I have a rather large dataframe with a lot of environmental variables which I fitted as vectors onto an ordination in R.
I know I can cutoff bad results using e.g. p.max=0.005 for significance. But frankly, this still leaves me with too many vectors and also it is sometimes more desireable to exclude weakly correlated variables from the plot.
I cannot submit my data, but the example dataset from package "vegan" is similar in structure.
P.S.: I've checked all documentation on this to the best of my knowledge. Maybe I'm just not using the right terminology
I've found the solution to that problem. You have to create a new object and apply a formula to select the variables you want to display. Like this:
Given objects:
- df: Your dataframe (rows=samples, columns=species/variables)
- mds: NMDS-results
- Var1; Var2; Var3: the variables to be displayed, they are columns in object df
ef
Niklas, I had the same question as you, I had too many significant response variables to pick from, it would be nice to have a solution that wasn't hard-coded (Var1 + Var2 + var3) as you have done... Any luck with finding a way that just picks out the variables with the best R-squared?
Thanks for the solution anyway, it's the best I could find so far....
To close this -now very old- topic. Jenny and I came to a solution. Find the script below. Of course, having too many significant predictors to display in an ordination points to a mulitcollinearity problem in the predictors. Be advised to check this beforehand.
This should work, some lines may be split by the comment format in researchgate though:
#Function: select.envfit - Setting r2 cutoff values to display in an ordination.r.select
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology