Polychromatic plots: Graphical display of multidimensional data


  • This article is a US Government work and, as such, is in the public domain in the United States of America.


Limitations of graphical displays as well as human perception make the presentation and analysis of multidimensional data challenging. Graphical display of information on paper or by current projectors is perforce limited to two dimensions; the encoding of information from other dimensions must be overloaded into the two physical dimensions. A number of alternative means of encoding this information have been implemented, such as offsetting data points at an angle (e.g., three-dimensional projections onto a two-dimensional surface) or generating derived parameters that are combinations of other variables (e.g., principal components). Here, we explore the use of color to encode additional dimensions of data. PolyChromatic Plots are standard dot plots, where the color of each event is defined by the values of one, two, or three of the measurements for that event. The measurements for these parameters are mapped onto an intensity value for each primary color (red, green, or blue) based on different functions. In addition, differential weighting of the priority with which overlapping events are displayed can be defined by these same measurements. PolyChromatic Plots can encode up to five independent dimensions of data in a single display. By altering the color mapping function and the priority function, very different displays that highlight or de-emphasize populations of events can be generated. As for standard black-and-white dot plots, frequency information can be significantly biased by this display; care must be taken to ensure appropriate interpretation of the displays. PolyChromatic Plots are a powerful display type that enables rapid data exploration. By virtue of encoding as many as five dimensions of data independently, an enormous amount of information can be gleaned from the displays. In many ways, the display performs somewhat like an unsupervised cluster algorithm, by highlighting events of similar distributions in multivariate space. Published 2008 Wiley-Liss, Inc.