Диссертация (1136702), страница 21
Текст из файла (страница 21)
Анализ психометрических свойстврусскоязычной версии Утрехтской шкалы увлечённостиработой (UWES-9)8Participants and ProcedureThe sample included 1783 employees: 516 (29%) men, 1213 (68%) women, and54 participants who did not specify their gender) of a major Russian energy company,which engages in the production and sale of heat and electricity in several regions ofRussia.Scales1. Work engagement was measured by the Russian version of the short UtrechtWork Engagement Scale (UWES-9) (Kutuzova, 2006; Schaufeli & Bakker,2004b).2. Burnout was measured by the Russian analogue of the MBI (Maslach & Jackson,1986), which was developed by Vodopiyanova and Starchenkova (2009).3.
Job satisfaction was measured by the Brief Index of Affective Job Satisfaction(Thompson & Phua, 2012).4. Life satisfaction was measured with the use of the Russian version of theSatisfaction with Life Scale (Diener, Emmons, Larsen, & Griffin, 1985), whichwas translated and adapted by Osin and Leontiev (2008).5. Turnover intention was measured by a single item (“I will look for a new job inthe near future”) (Bozeman & Perrewe, 2001).Data AnalysisTo check the factor structure of the Russian version of the UWES-9, a series ofconfirmatory factor analyses (CFA) were performed with the lavaan R packageВ данном приложении приведены фрагменты текста опубликованной статьи: Lovakov, A., Agadullina, E.
R., &Schaufeli, W. B. (2017). Psychometric properties of the Russian version of the Utrecht Work Engagement Scale(UWES-9). Psychology in Russia: State of the Art, 10(1), 145–162. http://doi.org/10.11621/pir.2017.01118113(Rosseel, 2012), using the maximum likelihood parameter estimates with standarderrors and chi-square test statistics that are robust to non-normality (MLR). Wecompared the two alternative models: the 1-factor model, in which all 9 items wereassessed as one common scale of work engagement, and the 3-factor model, in whichitems were divided into three factors (each including three items) reflecting the threesubscales of work engagement. The CFA results were evaluated by using severalindicators: χ2, the root mean square error of approximation (RMSEA); thestandardized root mean square residual (SRMR); the comparative fit index (CFI); andthe Tucker Lewis Index (TLI). Models showing values of greater than .93 for the CFIand TLI, up to .08 for the RMSEA, and up to .06 for the SRMR were judged asshowing a good fit (Byrne, 2011; Hu & Bentler, 1999).In order to investigate the measurement invariance of the UWES-9, weperformed a series of multi-group CFAs, which tested the configural (same structureacross groups), metric (same factor loadings across groups), scalar (same factorloadings and item intercepts across groups), factor variance, factor covariance, andfactor means invariances of the model across the gender and age groups (Vandenberg& Lance, 2000).
The differences between the nested models were evaluated by usingΔχ2 (Satorra & Bentler, 2001) and ∆CFI. We relied on the ΔCFI > .01 criterion ofsignificant difference (Cheung & Rensvold, 2002). For evidence of convergent anddiscriminant validity, we used the Spearman correlation coefficients between theUWES-9 and other dimensions such as burnout, and job and life satisfaction.To examine the incremental validity, we conducted a series of hierarchicalregression analyses (ordinary least squares method) to determine the additional powerof the UWES-9 in predicting job and life satisfaction and turnover intention,controlling for burnout subscales.ResultsDescriptive analysis of the items of the Russian version of the UWES-9Table 1 shows the descriptive statistics for each item of the Russian version ofthe UWES-9.
The average values of the individual items lie within the range of 4.23–1145.36. For all items, there is a small but non-significant negative skew, reflecting theaverage value of displacement in the direction of high scores. Items 1, 2, and 8 showa small positive kurtosis, and item 5 shows a small negative kurtosis. The distributionof the scores is “approximately normal” because the skewness and kurtosis are notsignificant.Table 1. Descriptive statistics of the items of the Russian version of the UWES-9ItemMSDMedianSkewKurtosis1. VI-1 [1]4.631.135-0.471.052. VI-2 [4]4.851.205-0.400.483.
