R help - Apr 2008 - Correspondence and detrended correspondence analysis

Hi,

I hope someone knows the answer to this or has a real good reference about it (I
am using Legendre & Legendre, Numerical Ecology, 1998).... My data is a
data.frame with locations as rows and vegetation assemblages / species as
columns. I've done a PCA, a correspondance analysis (CA) using ca in ca
package and a detrended correspondance analysis (DCA) using decorana from vegan
package. As far as i understand what i've done .... the relationships
between vegetation categories and locations should be similar in all 3 analysis,
although the 'arch' in locations / sites seen in the CA plot should be
flatten out by the DCA - which it is!.

My problem is that the locations / sites are flipped in the DCA plot versus PCA
or CA plots (locations with positive coordinates now have negative coordinates
in DCA versus CA or PCA), although the vegetation categories keep same
relationships between them in all three plots - which is expected. If some code
is required i can provide it but i didn't want to clutter the message too
much. But i am providing the resulting plots in case they will make my problem
clearer.

PCA plot: ftp://stpfiles.er.usgs.gov/Monica/R_Data/pca_veg_lidarclass_phase2.pdf
CA plot: ftp://stpfiles.er.usgs.gov/Monica/R_Data/corresp_anal_phase2.pdf
DCA plot:
ftp://stpfiles.er.usgs.gov/Monica/R_Data/detrend_corresp_anal_phase2.pdf
DCA how i would expect it:
ftp://stpfiles.er.usgs.gov/Monica/R_Data/detrend_corresp_anal_phase2_a.pdf

Any thoughts on this will be very much appreciated,

Monica


_________________________________________________________________


esh_getintouch_042008

On Fri, 2008-04-18 at 18:45 +0000, Monica Pisica wrote:
> Hi,
> 
> I hope someone knows the answer to this or has a real good reference
> about it (I am using Legendre & Legendre, Numerical Ecology, 1998)....
> My data is a data.frame with locations as rows and vegetation
> assemblages / species as columns. I've done a PCA, a correspondance
> analysis (CA) using ca in ca package and a detrended correspondance
> analysis (DCA) using decorana from vegan package. As far as i
> understand what i've done .... the relationships between vegetation
> categories and locations should be similar in all 3 analysis, although
> the 'arch' in locations / sites seen in the CA plot should be
flatten
> out by the DCA - which it is!.
> 
> My problem is that the locations / sites are flipped in the DCA plot
> versus PCA or CA plots (locations with positive coordinates now have
> negative coordinates in DCA versus CA or PCA), although the vegetation
> categories keep same relationships between them in all three plots -
> which is expected. If some code is required i can provide it but i
> didn't want to clutter the message too much. But i am providing the
> resulting plots in case they will make my problem clearer.
The sign of the eigenvectors (axis scores) is not defined, just the
magnitude; using the same software on different machines may produce
"flipped". Furthermore, the same analysis, PCA say, done using
different
software but on the same machine may not have the same sign, i.e. be
flipped, depending on algorithm used etc.

There is nothing wrong here. You'd be better off asking yourself which
is the appropriate technique to fit for your data rather than fitting
them all...

HTH

G
> 
> PCA plot:
> ftp://stpfiles.er.usgs.gov/Monica/R_Data/pca_veg_lidarclass_phase2.pdf
> CA plot:
> ftp://stpfiles.er.usgs.gov/Monica/R_Data/corresp_anal_phase2.pdf
> DCA plot:
> ftp://stpfiles.er.usgs.gov/Monica/R_Data/detrend_corresp_anal_phase2.pdf
> DCA how i would expect it:
> ftp://stpfiles.er.usgs.gov/Monica/R_Data/detrend_corresp_anal_phase2_a.pdf
> 
> Any thoughts on this will be very much appreciated,
> 
> Monica
> 
> 
> _________________________________________________________________
> 
> 
> esh_getintouch_042008
> ______________________________________________
> 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.
-- 
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
 Dr. Gavin Simpson             [t] +44 (0)20 7679 0522
 ECRC, UCL Geography,          [f] +44 (0)20 7679 0565
 Pearson Building,             [e] gavin.simpsonATNOSPAMucl.ac.uk
 Gower Street, London          [w] http://www.ucl.ac.uk/~ucfagls/
 UK. WC1E 6BT.                 [w] http://www.freshwaters.org.uk
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%

