The data were collected from 512 respondents on their self-reports of job engagement, which is defined as the extent to which an employee feels passionate about their jobs and put discretionary efforts into their work. The measure consists of 18 items with three theoretical factors (physical engagement, emotional engagement, and cognitive engagement). Although job engagement is theorized to include three different factors, many psychologists also argue that there is only one factor underlying all the items.
Figure 1 describes the notation for the CFA model for a single factor model. A Confirmatory Factor Analysis was conducted to verify the 1-latent factor structure using full-information data.
Figure 1. Confirmatory Factor Analysis. Model 1
Notation Single Factor Model
Table 1, presents the standardized loading factors \(\lambda_{i}\) that varies from 0.615 (JE6_p6) to 0.864 (JE18_c6), while Figure 2 shows the estimation of loading factors and variances for each item in the single factor.
| Item | Loading | SE | LL | UL |
|---|---|---|---|---|
| JE1_p1 | 0.770 | 0.019 | 0.733 | 0.807 |
| JE2_p2 | 0.776 | 0.018 | 0.740 | 0.812 |
| JE3_p3 | 0.763 | 0.019 | 0.725 | 0.801 |
| JE4_p4 | 0.733 | 0.021 | 0.692 | 0.775 |
| JE5_p5 | 0.724 | 0.022 | 0.681 | 0.767 |
| JE6_p6 | 0.615 | 0.028 | 0.560 | 0.671 |
| JE7_e1 | 0.745 | 0.021 | 0.705 | 0.786 |
| JE8_e2 | 0.684 | 0.024 | 0.636 | 0.732 |
| JE9_e3 | 0.633 | 0.027 | 0.580 | 0.687 |
| JE10_e4 | 0.635 | 0.027 | 0.581 | 0.688 |
| JE11_e5 | 0.663 | 0.026 | 0.613 | 0.713 |
| JE12_e6 | 0.693 | 0.024 | 0.647 | 0.740 |
| JE13_c1 | 0.815 | 0.016 | 0.784 | 0.846 |
| JE14_c2 | 0.861 | 0.012 | 0.837 | 0.885 |
| JE15_c3 | 0.875 | 0.011 | 0.853 | 0.897 |
| JE16_c4 | 0.777 | 0.018 | 0.741 | 0.813 |
| JE17_c5 | 0.812 | 0.016 | 0.781 | 0.843 |
| JE18_c6 | 0.864 | 0.012 | 0.840 | 0.888 |
Figure 3 presents the notation the 3-latent factor structure model theorized for the self-reports of job engagement construct. Specifically, the single factor JE1 in previous model is replaced for with LV representing physical engagement, emotional engagement, and cognitive engagement. Thus, three measurement models are included in this SEM, where \(\lambda\) coefficients link the observed variables with latent variables, \(\epsilon\) represents the measurement error of each observed exogenous variables, and \(\Phi\) the correlation between latent variables. Additionally, \(\theta_{i}\) indicates the residual variance for each latent variables, which will be a free parameter in estimation (equal to 1) for allowing model identification.
Figure 3. Confirmatory Factor Analysis. Model 2
Notation Three-Factor Structure Model
Table 2, presents the standardized loading factors \(\lambda_{i}\) that varies from 0.656 (JE6_p6) to 0.841 (JE2_p2) in the psychical engagement factor, from 0.775 (JE8_e2) to 0.922 (JE12_e6) in emotional engagement, and 0.783 (JE16_c4) to 0.932 (JE15_c3) in cognitive engagement.
| Item | Loading | SE | LL | UL |
|---|---|---|---|---|
| Factor 1. Physical Engagement | ||||
| JE1_p1 | 0.788 | 0.019 | 0.751 | 0.824 |
| JE2_p2 | 0.841 | 0.015 | 0.811 | 0.871 |
| JE3_p3 | 0.806 | 0.018 | 0.771 | 0.840 |
| JE4_p4 | 0.818 | 0.017 | 0.785 | 0.851 |
| JE5_p5 | 0.804 | 0.018 | 0.770 | 0.839 |
| JE6_p6 | 0.656 | 0.027 | 0.603 | 0.709 |
| Factor 2. Emotional Engagement | ||||
| JE7_e1 | 0.871 | 0.012 | 0.847 | 0.895 |
| JE8_e2 | 0.775 | 0.019 | 0.738 | 0.812 |
| JE9_e3 | 0.825 | 0.015 | 0.794 | 0.855 |
| JE10_e4 | 0.816 | 0.016 | 0.785 | 0.848 |
| JE11_e5 | 0.899 | 0.010 | 0.880 | 0.919 |
| JE12_e6 | 0.922 | 0.008 | 0.905 | 0.938 |
| Factor 3. Cognitive Engagement | ||||
| JE13_c1 | 0.836 | 0.014 | 0.808 | 0.864 |
| JE14_c2 | 0.924 | 0.008 | 0.908 | 0.939 |
| JE15_c3 | 0.932 | 0.007 | 0.918 | 0.946 |
| JE16_c4 | 0.783 | 0.018 | 0.748 | 0.819 |
| JE17_c5 | 0.840 | 0.014 | 0.812 | 0.867 |
| JE18_c6 | 0.887 | 0.011 | 0.866 | 0.908 |
Furthermore, Figure 4 presents the pattern of coefficients, error variances, and correlations among factors. The correlation between scales in Model 2 ranged from 0.63 < r < 0.85.
The goodness of model fit for measurement models conducted above was
evaluated using overall and individual fit indices: (1) overall fit
assessed using the \(\chi^{2}\)
statistic that evaluates the magnitude of discrepancy between the sample
and the model-estimated, largest values of \(\chi^{2}\) indicate a bad fit; (2) the
Comparative Fit Index (CFI), Tucker-Lewis index (TLI), and Normed Fit
Index (NFI) with values >.90 indicating an acceptable fit and values
>.95 indicating a good fit; (3) the Standardized Root Mean Square
Residual (SRMR) <.08 being indicative of good fit; and (4) the Root
Mean Square Error of Approximation (RMSEA) with values <.08 being
indicative of reasonable fit, values <.05 indicating a good fit and
>.10 unacceptable.
| Model | Model.Specif | DVs | Chisq | df | p-value | CFI | TLI | NFI | SRMR | RMSEA | LL | UL |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Model 1 | 1 Factor | 36 | 2679.429 | 135 | 0 | 0.712 | 0.674 | 0.702 | 0.102 | 0.192 | 0.186 | 0.198 |
| Model 2 | 3 Factors | 39 | 849.307 | 132 | 0 | 0.919 | 0.906 | 0.906 | 0.049 | 0.103 | 0.096 | 0.110 |
Model 1 included all 18 items distributed in a unique factor. Based on comparative indices (CFI = .712, TLI = .674, and NFI = .702) the model showed an inadequate fit, additionally the absolute indices (SRMR = .102 and RMSEA = .192) take values above the cut-off criteria for reasonable fit (<.08). Model 2 includes the 18 items distributed in the original 3-factor model (physical, emotional, and cognitive engagement). According to comparative indices (CFI = .919, TLI = .906, and NFI = .906) the model with 3-latent factor showed an acceptable fit, also the absolute indices (SRMR = 0.049) indicative of reasonable fit. However, the RMSEA and their 90% confidence interval exceed the cut-off criterion (RMSEA < .08). In sum, the original theorized model with 3-factors presents a better fit. Furthermore, \(\chi^{2}\) statistics is lower, suggesting that the magnitude of discrepancy between the sample and the model-estimated variance/covariance matrices is lower in the multidimensional model than in the single-factor model.