Appendix A
Preliminary analyses results on factor structure using EFA and CFA.
Using Principal Component Analysis (PCA), we performed a set of exploratory factor analyses in which we extracted a fixed number of factors. We did this separately for all items on job demands (expected to result in three factors: task load, team process load, and conflict with team members), job resources (expected to result in the factors of team social support and job autonomy), and well-being (expected to result in two factors: job strain and work engagement). The pattern matrices that followed from the PCA with the direct Oblimin rotation method are shown below. As can be seen, the results provided preliminary support for our hypothesized factor structure. All items had factor loadings above .30 on their respective factors and no cross-loadings were found.
To further test the proposed underlying dimensions of our instruments, we next followed a confirmatory approach with five distinguishable measurement models. We first tested the unidimensional model in which all items were seen as indicators of a single underlying factor. We then compared this model to a three-factor model in which we distinguished between job demands, job resources, and well-being as three latent factors. The single factor model was nested in this three-factor model (Kline, 2005) and we could therefore compare the global model fit statistic (χ²) of the models to test whether the constrained single factor model showed a significantly worse fit to the data compared with the three-factor model. The chi-square difference test was significant (Δχ²(3)= 912, p< .001), pointing to the three-factor structure as the best fitting model.
Our further analyses involved three steps, namely distinguishing between 1) three job demands, 2) two job resources, and 3) two indicators of well-being. The five-factor model that distinguished between task load, team process load, and conflict with team members as three separate job demands fitted the data significantly better than the three-factor model that clustered these job demands in one latent factor (Δχ²(7) = 360.9, p< .001). Moreover, the six-factor model that included social support from team members and job autonomy as distinguishable latent factors provided a better fit to the data than the model that clustered these factors in one ‘job resource’ dimension (Δχ²(5) = 723.2, p < .001). Finally, the chi-square difference test comparing the six-factor model to a seven-factor model showed that job strain and work engagement should be treated as separate indicators of well-being (Δχ²(6) = 321.5, p < .001).
The Akaike Information Criterion (AIC) compares nested models and aims at choosing a parsimonious model that is still acceptable in terms of the chi-square value. The AIC statistic supported the chi-square difference tests in all instances above because the models with more factors were always preferred over the models with fewer factors on the basis of the AIC (the lower the better). We can conclude that our hypothesized seven-factor model provided the best fit to the data. This is an important finding, given that previous research on the JD-R model has generally clustered job demands on the one hand and job resources on the other hand to test their composite effects on employee well-being (see e.g., Balducci et al., 2011; Schaufeli & Bakker, 2004).
We found significant latent correlations between the job demands task load and team process load (r = .44) and between team process load and conflict with team members (r = .25). Also, the job resource social support from team members and the job demand conflict with team members had a significant negative correlation (r = -.25). Team social support and job autonomy covaried significantly as job resources (r = .31). Furthermore, the latent factors of job strain and work engagement had a significant correlation of r = -.19. Finally, we found significant correlations between the factors that we hypothesized to be causally related.
Table A1
Overview of Items and Results of the Factor Analysis
Item wording / Task load / Team process load / Conflict with team membersMy work requires working very fast. / .58
My work requires working very hard. / .70
My work makes almost no demands on me. (R) / .61
I have enough time to do everything. (R) / .70
How much communication is required between you and other team members in order to do the job? / .37
Is communicating with other team members easy or demanding? / .58
How much monitoring of team members does the job require? / .65
Is attending to team members easy or demanding? / .60
How much correction of other team members does the job require? / .73
Is correcting other people simple or complex? / .54
How much correction or adjustment of your own actions is required in order to coordinate with your team members? / .72
Is adjusting your actions to improve coordination simple or complex? / .58
How much leadership is required of you? / .64
Is being a leader easy or demanding for you? / .32
How often do you get into arguments with team members at work? / .58
How often do team members yell at you at work? / .88
How often are team members rude to you at work? / .93
How often do team members do nasty things to you at work? / .86
Table A1 (continued)
Item wording / Social support from team members / Job autonomyMy team members adjust their individual task responsibilities to prevent work overload for me. / .32
My team members go out of their way to do things to make my job easier. / .51
My team members help me when I have a job-related problem. / .72
My team members are willing to help me when I need a special favor. / .80
My team members bring to my attention information that may be useful for my work. / .78
My team members make suggestions about how I can do my job best. / .73
My team members appreciate my contribution to the teamwork. / .72
My team members are proud that I am a part of this team. / .63
