Re-specification of the Servicescape Construct

The five factor re-specified structure for the 355 respondents were obtained. This model retains 17 of the 23 original items and obtains an excellent fit. Given the large sample size it is not surprising that a significant chi square was obtained (χ² = 205.15, df = 99, p

156

= .00) however, all other indices point towards a well fitting model (RMSEA = 0.05, CFI

= .94, NNFI = .92, NFI = 0.89, IFI = .93, SRMR = 0.054). It is not recommended that all fit indices that are produced by the LISREL program are reported (Hooper et al, 2008) however for those wishing to see a more detailed exposition of the fit indices can refer to Appendix 4G. The parameters for the model (seen in Figure 4.4 and Table 4.12) were all strong and significant with standardised loadings greater than .50. As with the previous model, further tests of convergent validity were made using Fornell and Larcker’s (1981) measure for Composite Reliability and their Average Variance Extracted formula.

Details of these calculations can be found in Appendix 4H while the results are presented in Table 4.12 below. Bagozzi and colleagues (Bagozzi et al, 1991) recommend that composite reliability statistics must be above .60 if reliability is to be assumed. All of the constructs apart from Space and Hygiene were found to meet this criterion. However, the composite reliability for Space and Hygiene was .57 which is only marginally below the cut-off point. The results for AVE were less positive with only two constructs meeting the recommended threshold. As with the previous model, discriminant validity was assessed using Fornell and Larcker’s (1981) stringent test. The calculations for these can be found in Appendix 4I. Discriminant validity was found between all constructs, apart from between the ‘Equipment’ and ‘Space’ dimension and also between the ‘Space’ and

‘Design’. Given that Fornell and Larcker’s (1981) test for discriminant validity is widely regarded as being very difficult to satisfy, further tests of discriminant validity were employed in order to determine whether these violations are to be considered worrying.

The first of these was Bagozzi’s, which states that discriminant validity has been found if the correlation between two constructs (phi φ) is significantly less than 1.0. The second

test of discriminant validity was to set the correlation between the two constructs of interest to 1.0 and to determine whether the model fit significantly improves. A significant improvement to the model would indicate that discriminant validity has not been met. In order to utilise these tests it was necessary model the two constructs in tandem using CFA.

The first model between Space and Design had a Phi (φ) of .34 with a standard error of .05. Using the formula described in Appendix 4D it was found that the Phi between these two constructs is significantly less than 1.0. Prior to fixing the correlations between the two latent constructs to 1.0 the model fit was χ² = 21.13 with 19 degrees of freedom.

Setting the covariances to 1.0 severely effected the fit increasing the Chi square (χ²) to 61.75 with 20 degrees of freedom. Both of these tests indicate that discriminant validity can be found for the two constructs and were therefore left as discrete dimensions.

Following this, Space and Equipment were also re-assessed using the above formulae.

The initial CFA in which no constraints were placed on the intercorrelations between the two latent constructs had an excellent model fit: Chi-Square = 19.13, df = 13, p –value = 0.11920. When the two latent constructs were fixed to 1.0 the model fit significantly worsened with a Chi square difference of 47.29 with 1 degree of freedom. Similarly, an unconstrained model also produces a Phi (φ) of .35 with a standard error of 0.06 which also indicates discriminant validity between the two constructs (Bagozzi and Yi, 1991).

As discriminant validity was found between all of these constructs they were all left as originally hypothesised.

Table 4.12: Results from the Second-Order Factor Analysis of Study Constructs

The background music was pleasant .60 —^a

The background music was appropriate .97 14.58*

Design .78 .80 .50

I found the interior design visually appealing .81 10.87*

The interior design was attractive .81 10.86*

The colour schemes were pleasant .59 —^a

The architecture was attractive .58 8.70*

Equipment .69 .70 .44

The equipment was modern looking .67 10.37*

The electronic equipment was excellent .55 8.82*

The equipment was of high quality .75 —^a

Hygiene and Space .73 .57 .43

The store was very clean .63 9.35*

The service station appeared to be hygienic .72 10.20*

The flooring was appropriate .63 9.35*

I found my way around quite easily .63 —^a

a = The corresponding coefficient was fixed to set the metric of the latent construct.

* = significant at the p = 0.01 level

Model Discussion

As mentioned above, all parameters in the above model were significant and loaded in the direction expected. In addition, the model fit was found to be acceptable, providing empirical confirmation that the servicescape consists of a multi-dimensional structure.

As was expected, all of the servicescape subdimensions significantly contribute to consumers’ overall perceptions of the servicescape. In particular it was found that the Space and Hygiene dimension explained more of the variance in the servicescape construct than any of the other constructs in the model. Shortly following behind this is the Equipment dimension which contributed .84 of the variance in the model. The constructs that contributed the least were Employees (γ = .66) and Ambience (γ = .53).

As this model achieved excellent fit, it indicates that a multidimensional model of the servicescape exists. This therefore supports Hypothesis 1 which was put forward earlier in this chapter. However, as the Hygiene and Space dimensions have not remained as separate dimensions, this part of the hypothesis remains unproven.

4.10.4 Alternative Conceptualisations of the Servicescape: Testing their