Диссертация (1136702), страница 24
Текст из файла (страница 24)
Each of the five components loaded ontothe expected second-order factor. All the second-order factor loadings exceeded .60,with many above .80, and differed significantly from zero (p < .05). The secondorder factors of self-definition and self-investment were strongly and significantlyrelated in all samples (.66–.84, all p < .05). All of these parameters confirm thatModel E, with five components and two second-order factors, was well defined by itsitems. The fit indices indicate that Model E fitted the data better, compared to the twoalternative models (Model F = five-components: individual self-stereotyping, ingroup homogeneity, solidarity, satisfaction and centrality, and one dimension:identification; Model G = alternative five-component/two-dimensional: selfdefinition (i.e., individual self-stereotyping, in-group homogeneity, and centrality)and self-investment (i.e., satisfaction, solidarity).129In order to investigate the factorial invariance of the instrument, we performed aseries of multi-group CFA analyses testing configural, metric, and scalar invarianceof the model across the 4 samples.
Latent factors were identified by fixing variance to1. Because the complete first- and second-order configural invariance model hadconvergence issues, we started by establishing the invariance of the first-orderstructure and then used the scalar-invariant first-order model as a baseline for thesecond-order models. This strategy resulted in a sequence of nested models, allowingto use the DIFFTEST function in order to compare the fit.
As the chi-square test isoverly sensitive in large samples, we relied on the ΔCFI > .01 criterion of asignificant difference between nested models (Cheung & Rensvold, 2002). At eachstage, we first evaluated the fit of a completely invariant model and then proceededby establishing partial measurement invariance (Byrne, 2011; Byrne, Shavelson, &Muthen, 1989). The results of model fit tests are summarized in Table 2 and those ofmodel comparison are presented in Table 3.Table 2.
The Fit of Multi-Group Models of In-Group IdentificationModelχ2, pdfRMSEA [90% CI] CFI TLI SRMRFull measurement invariance approachStage 1: first-order modelsModel 1 (configural)422.27, p < .001 268 .053 [.043, .062].953 .936 .051Model 2a (metric)485.89, p < .001 295 .056 [.047, .065].941 .928 .077Model 3a (scalar)693.58, p < .001 322 .075 [.067, .082].886 .871 .099Stage 2: second-order modelsModel 4a (configural)716.81, p < .001 338 .074 [.066, .081].884 .875 .102Model 5a (metric)728.54, p < .001 347 .073 [.066, .080].883 .877 .106Model 6a (scalar)777.90, p < .001 353 .076 [.069, .084].869 .865 .118Model 7a (factor cov)806.97, p < .001 356 .078 [.071, .086].861 .858 .188Partial measurement invariance approachStage 1: first-order modelsModel 2b (metric)471.59, p < .001 294 .054 [.045, .063].945 .932 .073Model 3b (scalar)532.00, p < .001 317 .057 [.049, .066].934 .924 .079Stage 2: second-order modelsModel 4b (configural) 556.37, p < .001 333 .057 [.049, .065].931 .925 .082Model 5b (metric)570.11, p < .001 342 .057 [.049, .065].930 .925 .088Model 6b (scalar)575.86, p < .001 346 .057 [.048, .065].929 .926 .088Model 7b (factor cov)575.45, p < .001 347 .056 [.048, .065].930 .926 .088Note.
df = degrees of freedom; RMSEA – root-mean-square error of approximation; CFI –comparative fit index; TLI – Tucker-Lewis index; SRMR – standardized root-mean square residual.130Table 3. Multi-group model comparison resultsModel comparisonΔχ2, pΔdfΔRMSEAΔCFIFull measurement invarianceModel 2a vs. Model 183.28, p < .00127.003.012Model 3a vs. Model 2a381.81, p < .00127.019.055Model 4a vs.
Model 3a31.28, p = .01216.001.002Model 5a vs. Model 4a16.10, p = .0659.001.001Model 6a vs. Model 5a88.21, p < .0016.003.014Model 7a vs. Model 6a30.09, p < .0013.002.008Partial measurement invarianceModel 2b vs. Model 165.36, p < .00126.001.008Model 3b vs. Model 2b96.35, p < .00123.003.009Model 4b vs. Model 3b31.40, p = .01216<.001.003Model 5b vs.
Model 4b17.78, p = .0389<.001.001Model 6b vs. Model 5b7.96, p = .0934<.001.001Model 7b vs. Model 6b0.00, p = .9871.001<.001Note. Δdf = difference in degrees of freedom; ΔRMSEA – change in root-mean-square error ofapproximation; ΔCFI – change in comparative fit index.The fit of the first-order configural invariance model (1) was good. Though thefit of the full metric invariance model (2a) was statistically worse, the absolutedifference was not large and there were no pronounced outliers among modificationindices (MI). The strongest MI (χ2 = 12.26) referred to the loading of item 2 in theOrthodox sample; removal of the respective constraint resulted in a statisticallysignificant, but marginal improvement of the model (2b). The fit of the first-order fullscalar invariance model (3a) was much worse, compared to the model 2a, suggestingsome strongly non-equivalent item intercepts. Based on MI, we relaxed interceptconstraints (one at a time) for item 3 in the Gender sample (χ2 = 44.94), item 2 in theStudent sample (χ2 = 38.55), item 5 in the Orthodox sample (χ2 = 26.34), and item 8in the Student sample (χ2 = 22.54).
