Some level of multicollinearity characterizes virtually
every multiple regression model. The
problem, of course, is that while the regression parameter estimates remain
unbiased with minimum variance properties, the variances are often
substantially larger than those obtained when multicollinearity is absent.
A variance inflation factor
(VIF) is the coefficient of multiple determination that is
obtained by regressing an independent variable on all of the other explanatory
(independent) variables in the analysis.
When no inter-corrleations exist, this coefficient of multiple
determination is zero (0) and the VIF is at its lower bound of one (1.0). As the variables become more correlated, the
VIF increases at an increasing rate towards infinity. Variance inflation factors that exceed ten (10.0) are considered
problematic. That is, this is a
sign that the variable has high and significant multicollinearity with one or
more other variables in the analysis.
Be sure to check the signs of the variables as well.