Numerous multivariate visualization techniques and systems have been developed in the past three decades to visually analyze and explore multivariate data being produced daily in application areas ranging from stock markets to the earth and space sciences. However, traditional multivariate visualization techniques typically do not scale well to large multivariate data sets, with the latter becoming more and more common nowadays. This paper proposes a general framework for interactive hierarchical displays (IHDs) to tackle the clutter problem faced by traditional multivariate visualization techniques when analyzing large data sets. The underlying principle of this framework is to develop a multi-resolution view of the data via hierarchical clustering, and to use hierarchical variations of traditional multivariate visualization techniques to convey aggregation information about the resulting clusters. Users can then explore their desired focus region at different levels of detail, using our suite of navigation and filtering tools. We describe this IHD framework and its full implementation on four traditional multivariate visualization techniques, namely, parallel coordinates (Inselberg and Dimsdale, Proceedings of Visualization (1990) 361; Wegman, J. Amer. Statist. Assoc. 411(85) (1990) 664), star glyphs (Siegel et al., Surgery 72 (1972) 126), scatterplot matrices (Cleveland and McGill, Dynamics Graphics for Statistics (1988)), and dimensional stacking (LeBlanc et al., Proceedings of Visualization 90 (1995) 271), as implemented in the XmdvTool system (Ward, Proceedings of Visualization 94 (1994) 326; Martin and Ward, Proceedings of Visualization 95 (1995) 271, Fua et al., Proceedings of Visualization 99 (1999) 43; Proceedings of Information Visualization 99 (1999) 58). We also describe an empirical evaluation that verified the effectiveness of the interactive hierarchical displays. (C) 2003 Elsevier Science Ltd. All rights reserved.
