Given the finding described above - that these paths are statistically equivalent
across groups - the non-statistical significance of these three parameter estimates
is likely to have arisen as a consequence of the smaller sample sizes (Group 1 =
106 and Group 2 = 107 cases) as described by Kline (2005: 41).
In summary, the post-development model validations have shown that:
the model-based imputation method is robust; and
that the final model (Model 5b) is likely to cross-validate in an independent
sample drawn from the same population.
7.13 Post-modification construct identities
The model development process has resulted in the selection of Model 5b (see
Section 7.9 above) as the optimal model that explains the observed data in
accordance with the proposed theoretical linkages between the latent constructs.
For reasons of parsimony, one of the measured constructs (Work Values) was not
retained in the model as its effect on Employee Positive Attitudes was not
statistically significant. A further two of the measured constructs, Social Support
and Service Quality, were not able to be included in the model as they were found
not to covary at a statistically significant level with the other constructs.
Aside from the three constructs that were not included in the final model, other
changes to the theorised model included modifications with minor substantive
implications that were made to the Motivational Leadership, Empowerment, Job
Performance and Discretionary Service Behaviour constructs. Modifications with
major substantive changes were made when the three Employee Attitude
variables (Job Satisfaction, Work Meaning and Affective Organisational
Commitment) were merged to form the new Employee Positive Attitudes
construct. The substantive implications of the modifications to construct
composition are discussed below.
Section 7.2.5 above describes content validity (also known as face validity) as the
correspondence between the observed (indicator) variables and the construct that
is intended to be measured by the latent construct.
Following the model generating approach, as Models 1 to 5b were being
developed, various indicator variables were removed from the analysis in order to