Detecting Nonlinearity in 3D Dynamic Graphs

Three-dimensional dynamic scatterplots can reveal characteristics of data – such as certain kinds of clustering and nonlinearity – that are not apparent in marginal two-dimensional views of the data. The experiments reported in this paper address the detection of nonlinearity in 3D dynamic scatterplots. Employing graduate-student subjects, we designed datasets that incorporated varying degrees of nonlinearity and varying degrees of systematic information (‘signal’). We also manipulated various aspects of the displays, including the display of regression planes and residuals; tying points visually to the horizontal (predictor) plane; the motion of the display; and the use of perspective and depth-cueing. We found that most (but not all) subjects were able to respond in a reasonable manner to properties of the data, so that the probability of detection of nonlinearity increased with its level, particularly when the signal was strong. Subjects’ performance was also affected, though to a lesser extent, by characterisics of the displays. We found, for example, that spinning the display horizontally in the regression plane was particularly effective. [