Biology Reference
In-Depth Information
Table 15.4. Changes in cluster assignment caused by the in-
clusion of additional variables, relative to the basic clustering
that uses variables
V
1
,
V
2
,and
V
3
.
Variables
Drug
Basic
New
Added
Cluster
Cluster
V
9
aldioxa
2
1
mycophenolate
3
1
zolpidem
1
2
V
7
,
V
9
aldioxa
2
1
dutasteride
3
2
mycophenolate
3
1
tamsulosin
2
1
zolpidem
1
2
V
7
,
V
9
,
V
10
aldioxa
2
1
insulin
1
2
mycophenolate
3
1
zolpidem
1
2
Table 15.5. Cluster assignments for seven drugs in the
basic clustering (the clustering for
k
=3based on vari-
ables
V
1
,
V
2
,and
V
3
), along with those for
k
=3and
k
=4based on variables
V
1
,
V
2
,
V
3
,
V
7
and
V
9
.
Drug
Basic
k
=3
k
=4
aldioxa
2
1
2
dutasteride
3
2
3
fluoxetine
2
2
4
mycophenolate
3
1
1
oxycodone
2
2
4
tamsulosin
2
1
2
zolpidem
1
2
2
along with the corresponding assignments in the basic clustering. Five drugs ex-
hibit different assignments between the five-variable clusterings with
k
=3and
k
=4: aldioxa, dutasteride, and tamsulosin move back into their assigned parti-
tions in the basic clustering, while fluoxetine and oxycodone form the new cluster
for
k
=4.
Fig. 15.10 gives a boxplot summary of the results for the five variable cluster-
ing with
k
=4, in the same format as Fig. 15.9. It is clear from this plot that the
key characteristic of the fourth partition in this clustering is the high correlation
between the two aggregate measures
R
ab
and
U
ab
and both the subjective measure
S
ab
and the case-specific measure
ψ
ab
. In particular, the correlation between
ψ
ab
and the two aggregate measures is markedly greater than that seen for any of the
other drugs in the case study, while the correlation between
S
ab
and these aggre-
gate measures is among the highest seen for any of these drugs. Also, note that
Search WWH ::
Custom Search