Perceiving Patterns in Parallel Coordinates: Determining Thresholds for Identification of Relationships

Abstract

This article presents a study that investigates the ability of humans to perceive relationships (patterns) in parallel coordinates, an ability that is crucial to the use of this popular visualization technique. It introduces a visual quality metric, acceptable distortions of patterns, which establishes the level of noise that may be present in data while allowing accurate identification of patterns. This metric was used to assess perceptual performance of standard 2D parallel coordinates and multi-relational 3D parallel coordinates in two experiments. In multirelational 3D parallel coordinates the axes are placed on a circle with a focus axis in the centre, allowing a simultaneous analysis between the focus variable and all other variables. The experiments aimed to determine the maximum number of variables that can be, from a user's point of view, efficiently used in a multi-relational 3D parallel coordinates display and to present a first attempt to study users' ability to analyse noisy data in parallel coordinates. The results show that, in terms of the acceptable level of noise in data, a multi-relational 3D parallel coordinates visualization having 11 axes (variables) is as efficient as standard 2D parallel coordinates. Visualizing a larger number of variables would possibly require a greater amount of manipulation of the visualization and thus be less efficient.