Gavin,
 
Thank you very much for the explanation .... it is very assuring ..... I knew
that the sign of the eigenvectors can flip but i was puzzled that always PCA and
CA came one way and DCA the other way - i mean flipped every time i've run 
the analysis .... the idea was to use different ordinations to see which might
explain the best the vegetation assemblages and their correlation with the
unsupervised classification i've done for the locations. For now DCA is very
promising in spite of the criticism some ecologists have against it. So you are
right - i don't need all of them - but i wanted to do all so if i choose one
and i am asked why i didn't do some other one - i can have an answer ready
;-)
 
Again, thanks,
 
Monica> Subject: Re: [R] Correspondence and detrended correspondence
analysis> From: gavin.simpson@ucl.ac.uk> To: pisicandru@hotmail.com>
CC: r-help@r-project.org> Date: Sat, 19 Apr 2008 11:18:25 +0100> > On
Fri, 2008-04-18 at 18:45 +0000, Monica Pisica wrote:> > Hi,> > >
> I hope someone knows the answer to this or has a real good reference>
> about it (I am using Legendre & Legendre, Numerical Ecology,
1998)....> > My data is a data.frame with locations as rows and
vegetation> > assemblages / species as columns. I've done a PCA, a
correspondance> > analysis (CA) using ca in ca package and a detrended
correspondance> > analysis (DCA) using decorana from vegan package. As far
as i> > understand what i've done .... the relationships between
vegetation> > categories and locations should be similar in all 3
analysis, although> > the 'arch' in locations / sites seen in the
CA plot should be flatten> > out by the DCA - which it is!.> > >
> My problem is that the locations / sites are flipped in the DCA plot>
> versus PCA or CA plots (locations with positive coordinates now have>
> negative coordinates in DCA versus CA or PCA), although the vegetation>
> categories keep same relationships between them in all three plots ->
> which is expected. If some code is required i can provide it but i> >
didn't want to clutter the message too much. But i am providing the> >
resulting plots in case they will make my problem clearer.> > The sign of
the eigenvectors (axis scores) is not defined, just the> magnitude; using the
same software on different machines may produce> "flipped".
Furthermore, the same analysis, PCA say, done using different> software but
on the same machine may not have the same sign, i.e. be> flipped, depending
on algorithm used etc.> > There is nothing wrong here. You'd be better
off asking yourself which> is the appropriate technique to fit for your data
rather than fitting> them all...> > HTH> > G> > > >
> PCA plot:> >
ftp://stpfiles.er.usgs.gov/Monica/R_Data/pca_veg_lidarclass_phase2.pdf> >
CA plot:> >
ftp://stpfiles.er.usgs.gov/Monica/R_Data/corresp_anal_phase2.pdf> > DCA
plot:> >
ftp://stpfiles.er.usgs.gov/Monica/R_Data/detrend_corresp_anal_phase2.pdf>
> DCA how i would expect it:> >
ftp://stpfiles.er.usgs.gov/Monica/R_Data/detrend_corresp_anal_phase2_a.pdf>
> > > Any thoughts on this will be very much appreciated,> > >
> Monica> > > > > >
_________________________________________________________________> > >
> > > esh_getintouch_042008> >
______________________________________________> > R-help@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.> --
> %~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%>
Dr. Gavin Simpson [t] +44 (0)20 7679 0522> ECRC, UCL Geography, [f] +44 (0)20
7679 0565> Pearson Building, [e] gavin.simpsonATNOSPAMucl.ac.uk> Gower
Street, London [w] http://www.ucl.ac.uk/~ucfagls/> UK. WC1E 6BT. [w]
http://www.freshwaters.org.uk>
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%>
_________________________________________________________________


esh_getintouch_042008
	[[alternative HTML version deleted]]