steve.kembel at gmail.com
Wed May 12 08:56:27 PDT 2010
> Does anyone know if/when Phylocom comtrait capabilities will be implemented in Picante? I think one can use ses.mntd/mpd with a trait distance matrix instead of a phylo distance matrix within Picante, but I wonder if the full capabilities of comtrait (trait range, variance, mntd, mpd) will be implemented in Picante soon?
Those other metrics are not currently implemented in picante. I will add them to my list of requested features but I'm afraid I can't make any promises as to when they'll be implemented. For future reference, if you have a feature request for picante the best thing to do is to add it to the feature request tracker at the picante website:
> Any solutions to comtrait as of yet? I am still having problems in that the number of iterations for "rankLow" and "rankHigh"in comtrait do not match the total iterations. Is it possible that the metrics are still fine, but for some reason the number of iterations is being counted wrong (i.e., there were actually 999 iterations if 999 were asked for)?
If I understand your question, rankLow and rankHi do not always add up to the total number of iterations. This is not a bug, it's a 'feature'. As stated in the Phylocom manual:
"Note that if the sum of rankLow and rankHi for MPD or MNTD is not close to the number of runs, there must be a large number of ties between observed and null community values and results should be interpreted with caution. This situation may arise when using very small phylogenies or numbers of samples."
Basically when rankLow and rankHigh do not add up to the number of runs it's a red flag that either your sample has too few species, or your species pool has too few taxa, and you shouldn't be interpreting P-values as meaningful, because you don't really have the number of independent randomizations you expected. For example, if you have two species in a community, and four species in your pool, there are only six possible combinations of those species in the null communities for that sample, meaning that even if you use 999 randomizations you will have a lot of ties in metrics measured in those null communities, because the metric can only take on six different values across all 999 runs. In this case you would find that the sum of rankLow and rankHi is not anywhere close to the total number of runs, and you couldn't possibly find a P-value less than 0.15 or so. This holds true both for the comstruct and comtrait functions.
Have a look at the web tutorial below for a specific example of what these numbers mean and how to interpret them:
More information about the phylocom-user