# Select species with goodness-of-fit at least 0.6 in the
# ordination plane formed by axes 1 and 2
spe.good <- goodness(spe.rda)
sel.sp <- which(spe.good[, 2] >= 0.6)
# Triplots with homemade function triplot.rda(), scalings 1 and 2
triplot.rda(spe.rda,
site.sc = "lc",
scaling = 1,
cex.char2 = 0.7,
pos.env = 3,
pos.centr = 1,
mult.arrow = 1.1,
mar.percent = 0.05,
select.spe = sel.sp
)
triplot.rda(spe.rda,
site.sc = "lc",
scaling = 2,
cex.char2 = 0.7,
pos.env = 3,
pos.centr = 1,
mult.arrow = 1.1,
mar.percent = 0.05,
select.spe = sel.sp
)
Hint Argument mar.percent can receive a negative value, e.g. −0.1, which shrinks
the margins of the graph. In some graphs, the arrows (species and/or
environmental variables) must be shortened to fit the size of the cluster of site
points. Shrinking the margin compensates for that, allowing all elements of the
graph to occupy the whole plotting area.
Display the documentation in the upper part of the file of function
triplot.rda() and explore the various possibilities to improve the look of
your triplots.
6.3.2.4 Permutation Tests of RDA Results
Due to widespread problems of non-normal distributions in ecological data, classical
parametric tests are often not appropriate in this field. This is why most methods of
ecological data analysis nowadays resort to permutation tests whenever possible. In
RDA, the use of parametric tests is possible only when the response variables are
standardized and the error distribution is normal (Miller 1975, Legendre et al. 2011);
this is clearly not the case for community composition data, for example. So all RDA
programs for ecologists implement permutation tests. The principle of a permutation
6.3 Redundancy Analysis (RDA)
219