standardised beta weight of 0.449 (p <0.001), this was followed in magnitude by
Motivational Leadership ( = 0.261, p <0.001) and Work Values ( = 0.139, p
<0.05).
The multicollinearity diagnostics produced by SPSS are the tolerance, the variance
inflation factor (VIF, which is the inverse of the tolerance value) and the condition
index. There are no specific statistical criteria for thresholds for any of these
multicollinearity diagnostics (Cohen et al. 2003: 424-425) and opinions differ
regarding rules of thumb for what is acceptable for each method of ascertaining
the likelihood of multicollinearity.
For VIF, Hair et al. (2006: 230) suggest that values >10 are problematic although
this threshold should be lowered when sample sizes are smaller; Garson (2012)
writes that VIF values >4.0 indicate problems with multicollinearity; and Cohen et
al. (2003: 423), with reference to the >10 rule of thumb, write that “We believe
that this common rule of thumb guideline is too high (lenient) for most behavioral
science applications”.
Regarding the condition index approach to assessing multicollinearity, Cohen et
al. (2003: 424) reiterate the “traditional rule of thumb” that condition index
values >30 “indicate highly severe problems of multicollinearity” - while also
noting that “no strong statistical rationale exists for this choice”. Garson (2012)
writes that condition indices >30 suggest serious collinearity problems and indices
15 are indicative of possible collinearity problems. The guideline that condition
indices >15 indicate multicollinearity problems is also noted by Belsley (1991,
cited in van Vuuren et al. 2007: 121).
For the regression model estimated with the summated scales, none of the VIF
values exceeds 4 (EM = 1.569, WV = 1.325 and ML = 1.385). However, two
condition indices (one condition index, or dimension, is generated for each
independent and the constant – therefore four in total for this model) exceed 15
(at 15.998 and 17.063 respectively). An examination of the related variance
proportions, following Garson’s (2012) guidelines, reveals that because two
variables (EM and WV) have variance proportions >0.5 (EM = 0.98; WV = 0.72)
on the two dimensions with condition indices >15, then these variables (EM and