reference for importance() / documentation of variation partitioning in general

[florianhartig]

Hi Max,

for another paper, we are just discussing where the idea for the importance() function originated. Is there are reference for this? I think you mentioned to me once that you saw it first in Gleb's HMSC implementation?

Maybe add whatever detail you can provide (also regarding advantages / disadvantages wrt to our ANOVA implementation) in the help? This might also have prevented this recent question with the ANOVA for other variables.

Could possibly also extend the Vignette by adding / extending the discussion on how to compare importance of different variables / compartments.

[kjmtaylor22]

Hi both,

On a related note, I find that I've been hitting my head against a wall trying to make sense of both the Rsquared and anova implementations. I've scoured the code and the Nakagawa (2013) article and I still feel like I'm missing some important assumption. So, my questions persist:

1. How are the individual (species) R2 so unbelievably small? (And, for that matter, in anova, why are the R2 so small?)
2. How do some of the R2 in anova end up negative? I realize that the R2 of the modules has to sum up to whatever the estimate is for the full model, but even so, a negative R2 doesn't make sense to me.

Some extra documentation to guide the user through the interpretation of these outputs would be really helpful.

Thanks, Kara

[mtva0001]

Hi,

Thanks for bringing up this issue, I'm also having some difficulties to understand what the anova results mean. Since my spatial fraction has a number of -2.2, I assume that the interpretation of the Venn diagram is not the same as in case of a traditional variation partitioning where the sum of each fractions should below 1 and the residuals (unexplained) is 1-Rsquare. How should we interpret these results and what is the link between your anova plot and a traditional variation partitioning? Thanks for your help in advance! /Máté