Testing for multicollinearity in EViews 7

[ralf]

Hi,

I'm doing LS regressions with fixed and random effects in EViews 7 and I would like to test my models for multicollinearity. I calculated variance inflation factors, but would also like to calculate the condition index/ condition number for my variables. Is there a way to 'automatically' calculate the condition number for all variables used in an equation? Can anybody help?

Thanks!

[EViews Gareth]

You can use the Coefficient Variance Decomposition diagnostic (which is beneath the VIFs).

[ralf]

Thanks for your quick reply Gareth! I have calculated the coefficient variance decomposition table as suggested. As I use much more independent variables in my regressions than in the example I found in the EViews User Guide II I am now not really sure how to interprete my results though.

Can anybody help me to interprete the results of the coefficient variance decomposition table? Maybe somebody can help with a detailed interpretation of the results in a published study (using eviews)?

The coefficient variance decomposition table looks like this:
* 16 out of 17 eigenvalues have condition numbers smaller than 0.001, the smallest condition number is 8.15E-7
* Decomposition proportions: The largest values in column one are 0.73, 0.16 and 0.10 (intercept c has a value of ~1.000- do I have to remove the intercept?)
* The largest values in the other columns reach ~0.9, but I do not observe any 'clusters' of large values (i.e., I do not find a column with more than one value >0.5)
* My panel has 1580 observations (~180 cross sections)
* I use an intercept (c) and a few dummy variables in my regression models

Thanks!

Ralf