Re: multicollinearity in regression

Bob,

Thanks for the reply.
Using SPSS as the package (and SAS also to test for
Heteroscedasticity).

Further to the Condition Index value of 67 and the highest Variance
Proportions are Constant .99, LOGSIZE .99 and GASHARE .28.
With regards VIF, LOGSIZE has VIF of 2.016 while the highest VIF is
2.38 for INDUSTRY.

The significance of GASHARE changes depending on whether LOGSIZE is
included in the model.
In the model with LOGSIZE included, GASHARE has coefficient 13.1 (SE of
7.9) and sig of .102.
When LOGSIZE is excluded, GASHARE has coefficient 18.6, (SE of 6.7) and
sig of .008.

There is one other coefficient FORSHARE whose coefficient sign changes
from -2.8 (St.Err of 15.9) when LOGSIZE is included to 1.54 (St.Err of
15.5) when LOGSIZE is excluded. FORSHARE does not indicate significance
though.

As to the size of the LOGSIZE coefficient you mentioned, this is 1.693
which is small but not the smallest in the model.

To begin with, I will look at the various relationships to see what the
underlying cause of the multicollinearity is and how it can be
explained by the theory of the model.

Thanks again,
Paul

.

Relevant Pages

  • Re: multicollinearity in regression
    ... Further to the Condition Index value of 67 and the highest Variance ... LOGSIZE .99 and GASHARE .28. ... In the model with LOGSIZE included, GASHARE has coefficient 13.1 (SE of ...
    (sci.stat.consult)
  • Re: multicollinearity in regression
    ... I want to make it clear to other readers that the Bob you thanked ... LOGSIZE .99 and GASHARE .28. ... In the model with LOGSIZE included, GASHARE has coefficient 13.1 (SE of ...
    (sci.stat.consult)