- Top of page
Background: Engagement is an emerging job attitude that purports to measure employees' psychological presence at and involvement in their work. This research compares three academic approaches to engagement, and makes recommendations regarding the most appropriate conceptualisation and measurement of the construct in future research. The current research also investigates whether any of these three approaches to engagement contribute unique variance to the prediction of turnover intentions above and beyond the predictive capacity of alternative constructs. Methods: An online survey was taken by 382 employees and managers from a mid-sized financial institution. Results: Results failed to support either a multi- or unidimensional factor structure for the Utrecht Work Engagement Scale (UWES) engagement measure. For the Shirom-Melamed Vigor Measure (SMVM), a multi-dimensional structure was identified as a good fit, while a unidimensional structure fit poorly. The uni-factorial structure of Britt's engagement measure was confirmed. The Schaufeli measure of engagement was a strong predictor of work outcomes; however, when controlling for job satisfaction and affective commitment, that measure lost its ability to predict intentions to leave. Two components of the Shirom vigor measure held their predictive validity. Conclusions: Collectively, these findings suggest that the Shirom vigor measure may provide better insight into whether and how much a person is ‘into’ his or her job. The Schaufeli measure was a good predictor of important work outcomes, but when job satisfaction and affective commitment were controlled, it lost its predictive validity. We were not able to confirm the three-factor structure of the Schaufeli measure. Two components of the Shirom vigor measure predicted turnover intentions after controlling for job satisfaction and affective commitment, suggesting less overlap with those constructs than the Schaufeli measure of engagement. This research adds important information on the nature of engagement and is expected to contribute toward a better understanding of the construct itself, as well as its measurement.
- Top of page
There are multiple diverging approaches to the concept of engagement in the academic realm. Three common approaches to engagement have been developed by the following researchers: Schaufeli (Schaufeli, Martinez, Pinto, Salanova, & Bakker, 2002a; Schaufeli, Salanova, Gonzalez-Roma, & Bakker, 2002b), Shirom (2003), and Britt (Britt, Thomas, & Dawson, 2006). These multiple and divergent approaches to engagement highlight the lack of focus and clarity surrounding the construct of engagement. Indeed, Bakker, Albrecht, and Leiter (2011a) recently suggested that the unclear measurement of engagement was one of the most problematic issues surrounding this otherwise popular construct. While the three approaches included here are similar in some important ways, we believe that there are also crucial distinctions between these three approaches to conceptualising and measuring engagement that are important for researchers to consider. These differences have created a lack of clarity regarding the concept of engagement. Luthans, Avolio, Avey, and Norman (2007) suggested that for a new construct to have practical and theoretical utility, it is important for it to be (a) theoretically sound, (b) empirically distinct, and (c) of practical importance to both academics and practitioners. This research aims to clarify and evaluate these three approaches to engagement to ultimately make recommendations as to the most appropriate conceptualisation and measurement of the construct in future research.
By far the most widely used measure of engagement in the academic literature is from Schaufeli and his colleagues (Christian & Slaughter, 2007). Schaufeli et al. (2002b) defined engagement as a persistent and positive affective-motivational state of fulfillment in employees, characterised by vigor, dedication, and absorption. According to Schaufeli and colleagues (2002a, 2002b) vigor is seen as high energy, resilience, willingness to invest effort on the job, ability to not be easily fatigued, and persistence in the face of difficulties. Dedication is characterised by strong involvement in work, enthusiasm, and a sense of pride and inspiration. Finally, absorption is characterised by a pleasant state of being immersed in one's work, time passing quickly, and being unable to detach from the job or task.
On the whole, previous findings for this three-factor Schaufeli model have been strong (Schaufeli, Bakker, & Salanova, 2006); however, the correlations among the three factors have been moderate to high; .71–.94 (Schaufeli et al., 2002a) and .56–.89 (Schaufeli et al., 2002b). Likewise, a recent meta-analytic review of the engagement research also found Schaufeli et al.'s three factors of engagement to be very highly related with corrected correlations ranging from .88 to .95 (Christian & Slaughter, 2007), suggesting possible multicollinearity between the dimensions. Given these and other findings, Schaufeli and Salanova (2007) noted that a one-factor conceptualisation of their measure is also supported (see Sonnentag, 2003, for an example of research utilising the measure unidimensionally). Despite potential limitations regarding the dimensionality of the UWES, on the whole, the three-factor conceptualisation of the UWES has received strong support (Schaufeli et al., 2002a; Schaufeli et al., 2002b; Schaufeli et al., 2006).
Hypothesis 1: A three-factor structure will be confirmed for the Schaufeli model of engagement.
