[adegenet-forum] Testing how many vectors are significant using R test

[Jombart, Thibaut]

Dear Nevil, 

here you're quite lucky, there are clearly two global structures in your data - no test needed.

Cheers

Thibaut

________________________________________
From: Nevil Amos [nevil.amos at gmail.com]
Sent: 20 August 2010 15:40
To: Jombart, Thibaut
Cc: adegenet-forum at lists.r-forge.r-project.org
Subject: Testing how many vectors are significant using R test

Thibaut,
Thanks for your response to my query about the missing values and Global R test.

I have a number of datasets where more than one lambda stands out on the screeplot, thus there may be more than one global structure present.  The Global R test allows me to test whether there is significant global structure in the data, however it is not clear to me how to determine the number of axes that should be interpreted, where more than on stands out from the rest.  A screeplot from the dataset appended to my last query is attached below, here it seems that both lambda 1 and 2 may stand out? In other cases the values are more marginal and a formal test of each would be useful.



Thanks

Nevil Amos

[cid:part1.00000102.00060307 at sci.monash.edu.au]




 > myspca<-spca(EYRM.genind)

Choose a connection network:
         Delaunay triangulation (type 1)
         Gabriel graph (type 2)
         Relative neighbours (type 3)
         Minimum spanning tree (type 4)
         Neighbourhood by distance (type 5)
         K nearest neighbours (type 6)
         Inverse distances (type 7)
Answer: 1

Keep this graph (y/n)? y
Select the first number of axes (>=1): 1
Select the second number of axes (>=0): 0
Warning message:
In spca(EYRM.genind) :
  NAs in data are automatically replaced (to mean allele frequency)
 > plot(myspca)
 > myGtest <- global.rtest(EYRM.genind$tab, myspca$lw, nperm = 999)
Error in global.rtest(EYRM.genind$tab, myspca$lw, nperm = 999) :
  NA entries in X
 > save.image("d:\\nevs_docs\\Rwork\\spca\\NA_data.RData")