attachment.
628
H. J. Kim
Focus, Prior Customer Relationship, Organizational Support, and Service
Under Pressure); (b) six from service quality (Tangibles I, Tangibles II,
Tangibles III, Reliability, Combination of Responsiveness and Assurance,
and Empathy); (c) two from customer satisfaction (two satisfaction items);
and (d) two from customer loyalty (Future Buying Intentions and Emotional
Attachment). The value of each indicator (subfactor) for the two constructs
(Service Quality and Service Orientation) was a composite score obtained
by averaging all variables in its respective subdimension; the indicators for
the two remaining constructs (Customer Satisfaction and Customer Loyalty)
used raw scores.
The paths connecting latent variables and their respective indicators
are called the measurement model, and the paths connecting the sets of
latent variables are called the structural model. As recommended by SEM
researchers (Hair, Anderson, Tatham, & Black, 1998; Joreskog, 1993), two
steps were used to test the proposed model. The first step involved con-
firmatory factor analysis to assess the validity of the measurement model.
The second step involved the generation of a structural model that tests the
research hypotheses.
To evaluate fits of both the measurement and structural model, several
fit indices were used including the chi-square statistic; the goodness-of-fit
index (GFI: a measure of the correspondence of the actual covariance matrix
with that predicted from the proposed model); the adjusted GFI index (AGFI:
a GFI index adjusted for sample size); the standardized root mean square
residual (standardized RMSR: average standardized residual value derived
from the fitting of the covariance matrix for the proposed model); and the
normed fit index (NFI: a measure that provides the incremental improvement
of the fit of the proposed model from a baseline model). After the evaluation
of the fit indices, the parameter estimates with their associated significance
levels using t values (parameter estimate or standard error) were reported
for the proposed model.
RESULTS
Model Fit
The overall chi-square for the measurement model was 107.12 with 71
df and a
p value less than .0057. When the chi-square is not significant
(p
> .05), the model fit is appropriate; that is, there is no significant differ-
ence between the actual matrix and the predicted matrix (Loehlin, 1992).
The small p value of the model indicated a significant difference between
the actual matrix and the predicted matrix. The chi-square statistic is known
to be very sensitive to sample size and the number of parameters estimated;
thus, using the normed chi-square (
χ
2
/df ) is appropriate (Hair et al., 1998;
Wheaton, Muthen, Alvin, & Summers, 1977). The normed chi-square had
a value of 1.51 (107.12
/71) for the measurement model. This falls well