Re: does very high or perfect correlation among variables cause problems for principal component analysis?
- From: "David Jones" <dajxxx@xxxxxxxxx>
- Date: Thu, 11 Jan 2007 10:35:55 -0000
pbrewster@xxxxxxxxxxx wrote:
Does very high or perfect correlation among variables cause problems
for principal component analysis?
No, given that PCA can be used to identify such things. It is up to
the user to decide which of the principal components are the
"interesting" ones. However, if there are some known linear
relationships that don't need to be estimated from the data it might
be better to redefine the set of variables to be analysed to take this
into account. This might be done using a residual from the known
relationship but should aim to define some externally meaningful
quantities. A raw PCA would itself attempt to estimate the known
relationship ... using the known relationship should remove some
"noise" from the data-set and so give a clearer analysis (but the
effect is probably small).
But perhaps you are thinking of the case where there is a near-perfect
non-linear relation among variables. PCA is not particuarly attuned to
this problem.
David Jones
.
- References:
- Prev by Date: Re: S-Plus multicomp function
- Next by Date: Re: interaction terms in regression model
- Previous by thread: does very high or perfect correlation among variables cause problems for principal component analysis?
- Index(es):
Relevant Pages
|
|