Even if I did the best job possible, my team members would fail to notice. (R) / .56
I frequently receive positive feedback from my team members. / .64
When my team members give me performance feedback, they are usually considerate of my feelings. / .39
The feedback I receive from my team members helps me do my job. / .54
I feel comfortable asking my team members for feedback about my work performance. / .56
My team members care about my general satisfaction at work. / .72
My team members are concerned about my well-being. / .69
There is a team member with whom I can share my joys and sorrows. / .42
I get the emotional help and support I need from my team members. / .64
My team members take my opinions into account. / .80
My team members would fail to understand my absence due to a personal problem. (R) / .56
My team members would forgive an honest mistake on my part. / .57
The job allows me to make decisions about what methods I use to complete my work. / .77
The job gives me considerable opportunity for independence and freedom in how I do the work. / .79
The job allows me to decide on my own how to go about doing my work. / .77
The job gives me a chance to use my personal initiative or judgment in carrying out the work. / .78
The job allows me to make a lot of decisions on my own. / .78
The job provides me with significant autonomy in making decisions. / .78
The job allows me to make my own decisions about how to schedule my work. / .81
The job allows me to decide on the order in which things are done in the job. / .69
The job allows me to plan how I do my work. / .78
Table A1 (continued)
Item wording / Job strain / Work engagementTense / .84
Uneasy / .92
Worried / .91
Calm (R) / .50
Contented (R) / .35
Relaxed (R) / .53
At my work, I feel bursting with energy. / .82
At my job, I feel strong and vigorous. / .83
I am enthusiastic about my job. / .85
My job inspires me. / .83
When I get up in the morning, I feel like going to work. / .79
I feel happy when I am working intensely. / .74
Table A2
Model / Chi-squarea (df) / TLIa / CFIa / RMSEAa / AIC / Chi-square difference testM1: 1-factor / 5664.8 (1652) / .22 / .27 / .120 / 6018.8 / -
M2: 3-factors (JD + JR + WB) / 4752.8 (1649) / .40 / .44 / .106 / 5112.8 / M2-M1 = 912***
M3: 5-factors (TL + TPL + Co + JR + WB) / 4391.9 (1642) / .46 / .50 / .100 / 4765.9 / M3-M2 = 360.9***
M4: 6-factors (TL + TPL + Co + TSS + Au + WB) / 3668.7 (1637) / .60 / .63 / .086 / 4052.7 / M4-M3 = 723.2***
M5: 7-factors (TL + TPL + Co + TSS + Au + JS + WE) / 3347.2 (1631) / .66 / .69 / .079 / 3743.2 / M5-M4 = 321.5***
Nested Model Comparisons in Confirmatory Factor Analysis
Note.JD = job demands. JR = job resources. WB = well-being. TL = task load. TPL = team process load. Co = Conflict with team members. TSS = team social support. Au = job autonomy. JS = job strain. WE = work engagement. TLI = Tucker-Lewis Index. CFI = Comparative fit index. AIC = Akaike Information Criterion.
aWhereas covariances between exogenous latent factors were included in our measurement models, we did not allow covariances between the error terms of the reflective indicators. Consequently, the five measurement models did not show an acceptable fit to the data as reflected by the significant chi-square values and fit indices. Adding such covariances on the basis of modification indices is an empiricist approach to increase the model fit. In our confirmatory factor analysis, we were interested in relative rather than absolute model fit because we compared nested models to identify the most optimal number of underlying dimensions.
*** p.001.
Appendix B
Supplementary analysis on causality using a reversed causation mediation test.
Our data come from a cross-sectional design and therefore causal claims are not warranted.Although the causal sequencing from job demands and resources to job strain and work engagement is aligned with a well-tested theoretical model (i.e., JD-R model), we decided to 1) use a method described by Antonakis, Bendahan, Jacquart, and Lalive (2010) and 2) test a reversed mediation model to back up our causal claims with respect to the hypothesized paths in the model. We should emphasize that our independent variable (MTM) is a work design feature unlikely to be influenced by the other variables included in the model and unlikely to be correlated with omitted causes of our dependent variables.
Antonakis and colleagues (2010) presented a method that would enable researchers to make explicit causal claims when estimating models from correlational data. As mentioned before, causal interpretations in our model are problematic for the associations between job demands and job strain and between job resources work engagement. According to Antonakis and colleagues (2010), however, the coefficients of job demands and resources “could be interpreted causally if an exogenous source of variance . . . were found that strongly predicts x [job demands and resources] and is related to y [job strain and work engagement] via x only” (p. 1101). MTM is such an exogenous variable and therefore made for an appropriate instrumental variable that enabled us to make causal claims regarding the associations between job demands and job strain on the one hand and job resources and work engagement on the other hand.
In addition, we tested the reversed causal model in which job strain and work engagement mediated the influence of MTM (fragmentation of time across teams) on job demands and job resources. The results of the reversed causation check are presented in Figure B. As indicated by the fit indices, this model differed significantly from the data and it had a worse fit than the original hypothesized model. Moreover, the standardized path coefficients from MTM to both job strain and work engagement were not significant. Taken together, these observations refuted the alternative reversed mediation model and supported the conclusion that the effect of time fragmentation on job strain and work engagement occurred through job demands and job resources.
