The sites are nicely clustered along their major ecological gradients. Remember
that this is an analysis of the fish community data.
# Complete CCA 3D triplot
ordirgl(spe.cca.pars, type = "t", scaling = 2)
orgltext(spe.cca.pars,
display = "species",
type = "t",
scaling = 2,
col = "cyan"
)
# Plot species groups (Jaccard dissimilarity, useable in R mode)
gs hclust(vegdist(t(spe), method = "jaccard"), "ward.D2"),
k = 4)
ordirgl(spe.cca.pars,
display = "species",
type = "t",
col = gs + 1)
Hint Three-dimensional plots have many options. Type ?ordirgl to explore some of
them. It is also possible to draw 3D plots of RDA results, but there is no simple
means to draw arrows for the response variables.
6.5 Linear Discriminant Analysis (LDA)
6.5.1 Introduction
Linear discriminant analysis differs from RDA and CCA in that the response
variable is a single variable classifying the sites into groups. This grouping may
have been obtained by clustering the sites on the basis of a data set, or it may
represent an ecological hypothesis. LDA tries to determine to what extent an
independent set of quantitative variables can explain this grouping. We insist that
the site typology must have been obtained independently from the explanatory
variables used in the LDA; otherwise the procedure would be circular and the tests
would be invalid.
LDA can provide two types of functions. Identification functions are obtained
from the original (non-standardized) descriptors and can be used to find the group to
which a new object should be attributed. Discriminant functions are computed from
standardized descriptors. These coefficients quantify the relative contributions of the
6.5 Linear Discriminant Analysis (LDA)
263