagged effect on depression at T2 (β = −0.12) and anxiety T2 (β = −0.09) as expected. Similarly, the hypothesized reversed negative effect was supported since depression at T1 had a negative cross-lagged effect on vigour (β = −0.11) and dedication (β = −0.10) at T2. Moreover, people high in symptoms of anxiety at Time 1 reported less vigour (β = −0.06) and dedication (β = −0.07) 2 years later. In contrast to what was hypothesized, dedication at Time 1 had a positive lagged effect on anxiety Time 2 (β = 0.06). The association between dedication Time 1 and depression Time 2 was not significant.

Table 3. Standardized path coefficients from the three models (baseline, occasion and LRSCF)
 Causality modelReversed causality model
  1. Note: *Indicate statistical significance determined by the t value of the beta coefficients. T1: Time 1; T2: Time 2; VIG: vigour; DED: dedication; DEPR: depression; ANX: anxiety.
Baseline model−0.12*0.01−0.09*0.06*−0.11*−0.10*−0.06*−0.07*
Occasion factor model−0.08*0.00−0.050.04−0.09*−0.07*−0.03−0.04
LRSCF model−0.050.00−0.030.04−0.05−0.03−0.02−0.02

Accounting for unmeasured third variables

In the next step, we wanted to test whether the hypothesized relationships remained significant when controlling for stable (occasion factors) and unstable (LRSCF) unknown third variables. The fit indices of these models are reported in Table 2. In general, the occasion factor models and the LRSCF models provided better fit indices than the baseline models not including unknown third variables. A series of chi-square differences tests between the occasion factor models and the LRSCF models, which are nested, indicated that none was significantly superior to the other. Thus, both stable and unstable unmeasured third variables are likely to explain some of the relationship between vigour and dedication and symptoms of anxiety and depression. Inspection of the standardized path coefficients in Table 3 revealed that all coefficients were smaller than the baseline model, and only the path linking vigour Time 1 to depression Time 2 (β = −0.08), and depression Time 1 to vigour (β = −0.09) and dedication (β = −0.07) at Time 2 remained significant in the occasion factor model.


The aim of the present study was to explore the longitudinal association between work engagement (vigour and dedication) and symptoms of depression and anxiety. By doing so, we sought to contribute to occupational health psychology in three ways. Firstly, by exploring work engagement, the positive antipode to burnout, the present study supplies empirical knowledge to the rather new field of ‘positive psychology’ where the concern is the positive aspects of employee health (Seligman & Csikszentmihalyi, 2000; Seligman et al., 2005). Secondly, research focusing on engagement is still in its infancy. In general, there are few longitudinal studies on the health impairment process with depression and other indicators of mental health as outcome variables (Hakanen, Schaufeli, et al., 2008). There is a dearth of research particularly on the relationship between work engagement and health (Bakker et al., 2008). By examining the longitudinal effects of work engagement on health, new approaches to health promotion can emerge. Thirdly, by means of a large panel sample and advanced statistical analyses, the present study could rule out alternative explanations for the hypotheses under study such as reversed causation and unknown third variables. This often has been considered a shortcoming in many published longitudinal studies (Dormann & Zapf, 2002).

In general, the model suggesting work engagement to be the antecedent variable and symptoms of depression and anxiety to be the outcome variables was superior to the hypothesized reversed causation model. Even though we found significant paths in the reversed causal model, this model did not fit significantly better than the model without reversed effects. This finding is in line with a related study that found that burnout predicted depression but not vice versa (Hakanen, Schaufeli, et al., 2008).

Specifically, the lagged negative effect of vigour on symptoms of depression 2 years later indicates that work engagement may be an important modifier of health. These results add to the mounting evidence of the benefits of subjective well-being at work (for a review, see Russell, 2008). The negative association between vigour and symptoms of depression also provides a positive approach to health promotion. Therefore, in addition to fixing what is wrong (risk prevention), human resource management interventions should focus on means to make workers flourish (health promotion) as suggested by Taris, Cox, and Tisserand (2008). The WHO's World Health Report (2001) estimated that by 2020, depression is expected to be the highest ranking cause of disease in the developed world. Therefore, finding new and alternative approaches to weaken this detrimental development is highly warranted. The present study suggests that finding ways to enhance the individual's vigour at work might be beneficial in this matter. The link between dedication and symptoms of depression 2 years later were not significant, even though the inspection of the correlation coefficients indicate a negative association between these two variables. The lack of significant findings may be due to the quite high standardized stability coefficients of the variables over time (ranging from 0.61 to 0.68), which does not leave much room for additionally explained variance.

