[adegenet-forum] SPCA with cost distances as matrix of spatial weights

[Jombart, Thibaut]

Hello, 

sPCA is meant to study spatial structures given a measure of spatial proximity, not the other way around. Here, you make the assumption that spatial structures (as measured by Moran's I and the variance) should be maximum, and somehow derive a measure of 'space' from this point. In most cases, it is probably very difficult to find support for such assumption. 

Even if this assumption was true, this approach cannot be used to compare different spatial models - the 'statistics' (lambda) from one model to another are not, as a rule, comparable (their range of variation differs). 

Cheers

Thibaut

--
######################################
Dr Thibaut JOMBART
MRC Centre for Outbreak Analysis and Modelling
Department of Infectious Disease Epidemiology
Imperial College - School of Public Health
St Mary’s Campus
Norfolk Place
London W2 1PG
United Kingdom
Tel. : 0044 (0)20 7594 3658
t.jombart at imperial.ac.uk
http://sites.google.com/site/thibautjombart/
http://adegenet.r-forge.r-project.org/
________________________________________
From: Nevil Amos [nevil.amos at gmail.com]
Sent: 22 June 2012 16:23
To: adegenet-forum at lists.r-forge.r-project.org
Cc: Jombart, Thibaut
Subject: Re: SPCA with cost distances as matrix of spatial      weights

I have run a series of spca with  square matrix of 1/(resistance+1)  as the spatial weight for a range of alternative landscape resistance matrices.  (pdf output attached, model names are in the second line of the plot of Global.rtest).   GEOG represents 1/(geographic distance +1).  Am I correct in interpreting those showing the greatest positive spatial autocorrelation in lambda 1 as  the "best landscape model" .  Thus  if any of the models with greater s.a. for lambda 1 than the GEOG model are explaining more than  geographic distance alone?( in the attached case none have  a greater s.a than GEOG.

cheers

Nevil Amos

MERG
Monash University