R help - Sep 2009 - PCA or CA

Dear all

I have a data set for which PCA based between group analysis (BGA) gives
significant results but CA-BGA does not.

I am having difficulty finding a reliable method for deciding which ordination
technique is most appropriate.

I have been told to do a 1 table CA and if the 1st axis is>2 units go for CA
if not then PCA.

Another approach is that described in the Canoco manual - perform DCA and then
look at the length of the axes.? I used decorana in vegan and it gives axis
lengths.? I assume that these are measured in SD units. Anyway the manual say if
the axis length is <3 go for PCA,>4 use CA and if intermediate use either.

Are either of these approaches good/valid/recommended or is there a better
method?

Thanks

Paul? 
 		 	   		  
_________________________________________________________________
Get the best of MSN on your mobile
http://clk.atdmt.com/UKM/go/147991039/direct/01/

Hi Paul,
>> I have a data set for which PCA based between group analysis (BGA)
gives
>> significant results but CA-BGA does not. 
>> I am having difficulty finding a reliable method for deciding which
>> ordination 
>> technique is most appropriate.
Reliability really comes down to you thinking about and properly defining
what _information_ you want to extract from your data set, which we know
nothing about. PCA and CA are fundamentally different. The classical use of
CA lies in the analysis of count-data (contingency tables), for which it
remains a brilliant method. It is also widely applied to analyzing normal n
x p matrices of the type usually analyzed by PCA. A doubly-centered PCA
would get you close to a CA of the normal n x p matrix (i.e. of an analysis
of the same matrix).

This is a biggish area, so grab your specs, and perhaps start with Jolliffe
(PCA) and Benzecri/Greenacre (CA).

Regards, Mark.


Paul Dennis-3 wrote:
> 
> 
> Dear all
> 
> I have a data set for which PCA based between group analysis (BGA) gives
> significant results but CA-BGA does not.
> 
> I am having difficulty finding a reliable method for deciding which
> ordination technique is most appropriate. 
> 
> I have been told to do a 1 table CA and if the 1st axis is>2 units go
for
> CA if not then PCA.
> 
> Another approach is that described in the Canoco manual - perform DCA and
> then look at the length of the axes.? I used decorana in vegan and it
> gives axis lengths.? I assume that these are measured in SD units. Anyway
> the manual say if the axis length is <3 go for PCA,>4 use CA and if
> intermediate use either. 
> 
> Are either of these approaches good/valid/recommended or is there a better
> method?
> 
> Thanks
> 
> Paul? 
>  		 	   		  
> _________________________________________________________________
> Get the best of MSN on your mobile
> http://clk.atdmt.com/UKM/go/147991039/direct/01/
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 
> 
-- 
View this message in context:
http://www.nabble.com/PCA-or-CA-tp25668667p25670451.html
Sent from the R help mailing list archive at Nabble.com.

On Tue, 2009-09-29 at 17:02 +0000, Paul Dennis wrote:
> Dear all
> 
> I have a data set for which PCA based between group analysis (BGA) gives
significant results but CA-BGA does not.
> 
> I am having difficulty finding a reliable method for deciding which
ordination technique is most appropriate.
> 
> I have been told to do a 1 table CA and if the 1st axis is>2 units go
for CA if not then PCA.
> 
> Another approach is that described in the Canoco manual - perform DCA and
then look at the length of the axes.  I used decorana in vegan and it gives axis
lengths.  I assume that these are measured in SD units. Anyway the manual say if
the axis length is <3 go for PCA,>4 use CA and if intermediate use either.
> 
> Are either of these approaches good/valid/recommended or is there a better
method?
> 
> Thanks
> 
> Paul  
Hi Paul 

I think that Ca is Correspondence Analysis and PCA is Principal
Component Analysis, right?

In this case, if all variables is numeric do you must use PCA, if all
variables is factor do you must use CA.

If you have a mixed  variables do you have a problem, in most case I
convert numeric variables to factor (with loss of information) and make
CA
-- 
Bernardo Rangel Tura, M.D,MPH,Ph.D
National Institute of Cardiology
Brazil