Understanding consumer online shopping behaviour from the perspective of transaction costs

Figure 5.2 The Structural Model (Used for Testing of H1a-5) 
Control Variables
 
H5: + 
Online store- and product-
related characteristics 
Product Quality Concern 
E-Service Quality 
Reputation of Online Store
Online channel-related 
characteristics 
Perceived Convenience
Privacy and Security Concerns
Perceived 
Consumer TCs of 
Online Shopping
Online 
Purchase 
Behaviour 
Customer 
Satisfaction
Customer 
Loyalty
H2a: +
H3a:
−
Consumer-related characteristics 
Internet Access Availability 
Perceived Internet Expertise
Online Buying Frequency
H1a:
−
H1b:
−
H1c:
−
H2d:
−
Age
Gender 
Income 
Education 
H2c: 
−
H3b: +

---

259
5.7.1 Validity of the Structural Model 
To assess the validity of the structural model for controlling for age, gender, income and 
education level, SEM model fit and structural parameter estimates were examined. The 
results indicated a moderate fit for this structural model consisting of composite measures. 
Some indices met the recommended threshold level. For example, 
χ
2
/df (4.189) was less than 
5, GFI (.959), TLI (.953) and CFI (.976) were above .90, and RMSEA (.058) was less 
than .08.
Nevertheless, since RMR (.052) was not within the acceptable fit criterion,
it was apparent 
that some modification in specification was needed to determine the structural model that 
better represented the sample data. Therefore, both the modification indices suggested by the 
data analysis output and theoretical foundation about the proposed model were taken into 
account when modifications were conducted. Any modification should be substantively 
meaningful in improving model fit. It was found that seven path related modification indices 
were approximately 4 or greater (Hair
 et al.
2006). Of the seven corresponding paths, two 
were not justified by theory (path from reputation of online store to online purchase 
behaviour, and path from online buying frequency to online purchase behaviour). The other 
five paths, which were justified by theory, were freed one at a time. The modified models 
were then tested using SEM. However, no SEM results indicated that the model fit was 
improved significantly by freeing any path. Thus, no change was made for the structural 
model.
Table 5.21 refers to H1a, H1b, H1c, H2a, H2c, H2d, H3a, H3b, H4a, H4b, H4c and H5 when 
controlling for age, gender, income and education level of online shoppers, and shows the 
path coefficients between the exogenous and endogenous variables, average variance 

---

260
accounted (AVA) for, T value and R
2
. The AVA for the endogenous variables was .80 and 
the individual R
2
were greater than the recommended .10 (Falk and Miller 1992) for all of the 
predicted variables. As all of these R
2 
were larger than the recommended levels, an 
examination of the paths’ significance associated with the variables was undertaken. As 
shown in Table 5.21, all but one structural path estimates were significant in the expected 
direction. Specifically, eight paths were significant at the .001 level, two were significant at 
the .01 level, one was significant at the .05 level and one was not significant. In the next 
section, the causal research hypotheses (i.e., H1a to H5) with control variables were each 
tested in turn.

---

261