Skip to content

cca standard error species

2 messages · prb, Jari Oksanen

Original
Jari Oksanen
# 
Paloma,

Firstly, I explain what the code snippet is supposed to do, and then I
wonder your specific error message.
On 3/05/10 14:41 PM, "prb" <eco.prb at gmail.com> wrote:

            

      
      
      
    

        
The ordiellipse function adds so-called "species sd" to the graph. If
species have a Gaussian response to the ordination space, this "species sd"
gives an estimate of Gaussian width parameter. In the example above this
"species sd" is drawn for species Cet.niv (a lichen that used to be called
Cetraria nivalis until taxonomists changed all the names). This happens so
that ordiellipse() finds weighted variances and covariances for scores using
species abundances as weights ('w = Cet.niv'). For this we must tell that
all observations belong to one single group ('groups = rep(1, 24)' when the
data have 24 rows, and 'groups' vector will be 24 elements of ones). In
order to have these covariance ellipses to coincide with species scores we
must base the analysis on LC (linear combination) scores ('display = "lc"')
and appropriate scaling where species are weighted averages of sites
('scaling = 2'). This was behind all the fine details of arguments above
('show' is actually not necessary, and 'col' only sets the colour of the
ellipse).

Was this what you meant with "species sd"?

            

      
      
      
    

        
The usual practice is to show species sd, not their "se". The "se" can be
calculated, but it does not have the meaning of standard error of species
scores. Finding this needs much more complicated code and some unpublished
science (but can be approximated). Species sd refers to the width of species
response and not to the reliability or accuracy of species ordination
scores.

            

      
      
      
    

        
One thing that you must adapt is to use the correct length of 'groups'
vector (I should have been more explicit here). So the command is better
written as:

with(varespec, ordiellipse(vare.cca, groups = rep(1, nrow(varespec)), ...)

That is, you cannot use rep(1, 24) for any data, but the latter number (24)
should be equal to the number of rows in your data sets. If you didn't adapt
the command to your data dimensions, you may get the error message you
reported. This happens if the species has only zeros in the first 24 rows.
If you adopted the command to use the correct number of rows in your data
and still got the error message, then believe what R says: you should
non-negative and not all zero weights, where weights are the abundances for
the species you want to plot.

Cheers, Jari Oksanen