CAREER: HCC: Developing Perceptually-Driven Tools for Estimating Visualization Effectiveness

University Of Colorado At Boulder, Boulder CO

Investigators

Abstract

Diverse people use visualizations to explore and communicate data across applications ranging from personal finance and activity tracking to public policy and scientific communication. New visualization techniques emerge as society's reliance on data grows. The ways these techniques represent data affect what people see in their data: different designs support different statistical insight and can even mislead people by appearing to show patterns that are not there. Designers lack grounded, actionable guidance for when and how their visualizations might most effectively communicate the patterns they see that matter most in their data. This project will work to develop data-driven models, metrics, and tools that describe what people see in a visualization and help them assess what it means and how it can help. Outcomes are expected to help people creating or using visualizations to rapidly predict what kinds of statistical patterns a visualization might communicate and the potential for biases in these patterns. This includes significantly reducing the time and expertise barriers for creating effective visualizations, leading to more precise and trustworthy public communication through data. The project will work to develop educational materials that teach best practices in visualization evaluation. These activities will collectively make it easier to develop honest and effective data visualizations across a range of goals and disciplines. The project investigates three core technical objectives: (1) modeling how people perceive different statistical properties of data across common representations, by performing a series of experiments modeling how people interpret statistical quantities in visualizations, leading to a curated corpus of experimental data for visualization perception; (2) generating metrics that describe the information people are most likely to gain from a visualization by conducting a series of mixed-methods experiments, paired with data from the first objective, to build models for probabilistically predicting visualization effectiveness; and (3) integrating these metrics into an interactive web tool for automating evaluation, by rapidly estimating the effectiveness of a given visualization for communicating a target set of properties. A crossover between visualization and vision science through interdisciplinary initiatives is planned, which includes the development and dissemination of educational materials released through open-source curriculum, an online course, and a textbook. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.