You could frame the question a little differently and use a logistic regression perhaps. Model the incidence of trees with and without epiphytes with tree abundances as a covariate. This essentially models the probability of a tree having epiphytes as a function of abundance. That seems like a clean approach to me.

You could frame the question a little differently and use a logistic regression perhaps. Model the incidence of trees with and without epiphytes with tree abundances as a covariate. This essentially models the probability of a tree having epiphytes as a function of abundance. That seems like a clean approach to me.

The best method, IMO, is use mutual information (MI) metric. Rank your data on a 2-d grid (5-10 bins on each axis would be convenient, if number of data points > ~500) and find the MI for each grid. Then you may use goodness of fit (Pearson's chi-squared test) or shuffle your data points-produce surrogate data to test the significance of the resulting MI..

if you deal with proportions of a whole, log-ratio transformation of your data is one possiblity. For example, for each observations/row and each value take the

log (value / geometricmean(row)). Afterwards you can build the tree and measure the correlations. The general approach (not the tree fitting) is content of Aitchison 1986 "the Analysis of compositional data" and the robCompositions R-package might be your friend (cenLR()).

Thank you all for your input; I really appreciate it.

Dear Corey,

I agree with the previous comments by Matthias Templ and Dan Gwin. But I have a doubt on your problem: is epiphyte given as proportions or as a classification? In the second case logistic regression seems to be an adequate way of dealing with the problem once you have tranformed proportions of trees into coordinates of the simplex. In the first case you are confronted with a regression from a simplex (epiphyte) on another simplex (tree species composition).

I think Matthias Templ has publish something in this direction. Here a reference on regression in the simplex. And, as Matthias told us: use R packages.

Egozcue, J. J., J. Daunis-i-Estadella, V. Pawlowsky-Glahn, K. Hron, and

P. Filzmoser, 2012: Simplicial regression. the normal model. Journal of

Applied Probability and Statistics (JAPS), 6, 87-108.

Spanish moss was measured in terms of rank abundance on a tree (zero to five)....so the data variable can be either rank or binary. I identified the species of tree on which Spanish moss was found and I also have a GPS location for each tree. I have been focusing on two types of forest communities: burn managed pine forest and mixed pine-hardwood secondary forest. I started with five 50 x 50 m quadrats in each community (sampling every tree in a quadrat)...and I also have some linear transect data.

I should note that the long term goal of this study is to tease apart the factors that influence the local spatial distribution of Spanish moss (including aspects of dispersal and host tree preference); however, the basic spatial distribution of Spanish moss at this scale had never been adequately characterized. I am interested in whether there is a connection between host tree abundance and Spanish moss incidence because it might provide some clues about dispersal. I suspect that seed dispersal might be somewhat random and that asymmetry in Spanish moss abundance might develop subsequently. There seems to be several species that have a much higher incidence of Spanish moss, but, otherwise, incidence looks to be somewhat proportional to relative abundance. I want to test this somehow.