DE-1 [5]5.051.185-0.310.244. DE-2 [7]4.731.315-0.230.045. VI-3 [8]4.231.494-0.21-0.346. AB-1 [9]4.761.375-0.400.047. DE-3 [10]5.161.355-0.490.018. AB-2 [11]5.361.165-0.530.589. AB-3 [14]4.401.434-0.310.03Note. VI = Vigor. DE = Dedication. AB = Absorption. The square brackets indicate the number ofthe item in the full version of the UWES-17.Factorial Structure of the Russian version of the UWES-9Table 2 shows the χ2 statistics and other fit indices obtained by the CFA of theRussian version of the UWES-9. The 1-factor model with all 9 items loading on thecommon factor of work engagement did not show a good fit with the data.
The valueof the CFI and the TLI were below the threshold of .93, and the value of the RMSEAwas above .08. The analysis of the modification indices revealed an error covariancebetween three pairs of items: VI-1 (1) (“At my work, I feel bursting with energy”)and VI-2 (2) (“At my job, I feel strong and vigorous”); VI-3 (5) (“When I get up inthe morning, I feel like going to work”), and AB-1 (6) (“I feel happy when I amworking intensely”); and AB-2 (8) (“I am immersed in my work”) and AB-3 (9) (“Iget carried away when I’m working”) (value varies from .26 to .41).
Errorcovariances between VI-1 and VI-2 and AB-2 and AB-3 were also noted in theUWES-9 versions in other languages (Balducci et al., 2010; Littman-Ovadia &Balducci, 2013; Seppälä et al., 2009; Zecca et al., 2015).115Since the covariances can be meaningfully explained (sequence, similarwording), we followed the practice of previous authors and decided to include themin the model; we modified the 1-factor model by freeing error covariances VI-1/VI-2and AB-2/AB-3. The error covariance between VI-3 and AB-1 is not in line withprevious literature, and both items belong to different subscales; thus it was notincluded in the model.
The modified 1-factor model showed an acceptable fit withdata (see Table 2). All factor loadings differed significantly from zero (p < .001) andexceeded .55. The general 9-item work engagement scale has high internalconsistency (α = .92 [95% CI: .90–.93]). Relatively high correlation coefficientsbetween the UWES-9 subscales (r varies from .68 to .79) (see Table 3) also confirmthe similarity in scores on the UWES-9 items. In other words, the UWES-9 can beconsidered a single work engagement scale.The 3-factor model, in which the items are grouped into three subscales (Vigor,Dedication, and Absorption), also did not show good fit with the data.
The values ofthe CFI and the TLI were below the threshold of .93, and the value of the RMSEAwas above .08. The analysis of the modification index showed the same errorcovariances as in the 1-factor model (values ranging from .29 to .43). Once modifiedby freeing the same VI-1/VI-2 and AB-2/AB-3 error covariances, the 3-factor modelshowed an acceptable fit with the data. All factor loadings differed significantly fromzero (p < .001) and exceeded .58.
The correlations between latent factors exceeded.90. All three subscales (Vigor, Dedication, and Absorption) had good internalconsistency (α varies from .75 to .87). In other words, the UWES-9 can be considereda three-component measure with three separate subscales. Therefore, both the 1factor and the 3-factor models fit well with the data. However, the 3-factor modelshowed a significantly better fit with the data (Δχ2 = 10.75, Δdf = 3, p = .013).116Table 2.
Goodness of fit indicators for measuring models of the UWES-9Models1-factor model1-factor model (modified)3-factor model3-factor model (modified)χ2pdfRMSEA[90% CI]CFITLISRMRAIC630.39323.66616.75319.73< .001< .001< .001< .00127252422.11 [.11–.12].08 [.08–.09].12 [.11–.13].09 [.08–.09].90.95.90.95.87.93.85.92.05.03.05.0344475.7944001.9944409.6043987.83Note.
df – degree of freedom; RMSEA – root mean square error of approximation; CFI –comparative fit index; TLI – Tucker Lewis index; SRMR – standardized root mean square residual;AIC – Akaike information criterion.A series of multi-group CFAs across gender and age groups was conducted toprovide evidence of the UWES-9 measurement invariance across different groups.The results of the model fit tests and the model comparison are summarized in Table4. First, a multi-group CFA was conducted across male and female samples. Modelsof configural invariance (Model 1a), metric invariance (Model 2a), and scalarinvariance (Model 3a) showed an acceptable fit and did not statistically differ fromtheir predecessor models (Δχ2 test not significant, ΔCFI < .01). These resultssuggested an invariant 3-factor structure of the UWES-9 in both samples so that it ispossible to compare scores across gender groups. Models of factor varianceinvariance (Model 4a) and factor covariance invariance (Model 5a) also demonstratedan acceptable fit and did not statistically differ from Model 3a.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
