[R-sig-eco] nmds -- "Error in class(S) ...", package conflict between vegan and ??

[Jari Oksanen]

On Wed, 2010-05-26 at 10:44 +0200, Kim MILFERSTEDT wrote:
> Hello,
> 
> I am attempting to use nonmetric dimensional scaling (metaMDS) in vegan 
> (1.17.2).
> 
> My data are 105 samples in rows and 1271 variables (peak positions for 
> community fingerprinting) in columns.
> 
> When starting metaMDS in the default settings, the analysis goes past 
> Run 0 but after Run 1, I get the error message:
> 
> Error in class(S) <- "scores" : attempt to set an attribute on NULL
> 
> I tried several other dissimilarity indices (e.g. euclidean) and the 
> analysis goes through various runs but eventually always crashes with 
> the same error message.
> 
> I have also tried the example from ?metaMDS which also failed. This made 
> me think that there may be some kind of conflict between packages. And 
> voilà, after closing and restarting R, just using vegan and MASS, the 
> example worked fine.
> 
> For aligning the community fingerprints, I use StatFingerprints which 
> depends on a ton of other packages.
> 
> Does anybody know how to read the error message? What is it pointing at 
> and how could I find out which package is causing me trouble?
> 
Kim,

After the crash, say 

traceback()

and send the output to me.

I have never used StatFingerprints, nor have I installed the package in
my system. This means that probably I cannot reproduce the problem.
However, the traceback can help me to focus on the problem. 

My guess is that that StatFingerprints (or some other package) has
function called scores() that is used instead of the one in vegan. Which
one of the competing versions of scores() is used depends on the order
they are loaded and attached, and how the different scores() are located
in the search path. -- If scores() is the problem.

Cheers, Jari Oksanen