Figure B. Results of the reversed mediation test.
Model fit: Chi-square = 36.34 (p = .003); TLI = .68; CFI=.86; RMSEA=.09.
Standardized path coefficients are shown.
* p < .05. ** p < .01.*** p < .001.
Appendix C
Correcting for measurement error and common-source biasusing a latent variable approach.
To correct for measurement error and common-source variance, we also tested our hypothesized model using a latent variable approach. MIMIC (multiple indicators multiple causes) models are a special case of structural equation models involving latent variables that are predicted by observed variables (Jöreskog & Goldberger, 1975). MIMIC models have both a structural component, which specifies the causal relations between exogenous and endogenous variables, and a measurement component, which relates each latent variable to a set of indicators. As such, the disturbances of the latent endogenous variables reflect only omitted causes and not measurement error (unlike ordinary path models) (Kline, 2005). MIMIC models are therefore said to yield estimates of path coefficients in the structural part of the model that are corrected for measurement unreliability in the independent and dependent variables.
Another advantage of MIMIC modeling is its ability to correct for common-source variance. Our data on demands, resources, job strain, and work engagement stemmed from a single source, and the endogeneity bias (i.e., variables are affected by a common source factor) was therefore a problem. Researchers generally model a latent common factor to account for common variance, yet Antonakis and colleagues (2010) argued that one cannot remove the common-source bias with this procedure and they provided an alternative solution. One statistical way to control for the common-source or common-method problem is to model exogenous sources of variance (Antonakis et al., 2010). These so-called instrumental variables should relate strongly to the independent variables and only to the dependent variables via the independent variables. In our case, multiple team membership is such an exogenous variable that is also unlikely to correlate with omitted common causes and could therefore correct the estimates for the relationships between job demands and job strain and between job resources and work engagement.
Based on the logic as described by Antonakis and colleagues (2010), we used a latent variable approach with multiple team membership as an instrumental variable. Our MIMIC model contained one exogenous observed variable (MTM) and seven latent factors with a set of reflective indicators (i.e., items) for each. We modeled three latent factors for job demands, namely task load (with four indicators), team process load (10 indicators), and conflict with team members (four indicators). In addition, two latent factors were modeled for job resources, namely team social support (20 indicators) and job autonomy (nine indicators). Finally, job strain and work engagement were modeled as latent factors with six indicators each. The model and analysis results can be found in Figure C1. Our earlier reported results from the path analysis were found to be robust because we found significant effects of time fragmentation on team process load (p = .002), conflict with team members (p = .004), and team social support (p = .022), but not on task load (p = .57) and job autonomy (p = .42). Also, we found that job strain was significantly predicted by task load (p < .001), team process load (p = .029) as well as conflict with team members (p = .048), while work engagement was significantly influenced by team social support (p < .001) and job autonomy (p = .002).
Figure C1: Test of the full MIMIC model.
Note. Observed variables are depicted in rectangles. Latent (unobserved) variables are depicted in circles. For reasons of parsimony, error terms of the endogenous latent factors and reflective indicators are not depicted in the model. Standardized path coefficients are shown.
†p < .10.* p < .05. ** p < .01.*** p < .001.
The test of the MIMIC model provided us with an important robustness check of our findings because 1) using a latent variable approach with a set of reflective indicators for each latent factor yieldsconsistent estimates that are free from measurement error (Kline, 2005) and 2) including multiple team membership as an instrumental variable purges the common-source bias from endogenous variables (Antonakis et al., 2010). Nevertheless, multiple team membership was not a significant exogenous source of variance for task load and job autonomy, and we could therefore not rule out the problem of endogeneity bias for the relationships involving these variables. To strengthen our claims on the associations between job demands and job strain on the one hand and job resources and work engagement on the other hand, we ran a supplemental analysis in which we tested an additional MIMIC model that clustered the five demands and resources in two latent factors: job demands (18 indicators) and job resources (29 indicators). The results of this MIMIC model test are presented in Figure C2. Because we found that multiple team membership (fragmentation of time across teams) was an exogenous source of variance for both job demands and job resources, we can conclude from the results that the associations between job demands and job strain (p < .001) and between job resources and work engagement (p < .001) were not explained by common-source variance.
Figure C2: Test of the clustered MIMIC model.
Note. Observed variables are depicted in rectangles. Latent (unobserved) variables are depicted in circles. For reasons of parsimony, error terms of the endogenous latent factors and reflective indicators are not depicted in the model. Standardized path coefficients are shown.
†p < .10. *** p .001.