Understanding consumer online shopping behaviour from the perspective of transaction costs

Table 5.5 Preliminary Data Analysis- Site Design (DES) 
Step One - Exploratory 
Indicators Wording 
EFA 
 
Loadings 
DES1 
This site is visually attractive 
.78 
DES2 
This site is easy to use 
.85 
DES3 
This site is easy to navigate 
.86 
DES4 
This site quickly loads all the text and graphics 
.84 
DES5 
This site provides me with sufficient information 
.84 
DES6 
This site provides me with accurate information 
.85 
DES7 
This site provides me with up-to-date information 
.83 
DES8 
This site is quick and easy to complete a transaction 
.80 
Reliability 
0.94 
KMO 
.901 
Variance Explained 
69.08 
Bartlett’s 
.000 
Step Two – Confirmatory 
Indicators CFA 
 
Stand. Loadings 
Stand. 
Error 
Critical 
Ratio
AVE CR 
DES1 .81 
-- 
-- 
.64 
.914 
DES2 n/a 
n/a 
n/a 
DES3 .77 
.049 
25.21
*
DES4 .80 
.060 
21.95
*
DES5 .93 
.066 
22.95
*
DES6 .82 
.059 
22.57
*
DES7 n/a 
n/a 
n/a 
DES8 .77 
.061 
20.90
*
Model Fit 
Statistic 
Model Fit 
Statistic 
Chi-square 5.527 
TLI 
.999 
df
4 
CFI 
.998 
P .237 
RMSEA 
.020 
GFI .998 
RMR 
.015 
* 
p 
< .001
 
5.5.5 Preliminary Analysis – E- Service Quality 
The e-service quality construct was measured using twelve items. These were subjected to the 
two-step preliminary data analysis process shown in Figure 5.1, the results of which are 
displayed in Table 5.6. Evaluation of the correlation matrix through the KMO and Bartlett’s 
DES1 DES2 DES3 DES4 DES5 DES6 DES7 DES8 
DES1 
1.00 
DES2 
.767 
1.00 
DES3 
.680 
.813 
1.00 
DES4 
.551 .662 .739 1.00
DES5 
.547 .593 .645 .750 1.00
DES6 
.571 .607 .625 .662 .767 1.00
DES7 
.566 .601 .627 .595 .664 .735 1.00
DES8 
.544 .587 .594 .590 .603 .641 .760 1.00 

---

223
Test results in a high KMO statistic (.918) and a significant probability level (p< .001) for the 
Bartlett’s test. These results indicate that sufficient correlations were found within the 
correlation matrix for factor analysis to proceed. In addition, bivariate correlations were 
inspected and all coefficients fell within the acceptable range for factor analysis of .30 
and .90. EFA was then conducted which indicated that REL1 had a cross-loading problem. 
After eliminating this item, EFA produced a three-factor structure for the eleven remaining 
items used to measure e-service quality. The factor loadings ranged from .70 to .88 and the 
cumulative variance explained by three factors was 79.59%, exceeding the recommended 
criterion of 60% (Hair
 et al.
2006). Cronbach’s alpha of .93 was then computed indicating 
good reliability of the scale.
The first factor, composed of four items, accounted for 28.30% of the variance explained. 
The four items loading on this factor reflected reliability with item loadings ranging from .85 
to .92. Cronbach’s alpha for Factor 1 was .92, indicating good reliability of the scale. 
Accounting for 27.78% of the variance explained, Factor 2 included four items related to 
responsiveness. Factor loadings of these four items ranged from .85 to .90 and Cronbach’s 
alpha of .89 indicated an internal reliability. The three items loading on Factor 3 were related 
to personalization. The variance explained by this factor was 23.51% with strong item 
loading ranging from .89 to .93, and Cronbach’s alpha was .90, exceeding the threshold 
of .70 (Nunnally 1978). At this point, as all the remaining eleven items met the criteria of 
Step One, they were retained for CFA analysis in Step Two. 
The eleven items (retained from Step One) measuring e-service quality were subjected to 
CFA in Step Two (refer Figure 5.1) of the preliminary analysis. The model fit indices 
indicated that this measurement model did not fit quite well to the data (
χ
2
/df = 9.763 > 5, 

---

224
RMSEA = .095 > .08, and RMR =.101> .05). In order to improve the model, indicators 
which were related to larger reductions of chi-square or problematic standard residuals (2.5 as 
cut-off) (Anderson and Gerbing 1988) were identified and eliminated one by one. After 
removing item REL5 and item RESP4, an acceptable model was achieved. All overall 
goodness-of-fit statistics were within acceptable ranges. For example, 
χ
2
/df (2.183) was less 
than 5, GFI (.989), TLI (.993) and CFI (.996) were above .90, RMSEA (.035) was less 
than .08 and RMR (.044) was less than .05. Parameter estimates of the final measurement 
model were inspected and no problems were found. Table 5.6 presents the second-order 
measurement model in which
 
e-service quality was the second-order construct with three 
first-order constructs (reliability, responsiveness, personalization) as indicators. The 
standardized factor loadings of the three first-order constructs used to measure the second-
order construct (e-service quality) ranged from .75 to .83, all exceeding the preferable 
criterion of .70 (Hair et al., 1995). The AVE for the second-order construct was .64, 
exceeding the recommended level of .50 (Hair
 et al.
2006). The CR of the second-order 
construct was .840, exceeding the threshold of .70 (Nunnally 1978). More specifically, by 
examining each first-order construct and its corresponding indicators, it can be found that the 
standardized factor loadings of all items were relatively high and significant, ranging 
from .82 to .91, and the AVE for all first-order constructs exceeded the recommended 
criterion of .50, ranging from .73 (responsiveness) to .80 (reliability). CR of all first-order 
constructs exceeded the recommended benchmark of .70, ranging from .891 (responsiveness) 
to .924 (reliability).Therefore, items REL2, REL3, REL4, RESP1, RESP2, RESP3, PERS1, 
PERS2, and PERS3 were considered to constitute a reliable and valid first-order 
measurement scale. Taken together, this three-factor model structure was retained for 
constructing the overall measurement model.

---

225