Cluster Analysis 9

Partitioning methods:

Use the squared Euclidean distance from the (dis) similarity menu.
Deciding on the number of clusters	Hierarchical clustering:
	Examine the dendrogram:
	Statistics ► Multivariate analysis ► Cluster analysis ► Postclustering ► Dendrogram
	cluster dendrogram wards_linkage, cutnumber (10) showcount
	Examine the VRC and Duda-Hart indices:
	Statistics Multivariate analysis ► Cluster analysis ► Postclustering ► Cluster analysis stopping rules.
	For VRC: cluster stop wards_linkage, rule (calinski) groups(2/11)
	For Duda-Hart: cluster stop wards_linkage, rule (duda) groups(1/10)
	Include practical considerations in your decision.
	Partitioning methods:
	Run a hierarchical cluster analysis and decide on the number of segments based on a dendrogram, the VRC, and the Duda- Hart indices; use the resulting partition as starting partition.
	Statistics Multivariate analysis ► Cluster analysis ► Postclustering ► Cluster analysis stopping rules.
	cluster kmeans e1 e5 e9 e21 e22, k(3) measure (L2squared) name(kmeans) start(group (cluster_wl))
	Include practical considerations in your decision.

(continued)

Table 9.12 (continued)

Theory

Action

Validating and interpreting the cluster solution

Stability	Re-run the analysis using different clustering procedures, linkage algorithms or distance measures. For example, generate a cluster membership variable and use this grouping as starting partition for k-means clustering.
	cluster generate cluster_wl ¼ groups(3), name (wards_linkage) ties(error)
	cluster kmeans e1 e5 e9 e21 e22, k(3) measure (L2squared) name (kmeans) start(group (cluster_wl))
	Examine the overlap in the clustering solutions. If more than 20% of the cluster affiliations change from one technique to the other, you should reconsider the set-up.
	tabulate cluster_wl kmeans
	Change the order of objects in the dataset (hierarchical clustering only).
Differentiation of the data	Compare the cluster centroids across the different clusters for significant differences.
	mean e1 e5 e9 e21 e22, over(cluster_wl)
	If possible, assess the solution’s criterion validity.
Profiling	Identify observable variables (e.g., demographics) that best mirror the partition of the objects based on the clustering variables.
	tabulate cluster_wl flight_purpose, chi2 V
Interpretating of the cluster solution	Identify names or labels for each cluster and characterize each cluster by means of observable variables.

4.