The process of measuring the latent variables is known as the measurement
model. Having satisfactorily established the measurement model (according to a
range of statistical criteria) , the SEM analyst then goes on to evaluate the inter-
relationships between the latent (unobservable) variables – this process is known
as the structural model. Structural models typically test and measure causal
relationships between latent variables.
Figure 6 - 1 illustrates this process with an example SEM model consisting of three
latent variables, each of which is measured with three indicator variables. Boxes
A, B and C highlight the confirmatory factor analysis (CFA) measurement models
for the Job Satisfaction, Job Performance and Intention to Quit factors. Box D
highlights the structural part of the model, that is, where the relationships
between the factors are measured. The nature of the relationships between
variables is indicated by the direction of the linking arrows.
Figure 6-1 Example structural equation model
In this example, Job Satisfaction is hypothesised to have a causal effect on both
Job Performance (H 1 ) and Intention to Quit (H 2 ). The nature of the hypotheses
are usually informed by theory or previous empirical research and are articulated
in such a way that the specific effect is made clear. In this example, H 1 may
suggest that Job Satisfaction exerts a positive effect on Job Performance and H 2
states that Job Satisfaction has a negative effect on Intention to Quit – that is,
when Job Satisfaction is higher, Intention to Quit is lower.
In SEM, oval or round boxes are used to represent unobserved (latent) variables
and rectangular (or square) boxes represent observed (indicator) variables. The