Shirom (2003) proposed a similar construct to engagement, which he termed ‘vigor’. Although Shirom's vigor is conceptually similar to the above-mentioned construct of engagement, Shirom (2003) criticised the Schaufeli model of engagement by suggesting that the vigor component of engagement is the only non-confounded construct in Schaufeli et al.'s three-factor engagement model; in addition, he critiqued engagement's close relationship with alternative psychological constructs including psychological presence, job involvement, and resiliency. In an attempt to overcome this supposed problem, Shirom developed a multi-dimensional conceptualisation of vigor to reflect individuals' feelings of physical strength, emotional energy, and cognitive liveliness. The concept of vigor and its three components reflect Kahn's (1990) original conceptualisation of engagement and the use of physical, emotional, and cognitive energetic capacities. Shirom (2011) acknowledges that vigor and engagement are similar, yet he proposes that engagement as measured by the UWES, specifically the vigor component measured in the UWES, refers to a person's motivation and resiliency. He contends that people can experience vigor regardless of their resiliency, a subtle distinction between the two constructs. The three-factor Shirom-Melamed Vigor Measure (SMVM) conceptualisation is similar to the three-factor conceptualisation of Schaufeli's engagement concept in that they both have three dimensions and both measure cognitive, affective, and behavioral components of a larger job attitude. Recently, Bakker, Albrecht, and Leiter (2011b) wrote a position article on the state of the engagement concept. They proposed that engagement is a “combination of capability to work (energy, vigor) and willingness to work (involvement, dedication)” (p. 75). Although the vigor concept appears to focus on the capability to work aspect of engagement, a person's emotional energy and cognitive liveliness also hint at a person's willingness to work. Theoretically this illustrates the overlap and overall similarity of vigor and engagement. Given the close nature of the vigor concept to engagement, we position it as an alternative conceptualisation of job engagement.
To be clear, however, we are not suggesting that the vigor concept is necessarily redundant with engagement itself. Rather, we consider vigor to be an alternative conceptualisation of engagement. Such a proposition would indicate that the concepts are similar and therefore there is value in comparing them. This is particularly true given the current popularity of the engagement construct, in addition to the consideration that the vigor concept was developed in response to the engagement concept. Our justification for using the Shirom measure of vigor as an alternative conceptualisation to engagement is due to the fact that they are quite similar in their definitions, conceptualisations, and the items themselves. All three measures are indicative of a positive, affective state directed towards one's work. Further, Shirom explicitly listed Kahn's (1990) seminal work, and, in particular, its focus on cognitive, physical, and emotional capabilities, as a theoretical basis for developing the SMVM. In addition, Shirom theoretically suggested that vigor was an antecedent of engagement, although he did not empirically test this idea. Considering the theoretical similarities between Kahn's engagement and Shirom's vigor, we felt that it was important to also evaluate the SMVM. The concept of vigor stems from Hobfoll's (1989) conservation of resources theory. Vigor is related to an individual's energetic resources (cognitive, emotional, and physical), and these resources are possessed individually and are all connected and continually influencing each other. Shraga and Shirom (2009) found support for the three-factor model of vigor, but also found support for a one-factor conceptualisation. Therefore, for both Schaufeli's measure of engagement and Shirom's measure of vigor, both one- and three-factor conceptualisations have both theoretical and empirical support. Nevertheless, both authors favor the original three-factor conceptualisation of their respective measures.
Hypothesis 2: A three-factor structure will be confirmed for the Shirom vigor model.
A third approach to engagement offers a much simpler view of the construct. This approach (Britt, 1999) defines engagement as feeling responsible for and committed to one's work performance such that one's job performance truly matters to the individual (Britt, 1999; Britt & Bliese, 2003). Britt, Dickinson, Greene-Shortridge, and McKibben (2007) suggest that when people are engaged in their work, they feel a sense of personal responsibility for their work performance that influences their identity. Unlike Schaufeli and Shirom, Britt defines engagement as a one-factor construct, and measures it as such. Nevertheless, the items included in Britt's measure examine a variety of conceptual categories, including perceived responsibility for job performance, commitment to job performance, and whether performance matters to the person (Britt, 1999, 2003; Britt, Adler, & Bartone, 2001; Britt & Bliese, 2003; Britt, Castro, & Adler, 2005).
Hypothesis 3: A single-factor structure will be confirmed for the Britt engagement model.
