Work and Stress Research Group, Centre for Applied Psychological Research, University of South Australia, Adelaide, Australia
Correspondence should be addressed to Professor Maureen F. Dollard, School of Psychology, University of South Australia, City East Campus, GPO Box 2471, Adelaide, South Australia 5001, Australia (e-mail: email@example.com).
Correspondence should be addressed to Professor Maureen F. Dollard, School of Psychology, University of South Australia, City East Campus, GPO Box 2471, Adelaide, South Australia 5001, Australia (e-mail: firstname.lastname@example.org).
We propose and test a comprehensive theory designed to explain seemingly contradictory relations between job demands, emotional exhaustion, and work-family conflict (WFC) reported in the literature. Using job demands-resources theory, effort-recovery theory, and personal resources theory we hypothesized that job demands would spillover to emotional exhaustion as mediated by WFC (causality model), and alternatively that job demands would also spillover to WFC as mediated by emotional exhaustion (reverse causal model). Further, we also hypothesized using loss spiral theory that a more comprehensive model representing reciprocal and cross-linked effects (causal and reverse causal simultaneously) would best fit the data. The hypotheses were tested in a longitudinal study of 257 Australian (Victorian) frontline police officers at two time points approximately 12 months apart. We used structural equation modelling and found in support of the simultaneous reciprocal effects hypothesis, that the more comprehensive model fitted the data better than either the causality or the reverse causal model. Future research should more comprehensively model the important relationships between job demands, emotional exhaustion, and WFC to reflect their complex interplay. Interventions to reduce work demands arising from work pressure and emotional demands are indicated to prevent conflict at home and emotional exhaustion in police officers.
Police work is often associated with high strain as evidenced by the severe psychological and physiological problems experienced by officers, such as burnout (including emotional exhaustion), marital problems, alcohol and drug abuse, heart disease, and suicide (Gaines & Jermier, 1983; Hart, Wearing, & Headley, 1995). Furthermore, the crossover (i.e. to spouses) and spillover (i.e. into other domains) effects of job demands (Bakker, Demerouti, & Dollard, 2008) and emotional exhaustion can be pervasive, negatively affecting police officers' personal lives, and contributing further to work-family conflict (WFC; Anshel, 2000). For police officers in particular, WFC is shown to affect their ability to perform their duties, with WFC contributing to increased absenteeism and turnoverrates (Howard, Donofrio, & Boyles, 2004; Johnson, Todd, & Subramanian, 2005; Mikkelsen & Burke, 2004). We therefore investigate the complex interplay between job demands, emotional exhaustion, and WFC in this longitudinal study of Australian frontline police officers.
WFC refers to ‘a form of interrole conflict in which the role pressures from work and family domains are mutually incompatible in some respect’, and is thought to comprise two separate domains: (1) WFC and (2) family-work conflict (Carlson, Kacmar, & Williams, 2000; Greenhaus & Beutell, 1985, p. 77). Because previous literature has shown that work factors are more strongly related to WFC than to family-work conflict, we focused on WFC in this study (e.g. Byron, 2005; Carlson et al., 2000).
Consistent with other research, studies on burnout within police find that emotional exhaustion, defined as feelings of being emotionally over-extended and exhausted, is the core explanatory dimension of burnout (Maslach-Pines & Keinan, 2006; Michinov, 2005; Shirom, 1989; Wright & Bonett, 1997). In this study, we start from the premise that WFC and emotional exhaustion have their genesis in the work environment. Hence all our propositions begin with specific job demands and predict how they spillover to WFC or emotional exhaustion among police officers. However, there are seemingly contradictory findings in the literature and we intend to sharpen the interpretation of these in a coherent theoretical framework.
Most empirical evidence in support of the propositions is based on cross-sectional studies and therefore does not assess possible reverse causal and reciprocal relationships. Indeed, because stressors can sometimes be affected by strain and vice versa, the issue of confounded antecedents and consequences is a valid criticism within the work-family interface literature (e.g. Frone, Quick, & Tetrick, 2003). This is exemplified in the relationship between social stressors and depression where depressed people, due to their negative mind-set, can inadvertently interact negatively with their environment and contribute to a more negative culture (Beck, 1972; Zapf, Dormann, & Frese, 1996). However, little research has investigated the influence of reciprocal effects in stress research generally, and specifically in relation to WFC. Furthermore, Demerouti, Bakker, and Bulters (2004) recommend models incorporating reciprocal relations between all constructs of the work-family interface as far more plausible representations of reality. Next, we explain the theoretical and empirical rationale for the propositions, and for reciprocal links between them.