The hypothesized relationship between work engagement (vigour and dedication) and anxiety was partly supported. As hypothesized, vigour at T1 was negatively associated with anxiety at T2. Therefore, high level of vigour at T1 was associated with low levels of anxiety 2 years later. Just like vigour might ease symptoms of depression, our study suggests that the feeling of vigour at work might have beneficial longitudinal effects on anxiety. In contrast to what was hypothesized, dedication at T1 was positively associated with symptoms of anxiety 2 years later. Thus, being highly dedicated to work may produce more symptoms of anxiety and not less, as hypothesized. This might indicate a ‘golden girl’ phenomenon (see Taris, 1999), which implies that a strong dedication to work might inflate future expectations unrealistically, causing unmet expectations, stress and ultimately impaired mental health. This indicates that there might be different processes operating in vigour and dedication, and that they may have different outcomes. However, when accounting for unmeasured third variables, this link did not remain significant. In fact, the models including unmeasured third variables—both unstable occasion factors and LRSCF—fit better than baseline models without unmeasured third variables. This indicates that unknown third variable explanations apply in the present study. In line with previous studies (i.e. Dormann & Zapf, 2002), controlling for unmeasured third variables leads to a reduction of effects. Only three of eight hypothesized relationships remained significant when occasion factors were accounted for. In the LRSCF model, no effects remained significant.

Strengths and limitations

Few longitudinal studies in occupational health psychology have used the potential that lies in the design to test for reversed causation and unmeasured third variables (de Lange et al., 2004; Dormann & Zapf, 2002). Studies on longitudinal relationships between engagement and health are even fewer (Bakker et al., 2008). The present study expands previous studies by testing reciprocal relationships between work engagement and symptoms of anxiety and depression in a large panel sample using structural equation modelling. Nevertheless, the findings of this study must be discussed with some limitations in mind.

Firstly, although superior to cross-sectional studies, longitudinal designs also have drawbacks such as testing effects (i.e. the respondents lose interest or are more sensitive the second time) and selective attrition (Taris & Kompier, 2003). The analysis may be biased by a healthy worker effect, since only healthy workers might have remained in the second survey. If this is the case, the strength of the associations found among these variables may have been underestimated. A second limitation concerns the use of self-reported data, which implies a certain risk that the findings are based on common-method variance (Podsakoff, MacKenzie, Jeong-Yeon Lee, & Podsakoff, 2003). Thirdly, a 2-year interval was used between the two waves. Even though Dormann and Zapf (2002) found a 2-year interval to be the most optimal time lag between social stressors and depression, we have little evidence regarding the corresponding length of the time between work engagement and mental health. Therefore, we can only generalize our results in relation to this measurement interval (Taris, 2000). Ideally, in a study with different time lags, a more complete understanding of the nature of such effects would be explored. Finally, due to identification problems involved in such complex models, it was impossible to estimate anxiety and depression in the same analysis. In the present study, the effects found were rather small or non-significant. However, the relative high stability of the variables across time (ranging from 0.61 to 0.68) leaves only a small part of the variance to be explained by additional factors. Thus, a 2-year lagged effect up to −0.12, the same as the average of the best longitudinal studies published so far (Zapf et al., 1996), indicates that the positive effect of work engagement and vigour in particular is likely to be of much importance.

Implications and future research

According to the European Commission's Green Paper on mental health [COM (2005) 484], anxiety and depression are two of the most prevalent mental health problems facing European citizens today. Taking into account the highly chronic nature of both psychological states, we consider our findings noteworthy in their suggestion that engagement might impair the level of anxiety and depression. Knowledge of how engagement and depression relate to each other in this respect is important for researchers, work organizations, and clinicians. Firstly, support for a cross-lagged relationship between work engagement and health might engender theory building and provide valuable aspects to different stress and motivation models like the Job Demand–Resources model by suggesting a possible link between the motivation and impairment processes through work engagement (see Hakanen, Schaufeli, et al., 2008; Schaufeli & Bakker, 2004). Clinically, the present study suggests that to strengthen an individual's feeling of vigour at work may be a potential target of intervention because it not only enhances important organizational outcomes such as work performance and organizational commitment (Hakanen, Schaufeli, et al., 2008; Russell, 2008) but might also impair the development of anxiety and depression. Therefore, increasing employees' work engagement could be related to a host of benefits for the employee, the firm and the country. Building on this knowledge, interventions may be designed to improve employees' positive, fulfilling and affective motivational states of work-related well-being. There are several strategies to promote well-being at the workplace such as work design, teams and work groups, and transformational leadership (for a discussion, see Russell, 2008). More research on the effectiveness of such strategies is highly needed. Nevertheless, this study urges further researchers and practitioners to find positive approaches to obtain and maintain good mental health and well-being.


The data collection was funded by the Research Institute of the Norwegian Medical Association. We would like to thank Professor Dr. Christian Dormann for his interest and help with the analysis of the LRSCF model.