Britt et al. (2007) see the outcomes and predictors of engagement differently from other researchers. The Britt approach states that engagement is a motivational state created by beliefs of personal responsibility and caring. Therefore, contrary to both Shirom's and Schaufeli's conceptualisations of engagement, Britt views vigor, physical exertion, attention, effort, and absorption as outcomes of engagement, rather than as contributing dimensions of the construct. Further, Britt views the multi-factor conceptualisations of engagement as inherently confusing (Britt et al., 2007), and thus measures it as a unidimensional construct. Another primary difference between Britt's view of engagement and the views of other engagement researchers such as Shirom and Schaufeli is that Britt argues that engagement can have negative as well as positive consequences. For instance, Britt suggests that when workers are engaged, or highly motivated to do well, they can quickly lose their enthusiasm and motivation if they fail to perceive the meaningfulness of or pathways to success in their job due to lack of resources, lack of support, etc. (Britt, 2003). This compromised engagement, in turn, could actually lead to performance decrements and associated detrimental outcomes for the organisation and the employee alike.
The preceding conceptualisations of what we are generally referring to as engagement were chosen because they are the three most popular publically available (e.g. non-proprietary) measures of the engagement construct. Further, each had substantial theoretical and empirical evidence supporting the conceptualisation and the measure. All three approaches to engagement involve a measure of how into the job a person is, each refers to employee affect as it relates to one's work, and all are in agreement that engagement is best conceptualised and measured as a relatively transitory state, as opposed to a more enduring trait. For the sake of completeness, it is worth noting that a widely used engagement measure in practice is the Gallup Q12 (Harter, Schmidt, & Hayes, 2002). However, this measure is proprietary and as such has been unavailable for independent researchers to verify its validity. As a result, it was not included in the present study.
Both the Schaufeli and Shirom measures have three components that seem to conform to Kahn's (1990) original conceptualisation of engagement as having physical, emotional, and cognitive components. There are differences, however, that are worthy of note. Specifically, Schaufeli's components include vigor, dedication, and absorption, while Shirom's components consist of physical strength, emotional energy, and cognitive liveliness. Britt's conceptualisation of engagement is unidimensional and suggests that the construct focuses on a person's feelings of responsibility for performance and commitment to their job, and as such, Britt's conceptualisation does not conform to Kahn's (1990) three components of engagement. For a thorough review of the engagement construct as a whole, see Macey and Schneider (2008).
Research has consistently found that engaged employees have higher job satisfaction, greater organisational commitment, and lower turnover intentions than do those individuals who are less engaged in their work (e.g. Christian & Slaughter, 2007; Hallberg & Schaufeli, 2006; Llorens, Bakker, Schaufeli, & Salanova, 2006; Schaufeli & Bakker, 2004; Schaufeli & Salanova, 2007). Clearly, these outcomes are important to every organisation because of their direct impact on the financial bottom-line. Therefore, given engagement's established relationships with both individual-level and organisational-level outcomes, it is evident that engagement is a core organisational competency that organisations should aim toward cultivating in their employees.
Theoretically, engagement, job satisfaction, job involvement, and organisational commitment are distinct constructs (Hallberg & Schaufeli, 2006). However, there is some overlap in the definitions of the constructs—namely affective reactions to the job are present in the definitions of each. However, the research methodology utilised to reach these conclusions consisted only of exploratory and confirmatory factor analyses. We believe that such analyses alone are insufficient in order to determine whether a new concept adds anything (i.e. unique variance) beyond that of more established and similar concepts (i.e. job satisfaction) to the prediction of critical bottom-line organisational outcomes, such as performance and turnover intentions.
The framing of engagement under the umbrella of job attitudes puts a burden on the newer concept of engagement to demonstrate its differentiation from more established attitudes. In this framework it is important to determine if engagement adds to our understanding of job attitudes and to the prediction of important outcomes such as job performance above and beyond what can already be explained by existing positive psychological constructs. For example, assuming that engagement and satisfaction are distinct constructs, we would expect engagement to uniquely predict outcomes such as job performance, above and beyond other similar predictor constructs. Some recent research using only Schaufeli's measure of engagement suggested that there may be a great deal of overlap between engagement and satisfaction and that engagement may not add any incremental validity beyond that of satisfaction (Wefald & Downey, 2009). However, a comparison of the three published and publicly available engagement measures discussed previously has not yet been conducted. Likewise, a side-by-side investigation of the three engagement measures to see which measure(s) differentially predict important organisational outcomes, specifically turnover intentions, has also not yet been done. This research sought to fill this void in the literature.
Hypothesis 4: All three engagement measures will add unique variance to the prediction of turnover intentions over and above other typical predictive constructs (job satisfaction and organisational commitment).
In sum, the goals of this research are to (a) empirically examine three of the engagement measures currently used in the academic and professional literature, (b) assess the factor structure of the Schaufeli, Shirom, and Britt models of engagement, and (c) examine which measure of engagement was a better predictor of intentions to leave the organisation.
