Working with the Cluster Matrix Window

The Cluster Matrix displays a two-dimensional projection of each cluster onto a specified number of effect pairs. These projections are useful for spotting cluster similarities and differences within the plotted effect pairs.

Each cluster is assigned a unique color. Although each cluster is unique in n-space, the two-dimensional projections will overlap. It is important to note that every observation can belong to exactly one cluster. However, because the Cluster Matrix displays a projection in just two dimensions, multiple clusters can overlap an observation.

When a heat map is not used, individual observations are color-coded to indicate cluster membership.

To view a larger plot for an effect pair, right-click inside that plot, and click Explore. In the Explore window, it is easier to view and select observations.

To view a box plot for a variable that segments the observations by cluster, right-click inside a plot that contains the variable of interest, and select Plot variable_name by Cluster ID. Each variable in the selected plot has a menu item. The box plot is used to determine how similar the clusters are for the selected variable.

To derive a cluster ID variable that can be used in other explorations, right-click inside a plot and select Derive a Cluster ID Variable. When you select this item, SAS Visual Statistics creates a category variable that contains the cluster ID for each observation. Cluster ID -1 is reserved for observations with missing values.