4.7.3.5 Comparison Between the Dissimilarity Matrix and Binary Matrices
Representing Partitions
This technique compares the original dissimilarity matrix to binary matrices computed from the dendrogram cut at various levels (and representing group allocations). The idea is to choose the level where the matrix correlation between the two is
the highest. Tests are impossible here since the matrices are not independent of one
another; indeed, the matrices corresponding to the partitions are all derived from the
original dissimilarity matrix.
To compute the binary dissimilarity matrices representing group membership, we
will use our homemade function grpdist() to compute a binary dissimilarity
matrix from a vector defining groups. The results are shown in Fig. 4.12.
0
5
10
15
20
25
30
0.0
0.2
0.4
0.6
Matrix correlation-optimal number of clusters
k (number of clusters)
Pearson's correlation
optimum
6
Fig. 4.12 Barplot showing the matrix correlations between the original dissimilarity matrix and
binary matrices computed from the dendrogram cut at various levels
4.7 Interpreting and Comparing Hierarchical Clustering Results
83