On the effective visualisation of dynamic attribute cascades

Abstract

Cascades appear in many applications, including biological graphs and social media analysis. In a cascade, a dynamic attribute propagates through a graph, following its edges. We present the results of a formal user study that tests the effectiveness of different types of cascade visualisations on node-link diagrams for the task of judging cascade spread. Overall, we found that a small multiples presentation was significantly faster than animation with no significant difference in terms of error rate. Participants generally preferred animation over small multiples and a hierarchical layout to a force-directed layout. Considering each presentation method separately, when comparing force-directed layouts to hierarchical layouts, hierarchical layouts were found to be significantly faster for both presentation methods and significantly more accurate for animation. Representing the history of the cascade had no significant effect. Thus, for our task, this experiment supports the use of a small multiples interface with hierarchically drawn graphs for the visualisation of cascades. This work is important because without these empirical results, designers of dynamic multivariate visualisations (in many applications) would base their design decisions on intuition with little empirical support as to whether these decisions enhance usability.