Understanding consumer online shopping behaviour from the perspective of transaction costs

Model Fit
 
χ
2
3839.698 
TLI 
.959 
df 1779 CFI 
.963 
P .000 RMSEA 
.035 
GFI .932 
RMR 
.102 
* 
p 
< .001

---

253
 
 5.6.5 Construct Validity for New Model 
Convergent validity of the new overall measurement model was rechecked. Similarly, the 
standardized factor loadings, AVE, and CR related to each construct were examined. As 
shown in Table 5.18, all estimated factor loadings were highly significant (p<0.001 by 
inspecting the critical ratios) as required for convergent validity. All the factor loadings were 
relatively high and greater than the recommended 0.5, which indicated that convergent 
validity was supported (Bagozzi
 et al.
1991). Specifically, the lowest loading obtained 
was .69 linking item4 of risk-bearing propensity, and highest loading was .94 linking item4 
of product quality concern. Moreover, the AVE for all constructs exceeded the recommended 
level of .50, ranging from .51 (risk-bearing propensity) to .78 (customer satisfaction) 
providing another support for convergent validity (Bagozzi
 et al.
1991). CRs of all constructs, 
ranging from .839 (e-service quality and risk-bearing propensity) to .938 (product quality 
concern), exceeded the recommended benchmark of .70, which also demonstrated the 
reliability of the proposed measures.
Discriminant validity was assessed in two ways. Firstly, following the method adopted by 
Fornell and Larcker (1981), discriminant validity is achieved if the square root of the AVE is 
higher than the correlations between two composite constructs. As indicated in Table 5.19, 
the constructs’ square roots of AVE are consistently greater than the off-diagonal correlations. 
Secondly, based on the work of O’Cass and Ngo (2007), it is suggested that discriminant 
validity is exhibited when the corresponding correlation (the off-diagonal entries) is not 
greater than two constructs’ respective reliability value. The results presented in Table 5.19 
show that all individual correlations were lower than their respective reliabilities, which 
shows that discriminant validity is supported.