0

In R: I'm doing a partial RDA, using a forward selection procedure to identify the explanatory variables which are most important for a coleoptera community.

But how do I get not only adjusted R2 values, but also p-values for all tested variables when using function ordiR2step?

And why is function ordiR2step so much faster than ordistep in doing so??

Conny
  • 41
  • 6

1 Answers1

0

I first answer to the last question: ordiR2step is so much faster than ordistep because it does not estimate all p-values.

The p-values are found with permutation tests, and running those tests to all variables takes time.

There is no way of getting all so-called p-values in ordiR2step because the method does not use those, as is explained in the source paper cited (Blanchet et al. 2008 in the help page) and the help page. That is the reason why the method is called ordiR2step: it uses primarily R2 instead of p-values. If you want to get all p-values, you must use a method that was written to evaluate those, that is ordistep.

Jari Oksanen
  • 3,287
  • 1
  • 11
  • 15
  • Thank you Jari. One last question: I'm interested in how much of the variance is explained by each variable (not only the final variables of the "best model"). Can I just enter the variables by their order of importance (determined by ordistep) into a new RDA, doing an RDA.model$anova to get these values? I saw these outputs gained by stepwise variable selection done in CANOCO e.g. https://www.nature.com/articles/srep18496/tables/3 – Conny Feb 15 '21 at 11:46