- Top of page
The data were first examined using visual scans of data plots, means, standard deviations, skew, kurtosis, and scale minimums and maximums, and were found to be normally distributed (see Table 1). The data were imputed using SPSS's linear trend at point method. The linear trend at point data imputation method replaces missing values with the linear trend for that point using regression such that missing values are replaced with predicted values (SPSS, Inc., 2006). With the exception of demographics, 11 of the items on the survey were required; however, the last section of the survey dealing with demographics was optional.
Table 1. Means, Standard Deviations, Skewness, Kurtosis, Correlations, and Reliabilities (Cronbach's alpha) (N = 382, df = 380)
| 1. Job satisfaction||(.84)|| || || || || || || || || || || |
| 2. Affective commitment||.58**||(.88)|| || || || || || || || || || |
| 3. Engagement (Schaufeli)||.70**||.53**||(.93)|| || || || || || || || || |
| 4. Vigor (Schaufeli)||.65**||.49**||.92**||(.87)|| || || || || || || || |
| 5. Dedication (Schaufeli)||.73**||.56**||.94**||.82**||(.84)|| || || || || || || |
| 6. Absorption (Schaufeli)||.52**||.39**||.88**||.67**||.76**||(.78)|| || || || || || |
| 7. Vigor (Shirom)||.49**||.32**||.66**||.69**||.62**||.49**||(.91)|| || || || || |
| 8. Physical strength||.46**||.28**||.66**||.73**||.58**||.46**||.88**||(.93)|| || || || |
| 9. Emotional energy||.35**||.26**||.45**||.46**||.47**||.36**||.77**||.46**||(.91)|| || || |
|10. Cognitive liveliness||.37**||.22**||.47**||.43**||.43**||.36**||.77**||.60**||.39**||(.81)|| || |
|11. Engagement (Britt)||.46**||.36**||.59**||.51**||.58**||.52**||.46**||.40**||.39**||.30**||(.82)|| |
|12. Intentions to leave||−.62**||−.54**||−.48**||−.43**||−.51**||−.36**||−.29**||−.23**||−.21**||−.29**||−.33**||(.92)|
Confirmatory factor analyses were conducted for both the one- and three-factor structures of the Schaufeli model, the one- and three-factor structures of the Shirom model using the original items, and the one-factor Britt model. In addition, confirmatory factor analyses were performed at the item and factor level using the three scales and their items. Goodness of fit indices were judged based upon common and well-established rules for acceptability. For instance, CFI and NFI should exceed .90 (Hoyle, 1995), and RMSEA should fall below .10 (Browne & Cudeck, 1993). Hu and Bentler (1999) have suggested values of less than .06 for the RMSEA and values exceeding .95 as acceptable for the NFI and CFI. GFI and AGFI having values close to 1.00 are indicative of a good fit. However, others have suggested more liberal values of less than .10 for the RMSEA and values exceeding .90 for the GFI, AGFI, NFI, and CFI as acceptable (Byrne, 2001). MacCallum, Browne, and Sugawara (1996) noted that RMSEA values ranging from .08 to .10 indicate mediocre fit and those greater than .10 indicate poor fit.
We also calculated the AVE score for engagement as measured by all three scales. The AVE score is a test of the discriminant validity and compares the various factorial solutions. Values above .50 are generally recommended (Fornell & Larcker, 1981).
Hierarchical regression analysis was used to determine the unique variance that the three engagement measures offer above and beyond one another and other job attitudes, such as job satisfaction and affective commitment. Regressions were also used to determine the predictive validity of the three measures on three important organisational outcomes (intentions to leave, job satisfaction, and affective commitment). A relative weights analysis was used to determine the relative importance of the predictors in this study (LeBreton, Hargis, Griepentrog, Oswald, & Ployhart, 2007; Johnson & LeBreton, 2004). A relative weights analysis examines the proportionate contribution each predictor makes to R2 considering both its individual effect and its effect when combined with other variables in a regression equation (Johnson & LeBreton, 2004). A relative weights analysis supplements a traditional multiple linear regression analysis and takes into account collinearity issues. Relative weights are calculated by creating a new set of uncorrelated predictors that are maximally related to the original set of correlated predictors and both sets of variables are used to estimate importance (Johnson, 2000). The estimates of relative importance sum to R2 and the estimates reflect effect size. A relative weights analysis examines the comparative usefulness of new variables, which variable or variables are primarily driving the R2, and how new variables contribute to the R2, which all add information to an analysis of incremental validity.
- Top of page
The means, standard deviations, skew, and kurtosis values for all study variables are presented in Table 1. All values for skew and kurtosis were within normal ranges except for Britt's Engagement scale where kurtosis was high. Note that while multicollinearity was not evident, the three factors of the Schaufeli model of engagement (vigor, dedication, and absorption) were statistically significantly correlated from .67 to .82, and the three factors of Shirom's vigor showed statistically significant correlations ranging from .45 to .66.
