(2003) Influence dimension – remains adequately measured by items EM1, EM3
and EM4.
The correlation between EM and ML is moderately high (r = 0.578, p < 0.001)
and the ML→EPA path has dropped in magnitude from = 0.446 in Model 4 to =
0.213. The moderately high correlation and the change in the ML→EPA structural
coefficient suggests that collinearity between the EM and ML constructs may be
continuing to influence the estimates of EM and ML on EPA. The implications of
this will be discussed further in the final chapter (Section 8.2).
Mediator and total effects for Model 5b
Testing the mediator effect of EPA was carried out in the same way as with
previous models and confirmed that EPA acts as a partial mediator between ML
and JP.
A constrained model with no direct effect (ML→JP) was estimated (SEM 5:2 1 ).
The chi square difference (Δχ^2 ) between models SEM 5:2 (χ^2 = 125.182) and SEM
5:2 1 (χ^2 = 116.829) was calculated at 8.353. With 1 degree of freedom this
yields a p value of 0.0039. Accordingly, (because p <0.01) full mediation of the
ML→JP path by EPA is not supported.
To determine whether or not Employee Positive Attitudes (EPA) has a partial
mediation effect on the ML→JP relationship it is necessary to estimate a third
model (SEM 5:2 2 ) in which there is no indirect effect (EPA→JP) on Job
Performance (JP).
The structural coefficient values for the ML→JP path are now compared between
model SEM 5:2 2 and the hypothesised model SEM 5:2. In the constrained model
(SEM 5:2 2 ) the value (effect size) for the ML→JP path is 0.424 while in the
hypothesised model (SEM 5:2) this value is 0.256. When comparing the ML→JP
path between the constrained model (SEM 5:2 2 ) with the hypothesised model,
because ML→JP is statistically significant in both but reduced in size in the
hypothesised model (0.424 in SEM 5:2 2 and 0.256 in SEM 5:2) we can conclude
that partial mediation is supported (see Hair et al. 2006 : 867).