Re: Multicollinearity !!!!!

On Wed, 6 Aug 2008 13:10:02 -0700 (PDT), "geetha.shree@xxxxxxxxx"
<geetha.shree@xxxxxxxxx> wrote:

Hi
I am doing a regression analysis. I have one response variable
(Y) and six predictor variables ( X1, X2, X3, X4, X5, X6).

The R-sq value obtained was about 32.1%.

I found out through correlation matrix , the correlation between
X1 and X2 = .96
X4 and X5 = .94

********************************************
Response is Median
The following variables are included in all models: X3, X6

Mallows
Vars R-Sq R-Sq(adj) Cp S x1 x2 x4 x5
1 25.2 2.8 1.7 2.2935 X
1 19.4 0.0 2.3 2.3812 X
2 28.3 0.0 3.4 2.3671 X X
2 28.1 0.0 3.4 2.3701 X X
3 30.6 0.0 5.2 2.4713 X X X
3 28.7 0.0 5.3 2.5049 X X X
4 32.1 0.0 7.0 2.6133 X X X
X
*********************************************************

I would have a the maximum R- Sq value only if I have all the
variables.
But the correlation matrix tells another story !

Could anyone please suggest me how to deal with this problem.

thanks.


What is your problem? Multicollinearity does not bias the coefficients
of a regression. Statistical inference is fine.

It may be of course, that you can drop some variables with little
reduction in R-sq. Whether this is okay depends on what your are
planning on using the results for.

-*** Startz
.

Quantcast