Hint See how the goodness-of-fit of individual sites is represented using the results of
the goodness() analysis by way of the cex argument of the points()
function.
As with the other ordination methods, it is possible to add information from a
clustering result to an NMDS ordination plot. For instance, compute a Ward
clustering of the percentage difference matrix, extract 4 groups and colorize the
sites according to these groups:
# Ward clustering of percentage difference dissimilarity matrix
# and extraction of four groups
spe.bray.ward spe.bw.groups <- cutree(spe.bray.ward, k = 4)
grp.lev <- levels(factor(spe.bw.groups))
# Combination with NMDS result
sit.sc <- scores(spe.nmds)
p main = "NMDS/% difference + clusters Ward/% difference")
for (i in 1:length(grp.lev))
{
points(sit.sc[spe.bw.groups == i, ],
pch = (14 + i),
cex = 2,
col = i + 1)
}
text(sit.sc, row.names(spe), pos = 4, cex = 0.7)
# Add the dendrogram
ordicluster(p, spe.bray.ward, col = "dark grey")
# Add a legend interactively
legend(
locator(1),
paste("Group", c(1:length(grp.lev))),
pch = 14 + c(1:length(grp.lev)),
col = 1 + c(1:length(grp.lev)),
pt.cex = 2
)
5.6.3 PCoA or NMDS?
Principal coordinate analysis and nonmetric multidimensional scaling often pursue
similar goals: obtain a meaningful ordination of objects on a small (usually 2 or 3)
number of axes. As seen in the previous sections, in PCoA the solution is found
196
5 Unconstrained Ordination