Abstract
The duck-rabbit and Necker cube illusions reveal that the visual system can lock into a single view of a multi-stable percept (Attneave, 1971). Such ambiguity is rare in the natural world, but ubiquitous in the artificial world of information visualizations. Graphs are one example of ambiguous figures because they contain many perceivable patterns despite providing the same visual stimulation. After viewers extract an initial set of visual statistics about the dataset (Szafir et al., 2016), they exercise top-down attentional control (Egeth et al., 2010) to extract relationships and patterns from the data values (Michal et al., 2016; 2017). This diversity of percepts extractable from patterns in graphs might lead to two people seeing different patterns in the same graph as a function of how they configure that top-down control. In this experiment, participants were shown line graphs and bar graphs depicting a student government election and asked to predict which party would win. They were also asked to indicate which graph features were most visually salient. Critically, the graphs were designed to be ambiguous so that viewers' predictions could differ depending on which features and patterns they attend to. For example, participants who selected certain features, such as more global trends, as the most salient tended to predict a certain party to win, while other participants who selected other features, such as local comparisons, tended to predict the other party to win. Even in the same image, viewers could set themselves to different attentional modes (e.g. global or local) to draw different conclusions from an identical dataset, providing a real-world example of an ambiguous figure.
Meeting abstract presented at VSS 2018