Confirmatory factor analyses were performed in order to examine the structure of the scales as proposed in the hypotheses (see Table 2). The three-factor structure of the Schaufeli engagement model showed good fit with the NFI and CFI; however, the GFI, AGFI, and RMSEA indicated a poor fit (χ2 = 250.15, df = 24, p < .01; GFI = .88, AGFI = .77, NFI = .90, CFI = .91, RMSEA = .16). While the NFI and CFI show acceptable fit with the data, the GFI, AGFI, and RMSEA did not show unequivocal good fit. That the fit indices do not converge to show good fit suggests that the factor structure is probably not an optimal model for the data. The one-factor structure suggested by Schaufeli and colleagues (2002a) was not a good fit with the data (χ2 = 341.61, df = 27, p < .01; GFI = .88, AGFI = .71, NFI = .86, CFI = .87, RMSEA = .18). The three-factor model of engagement was a better fit to the data than the one-factor model (χ2diff = 91.46, dfdiff = 3, p < .001). Therefore, Hypothesis 1 was not supported.
Table 2. Confirmatory Factor Analyses
|Model||df||χ2||GFI||AGFI||CFI||NFI||RMSEA (90% confidence interval)||p|
|Three-factor Engagement||24||250.15||.88||.77||.91||.90||.16 (.14–.18)||.001|
|One-factor Engagement||27||341.61||.83||.71||.87||.86||.18 (.16–.19)||.001|
|Three-factor Vigor||51||195.51||.93||.89||.96||.95||.09 (.07–.10)||.001|
|One-factor Vigor||54||1291.86||.61||.44||.66||.65||.25 (.23–.26)||.001|
|One-factor Britt Engagement||2||1.68||.99||.99||1.00||.99||.00 (.00–.10)||.431|
|All Engagement/Vigor items one factor||275||3200.66||.51||.42||.60||.58||.17 (.16–.17)||.001|
|All Engagement/Vigor items three factors||272||2136.97||.65||.59||.75||.72||.13 (.13–.14)||.001|
|All Engagement/Vigor items seven factors||254||823.46||.85||.81||.92||.89||.08 (.07–.08)||.001|
|All Engagement/Vigor sub-scales (one factor) including Britt's Engagement||14||195.34||.87||.74||.88||.87||.18 (.16–.21)||.001|
|All Engagement/Vigor sub-scales (two factors) not including Britt's Engagement||8||105.66||.91||.77||.93||.92||.18 (.15–.21)||.001|
A confirmatory factor analyses was also performed on Shirom's (2005) vigor scale. The three-factor model produced a good fit (χ2 = 195.51, df = 51, p < .01; GFI = .93, AGFI = .89, NFI = .95, CFI = .96, RMSEA = .09), therefore supporting Hypothesis 2. Shraga (2007) has also suggested a one-factor model, although this model was found to have a poor fit (χ2 = 1291.86, df = 54, p < .01; GFI = .61, AGFI = .44, NFI = .65, CFI = .66, RMSEA = .25). The three-factor model had a superior fit to the data compared to the one-factor model (χ2diff = 1096.35, dfdiff = 3, p < .001). Engagement as measured by the UWES yielded AVE scores above the .50 threshold (three-factor = .65; one-factor = .59), as did engagement as measured by Britt's measure (.56). For Shirom's measure, the three-factor model was well above the .50 threshold for vigor, although the one-factor model fell slightly short (three-factor = .71; one-factor = .45). Therefore Hypothesis 2 was supported.
A confirmatory factor analysis was conducted to evaluate the unidimensionality of the Britt engagement model. The single factor model produced a good fit (χ2 = 1.683, df = 2, p = .43; GFI = .99, AGFI = .99, NFI = .99, CFI = 1.00, RMSEA = .00), therefore supporting Hypothesis 3. See Table 2 to review complete model fit indices for each confirmatory model.
