Abstract
To interpret information visualizations, people infer how visual features map onto concepts. Evidence suggests this inference process is analogous to solving an assignment problem in optimization. People infer optimal mappings between colors and concepts, even if that requires assigning a concept to its most weakly associated color (Schloss, Lessard, Walmsley, & Foley, 2018). We developed and began testing a process model for how assignment inference works, asking how assignment inference is influenced by the relative association strength between two candidate colors with a target concept and a non-target concept. We selected the colors and concepts using data from an initial experiment in which participants rated the association strength between each of 58 colors (sampled uniformly over CIELAB space) and twelve fruits. We selected two fruits (cantaloupe and strawberry) and eight colors that varied systematically in association with each fruit. In the main experiment, participants interpreted bar graphs depicting fictitious fruit preference data. Each trial contained a graph with two unlabeled colored bars representing preferences for cantaloupes and strawberries and a target fruit name (cantaloupe/strawberry) above. Participants reported which color represented the target concept. Graphs were constructed from all 56 pairwise combinations of eight colors (left/right balanced) × 2 possible taller bars (left/right) × 2 target concepts (cantaloupe/strawberry) × 3 repetitions (total of 672 trials). Multiple linear regression demonstrated that 83% of the variance in response times was predicted by the association strength between the correct color and the target (70%) and the non-target (+13%). For accuracy, 79% of the variance was predicted by association strengths between the correct color and the target (41%), the incorrect, competitor color with the target (+28%), and between the correct color and the non-target (+10%). These results help refine our assignment inference process model and can be used to help design optimal visual encoding systems.
Acknowledgement: Office of the Vice Chancellor for Research and Graduate Education