Category Analysis Tool | Download | Installation | Operation | Results

Four Types of Results
  • Tree visualization of clustering: This component of the results report is implemented as a Java applet and is based in large part on IBM's EZ Sort application.

    The radio buttons at the top of the panel control whether a 'single linkage' or 'complete linkage' method is used in calculating the display parameters. When single linkage is used, the algorithm computes the minimum distance between clusters in order to determine which new element to add to the growing tree. In complete linkage, the maximum distance is used.

    Besides toggling the radio buttons, you can interact with the tree by dragging the red vertical line across the display. As cross-bars of the tree move from one side of the sliding bar to the other, you will see different colored clusters appear. The implementation of the coloring of the labels only provides for 8 distinct colors; therefore, adjacent, same-colored items are part of the same cluster while a same-colored item separated by 7 other colors is in a different cluster.

    While drawing a tree is science, interpreting a tree is an art. With experience you will be able to make reasonable inferences about your subjects' clustering patterns based on the tree display. Some heuristics are:

    • Tight clusters on the left side of the tree show consensus among your subjects.
    • The further the slider is to the right, the more distance was required between subclusters before commonality could be found by the algorithm. Therefore, when all the cross-bars are toward the right side of the tree, one can interpret that there were no tight clusters.

  • Link to raw data files (data.html): These are plain text files that can be used to cut-and-paste data into other desktop applications. They contain the raw item-by-item data matrix, the distance matrix (calculated using the squared Euclidean distance), and the amalgamation schedule (the order in which items were incorporated into a cluster). Each cell in the item-by-item matrix contains the number of subjects who put the pair of items for that row and column into the same bin; the diagonal contains an integer that indicates the number of subjects who performed the experiment; the upper-right and lower-left triangular submatrices are symmetrical. The number in parentheses next to an item name is the item ID created during the setup procedure. The height in the amalgamation schedule is the point at which a cross-bar appears in the tree visualization.

  • Tabular summary of comments lists the comments made by each of the subjects. This report component is suppressed if the design disallowed commenting. If the commenting option was changed part way through testing, you may recover those comments by entering the setup process, then continue through the pages until 'page 4 of 4'. Check the box for commenting. Then select to view the results. Blanks will be shown in cases where the subject did not make a comment, either because commenting was not allowed or because they just didn't make a comment.

  • Tabular summary of items in a subject X categories matrix: Inspection of this table can give a variety of insights dependent on the design of the study. It is easy to detect if subjects created roughly the same number of categories. It is also easy to detect subjects whose performance might cause them to be labeled as 'outliers'.

    When pre-named categories are used, common column headings are created in the table. This allows easy scanning of the categories for each subject.

    Note that when a category method is other than pre-named (e.g., unspecified, specified, fixed but not named), then common category names are not possible. If the subject is 'allowed' or 'required' to name the categories, then the name entered by the subject is shown in bold within the cell before the items associated with that cell. WebCAT will capture bins that have labels entered by the subject whether or not the subject placed any items in the bin. Similarly, WebCAT will capture the list of items from a bin regardless of whether or not the subject provided a name for the category. If the subject is 'required' to enter labels, there will be a label for each cell.



Overview | Installation | Operation | Results | FAQ
Version 2.5
Page last modified: 15 May 2002
National Institute of Standards and Technology (NIST)