The resulting partial scalar invariance model (3b,based on model 2b) had no pronounced MIs for intercepts (the strongest MI: χ2 =6.70) and exhibited satisfactory fit.When configural invariance second-order part of the model was introduced intothe models 3a and 3b, the fit did not become worse, according to the ΔCFI criterion,both for the model developed using full invariance approach (4a) and partialinvariance approach (4b).
Introduction of metric invariance constraints into thesecond-order part of the model did not lead to a deterioration of model fit either131(models 5a and 5b). There was empirical under-identification in the means part of thesecond-order scalar invariance model, and second-order factor means for Student andGender samples were constrained to 0 in order for the models 6a and 6b to converge.Comparison of model fit indices indicated some pronounced non-invariance of factorintercepts, and we relaxed constraints for the intercepts of second-order solidarityfactor in the Gender sample (χ2 = 54.03) and the individual self-stereotyping factor inthe Orthodox sample (χ2 = 21.70), after which the fit of model 6b did not differstatistically from that of model 5b.Finally, in model 7 we constrained the covariance of second-order factors to beequal across groups.
Because the model fit deteriorated, in the partially invariantmodel we relaxed this constraint for the Russian sample (χ2 = 28.04) and for theOrthodox sample (χ2 = 10.09), after which the fit indices of the models 6b and 7b didnot differ.Finally, we computed the scale scores for each component. All the five scaleswere of moderate or high reliability, Cronbach’s α ranged between .65 and .93 (seeTable 4). Correlations between the five components were moderate or high, but in allfour in-groups inter-correlations were higher for the components referring to thesame dimension. Satisfaction, solidarity, and centrality have higher correlations witheach other than with individual self-stereotyping or with in-group homogeneity, butcorrelations between individual self-stereotyping and in-group homogeneity werehigher than their correlations with satisfaction, solidarity, and centrality (see Table 4).This supports the hierarchical conceptualization proposed by Leach et al.
(2008) andthe results of the CFA. Therefore, the results support the use of the subscales andconfirm the structural validity of the Russian version of the measure.132Table 4. Descriptive Statistics and Inter-correlations for Five Components of InGroup IdentificationIn-group / ComponentМSDα12345Russians1. Individual self-stereotyping2. In-group homogeneity3. Satisfaction4. Solidarity5. Centrality5.205.005.875.685.251.581.391.341.381.69.91.77.93.90.89–.67**.66**.69**.68**–.52**.55**.60**–.79**.74**–.72**–Orthodox (Christians)1. Individual self-stereotyping2.
In-group homogeneity3. Satisfaction4. Solidarity5. Centrality3.432.974.444.333.92.89.91.68.80.94.88.67.73.73.74–.56**.46**.44**.42**–.38**.44**.39**–.51**.54**–.44**–Students1. Individual self-stereotyping2. In-group homogeneity3. Satisfaction4. Solidarity5. Centrality4.404.035.895.274.931.311.181.121.131.33.90.65.92.83.82–.50**.41**.42**.44**–.31**.30**.40**–.67**.64**–.50**–Males / Females1.
Individual self-stereotyping 4.46 1.46 .91–2. In-group homogeneity4.24 1.39 .69.59** –3. Satisfaction5.84 1.20 .91.39** .24** –4. Solidarity4.96 1.08 .70.36** .22** .47** –5. Centrality5.34 1.27 .81.39** .23** .47** .37** –Note. In the Orthodox sample we used a 5-point scale. Therefore, means and standarddeviations are lower than in the student and Russian samples. Bold correlations are those of scalesthat refer to the same dimension. ** p < .01Examining ValidityParticipants and MeasuresWe performed additional studies to examine the convergent and discriminantvalidity of the Russian version of the measure of in-group identification. As part ofthe same survey session, two of the three Study 1 samples (undergraduate students ofthe Higher School of Economics and people who identified themselves as Russians)completed several additional measures related to the in-group identification.133Sample 1146 people who identified themselves as Russians in Study 1 completed severaladditional measures:1.
Multigroup Ethnic Identity Measure (MEIM). To gauge the convergent validity ofthe Leach et al. (2008) measure, we used MEIM (Phinney, 1992). This scaleconsists of 12 items scored on a 4-point frequency scale and comprises twosubscales: Affirmation and Belonging (α = .92), and Identity Search (α = .84).Participants completed the Russian version of the MEIM (sample items: “I amglad that I belong to Russians”, “I follow the traditions of Russians”) (Tatarko &Lebedeva, 2011). Participants were asked to indicate their identification asRussians (as opposed to immigrants from Central Asian countries).
As Leach et al.(2008) suggested, the Affirmation and Belonging subscale includes items similarto the solidarity, centrality, and satisfaction components and the Identity Searchsubscale includes items similar to the centrality and satisfaction components.2. Self–Group Overlap.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
