it can potentially contribute to the prediction of Y [the dependent
variable]
(Cohen et al. 2003: 419)Cohen et al. go on to describe how, under these conditions, individual regression
coefficients can change in magnitude, making them difficult to interpret.
Elsewhere, Hair et al. (2006) describe how the presence of multicollinearity:
...creates “shared” variance between variables, thus decreasing the
ability to predict the dependent measure as well as ascertain the
relative roles of each independent variable
(Hair et al. 2006: 228)Hair et al. (2006:228) describe two implications of multicollinearity that concur
with the current situation:
as multicollinearity increases, the ability to demonstrate that the estimated
regression coefficients are significantly different from zero can become
markedly impacted due to increases in the standard error; and
high degrees of multicollinearity can also result in regression coefficients being
incorrectly estimated.
Examining the correlations between independent variables (as undertaken above)
is a useful initial step in diagnosing multicollinearity, however, an absence of high
correlations does not confer that multicollinearity is not present (Hair et al. 2006:
227). A number of diagnostic procedures are available to identify and assess
multicollinearity, however, these are not available in the AMOS structural equation
modelling software, nor any other SEM software (McIntosh 2009). Accordingly, to
evaluate the degree of multicollinearity present, the observed variables for each
of three independent variables - Motivational Leadership (ML), Work Values (WV)
and Employee Empowerment (EM) and the relevant dependent variable,
Employee Positive Attitudes (EPA) - were summed to create four summated scale
variables suitable for use in a multiple regression model (SPSS’s regression
procedures can generate multicollinearity diagnostics).
The regression model was specified and estimated using SPSS and the overall
model fit was satisfactory (F = 68.436, d.f. = 3, p < 0.001) with the three
independent variables (ML, WV and EM) accounting for approximately half of the
variance in Employee Positive Attitudes (R^2 = 0.496). As with the SE model (SEM
5:2), Employee Empowerment was the most influential predictor with a