Detecting Clusters in 3D Dynamic Graphs
John Fox, Robert A. Stine · 2000
At least in principle, three-dimensional dynamic scatterplots can reveal certain features of data that cannot be apprehended in marginal two-dimensional displays of the data. Using graduate students in statistics as subjects, we seek to establish whether the probability and rapidity of detection of clusters in 3D plots vary by easily characterized properties of the clusters -- in particular, their separation and orientation. Next, we address whether the design of the display -- in particular, the use of perspective, depth-cueing, and subject-controlled (as opposed to automatic) motion --- influences subjects' ability to detect clusters. We find that probability of detection and rapidity of response increase smoothly with cluster separation, and that, at a fixed level of separation, `diagonally ' displaced clusters are easier to detect than `horizontally' displaced clusters. Accuracy and response latency in this task appear to be a#ected to a smaller extent by the design of the display. We discuss the potential implications of these results for the design of statistical software incorporating dynamic 3D scatterplots. [Keywords: 3D scatterplots, graphical perception, statistical graphics software] 1