Kushin Mukherjee¹ (), Holly Huey², Lauren A Oey³, Judith E Fan^1,4,5; ¹Department of Psychology, Stanford University, ²Department of Psychology, University of California, San Diego, ³Department of Psychology, Princeton University, ⁴Department of Computer Science, Stanford University, ⁵Graduate School of Education, Stanford University

Data visualizations are used to efficiently communicate patterns in quantitative data. While substantial work has investigated which visualization properties affect observers’ ability to make fast and accurate judgments about data (e.g., using scatterplots to expose correlations rather than grouped bar plots), less work has investigated which factors visualization designers consider when trying to anticipate what would be helpful to an observer. Are designers more likely to follow simple heuristics (e.g., “Use scatterplots because they are more informative than barplots.”) or do they choose different strategies depending on the question at hand? English-speaking adult participants (N=119; mean age: 38.87 years) performed a visualization design task where they decided which graph to show an observer to help them answer a question about a dataset (e.g., “What is the average departure delay of LaGuardia flights leaving at 2pm-6pm?”). Building on preliminary work (Huey et al., 2023), we devised scenarios where participants chose between different graph types (i.e., bar, line, scatter) and how many variables to show (i.e., between 3-5). We found that participants’ choices were far from uniform (χ2(8)=1760.40, p<.001), and that simple heuristics, such as a persistent preference for fewer variables or for a specific graph type, failed to explain these patterns. Instead, participants chose different visualizations depending on which question they were presented with, and they were more likely to choose visualizations that an independent sample of observers (N=1,229; mean age: 21.14 years) found useful for accurately answering that question. These results are consistent with the possibility that participants make their choices by simulating how another person would answer the question, given a visualization. More broadly, such studies suggest that people make flexible use of their own visualization understanding abilities to communicate information effectively to others.

Acknowledgements: This work was supported by NSF CAREER Award #2047191, NSF DRL award #2400471, and a Stanford Hoffman-Yee HAI Grant to JEF