Chapter 4 – Research Methodology
5.3 Data Analysis Interpretation
5.3.4 Data Analysis Interpretation of the Structural Equation Modelling for the
the Primary Dimensions Model
5.3.4.1Interpretation of the First-Order Measurement Model of the Primary Dimensions
Based on the research model in Figure 3.1, the first-order model for service quality as illustrated in Measurement Model 4 (see Figure 4.5) was designed to examine the relationships between the three primary dimensions (interaction quality, physical environment quality and outcome quality) and the measured items (see Figure 5.13).
The first-order model presented seven items. The number of measured items was
v = 28 pieces of information
7
71
228
and the number of estimated parameters wasp = 17 parameters (4 regression weights, 3 covariances and 10 variances); the model was over-identified with 11df (28 pieces of information - 17 parameters).The first-order model was statistically significant at the .05% level and model-fit-indices according to all indicators demonstrated that the model was acceptable based on the recommended thresholds presented in Table 4.5
112 32.243;χ2 df 2.931; GFI = 0.962; NFI = 0.981; RMSEA = .087; CFI = 0.987; and RMR= .018) (see Table 5A.8, Appendix 4). In reviewing the model-fit-indices, model modification was not necessary because the first-order model had model-fit-indices that were more than satisfactory.To verify the construct validity, the convergent and discriminant validities were measured (Fornell & Larcker, 1981; Hair et al., 2010; Janssens et al., 2008). Table 5A.8 (see Appendix 4) showed that each standardized factor loading ranged from 0.857 to 0.952, well above the acceptable value and statistically significant at the .001% level, indicating the unidimensionality of each scale, showing that convergent validity was obtained.
Reliability must always be verified (Cronbach, 1951; Hair et al., 2010). Reliability was verified with Cronbach alpha, composite reliability and AVE. Cronbach alpha for all constructs in the first-order model were 0.911, 0.917 and 0.917, all of which exceeded the threshold level of 0.60, indicating high internal consistency of the measurement scales (Churchill, 1979; Hair et al., 2006; Janssens et al., 2008). Using Equation 4.1, the composite reliability was calculated; interaction quality (0.913), physical environment quality (0.918) and outcome quality (0.920) all exceed the threshold level of 0.60 indicating high internal consistency of the measurement scales (Bagozzi & Yi, 1988; Fornell & Larcker, 1981; Hair et al., 2006). Using Equation 4.2, the AVE values for the three primary dimensions were 0.841, 0.848, and 0.792; all exceed the minimum criterion
AVE0.50
(Fornell & Larcker, 1981; Hair et al., 2010; Janssens et al., 2008). These three constructs indicated that the variance due to measurement error was less than the variance captured by the construct (Fornell & Larcker, 1981). This evidence supported the reliability of each scale which indicated that convergent validity was obtained, therefore showing that the measurement items had high reliability and validity (Bagozzi & Yi, 1988; Chen, 2008; Fornell & Larcker, 1981). Once convergent validity was achieved, it was appropriate to test for discriminant validity. The correlation estimates on all pairs of the three subdimensional factors of outcome quality were 0.708, 0.842, and 0.774 which areless than the recommended value
r0.85
of Kline (2005), indicating that discriminant validity existed (see Table 5A.8, Appendix 4).In summary, the analysis of the measurement model of the primary dimensions suggested that the scales used in the study adequately captured the latent constructs. The measurement model generally exceeded the criteria established in Table 4.5 for model-fit- indices. In addition, all conditions required for examining the convergent and discriminant validities recommended by several researchers (Anderson & Gerbing, 1988a; Byrne, 2009; Chinna, 2009; Kline, 2005; Nokelainen, 2009; Schumacker & Lomax, 2004) were satisfactorily met. The measurement model represented the best model fit for measuring the service quality structure with the present data; therefore, the model was used for the structural model.
5.3.4.2Interpretation of the Second-Order Model for the Primary Dimensions Model
Based on the research model illustrated in Figure 3.1 and the results of the modified first-order model of service quality, Structural Model 4 in Figure 4.10 was initially specified. Specifically, the second-order model for service quality was designed to test the relationships between the three primary dimensions (interaction quality, physical environment quality and outcome quality) and one independent second-order construct, service quality.
Before examining the validity of the second-order structural model, it was essential to address the identification issues in the second higher-order (Byrne, 2009). The first-order model was over-identified with 6df;however, with the second-order, model identification had to be re-specified to check the identification status (Byrne, 2009). In this case, the higher-order structure with three first-order factors was v = 6 pieces of information
331 26
and the number of estimated parameters was p = 6 parameters (3 factor loadings and 3 residuals); the model was just-identified (6 pieces of information = 6 parameters).The second-order model was required to be over-identified; therefore, to solve the just-identified problem, Byrne (2009) suggests placing equality constraints on particular parameters that are approximately equal. The two higher order residuals chosen for this were physical environment quality (0.77) and outcome quality (0.92). In a more detailed inspection of the variances, the CRDIFF between these two residuals was compared with the critical value of 1.96. From the CRDIFF listing, both residuals were less than the
critical value of 1.96, thus, the hypothesis that these two residuals’ variances were equal in the population was accepted (Kim, 2003). Given this information, it was reasonable to place equal constraints (var_a) on these two residuals (Byrne, 2009; Kim, 2003). In this case, the second-order structure was v = 6 pieces of information
3
31 26
and the number of estimated parameters was p =5 parameters (3 factor loadings and 2 residuals); the model was over-identified with 1df (6 pieces of information - 5 parameters). The second-order model is shown in Figure 5.14.Figure 5.14: Structural Model 4 – Second-Order Model for Primary Dimension
The model-fit-indices indicated that the present data fitted the model (see Table 5A.9, Appendix 4). The results were statistically significant at the .05% level and all model-fit-indices sufficiently satisfied their relevant recommended thresholds as reported in Table 4.5
212 34.553;2 df 2.879; GFI = 0.960; NFI = 0.979; CFI = 0.986; RMR = .021; and RMSEA = .086). As a result, the modified second-order model was determined as the final model to represent an adequate description of the primary dimension of structure in the present study. Thus, the modified second-order model was used to examine Hypotheses 4 to 6, which in turn, satisfies Research Objectives 1 and 3.5.3.4.3Hypothesis Testing
In this structural model, the second-order model for service quality was designed to examine the hypothesis that service quality is a multidimensional construct composed of three primary dimensional factors (interaction quality, physical environment quality and outcome quality), which in turn, partially satisfies Research Objective 1. In addition, Hypothesis 8 proposed that restaurant patrons will vary in their perceptions of the importance of each primary dimension, in turn, partially satisfying Research Objective 3.
As indicated in Table 5.21, all three primary dimensions positively affected overall service quality perceptions and were statistically significant at the .001% level with outcome quality
0.920
being the most important, followed by physical environment quality
0.890
and interaction quality
0.811
. Thus, the results supported Hypotheses 4, 5, 6 and 8 therefore satisfying Research Objectives 1 and 3. The second- order latent variable, represented by service quality, explained 85% of the variance for outcome quality, 80% of the variance for physical environment quality and 66% of the variance for interaction quality.Table 5.21: Structural Parameter Estimates Hypothesized Path Standardized
Coefficients Path
Critical Ratio2
R Assessment
H4: IQ SQ 0.812 13.967*** 0.659 Supported
H5: PEQ SQ 0.905 15.815*** 0.819 Supported
H6: OQ SQ 0.929 16.990*** 0.864 Supported
Note: Statistically significant at *** = .001; ** = .05; * = 0.10