Hierarchical cluster analysis may be combined with kmeans
clustering to combine tbe strengths of botb approaches
(Hair et al. 1998). Tbe agglomeration schedule
is able to indicate tbe most appropriate number of clusters,
and tbis provides tbe basis for tbe number of clusters
to be specified for the k-means cluster procedure. In
addition, the cluster centroids produced by the hierarchical
clustering procedure may be used as initial cluster
centers for the k-means procedure. We adopt tbis approach
for this study. Missing values for some cases reduced
the sample size for the clustering procedure from
404 to 371 firms. The agglomerative schedule from tbe
hierarchical clustering procedure indicated that a twoor
four-cluster solution would be most appropriate. After
we conducted k-means clustering solutions for both
Clusters 2 and 4, we found a four-cluster solution to be
tbe most interpretable. We then used several procedures
to establish cluster stability.