Understanding consumer online shopping behaviour from the perspective of transaction costs

5.10 TESTING FOR MODEL DIFFERENCES ACROSS PRDOCUT CATEGORIES 
In order to test the final hypotheses of this study (H9a and H9b) which postulate that the 
proposed model will differ across different product categories, the sample was split into two 
groups - data collected with reference to search products and data pertaining to experience 
products. This resulted in one sample containing 455 cases (search products) and the other 
sample containing 507 cases (experience products). Each sample was then analysed 
separately via AMOS 21.0 in the same fashion as the previous analysis for the overall model. 

---

275
The results of individual SEM analysis are initially presented prior to making a comparative 
analysis.
5.13.1 Model Results --- Search Products 
The model for search products exhibited a 
χ
2 
of 273.820 with df = 99, p < .01; 
χ
2
/df = 2.766 < 
5, GFI = .943 > .90; CFI = .950 > .90; TLI = .904 > .90; RMSEA = .062 < .08. All statistics 
for this model were at the appropriate levels, except the RMR, which was .063 > .05. Table 
5.29 presents the results of testing the structural model (Figure 5.4) for the sample pertaining 
to search products by providing the path coefficients between the exogenous and endogenous 
variables, average variance accounted (AVA) for, T value and R
2
. All numbered hypotheses 
contain an additional “s” which denotes the use of the sample pertaining only to search 
products in this analysis.
The AVA for the endogenous variables was .65 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 it was appropriate to examine the paths’ 
significance of the paths associated with these variables. The absolute value of the product of 
the path coefficient and T value were used to evaluate the significance of the individual paths 
(Falk and Miller 1992). As displayed in Table 5.29, nine out of twelve paths met the criterion 
of T value (greater than 1.96), which indicated that the nine paths had significant effects. 
Further examination of the standardized path coefficients found that all nine paths were 
significant in the hypothesized direction. Accordingly, H1b(s), H1c(s), H2c(s), H2d(s), 
H3a(s), H4a(s), H4b(s), H4c(s) and H5(s) were supported. Nevertheless, the results revealed 
that three paths failed to meet the T value’s recommended level (above 1.96), which implied 
that these three paths were not significant. Therefore, H1a(s), H2a(s) and H3b(s) were 

---

276
rejected. Furthermore, the data showed that 92 per cent of the variance in consumer TCs was 
explained by five antecedent variables (i.e., perceived Internet expertise, online buying 
frequency, e-service quality, reputation of online store, and perceived convenience), while 
consumer TCs explained 64 per cent of the variance in customer satisfaction. In addition, 
customer satisfaction and TCs accounted for 55 per cent of the variance in customer loyalty. 
48 per cent of the variance in online purchase behaviour was explained by TCs.
The preceding SEM analysis of the proposed model for search products provided support for 
all but three hypotheses, which indicated that the proposed model received mixed support in 
the context of search products. Figure 5.4 presents the proposed model for search products 
showing all path coefficients and R
2
values for endogenous variables.

---

277