Authors
Jing Li
Jack van Wijk
Abstract
Scatterplots and parallel coordinate plots can both be used to find correlation visually [2][3][4]. In this paper, we compare these two visualization methods in two user studies. In the first study, 25 participants indicated the correlation they observed in plots with varying correlation of sample data, sample size, and timing condition. In the second study, the effect of training on parallel coordinate plots was studied. Results show that the degree of correlation is under-estimated, particularly in parallel coordinate plots, and that the observations based on parallel coordinate plots are less constant under different conditions. Moreover, training of parallel coordinate plots for moderate and low correlations might confuse users. Therefore, we conclude that scatterplots are better than parallel coordinate plots for visual correlation analysis.
Acknowledgments
The project is supported by the VIEW program of the Netherlands Organization for Scientific Research (NWO) under research grant no. 643.100.502.