Information Technology Reference
In-Depth Information
Fun(96)
Fun(96)
Tem(82)
Tem(82)
Ran(94)
Ran(94)
T64(98)
T64(98)
Hun(96)
Cap(94)
Fun(96)
Dur(95)
Hob(98)
Spo(96)
T95(97)
T
e
m
(82)
T64(98)
T68(94)
Cap(94)
Ran(94)
Ear.H
Beg.H
Ran(94)
T64(98)
Fun(96)
Tem(82)
Kin(97)
Hun(96)
Ear.L
T68(94)
Tru.L
Box.H
B
o
x.L
Ear.L
Box.H
Tru.L
Box.L
D
ur(95)
Ear.H
Beg.H
Tru.H
Tru.H
Kin(97)
Cra.H
Cra.H
Fow.L
Cra.L
Beg
.
L
Edn.H
Cra.L
Beg.L
Edn.H
Edn.L
Edn.L
T95(97)
Fow.H
Fow.H
Fow.L
T68(94)
T68(94)
Dur(95)
Dur(95)
Kin(97)
Spo(96)
Kin(97)
T95(97)
T95(97)
Hob(98)
Hun(96)
Cap
Hun
Cap(94)
Fun
Fun(96)
Ran(94)
T64
T64(98)
T
68(94)
Tem(82)
Tem
Ran
T
6
8
D
u
r(95)
D
u
r
Kin
Ear.H
Ear.H(71)
Tru.L
Kin(97)
Box.H(34)
Tru.L(88)
T95
Box.L
Box.H
T95(97)
Cra.L(99)
Beg.L(99)
Ear.L(90)
Edn.H(99)
Tru.H
Ear.L
Tru.H(58)
Edn.H
Beg.H
Cra.L(99)
Box.L(80)
Cra.H(93)
Beg.H(50)
Fow.H(74)
Edn.H(99)
Cra.L
Cra.H
Edn.L
Fow.H
Edn.L(100)
Beg.L
Beg.L(99)
Edn.L(100
Spo(96)
Spo
Fow.L(90)
Fow.L
Hob(98)
Hob
Tru.L(88)
Figure 6.2
Biadditive biplots of wheat data (Table 6.1) after eliminating the overall
mean: (top left)
X
adj
V
used for plotting the sites and
V
for the varieties; varieties almost
on top of each other; (top right) biplot with lambda-scaling; (bottom left)
U
1
/
2
used for
1
/
2
plotting the sites and
V
for the varieties; (bottom right) biplot showing sites as biplot
1
/
2
1
/
2
axes with
U
used for plotting the varieties and
V
for the sites. Calibrations on
axes are in terms of deviations from the overall mean.
is that the two sets of points are approximately orthogonal. This is a diagnostic feature
that suggests that a main effects model suffices - we comment briefly on diagnostic
biplots towards the end of this chapter. We could have achieved a similar rescaling by
basing the plot on
U
1
/
2
1
/
2
. The biplot in the top left panel of Figure 6.2 is
and
V
distorted because
incorporates all information on size. This suggests that plots for
biadditive models should always be based on the
2
-scaling, possibly followed by
further lambda-scaling to improve presentation. because with our data
p
and
q
are of
similar size, additional lambda-scaling will have little effect. We emphasize that none of
these cosmetic changes affect the approximating inner product.
1
/
