I am analyzing the relationships between bacterial community structure (response variable) and a slew of soil chemical data (explanatory variables). I've been running redundancy analyses (RDA's), and playing with which explanatory variables to include in the model. I've removed quite a bit based on VIF, and also performed ordistep to find a more parsimonious model (which didn't really remove anything).
My main question is, the RDAs give me an overall proportion explained by the constrained variables I put in each model, but is there a way to find how much EACH variable is contributing to the total proportion explained in each model?
Thus far, I've been running individual RDA models with each variable separately (e.g. otu_mat ~ pH, otu_mat ~ C, etc...) in order to see the proportion they explain alone. Is this appropriate? Or is there a better way??
Thanks for any advice!!
Mike
Michael, you should use a forward selection approach in redundancy analysis (RDA). Values of Marginal (individual effect of each environmental variable) and Conditional effects (effect of each environmental variable in addition to the presence of other variables already selected) will be automatically displayed if you are using CANOCO for RDA.
https://www.nature.com/articles/srep18496#t3
https://www.nature.com/articles/srep12130#t4
Thanks Arideep! But, I am not using CANACO. I am currently using the ordistep function in the vegan package for R, which returns the following:
Inertia Proportion Rank
Total 74.56722 1.00000
Constrained 43.51069 0.58351 18
Unconstrained 32.00939 0.42927 242
Imaginary -0.95287 -0.01278 93
Is there another function in R that may give me all of the conditional effects without having to go through and supply each individual condition for each soil factor?
Hi Michael. How many factors do you have in total? I think partial RDA might solve your problem. Using pRDA (VPA), you can get the explanation rate of each category of your factors. See here:
https://mb3is.megx.net/gustame/constrained-analyses/variation-partitioning
Otherwise, another way to get the explanation rate of each factor is PERMANOVA or MRM (multiple regressions on matrices).
- Badges
- Science topic
- Similar topics
- Biological Science
- Bioecology
- Systems Ecology
- Landscape Ecology
- Connectivity