Regarding the first proposition, the job demands→WFC→emotional exhaustion, pathway is well established in the literature (see Byron, 2005, for a recent meta-analysis). Theoretically, the relationship can be explained by drawing on Hobfoll's (2002) personal resources perspective that proposes that high levels of demands at work require one to focus personal resources in this area leaving fewer resources to tackle demands in other areas, such as those in the family domain. In turn the added conflict at home can lead to emotional exhaustion (Kossek & Ozeki, 1999; Netemeyer, Boles, & McMurrian, 1996). This classical view that job demands spillover to WFC, and consequently strain, can be similarly explained by the effort-recovery (E-R) model (Meijman & Mulder, 1998). Because pressures at work hamper functioning at home, there may be insufficient opportunity for recovery outside of the workplace after exposure to high job demands with the stressor/strain processes accumulating in a loss spiral giving workers little chance to stabilize and recover (Demerouti et al., 2004).
There is also strong theoretical evidence for the second proposition, job demands→emotional exhaustion→WFC. The first component that high job demands lead to emotional exhaustion is modelled as a key assumption in the job demands-resources (JD-R) model of burnout (Bakker & Demerouti, 2007). The JD-R model draws on conservation of resource theory and specifically personal resource theory (Hobfoll, 2001, 2002) to explain an erosion process whereby job demands tax personal energy resources over time leading to emotional exhaustion (Bakker et al., 2007). Again Hobfoll's (2002) personal resources perspective can be used to propose that a state of emotional exhaustion will leave fewer resources to manage demands in other areas, such as in the family domain.
In fact, several studies indicate the possibility of WFC as an outcome of the negative effects of emotional exhaustion (e.g. Demerouti et al., 2004; Kelloway, Gottlieb, & Barham, 1999; Westman, Etzion, & Gortler, 2004). For example, Burke and Mikkelsen (2006) found that police officers who experienced higher levels of emotional exhaustion also reported more WFC in a large Norwegian sample. Also in a study of Australian policewomen, Thompson et al. (2005) found that ‘work stress impacted on the family environment through emotional exhaustion’ (p. 204). Both these studies suggest examination of emotional exhaustion could prove fruitful in the exploration of stress transmission from job demands to the family environment.
Theoretically, the link between Propositions 1 and 2, can be explained by the principle of loss spirals (Demerouti et al., 2004; Hobfoll, 2001). According to the conservation of resources theory, individuals strive to conserve valuable resources, such as conditions, rewards, and personal energy. However, initial resource loss renders them more vulnerable to future loss. This principle proposes that employees who lack resources in the work, personal, or home domains are most vulnerable to a downward spiral of additional losses in other domains, because resources are interlinked in a web-like structure (Demerouti et al., 2004). Specifically, Demerouti et al., found support for the loss spiral hypothesis in a sample of 335 employment agency employees.
Applied to our study, high levels of job demands could tax resources leading simultaneously to spillover in both the personal (emotional exhaustion) and family (WFC) domains as explained above. In turn, additional losses are expected, such that WFC adds an additional burden to employees facing a resource drain, leading to emotional exhaustion, and similarly emotional exhaustion leads to further WFC. Indeed, the negative influence of loss spirals was shown in a recent study that found reciprocal effects between WFC and burnout in a longitudinal study among a large Norwegian sample (N = 2,235) of respondents from eight different occupational groups (Innstrand, Langbelle, Espnes, Falkum, & Aasland, 2008).
Job demands at Time 1 will be positively related to emotional exhaustion at Time 2 mediated by WFC at Times 1 and 2 (causality model M2i).
Hypothesis 2 (H2)
Job demands at Time 1 will be positively related to WFC at Time 2 mediated by emotional exhaustion at Times 1 and 2 (reverse causal model M3i).
Hypothesis 3 (H3)
Job demands at Time 1 will be positively related to both emotional exhaustion and WFC at Time 2, mediated by WFC and emotional exhaustion at Times 1 and 2 (simultaneous reciprocal model M4) and this model will fit the data better than either the H1(M2i) or the H2 (M3i) model.
Participants and procedure
Data were collected via self-report questionnaires (Dollard, Saebel, Chrisopoulos, Tuckey, & Winefield, 2007) at two time points approximately 12 months apart. Participants were frontline police officers from the Australian state of Victoria who were sent 3,250 questionnaires of which 716 (22%) were returned at the first time point, and 518 (16%) at the second that contained usable data. The sample was 257 police officers (230 male, 89.5% and 27 female, 10.5%) who could be matched between the two time points. Their average age was 42.64 years (SD = 8.09), with 86% married or de facto, and 71% having dependants.
The officers held the following rank: Probationary Constable (N = 10, 3.9%); Constable (N = 13, 5.1%); Senior Constable (N = 77, 30%); Leading Senior Constable (N = 14, 5.4%); Sergeant (N = 112, 43.6%); and Senior Sergeant (N = 31, 12.1%). Because of the low response rate we sought to establish the representativeness of the sample by comparing it with published demographic data for the Victorian Police Force (Australian Institute of Criminology, 2003). Using chi-square test of association, we found no significant differences between the observed and expected distribution by gender and rank but found that sergeants were overrepresented and senior constables underrepresented in the sample.
To test the hypotheses, we used the rationale of the triple-match principle (TMP) to inform construct measurement choices (de Jonge & Dorman, 2006). The TMP proposes that when stressors, resources, and strains are matched effectively on the same qualitative dimensions (i.e. cognitive, emotional, and physical) the strongest interactive effects will be evident. To strengthen the possibility of effects, we adopted this principle and selected specific strain- and time-based job demand variables that matched the potential spillover effects of specific strain-based and time-based WFC. Behaviour-based WFC was not measured in the survey in the WFC Time 1 scale and so could not be matched to the WFC Time 2, and hence it was not used in this study.
WFC at Time 1 was measured using a four-item scale based on Holahan and Gilbert (1979). Items were measured on a five-point Likert scale ranging from 0 (very rarely/never) to 4 (very often/always), for example: ‘How often do the demands of your work interfere with your family life?’ To improve the domain sampling of the WFC measurement (Hinkin, 1995) to match the demand domains (see below), we measured WFC at Time 2 using domain specific scales (Carlson et al., 2000). We used six items representing strain-based (e.g. ‘Due to the pressures at work, sometimes when I come home I am too stressed to do the things I enjoy’), and six items representing time-based (e.g. ‘My work keeps me from my family activities more than I would like’) WFC. Responses were on a five-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). Together, these scales were indicators for the latent measure, WFC.
Emotional exhaustion was measured at Times 1 and 2 using five items from the Maslach Burnout Inventory (Schaufeli, Leiter, Maslach, & Jackson, 1996). Items were measured on a 7-point Likert scale, ranging from 1 (never) to 7 (always). An example item was: ‘I feel emotionally drained from my work’.
All demands were assessed at Times 1 and 2. Strain-based job demands were assessed with items from The Demand-Induced Strain Questionnaire(DISQ1.0; de Jonge et al., 2004) and were measured on a 5-point Likert scale, from 1 (very rarely never) to 5 (very often/always). Four psychological (e.g. ‘need to display high levels of concentration and precision at work’) and six emotional demand items (e.g. ‘have to deal with people such as clients, colleagues or supervisors who have unrealistic expectations’) were included.
Critical incidents were assessed separately as an indicator of emotional demands also measured on the same five-point Likert scale, and included three items such as: ‘Have you experienced prolonged exposure to any unpleasant duties?’ Time-based job demands included one item from the psychological demands section concerned with time: ‘Solve work related problems within a limited time frame’ (de Jonge et al., 2004), and three items regarding workload (Richman, Flaherty, & Rosependa, 1996), for example: ‘I work under a great deal of pressure and deadlines’. We also included an item regarding overtime: ‘How often do you work unpaid overtime?’ These items were measured on the same five-point Likert scale outlined above. Together, these measures were indicators for the latent measure, job demands.
To test the hypotheses, three plausible models were compared utilising SEM with AMOS 6 (Arbuckle, 1997). SEM is highly recommended to analyse longitudinal data (Zapf et al., 1996). The models testing the hypotheses were nested within a Null model (M1) that encompassed stabilities between all the indicator and latent variables by co-variation of all the latent residual and manifest error variances, respectively (e.g. Zapf et al., 1996). Further negative error variances incurred on the single indicator latent variables were accounted for by setting the error variances to the product of its variance multiplied by one minus the estimated reliability factor (Bollen, 1989).