A series of confirmatory factor analyses were performed to test the distinctiveness of the three measures of engagement. Initially, all 25 items were tested using a one-factor model, a three-factor model, and a seven-factor model. The one-factor model yielded a poor fit with the data (χ2 = 3200.66, df = 275, p < .001; GFI = .51, AGFI = .42, NFI = .58, CFI = .60, RMSEA = .17). The three-factor model had the Schaufeli engagement items on one factor, the Shirom vigor items on another factor, and the Britt engagement items on a third factor. This model yielded a poor fit to the data (χ2 = 2136.97, df = 272, p < .001; GFI = .65, AGFI = .59, NFI = .72, CFI = .75, RMSEA = .13). The seven-factor model used the scales' sub-factors so there were three factors for Schaufeli's engagement scale, three factors for the Shirom vigor scale, and one factor for the Britt engagement scale. This model had a mediocre fit with the data (χ2 = 823.46, df = 254, p < .001; GFI = .85, AGFI = .81, NFI = .89, CFI = .92, RMSEA = .08). The chi-square difference tests suggested that the seven-factor model had a superior fit to the data compared to the other two models: seven factors vs. one factor (χ2diff = 2377.20, dfdiff = 21, p < .001), seven factors vs. three factors (χ2diff = 1313.51, dfdiff = 18, p < .001), three factors vs. one factor (χ2diff = 1063.69, dfdiff = 3, p < .001). Next, all the engagement and vigor sub-scales (seven in total) were tested with a one-factor model. This model yielded a poor fit to the data (χ2 = 195.34, df = 14, p < .001, GFI = .87, AGFI = .74, NFI = .87, CFI = .88, RMSEA = .18). A two-factor model was also tested using the Schaufeli engagement and Shirom vigor sub-scales. Since the Britt engagement scale is a one-factor scale it was dropped from this analysis. This model had a mediocre fit to the data (χ2 = 105.66, df = 8, p < .001; GFI = .91, AGFI = .77, NFI = .92, CFI = .93, RMSEA = .18).
Regression analyses were used to examine how the three measures differentially predict those outcomes with Hypothesis 4. Hypothesis 4 proposed that all three measures of engagement/vigor would add unique variance to the prediction of turnover intentions beyond that of job satisfaction and affective commitment (see Table 3). Job satisfaction was entered in the first step of the regression followed by affective commitment in the second step. The third step of the regression included the Schaufeli sub-scales, the three sub-scales of Shirom's scale, and Britt's one-dimensional scale. The Schaufeli and Britt scales were found not to add any unique variance to the prediction of turnover intentions beyond that of job satisfaction and affective commitment. However, two facets of Shirom's scale (physical strength, cognitive liveliness) added unique variance to the prediction of turnover intentions in the third step (see Table 4). Therefore, Hypothesis 4 was partially supported. Although none of the Schaufeli measures of engagement nor the Britt measures predicted turnover intentions above and beyond that of job satisfaction and affective commitment, two components of the vigor measures did (physical strength and cognitive liveliness). Collinearity diagnostics yielded tolerance levels well above .10, suggesting that multicollinearity was not evident in any of the regression analyses. Likewise, these two facets of the vigor scale are not redundant with job satisfaction or affective commitment in predicting turnover intentions. Three regressions were performed to compare the engagement measures as they predicted intentions to leave, job satisfaction, and affective commitment (see Table 4). In these analyses job satisfaction and affective commitment were outcomes. Previous research has used engagement as both an antecedent and a correlate of other job attitudes (Christian & Slaughter, 2007; Harter et al., 2002; Macey & Schneider, 2008; Schaufeli & Salanova, 2007; Wefald & Downey, 2009). The first regression used intentions to leave as the criterion. The results indicated that the vigor and dedication components of Schaufeli's engagement as well as the physical strength and cognitive liveliness components of Shirom's vigor predicted intentions to leave (R = .545, R2 = .297, p < .001). See Table 4 for the regression results and beta weights. When job satisfaction was the criterion, only Schaufeli's vigor and dedication predicted the criterion (R = .744, R2 = .554, p < .001). When affective commitment was the criterion, Schaufeli's vigor and dedication and Shirom's physical strength predicted the criterion (R = .580, R2 = .336, p < .001). The series of results taken together indicate that the vigor and dedication components of the Schaufeli measure predict all three criteria, suggesting that the measure is a useful predictor of important organisational outcomes. When job satisfaction and affective commitment are considered as correlates of Schaufeli's engagement, it is no longer predictive, therefore indicating overlap with those constructs.
