[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: correlation constant and independent variables

From   "Nick Cox" <>
To   <>
Subject   RE: st: correlation constant and independent variables
Date   Sun, 19 Aug 2007 15:45:17 +0100

Roger's main point seems to me to be on target. 

On a detail: The problem with a constant 
is that it has zero variance, and that 
makes its correlation with anything else 
indeterminate, or undefined, depending on 
how you want to think of it. Zero 
covariance is itself not a problem. In 
other words, the difficulty is in the 
denominator, not the numerator. 


Roger Harbord
> -coldiag2- doesn't check for correlation, it checks for 
> collinearity. A 
> constant doesn't vary so its covariance with any variable is zero and 
> the correlation isn't defined.
> If you found high collinearity between an independent 
> variable and the 
> constant that would imply that the variable's mean is large 
> compared to 
> its variance. I'm not sure that would necessarily be a 
> problem, but you 
> could consider centering it or otherwise changing its origin to be 
> closer to its mean if you want to reduce the collinearity.
> frank palme wrote:
> > hello list,
> >
> > i have a simple question which is bothering me. i could not find any
> > answer to it in the literature.
> >
> > i used the coldiag2 command to check for correlation. i found high
> > correlation between an independent variable and the 
> constant. but the
> > theoretical part of my thesis clearly suggests to include a 
> constant.
> >

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index