User Tools

Site Tools


Section: Ordination analysis

Variable selection (constrained ordination)

R functions

  • ordistep (library vegan) - variable selection function, which is applicable with functions rda, cca or cmdscale. Offers forward, backward and combined step-wise selection algorithm (specified by argument direction). Comparison of variables is based on AIC criteria and P-values from Monte Carlo permutation test.
  • ordiR2step (library vegan) - similar to ordistep, but based on R2adj, and implements three stopping stopping rules when selecting the variables: if R2adj of the selected model starts to decrease, or R2adj of global model (defined in scope) is exceeded, or any remaining variable is not significant at predefined P-value. It can be used with rda, cca or capscale. (Before, this function used to be limited to only rda and capscale, but recently vegan implements the permutation method of estimating adjusted R2 for CCA (Peres-Neto et al. 2006) into the function RsquareAdj, and since then the ordir2step function can handle also formula with cca function.)
  • forward.sel (library adespatial) - variable selection method, which is implicitly using RDA ordination, and offers only forward selection (not backward or step-wise). Contains stopping rules from Borcard et al. 2008, the same as ordiR2step, but has different logic of setting the arguments. Eventually, however, it returns the same results as ordiR2step and could be favoured for rather easy use (no need to specify null and full models and the scope).
en/forward_sel_r.txt · Last modified: 2022/03/30 10:34 by David Zelený

Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki