Agglomerative cluster analysis

75

15

Agglomerative cluster analysis

Hierarchical agglomerative cluster analysis is selected by choosing

Clustering|Agglomerative from the drop down menu. The basic process is that all

the samples (columns) start as separate groups that are joined together one at a

time until a single group is formed. At each iteration, a number of different

procedures can be used to determine which groups should be joined. The

following methods for linking groups are available:

Ward s

75

Single linkage

75

.

Complete linkage

75

.

Average linkage.

76

McQuitty s

76

Gower s.

76

Centroid.

76

In the Setup For Cluster box, 

CAP

 then offers a choice of 9 distance measures:

Euclidean

91

, 

Geodesic

92

, 

Whittakers

93

, 

Ave. Distance

92

, 

Manhattan

92

,

Canberra

93

, 

Chord

92

,  

Mean Character Difference (Czekanowski)

91

 or 

Bray 

Curtis

94

 distance.

Once a measure has been selected the cluster analysis will immediately be run.

Output is presented under a number of tabbed components that can each be

viewed by clicking on the tab. These are described in turn below.

Cluster summary

77

Cluster groups

77

Dendrogram   CLUSTER ANALYSIS

77

15.1

Ward's

Also termed minimum variance or error sums of squares clustering. At each

iteration, all possible pairs of groups are compared and the two groups chosen for

fusion are those which will produce a group with the lowest variance.

15.2

Single linkage

Also termed minimum or nearest neighbour method. At each iteration, the

clusters are compared in terms of the similarity of their most similar samples

(columns) and the two clusters that hold the most similar samples are fused.

15.3

Complete linkage

This is also called furthest neighbour sorting. At each step, the clusters are

compared in terms of the similarity of their least similar members and the two

clusters that are most similar are fused.

Copyright 2004, PISCES CONSERVATION LTD