Redundant Coding Can Speed Up Segmentation in Multiclass Displays

Multiclass data visualizations allow viewers to compare one dataset to another. The visual marks that represent these datasets, or classes, are visually distinguished from one another by easily perceived visual feature differences, such as color or shape. A designer of a graph or map might encode one class of marks as either red, or circular, and another class as either blue, or triangular. One common technique is to combine these cues in a redundant fashion, encoding one class as red and circular, and the other as blue and triangular, under the assumption that a larger difference (via multiple differing features) should help. Recent work [6] has empirically demonstrated strengthened grouping and improved accuracy in segmentation of redundantly coded objects. Does this redundancy benefit generalize to more realistic displays, and to other measures such as segmentation speed? We demonstrate in an experiment that redundant coding can lead to a small improvement in speed of visual differentiation in a simulated dataset in a crowded display.