removed), it is necessary to re-estimate the measurement model and check the
construct validity.
The re-estimated measurement model (CFA 5b:9) does demonstrate both content
and discriminant validity. All AVE values (average variance extracted) are greater
than 0.5 and all CR (construct reliability) values are greater than 0.7.
Discriminant validity is achieved with the lowest AVE value (0.600 for EM)
exceeding the highest squared multiple correlation value (0.560 for EM EPA).
Figure 7-22 Estimates and model fit for SEM 5b:2
In the finalised structural model (SEM 5b:2, Figure 7 - 22 ) all parameters are
statistically significant and the model demonstrates good fit to the data. The
critical ratio (CR) for multivariate kurtosis was 17.7 and a bootstrapped model
estimation was run to check on the robustness of the parameter estimates in the
light of this. The bootstrapped estimates (for both the structural and
measurement models) indicated that all parameters are robust under the
conditions of multivariate non-normality (see Appendix IV for details).
The substantive implications of the item removals in the EM construct are dealt
with in greater depth in Section 7.13 below. In brief, the substantive content of
the EM factor that is of central interest in this research - Hancer and George’s