To avoid the problem of multivariate normality, maximum likelihood
method (Byrne, 2001) was used to estimate the measurement model.
The five-factor scale was not acceptable, since a solution could not be
achieved due to negative error variance of variables 00026 (staff courtesy)
and 00022 (service system 2). Therefore, these attributes were removed
from the model. Subsequently, staff courtesy and service system 2 became
single-attribute factors and were, therefore, removed. The resulting threefactor
scale was tested by CFA and the model could not be accepted due
to poor indices of fit (AGFI = 0.645; NFI = 0.582; CFI = 0.598; RMSEA
= 0.199). After repeated modification and removal of variables with crossloadings
whose correlations were adversely affecting the fit of the model
(based on modification indices readings), the final accepted modified service
quality measurement scale consisted of two dimensions (as given in
Fig. 2.0). The service quality dimensions and the attributes loading on
them are given in Table 2.0. The two-dimensional scale of service quality
was tested by CFA and was found to be of acceptable fit (AGFI = 0.999;
CFI = 1.0; NFI = 0.996; RMR = 0.005; RMSEA = 0.000), as given in Table
3. The t statistics were significant for each path, and the critical ratios
were more than twice the standard errors.