Table 3. Hierarchical Regression (Criterion = Turnover Intentions)
|Job satisfaction||−.621**|| || || |
|Job satisfaction||−.464**|| || || |
|Affective commitment||−.272**|| || || |
|Job satisfaction||−.431**|| || || |
|Affective commitment||−.256**|| || || |
|Vigor (Schaufeli)||−.058|| || || |
|Dedication (Schaufeli)||−.067|| || || |
|Absorption (Schaufeli)||.024|| || || |
|Physical strength (Shirom)||.200**|| || || |
|Emotional energy (Shirom)||.030|| || || |
|Cognitive liveliness (Shirom)||−.146**|| || || |
|Engagement (Britt)||−.032|| || || |
|Regression (Criterion = Turnover Intentions)|
|Vigor (Schaufeli)||−.198*|| || || |
|Dedication (Schaufeli)||−.468**|| || || |
|Absorption (Schaufeli)||.096|| || || |
|Physical strength (Shirom)||.267**|| || || |
|Cognitive liveliness (Shirom)||−.153**|| || || |
|Emotional energy (Shirom)||.006|| || || |
|Engagement (Britt)||−.073|| || || |
|Regression (Criterion = Job satisfaction)|
|Vigor (Schaufeli)||.193**|| || || |
|Dedication (Schaufeli)||.622**|| || || |
|Absorption (Schaufeli)||−.093|| || || |
|Physical strength (Shirom)||−.063|| || || |
|Cognitive liveliness (Shirom)||.037|| || || |
|Emotional energy (Shirom)||.026|| || || |
|Engagement (Britt)||.058|| || || |
|Regression (Criterion = Affective Commitment)|
|Vigor (Schaufeli)||.221*|| || || |
|Dedication (Schaufeli)||.523**|| || || |
|Absorption (Schaufeli)||−.125|| || || |
|Physical strength (Shirom)||−.155*|| || || |
|Cognitive liveliness (Shirom)||−.031|| || || |
|Emotional energy (Shirom)||.047|| || || |
|Engagement (Britt)||.062|| || || |
A relative weights analysis was performed to further test Hypothesis 4, specifically looking to determine the importance of the engagement measures in predicting turnover intentions (see Table 5). Collectively, the engagement measures yielded an R2 value of .23. The Schaufeli measure of engagement had the highest relative weight (.148) and accounted for 63.8 per cent of the R2 value. Next was Britt's Engagement (.050), or 21.6 per cent, and finally Shirom's vigor (.034), or 14.6 per cent. A relative weights analysis was also performed using the sub-scales of the Schaufeli engagement and Shirom vigor measures along with the Britt measure of engagement (Table 5). The three factors of the Schaufeli measure had the highest relative weights in predicting turnover intentions.
|Relative Weights Analysis of Overall Engagement Measures (Criterion = Intent to Leave; R2 = .231)|
| ||Raw relative weights||Relative weights as percentage of R2|
|Relative Weights Analysis of Engagement and Vigor Factors (Criterion = Intent to Leave; R2 = .297)|
| ||Raw relative weights||Relative weights as percentage of R2|
|Physical strength (Shirom)||.016||5.5%|
|Cognitive liveli (Shirom)||.031||10.5%|
|Emotional energy (Shirom)||.010||3.3%|
|Relative Weights Analysis of Job Satisfaction, Affective Commitment, Physical Strength, and Cognitive Liveliness (Criterion = Intent to Leave; R2 = .453)|
| ||Raw relative weights||Relative weights as percentage of R2|
Finally, considering that the physical strength and cognitive liveliness components of vigor provided unique predictive variance over and above job satisfaction and affective commitment for turnover intentions in the hierarchical regression, a third relative weights analysis was conducted using the vigor components as measures of engagement (see Table 5). The results revealed that collectively, job satisfaction, affective commitment, and Shirom's physical strength and cognitive liveliness dimensions yielded an R2 value of .45. Job satisfaction had the highest relative weight (.237) and accounted for 52.2 per cent of the R2 value. Next was affective commitment (.163), or 36.0 per cent, then cognitive liveliness (.037), or 8.2 per cent, and finally physical energy (.016), or 3.6 per cent.
- Top of page
The goals of this research were to examine the factor structure of three of the most frequently utilised measures of employee engagement and to examine whether any of these approaches to engagement added anything unique to the prediction of important organisational outcomes beyond that of other, more established measures of job attitudes. The most widely used engagement measure in the academic literature is the Schaufeli engagement measure, and part of our goal with this research was to compare that measure to other similar measures that may capture what the Schaufeli measure is trying to capture. Since very little research has been done in this area, the present research has attempted to begin filling this void in the literature. Previous research has suggested that there is substantial overlap between satisfaction, commitment, and engagement (Wefald & Downey, 2009). However, a comparison of the three published and publicly available engagement measures has not yet been attempted. The present study explores these three measures, comparing them side-by-side in order to determine which measure(s) differentially predict important organisational outcomes.
Several conclusions can be drawn from the above series of factor analyses. First, the three-factor structure of the Schaufeli engagement measure did not have an ideal fit with the data. Specifically, although the NFI and CFI were acceptable, the GFI, AGFI, and RMSEA had less than ideal fit indices, suggesting a less than optimal fit with the data. However, the three-factor structure, while still not an ideal fit, did indeed fit the data better than the one-factor model. Any multi-factor construct that has highly correlated factors can have problems. We believe our results offer a unique view of engagement, especially since the kind of comparison we attempted here has not previously been published. The Shirom measure fared better, as analyses revealed that the three-factor structure of that measure fit the data well, with all of the fit indices meeting or exceeding the standard cut-off points. Similarly, the unidimensional Britt model was confirmed.
