Understanding consumer online shopping behaviour from the perspective of transaction costs

4.9 DATA ANALYSIS TECHNIQUES: STRUCTURAL EQUATION MODELLING 
Structural equation modelling (SEM) is a powerful statistical technique that allows 
measurement analysis (specifying relationships among observed variables underlying latent 
variables) and structural analysis (specifying relationships among the latent variables) to be 
performed simultaneously (Kelloway 1995, Schumacker and Lomax 2004, Kline 2011). It 

---

196
allows more flexible assumptions, has the ability of testing models overall rather than 
coefficients individually, is able to model mediating variables, and can handle difficult data 
(Anderson and Gerbing 1988, Hair
 et al.
2006). Schumacker and Lomax (2004) describe the 
following five basic-building blocks of all SEM analyses: model specification, model 
identification, model estimation, model fit testing and model modification. These basic-
building blocks are absolutely essential to both the measurement and structural models. 
SEM is particularly valuable in inferential data analysis and hypothesis testing where the 
pattern of relationships among the study constructs is specified a priori and grounded in 
established theory. It allows the researchers to test prior theoretical assumptions against 
empirical data statistically (MacCallum and Austin 2000, Schumacker and Lomax 2004, 
Arbuckle 2011). For these reasons, the present study employed SEM (using the AMOS 21.0 
software program) to test the proposed hypothesized models. However, there are a number of 
issues that must be addressed when using the SEM technique. 
4.9.1 Analysis Approach 
The first issue in the application of the SEM technique is the sequence in which structural and 
measurement analysis should occur. Although SEM is capable of testing the measurement 
model and structural model simultaneously, Anderson and Gerbing (1988) recommend a two-
stage model-building approach and that the measurement model should be tested separately, 
using confirmatory factor analysis (CFA) in order to detect any inadequacy in fit, prior to 
testing the full structural model. They suggest that the measurement model provides an 
assessment of convergent and discriminant validity, while structural model provides an 
assessment of predictive validity. By using a sequential approach in analysis, the researcher is 
able to pinpoint where a model is misspecified (Anderson and Gerbing 1988). The present 

---

197
study adopts this approach by measuring the fit of a model with the current data, before 
testing the structural relationships among the constructs in the model. 
In addition, Mulaik and Millsap (2000) and Byrne (2009) suggest that it is best to have a few 
good indicators for each of the latent variables in order to check more easily how well each 
observed variable measures a latent variable. Therefore, rather than using individual items as 
indicator variables, the present study also uses EFA for each of the proposed sixteen latent 
variables and so reduce a large number of related items to a manageable number prior to 
using them in the measurement and structural analyses (See the next chapter).