arrows pointing from the latent variables (Job Satisfaction, Job Performance and
Intention to Quit) to the observed variables indicate the ‘effect’ that the latent
variable is having on the observed variables: thus, the observed variables
‘indicate’ the existence of the latent variable by representing the effects of the
unobservable variable. Employees who have high levels of Job Satisfaction, for
example, are expected to exhibit relatively high scores for JS1, JS2 and J3
compared with employee who have lower levels of Job Satisfaction. In
organisational psychology, observable variables, such as JS1, JS2 and JS3 in this
example, are often measured by asking respondents to indicate their thoughts,
feelings, perceptions etc. using a scalar measure.
In organisational psychology research, five point response scales are often used
(as are four point scales, seven point scales and sometimes even 10 point scales).
This research makes use of five point scales: (i) to satisfy the requirements for
the effective use of maximum likelihood estimation during the structural equation
modelling stage (see Section 6.3); (ii) following the extensive use of five-point
scales in the previous studies that have been drawn upon to generate the data for
the latent factors; and (iii) following the findings of Preston and Coleman (2000:
12) on the psychometric properties and respondent preferences for rating scales.
Job Satisfaction, in this example, may be measured (within a larger
questionnaire) by asking employees to respond to the following statements:
Please indicate your level of satisfaction with each of
the following aspects of your work
Your level of satisfaction
Very
unsatisfiedVery
satisfiedJS1 The actual daily tasks that you do 1 2 3 4 5
JS2 The pay (your wages / salary) 1 2 3 4 5
JS3 The people you work with 1 2 3 4 5
Surveys can be designed to includes responses using a range of scale semantics
including level of satisfaction (as in this example), level of agreement, or level of
frequency (e.g. ranging from ‘not at all’ to ‘very frequently’).
Referring once again to Figure 6 - 1 , the small circles labelled e1 to e9 represent
the proportion of the variance in each indicator variable that is not accounted for
by the latent variable. This ability to include unmeasured (or ‘error’) variance in
the model results in more accurate predictions about the size of effects (the
strength of inter-relationships) and is a key advantage of SEM over the more
traditional linear modelling approaches (e.g. regression analysis, and analysis of