Information Technology Reference
In-Depth Information
Alkali (0.46)
1.2
Burst (0.92)
TEA (0.71)
TotYield (0.73)
2.2
2.4
3
T
ensile (0.65)
1.6
P.max
Growth (0.43)
.
P.kes
2
P.pat
P.tae
P.ell
3.5
Density (0.95)
3.5
P.ell; n = 11
P.max; n = 6
P.kes; n = 5
P.pat; n = 9
P.tae; n = 5
T
ear (0.75)
Figure 5.34
AoD biplot of the pine data. After normalizing the data,
PCAbipl
was
called to construct a PCA biplot of the group means. Subsequently the colour-coded
samples were interpolated into the biplot. Axis predictivities are given in parentheses.
alternative hypothesis permutation of the sampling units should tend to increase the
B-value. The algorithm provided by Good (2000) is implemented in our function
Per-
mutationAnova
to randomly permute
n
samples into
K
groups of fixed sample sizes
summing to
n
. Applying
PermutationAnova
to the normalized pine data resulted in
an achieved significance level of 0.0001.
As a final example we present in Figure 5.35 an AoD plot of the pine data based
on Clark's distance. Comparing Figure 5.35 with Figure 5.34, we see that the use of
Clark's distance resulted in a shift of
P.tae
away from
P.kes
, while
P.ell
shifted to a
more central position to end up closer to
P.tae
. Figure 5.35 is the result of the following
function call to
AODplot
:
AODplot(X = Pine.data[,-1], dist = "Clark", X.new.samples =
Pine.data[,-1], group.vec = Pine.data[,1], exp.factor = 2.5,
label.size = 0.6, pch.samples = rep(0,length(unique
(Pine.data[,1]))), pch.samples.col = c("red",
"green","blue","orange","purple"), pch.samples.size = 0.9,
pch.means = rep(15, length(unique(Pine.data[,1]))),
pch.means.col = c("red","green","blue","orange","purple"),
pch.means.size = 1.5, pos.labs = 1, weight = "unweighted")
