The Δχ^2 p value of 0.259 indicates that the model fit has not statistically
significantly worsened and that we should therefore accept the constrained model
(without the WV→EPA) path as the best alternative.
In removing the WV→EPA path, however, we are effectively removing WV from
the model, since it now has no causal effect on any construct. This leads us back
to the same situation described following the implementation of remedy (iii) from
Cohen et al. (2003: 430).
The removal and the Work Values construct and the likely theoretical implications
of multicollinearity for the research are discussed in the concluding chapter
(Section 8.2).
7.9 Model 5b
Following the removal of the Work Values construct, a new model (Model 5b) is
specified as illustrated in Figure 7 - 21. The measurement model for Model 5b is
initially specified (as per previous models) with all of the measured (indicator)
variables loading on their respective factors.
Figure 7-21 Structural specification for Model 5b
This initial specification is used, firstly for the model-based data imputation
procedure (to replace the missing data) and, secondly, to estimate and test the
measurement model in the usual way. As with Model 5, EM-C (Competencies)
does not covary significantly with the ML (p = 0.151) or DSB (p = 0.202)
constructs; additionally, the covariance with JP is not significant (p = 0.057).