reported in Section 7.8, this removal is also supported by guidelines in the SEM
literature regarding parsimony in model building (Kline 2005: 145- 147 ; Raykov
and Marcolides 2006: 41-43; Byrne 2010: 185). Model 5b - containing the EM
construct but not WV - was specified and estimated as illustrated in Figure 8 - 4.
Figure 8-4 Model 5b
The removal of the Work Values construct allowed for the development of Model
5b, which is preferable to Model 5 insofar as it satisfies the principle of parsimony
(containing no non-significant parameters) and has a lower ECVI value indicating
that it is slightly more likely to be replicated in an independent sample than Model
5 (see Table 7 - 31 ).
The identification of collinearity between Employee Empowerment and Work
Values constructs and the subsequent removal of WV from the model does not,
however, mean that WV has no effect on EPA. It simply means that its effect is
not clear when set alongside the EM construct (in Model 5) because the conditions
of collinearity make the WV→EPA structural coefficient estimate unreliable.
As described in Section 7.8, when collinearity between independent variables is
present, large standard errors can lead to paths being measured as non-
significant and individual regression coefficients can change in magnitude, making
them difficult to interpret. How, then, should the WV→EPA estimate be
interpreted? The most straightforward interpretation is to accept the finding from
Model 4 where the WV→EPA relationship was estimated in the absence of any