To test mediation effects, pathways were nested within the Null model as per: job demands Time 1→WFC Time 1→WFC Time 2→emotional exhaustion Time 2 with the direct job demands Time 1→emotional exhaustion Time 2 constrained to zero to ascertain the significance of the mediation pathways (Holmbeck, 1997). Note that as this is a two wave longitudinal study examining the mediating relationships between three variables certain synchronous effects were unavoidable. Therefore, the main results of this study reported on the accumulative affects of the mediating variables WFC and emotional exhaustion at both Times 1 and 2 as per a mixed lagged-synchronous model (e.g. Frese et al., 2007). By examining the accumulative effects of emotional exhaustion at Times 1 and 2 as mediating the relationship between job demands at Time 1 and WFC at Time 2 we could ascertain the overall effect of the spillover of job demands to emotional exhaustion on to WFC and vice versa for WFC as the mediator between job demands and emotional exhaustion. Chi-square difference analysis (Joreskog & Sorbom, 1993) was used to ascertain whether the hypothesized models M2i, M3i, and M4 fitted the data better than the baseline Null model (M1). As there was no difference in degrees of freedom between the hypothesized models because mediation variables only alternated, fit indices were used to compare models M2i, M3i, and M4 that tested H1, H2, and H3.
Specifically, we used the: goodness of fit index (GFI), adjusted goodness of fit index (AGFI), comparative fit index (CFI), the root mean square error of approximation (RMSEA), and the Akaike information criterion (AIC). In general, models with fit indices >.90, and RMSEA <.08 indicate a good fit and adjusting the chi-squared for the number of estimated parameters, the AIC (lowest value indicates best fit) allows for comparison of models that need not be nested (Schermelleh-Engel, Moosbrugger, & Muller, 2003). As indicated by Byron (2005) demographic variables, such as sex and marital status are alone poor predictors of WFC. This was also found in this study, with no significant correlations between these variables, WFC and emotional exhaustion, hence demographic variables were excluded from the analysis.
Descriptive statistics and mediation effects
Table 1 shows the descriptive statistics including the means, standard deviations, intercorrelations, and reliability (Cronbach's alpha) coefficients. All but one of the correlations (WFC Time 2 - unpaid overtime Time 1) were significant, with most at the p <.01 level (two-tailed). Job demands at Time 1 correlated with WFC and emotional exhaustion at Times 1 and 2, and WFC and emotional exhaustion correlated at Times 1 and 2 indicating preliminary support for mediation analysis. The scales all had good reliability with all Cronbach's alpha coefficients exceeding .70 (Nunnaly & Bernstein, 1994).
Table 1. Means, standard deviations (SD), Cronbach's alpha (on the diagonal), and correlations between variables, N = 257
Model fit to data
All indicators loaded significantly in expected ways on their respective latent variables. Hypotheses H1, H2, and H3 were confirmed with models M2i, M3i, and M4 fitting the data better than the baseline Null model (M1) by the chi-square difference results: M2i (Δχ2(1) = 89.24, p <.001); M3i (Δχ2(1) = 94.80, p <.001); and M4 (Δχ2(1) = 123.5, p <.001). The fully mediated pathways of models M2i and M3i (direct paths constrained to zero) did not diminish with the direct paths job demands Time 1→emotional exhaustion Time 2: M2ii (Δχ2(1) = 1.71, ns), and job demands Time 1→WFC Time 2 M3ii (Δχ2(1) = 0.39, ns) not constrained to zero.
Inspection of parameter estimates also revealed that the direct paths were not significant (γ = .09, p = .185; γ = .05, p = .514) when not set to zero in the models including full mediation paths, thus demonstrating the strength of the mediated pathways. In summary, Hypotheses 1 and 2 were supported by models M2i and M3i (direct paths constrained to zero) that demonstrated the significance of the mediated pathways in comparison to the baseline Null model M1 and the direct paths models M2ii and M3ii (see Table 2). In this study, it was hypothesized in H1 and H2 that the accumulative affects of WFC and emotional exhaustion measured over two time-spans (Times 1 and 2) would mediate the relationship between job demands at Time 1 on each other at Time 2, respectively. Importantly, further analysis also showed that WFC at Time 2 fully mediated the relationship between job demands at Time 1 and emotional exhaustion at Time 2 with this model fitting the data better than the baseline null (Δχ2(1) = 66.32, p <.001). Similarly emotional exhaustion at Time 2 fully mediated the relationship between job demands at Time 1 and WFC at Time 2 (Δχ2(1) = 54.80, p <.001). However, as much as these results serve to back up the mediation hypotheses, the models accounting for the accumulative affects of the mediating variables at both time waves provided a better fit to the data.