We acknowledge that our results diverge from consistent findings supporting the three-factor structure of the Schaufeli engagement model. We also acknowledge that the Schaufeli measure of engagement is the standard by which other academic (and possibly even practice-based) measures of engagement are measured. It is by far the most widely used measure of engagement in the academic literature and there is abundant empirical support for the three-factor structure (Schaufeli et al., 2006). However, there is also some support for the one-factor structure (Schaufeli & Salanova, 2007; Sonnentag, 2003) and, as previously noted, the high intercorrelations of the three factors present an inherent problem as they would for any construct with multiple factors. The present study explored these three measures, comparing them side-by-side in order to determine which measure(s) differentially predict important organisational outcomes.
In addition, the present research suggests that Shirom's vigorous engagement is distinct from other constructs such as job satisfaction and affective commitment. More directly, when the vigor construct was tested at the dimensional level, the physical strength and cognitive liveliness components added unique variance to the prediction of turnover intentions beyond that accounted for by job satisfaction and affective commitment. While the additional variance explained by these two components amounts to slightly more than 2 per cent, even an apparently small amount of additional unique variance explained can be meaningful (see Abelson, 1985). Further, the confirmatory factor analyses suggested that Shirom's vigor approach to engagement has a better operationalisation and possibly a better construction than the structure of Schaufeli's measure. When job satisfaction and affective commitment are considered outcomes, then the Schaufeli measure provides considerable predictive validity, more so than the other two measures, for those outcomes. Interestingly, it is the vigor and dedication components of the Schaufeli measure that provide the predictive validity.
The findings from the relative weights analysis support previous findings that the Schaufeli measure of engagement may correlate more highly with the organisational outcome measures (job satisfaction, affective commitment, etc.) than does Shirom's measure. In contrast, the results suggest that Shirom's vigor measure contributes unique variance to the prediction of such outcomes that Schaufeli's measure is unable to provide. A review of the zero-order correlation may provide some insight into this finding. For example, the correlation between Schaufeli's engagement and job satisfaction is .70. This high level of shared variance is likely to limit the incremental prediction in organisational outcomes for Schaufeli's engagement above and beyond job satisfaction. Alternatively, vigor and job satisfaction are only correlated at .49. Thus, although both constructs share some variance, there are distinct and unique differences between the constructs that allow for unique variance prediction in organisational outcomes above and beyond satisfaction, thereby increasing the utility of Shirom's vigor construct.
The relative weights analysis reveals that Shirom's measure may be better able to predict the outcomes of job satisfaction and affective commitment than it is able to predict turnover intentions. This finding suggests that the effect of vigor on turnover intentions is likely mediated by job satisfaction and affective commitment. Future research will need to verify this finding.
Limitations of the present research include the use of a common method (survey research) and the lack of objective performance data to be used as an additional outcome variable. Furthermore, the organisation used in the present research limited survey length, thus restricting the inclusion of additional constructs as well as forcing attention to parsimony in the administered measures (e.g. using the nine-item measure of the UWES as opposed to the full measure). We were also unable to test for non-response bias due to the nature of the data set, employee confidentiality, and lack of access to employees who opted not to complete the survey. Despite the fact that our response rate was 57 per cent, we were unable to determine if our results were at least in part due to sampling bias. As previously mentioned, the organisation's focus on parsimony forced us to use the nine-item measure of the UWES as opposed to the full 17-item measure, the latter of which has been more frequently used and thus has more empirical support. However, it should be noted that initial research (e.g. Mills, Culbertson, & Fullagar, in press) has found that results yielded from the two versions are very highly correlated with each other, and that the nine-item version may actually be more representative of the engagement construct. Nevertheless, despite these limitations, the present research gains strength from the use of employed individuals in a large organisation whose employment situation is easily generalisable to a majority of white-collar office workers. Furthermore, the present study investigated the nature of engagement by tracking not only one but all three of the publicly available approaches to and measures of engagement. Such a comparison has never before been made in the literature, and is a substantial contribution to the future use of such engagement measures in both theory and practice. Nevertheless, an additional limitation is the lack of inclusion of a new measure of engagement proposed by May, Gilson, and Harter (2004). At the time of data collection for this study, the researchers were unaware of the May et al. (2004) measure of engagement and therefore it was not included in the present comparison. Future research would do well to include it in studies comparing or critiquing engagement measurement instruments. Such future research should also include the various other measures of engagement that have emerged since the present study was conducted (e.g. see Rich, Lepine, & Crawford, 2010).