An official website of the United States government
Surveillance Research Program

Multi-Group Clustering Results

Describes the results produced by the clustering process.

 NOTE:  The Multi-Group Clustering is considered an "Alpha" feature.

When a Multi-Group Clustering analysis is performed, two new tabs are added to the output viewer:

  1. Final Clusters
  2. Clustering Calculations

 

Final Clusters Tab

The Final Clusters tab will contain one table.  The table will contain Joinpoint model results for each of the best number of clusters for the cohort.  For Example, if the multi-group clustering process determined that 4 clusters was the best number overall, rates for each of the age-groups comprising the 4 clusters are computed and then modeled.  The intercepts, APC values, and APC ranges for the best fit model of each modeled cluster are displayed. 

 

Clustering Calculations Tab

The Clustering Calculations tab will contain up to four tables, each showing information about the final selected clusters and how they were determined.  There is also a button to display all the best fit model graphs for the final selected clusters.

View Cluster Graphs button

At the top of the tab, there will be a button labeled “View Cluster Graphs”.  When this button is clicked, a special dialog will appear showing the graphs for the best fit Joinpoint model for each of the final selected clusters.

 The View Cluster Graphs dialog will have the following two checkbox options:

  • Show APCs
    • When this option is selected, the APCs for each line segment are displayed within each graph on the dialog.
  • Start Y-Axes at Zero
    • When this option is selected, the Y-axis for each graph displayed on the dialog will start at zero.

 Tables Displayed

The following three tables will always be shown:

  1. Pre-Clustering Age Groups Table

This table contains the results of the forward or backward age group count compression for the given cohort.  The table will list the resultant compressed age groups along with the minimum count found for each group over all the data points (e.g., years) along with the average and maximum counts.  The age groups displayed in this table were used as a starting point for the multi-group clustering algorithm for the given cohort.

  1. Cluster Selection Table

For each possible cluster group (from 1 up to the maximum number specified by the user), this table will contain the best age groups for each cluster number.  All groups will have an associated BIC value.  The lowest BIC value determines the best overall cluster group.  E.g.  The group of 5 clusters had the lowest BIC value and therefore is the best overall for the cohort.   The best overall cluster group will be modeled, and the results show on the regular Joinpoint tabs. 

  1. Cluster Summary Table

This table contains information on the best selected cluster group for the given cohort (e.g. 5),  The table will  contain information such as the method used to select the best cluster group (e.g., BIC), the method for selecting the best joinpoint model fit for each cluster within the group (e.g. WBIC), the max number of clusters specified by the user, the best number of clusters selected by the clustering algorithm, and the Q statistic.  The Q statistic is the sum of the residual sum of squares from the best Joinpoint model for each cluster within each group.  

 If you selected the Clustering MADWD option, the following table will be displayed:

  1.  MADWD Process Table

This table will contain the results of the MADWD process on the Best Number of Clusters.  If the Best Number of Clusters (e.g. 5) fails, the MADWD process will look at the number of clusters below that (4) and will continue until it reaches Cluster 1, or a cluster passes the MADWD process.   If a MADWD test fails, the “MADWD Result” column will contain “Fail” and be colored red.