Multicollinearity
characterizes the model when two or more independent variables are linearly
related. The presence of
multicollinearity can have serious effects on the overall quality of the
regression model. See the discussion on
VIFs for measurement of multicollinearity.
The negative side of multicollinearity may be summarized as follows:
1.
The
independent, or direct, impact of the explanatory factors may not be estimated
correctly. Stated differently, the
precise meaning of each parameter estimated by be obscured.
2.
Parameter
estimates may not appear significantly different than zero. This condition may lead to the dropping of
an otherwise significant variable.
3.
Estimated
parameter estimates may be sensitive to the addition or the deletion of a few
observations.
4.
Estimated
parameter estimates may be sensitive to the addition or the deletion of an
apparently insignificant variable.