Open wevertonbio opened 2 years ago
Hi @WevertonBio,
Difference between importance and anova:
The overall structure should be similar between the importance and the anova function, however, our current anova is a type I anova where the unique components (A, B, and S) are affected by collinearity (
) i.e. the individual components are not corrected for the shared contributions (the spatial only model will estimate non-zero effects for space because the second spatial predictor is weakly correlated with the third environmental predictor) which is not an issue for the importance(...) where we use the full model (so the confounding effect of the shared contributions is accounted for, i.e. the spatial estimates are correctly estimated to zero). We are currently working on type II anova to improve this behaviour.
About the R Squareds of the return objects, I forgot to normalise the R2s of the plot.anova values. They are actually the same (if you use McFadden instead of Nagelkerke):
Hi Max, looking at this, I think we should have a go at the vignette in a quiet moment and expand the explanation. I can also give this a go if a quiet day comes up in summer! Your explanation here would already be a big improvement over the text that is currently in the vignette.
F
Hi @MaximilianPi,
Thank you very much for your quick and helpful answer!
Another doubt: in issue https://github.com/TheoreticalEcology/s-jSDM/issues/73, you said that "importance(model)
calculates the variation partitioning of the three groups (env, spatial, and biotic) for each site".
Where exactly is this stored? I was only able to find the importance of components to individual species (df.imp
in the reproducible example), not to individual sites.
Hi @WevertonBio,
the importance(....) can calculate the VP only for the species, not the sites. I made a mistake in #73.
Hi,
I'm trying to extract the three components of metacommunity internal structure (Environment, spatial and biotic), the same generated by
plot(anova(model), internal= TRUE)
.I know I can get it using:
I also tried it using the function
importance(model)
, however it gives a different result. What is the difference between them?I also noticed that the R2 from
plot.anova$data$Species$r2
are different from R2 froman$species$R2_Nagelkerke$Full
(wherebyan = anova(model)
). Where do the R2 inplot(anova (model), internal = TRUE)
come from?To illustrate my question, here's a reproducible example: