Re: Principle Components Analysis
- From: Paige Miller <pmiller5NOSPAM@xxxxxxxxxxxxxxxx>
- Date: Thu, 28 Jul 2005 16:30:48 GMT
On 7/26/2005 11:39 AM, RL wrote:
Hi I am trying to run a model with quantitative and qualitative factors. For avoiding collinearity and making sure the factors are with same variance before the model is run, should i do linear transformation first? Pls advise.
Can i do that by using stat of SAS or is there any software that can help?
SAS PROC PRINCOMP will do this for you. The linear transformation to unit variance I believe is the default. You don't have to specifically tell SAS to do that transformation. (If you don't want it to do that transformation, use the COV option)
Ray Koopman has written an excellent response. I would like to add one thing to what Ray said. He said:
To preprocess an m-category variable, treat it as a set of m dichotomous variables, and do a component analysis on those variables. The will be m-1 components; you want them all, not just the largest few. For future reference you will also need the m x m-1 matrix of component loadings.
I have found (and others have agreed with me) that if you have m dichotomous variables in this situation, you don't want each one of them to have a variance of 1. You want the sum of the variances of the m dichotomous variables to be equal to 1, or equivalently each dichotomous variable to have a variance of 1/m. This requires some SAS programming, but the benefit is that the m-category variable is treated equally to a quantitative variable by Principal Components, sinc each now have a variance of 1.
-- Paige Miller pmiller5@xxxxxxxxxxxxxxxx http://paiges-page.net
It's nothing until I call it -- Bill Klem, NL Umpire If you get the choice to sit it out or dance, I hope you dance -- Lee Ann Womack .
- References:
- Principle Components Analysis
- From: RL
- Principle Components Analysis
- Prev by Date: Confidence intervals to obtaine the number of runs
- Next by Date: Paper recommendation
- Previous by thread: Re: Principle Components Analysis
- Next by thread: NIPS 2005 Workshop and Demo Proposal Deadlines
- Index(es):
Relevant Pages
|