Table 2. Chi-square analysis of structural equation models, N = 257
Hypothesis 3 is supported by model M4 that shows when reciprocal mediated pathways are present simultaneously the model fits the data better than either of the separate mediated models M2i and M3i. M4 had better fit indices: RMSEA = .08; GFI = .92; IFI = .94; NFI = .92; CFI = .94; and an AIC substantially better at 274.96. As can be seen in Figure 1 illustrating model M4, the reciprocal paths are all positive and significant. Again, results of an additional model analyzing the non-accumulative affects of the mediating variables WFC and emotional exhaustion at Time 2 on each other at Time 2 revealed a significant difference from the baseline null (Δχ2(3) = 78.17, p <.001) with an AIC of 324.30, but it was not as significant as the reciprocal accumulative affects tested in the M4.
Additionally, we tested the model with reciprocal cross lagged paths between WFC Time 1 and emotional exhaustion Time 2 (γ = .43, p <.001) and emotional exhaustion Time 1 and WFC Time 2 (γ = .50, p <.001) and found significant positive pathways between them. However, the pathways of this model did not have better fit indices overall than the M4 model, further supporting Hypothesis 3 which includes the simultaneous reciprocal mediated effects showing the loss spiral of job demands, WFC and emotional exhaustion.
Despite growing research on the important impacts of job demands in the work-family interface there is a lack of coherent theory to explain the seemingly contradictory propositions that job demands lead to both WFC→emotional exhaustion and emotional exhaustion→WFC. Using conservation of resource theory and specifically personal resource theory (Hobfoll, 2001, 2002), JD-R theory (Bakker & Demerouti, 2007); effort-recovery theory (Meijman & Mulder, 1998), and loss spiral theory (Hobfoll, 2001) we proposed a comprehensive process whereby job demands simultaneously lead to WFC and emotional exhaustion in police officers (see also Demerouti et al., 2004). We utilised SEM to assess causal mediated, reverse causal mediated, and simultaneous reciprocal mediated pathways between job demands, WFC, and emotional exhaustion in a longitudinal study of Australian frontline police officers.
Other literature has also stressed the importance of reciprocal effects, and this study has shown with the use of longitudinal data the power of reciprocal mediating effects between job demands, WFC, and emotional exhaustion (e.g. Demerouti, et al., 2004; Ford et al., 2007; Frese et al., 2007). In accord with Demerouti et al. (2004) theoretically this means that WFC and emotional exhaustion are best explained by a complementary theory indicating reciprocal directions and cross-links between seemingly contradictory pathways, explained through a loss spiral. Rather than concluding one theory proposing stressor→strain directions only, implications of this study suggest future research should address the importance of reciprocal mediated relationships extending the traditional stressor→strain paradigm to include a stressor→strain→stressor paradigm in order to examine job demand spillover effects in more comprehensively.
Some limitations of this study should be noted. The data were collected using self-report questionnaires, which can lead to common-method effects (e.g. Podsakoff, MacKenzie, Lee, & Podsakoff, 2003), although longitudinal assessments are expected to offset this limitation. Also, we used a more comprehensive measure of WFC at Time 2. Relationships with WFC within time may have increased in size due to the domain coverage of the measures. Also, it is likely that the stability of the measure was reduced. This would have the effect of enabling stronger relationships with WFC at Time 2 in the study model, because less variance is accounted for by WFC T1. However, we do not think this has an impact on the overall conclusion as the best fitting model was the simultaneous reciprocal effects model. Although the sample was not representative by rank of sergeant and senior constable, overall we do not believe this would effect the relational conclusions drawn in this study.
Practical implications are that increased job demands not only spillover to WFC, preventing recovery and influencing emotional exhaustion, but that the strain of emotional exhaustion can also build-up at work influencing a potential loss spiral of spillover to home life contributing to WFC. Specifically, for police, more work needs to be done to prevent spillover from job demands, emotional exhaustion and WFC within the work-family interface in order to stem the high rates of marital discord and divorce (Howard et al., 2004). JD-R theory proposes that adequate resources, job-related support, and job control could reduce the experience of job demands in the first instance. The addition of resources in the comprehensive model as proposed would be a fruitful line of enquiry of significant practical importance.
This research is funded by the Australian Research Council (ARC Grant Number: LP0562310) and The Police Association Victoria. The authors also gratefully acknowledge Judith Saebel for her assistance with structural equation modeling.