The first step of any K-means clustering analysis is to define the initial location of the clusters. Once these locations have been specified, the K-means algorithm iteratively assigns observations to each cluster and updates the location of the cluster center. This process repeats until no observations change their cluster assignment. While the different algorithms for performing K-means clustering are different in the way that they iteratively perform this process, they all depend on the clusters to first be assigned to a given location before they can operate.

This process of assigning initial cluster locations is - appropriately enough - called initialization. Regardless of the algorithm chosen on the Options tab of the analysis parameters dialog, Prism uses the K-means++ method of initialization. This tab of the results provides the initial location of the center of each cluster. These locations are given as scaled values for each of the variables used in the analysis. Note that the initial cluster centers depend on the random seed used for the analysis. The default setting is to use an automatic random seed. That means that if you perform two different analyses with identical parameters on the same data, you may still end up with slightly different results due to the fact that the initial cluster locations will likely be different. page.