As with Model 4, there is a second mediation effect to be tested in Model 5b,
specifically, the hypothesised full mediation of the EM→JP path by EPA. To
confirm the full mediation effect of EPA - following Hair et al.’s guidelines (2006:
top of p. 868) – a direct path between EM and JP is estimated. Because this path
is not significant (p = 0.141) the full mediation of the WV→JP path by EPA is
supported.
Table 7 - 30 describes the total (combined direct and indirect) effects for Model 5b
(SEM 5:2).
Independent
variables
Dependent variablesEPA JP DSB
ML 0.213 0.315 0.403
EPA - 0.279 -
EM 0.626 0.174 -
Table 7-30 Total (standardised) effects for Model 5b
In comparison with previous models, the total effect of ML on JP has dropped:
in the latest model, the total effect, ML→JPTOT = 0.315;
in Model 4 ML→JPTOT = 0.380;
in Model 3b ML→JPTOT = 0.415;
in Model 2 ML→JPTOT = 0.415; and
in Model 1 ML→JP = 0.415 (there only is a direct effect in Model 1).
The reason for this effect size dropping as the models have become more complex
is that, as extra constructs have been added, these extra constructs (ME/EPA, WV
and finally EM) have been able to account for some of the effect on JP.
Comparison of Models 5 and 5b
Model 5b was arrived at following the finding (during the development of the
structural model for Model 5) that the WV→EPA path was not statistically
significant. Removing that non-significant path on the ground of parsimony was
tantamount to removing the WV factor entirely and thus the structure for Model
5b (i.e. Model 5 minus the WV construct) was arrived at.