Skip to content

I cannot get species scores to plot with site scores in MDS when I use a distance matrix as input. Problems with NA's?

5 messages · Edwin Lebrija Trejos, Brad Patrick Schneid, Gavin Simpson

Original
Edwin Lebrija Trejos
# 
Hi, First I should note I am relatively new to R so I would appreciate answers that take this into account.
 
I am trying to perform an MDS ordination using the function ?metaMDS? of the ?vegan? package. I want to ordinate species according to a set of functional traits. ?Species? here refers to ?sites? in traditional vegetation analyses while ?traits? here correspond to ?species? in such analyses.  
 
My data looks like this:
 
         Trait1   Trait2 Trait3  Trait4  Trait5  Trait?  
Species1 228.44   16.56   1.66   13.22     1     short 
Species2 150.55   28.07   0.41   0.60      1     mid
Species3     NA   25.89     NA   0.55      0     large
Species4 147.70   17.65   0.42   1.12     NA     large
Species? 132.68      NA   1.28   2.75      0     short

 
Because the traits have different variable types, different measurement scales, and also missing values for some species, I have calculated the matrix of species distances using the Gower coefficient of similarity available in Package ?FD? (which allows missing values). 
My problem comes when I create a bi-plot of species and traits. As I have used a distance matrix in function ?metaMDS? there are no species scores available. This is given as a warning in R: 
 
"> NMDSgowdis<- metaMDS(SpeciesGowdis)
Warning message:In ordiplot(x, choices = choices, type = type, display = display, :Species scores not available? 
 
I have read from internet resources that in principle I could obtain the trait ("species") scores to plot them in the ordination but my attempts have been unsuccessful. I have tried using the function ?wascores? in package vegan and ?add.spec.scores? in package BiodiversityR. For this purpuse I have created a new species x traits table where factor traits were coded into dummy variables and all integer variables (including binary) were coerced to numeric variables. Here are the codes used and the error messages I have got: 
 
?> NMDSgowdis<- metaMDS(SpeciesGowdis)
Error in if (any(w < 0) || sum(w) == 0) stop("weights must be non-negative and not all zero") : missing value where TRUE/FALSE needed? 
 
I imagine the problem is with the NA?s in the data. 
Alternatively, I have used the ?add.spec.scores? function, method=?cor.scores?, found in package BiodiversityR. This seemed to work, as I got no error message, but the species scores were not returned. Here the R code and results:
?> A<-add.spec.scores(ordi=NMDSgowdis,comm=TraitsNMDSdummies,method="cor.scores",multi=1, Rscale=F,scaling="1")
Warning message:In ordiplot(x, choices = choices, type = type, display = display, :Species scores not available?
 
Can anyone guide me to get the trait (?species?) scores to plot together with my species (?site?) scores?
Thanks in advance,
Edwin
Brad Patrick Schneid
# 
Try the daisy() function from the package "cluster", it seems to be able to
handle NAs and non-dummy coded character variables

metaMDS(daisy(df, metric="gower"))




Edwin Lebrija Trejos wrote
--
View this message in context: http://r.789695.n4.nabble.com/I-cannot-get-species-scores-to-plot-with-site-scores-in-MDS-when-I-use-a-distance-matrix-as-input-Pr-tp4103699p4104406.html
Sent from the R help mailing list archive at Nabble.com.
5 days later
Gavin Simpson
#
On Thu, 2011-11-24 at 07:57 -0800, B77S wrote:
That won't help the OP as the species scores (the species data, i.e. the
traits in this case) can not be computed from a site x site
dissimilarity matrix. This is has the same problem as the OPs existing
problem.

G

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Gavin Simpson
    

    

      
      #
    

  


  

      
On Thu, 2011-11-24 at 12:16 +0000, Edwin Lebrija Trejos wrote:

            

      
      
      
    

        
<snip />

            

      
      
      
    

        
I think you should pass metaMDS the species trait matrix and then get it
to use FD to compute the dissimilarities. Note from ?metaMDS there is a
distfun argument for metaMDSdit. metaMDS passes the community data to
metaMDSdist to compute the dissimilarities.

The only snag is that gowdis doesn't accept a `method` argument so we
need a wrapper:

wrapper <- function(x, method, ...) {
    gowdis(x, ...)
}

then do

metaMDS(SpeciesGowdis, distfun = wrapper, XXXXX)

where XXXX represents any other arguments you want to pass to gowdis via
our wrapper. Essentially the wrapper ignores the method argument, we
just need it as metaMDSdist wants to call the dissimilarity function
with a method argument.

This is not tested as there wasn't reproducible code, but hopefully
you'll be able to work it out from the above.

HTH

G