Subscribe Bookmark
danschikore

Staff

Joined:

Sep 21, 2012

Totally Rad Visualization with JMP

Everyone who works with multivariate data has dealt with the problem of how to show more than two dimensions at a time on a typical computer display. JMP has many options available for visualizing and analyzing multivariate data, including Scatterplot Matrix, Parallel Coordinates, Clustering and more.


Another graphical technique from the Visualization community is called “RadViz.” A RadViz add-in is now available on the JMP File Exchange to demonstrate this technique using JMP 9. (A free SAS login is required to download the add-in.)




RadViz is a shorthand for Radial Visualization. Given a set of n columns, you can arrange n points evenly distributed around a circle. Each observation in your data is mapped to a point in the circle, based on the relative proportions of the column values. You can think of the points on the circle as anchors for springs, with stiffness proportional to the normalized values. The coordinate for each observation is the point where all forces are equalized. As a result, you can see a visual clustering of your data that takes into consideration many different variables.




When applied to three-dimensional data, the result is the same as that produced by the Ternary Plot platform. Also, the choice of the order of points around the perimeter of the circle results in different mappings. Techniques like Principal Component Analysis and Partitioning can help you determine which columns best separate the data into distinct clusters. A useful extension to this add-in would be to calculate an optimal ordering of the columns based on these or other similar techniques.


I look forward to hearing about your Totally Rad Visualization!

Article Tags