In contrast, a CA groups similar data into clusters by attempting to minimize the variability within each cluster and maximizing variability between clusters. The kmeans method is nonhierarchical supervised partitioning CA (Rencher 2002, 482). The number of clusters (k) is predetermined or supervised and a vector specifies the mean of a cluster, or centroid, with each component being the average of a variable in the analysis. The algorithm uses one initial observation per cluster as the mean for that cluster and then evaluates each of the remaining observations for inclusion into a cluster. As each observation is included, the mean of each cluster is recalculated and previously clustered observations are reevaluated for appropriate clustering. Observations and k number of means are reevaluated at each step until no further improvement can be achieved and all observations have been clustered. Conversely, a hierarchical clustering analysis (HCA) begins with one cluster per observation and then successively agglomerates clusters using an appropriate measure of similarity between clusters. As the number of observation per cluster is increased, the number of clusters is reduced until one cluster with all observations is reached. In HCA, a clustered observation is not reevaluated for a better cluster fit, but no supervision in the number of groups is required.