confounding collinearity. Accordingly, the parameter estimate is accepted as =
0.334 (p <0.001).
Regarding Model 5b, it is also necessary to examine what, if any, effects
multicollinearity has on the parameter estimates in this model. Of particular
interest is the ML→EPA estimate that is reduced in Model 5b (ML→EPA = 0.213,
p <0.01) compared with Model 4 (ML→EPA = 0.446, p <0.001). This is a
considerable reduction and it is likely to have been caused by the moderately high
correlation (collinearity) between the independent variables EM and ML (r =
0.575, p = <0.001).
To further investigate any likely effect of multicollinearity in Model 5b, following
the same procedure described in Section 7.8 for Model 5, a multiple regression
model was specified using summated scales created from the observed variables
for the independents EM, ML and the dependent EPA. The regression model was
estimated using SPSS and the overall model fit was satisfactory (F = 68.436, d.f.
= 2, p < 0.001) with the two independent variables (ML and EM) accounting for
almost half of the variance in Employee Positive Attitudes (EPA) (R^2 = 0.476).
The standardised regression coefficient () for Employee Empowerment → EPA
was 0.496 (p <0.001) and the Motivational Leadership → EPA was 0.291 (p
<0.001). The collinearity diagnostics did not indicate any major issues with the
VIF values not exceeding 1.3 and the largest condition index estimated at 12.2.
As noted in Section 7.8, there are no specific statistical criteria for thresholds for
the VIF and condition index multicollinearity diagnostics (Cohen et al. 2003: 424-
425). Hair et al. (2006: 230) advise individual researchers to determine
acceptable degrees of collinearity for their models on the basis that “most defaults
or recommended thresholds still allow for substantial collinearity”. In this case, it
appears that the ML→EPA structural coefficient is being affected (reduced) by the
collinearity between ML and EM, albeit the magnitude of the collinearity (as
evidenced by the summated scale regression model collinearity diagnostics) being
moderate. The assessment here, therefore, is that the likelihood is that a more
accurate estimate for the ML→EPA coefficient lies somewhere between 0.213 (the
Model 5b finding) and 0.446 (the Model 4 finding), i.e. somewhere closer to the
ML→EPA coefficient that was estimated in the absence of collinearity between the
Motivational Leadership and Employee Empowerment constructs.
Relating this matter to the wider theory that is being scrutinised, the ML→EPA
path measures the effect that motivational leadership has on employee attitudes,