I'm studying differences in tree health among 5 species of trees across 3 different green infrastructure types. Here are the first few lines of data:
Species Type Dieback Discolor FvFm Transparency 1 Quercus palustris STree 0.00 0.05 NA 0.3870864 2 Quercus palustris ROWB 0.20 0.05 0.7146000 0.2957104 3 Ginkgo biloba STree 0.80 0.90 0.6514000 0.6012480 4 Gleditsia triacanthos SGS 0.05 0.35 0.8052500 0.4581673 5 Gleditsia triacanthos STree 0.05 0.50 0.8218000 0.6596218
All of the tree health metrics are proportions: Dieback=percentage of canopy dieback (in 5% increments), Discolor=percentage of canopy discoloration (in 5% increments), FvFm=a measure of photosynthetic efficiency (out of 100%), Transparency=percentage of canopy transparency. Canopy transparency is normally distributed, so I just used
lm in R to analyze it. Dieback and discoloration are really skewed because most trees are pretty healthy, so there are a lot of 0's. I also found a few dead trees, which means they have 100% canopy dieback. FvFm is similar: there were lots of trees with high FvFm's, and the dead trees had 0 FvFm.
I'm interested in how tree species, green infrastructure types, and the interaction of the two influence tree health. Using
lm, this would mean:
Model <- lm(Transparency ~ Species + Type + Species:Type, data=Trees)
I've read that proportion data can be analyzed using logistic regression (using
glm) or beta regression (using
betareg). This might be a silly question, but I thought that logistic regression is more for count data and
betareg only works for proportions between 0 and 1, NOT including 0 and 1. Is this correct? Or are there ways to use logistic regression or beta regression on my data? Are there other tests that would be more appropriate? Any advice would be greatly appreciated.