Understanding consumer online shopping behaviour from the perspective of transaction costs

5.4.2 Description of Step Two 
Step Two of Figure 5.1 is referred to as scale verification and involved conducing scale by 
scale CFAs in order to determine if the dimensions, as measured by the items selected in Step 
One, were truly convergent and unidimensional, and to examine if individual measurement 
model provided a good fit to the data. At this point, the empirical assessment of convergent 
validity of the scale was assessed using guidelines recommended by Hair et al. (2006). As 
they advocated, the guidelines are: all standardized loadings for a construct should be at least 
0.50, and preferably 0.70; the threshold for composite reliability (CR) is considered to be 
0.70; and average variance extracted (AVE) should equal or exceed 50 % (Hair
 et al.
2006). 
Therefore, the standardized factor loadings, CR, and AVE related to each construct were 
examined to determine convergent validity.
When examining the results of the CFA for each construct, the model fit statistics also need 
to be checked, as recommended by Holmes-Smith and Coote (2012). As outlined in Chapter 
4, this study reported the GFI, TLI, CFI, RMSEA and RMR in combination with reporting 
the 
χ
²
/df. The benchmark for GFI, TLI and CFI is 0.90. Their value should be more than 0.90. 
For RMSEA and RMR, their value should be less than 0.08 and 0.05, respectively. For 
χ
²
/df, 
the value should be less than 5. Cut-off scores are also shown in Step Two in Figure 5.1. 
Those scales that produced acceptable fit statistics through CFA (obtained using AMOS 21.0 
with fit being assessed via maximum likelihood) were then deemed to be both reliable and 
valid indicators of the measured constructs.

---

215
The two-step process of preliminary data analysis shown in Figure 5.1 provided a framework 
which involved the sequential examination of the data in order to produce both valid and 
reliable factor structures upon which composite measures were computed and used in 
structural equation modelling. The preceding discussion of the two-step model in Figure 5.1 
provides a detailed description of the preliminary data analysis undertaken and is the basis 
upon which the ensuing preliminary results are presented.