I'm trying to test how the amount of native forest in a landscape affects the number of bird species, and how this relationship varies with the amount of teatree (another forest type) in the landsacpe.
The relationship between native forest and birds is much stronger when there is little teatree (see coplot, bottom left panels) than when there's lots of teatree (upper right panels.
My concern is that native forest and teatree are correlated; landscapes with little native forest also have little teatree. This has caused datapoints to cluster on the left in the bottom panels of the coplot.
Can someone please tell me if this is an issue for testing this interaction, and if so, how I might address it?
Re: Interaction terms between correlated predictors
I would test for the correlation and see what your variance inflation factor is. If you look at Zuur et al. 2007 Mixed effects models and extensions in R, it has a good description on how to calculate variance inflation factors in R. People generally use cut-offs for VIFs to decide whether the terms are highly correlated and can be put in a model together. This is somewhat controversial. My general philosophy is if two values are strongly positively correlated (VIF>3), you might not want to keep both of them in. So I would check the level of the correlation first and then decide whether examining the interaction may be an issue.
Reply With Quote