Improving Data Visualization With Cognitive Science (2025-2026)
Background
Effective data visualizations allow audiences to perceive data patterns much more efficiently and accurately than written or verbal descriptions of those same patterns. When designed appropriately, they can also enhance people’s trust in the data and confidence in decisions based on that data. Best practices in data visualization have largely been developed by designers, statisticians and computer scientists. Surprisingly, there is still little actual evidence about whether humans perceive, interpret and interact with data visualizations the way best practices assume.
Many data visualization practices are inspired by predictions of how they will impact “cognitive load,” or our limited cognitive capacity for remembering and manipulating information during tasks. For example, one best practice — introduced by statistician Edward Tufte — is to maximize the “data-ink ratio,” or the proportion of ink that cannot be erased from a visualization without losing information. It is often assumed that reducing unnecessary visual complexity will reduce imposition on cognitive load, freeing up working memory to do things like correctly interpret data. However, evidence is lacking about whether the data-ink ratio actually impacts cognitive load or the viewer’s ultimate comprehension of a graph, and some redundancy of visual information may be helpful for learning and comprehension.
The gap between the hypotheses and the reality of how we process data visualization limits our ability to predict which data visualizations will be most effective or most misleading. This can have significant impacts in applications like making medical decisions or determining whether an artificial intelligence model is fair. Leveraging cognitive science tools to evaluate popular “best practices” could help resolve this problem.
Project Description
Building on the work a previous teams, this project team will use cognitive science techniques to address knowledge gaps related to data visualizations and human perception. Metrics such as pupil diameter, the number and duration of eye fixations on a visual stimulus and the number of eye movements associated with these fixations correlate with cognitive load. Thus, eye-tracking studies can evaluate how different versions of a graph impact cognitive load and subsequent understanding.
In this project, groups of team members will individual data best practices like “maximize the data-ink ratio” to the test. Each group will create relevant stimuli; collect relevant eye-tracking, facial expression, questionnaire and qualitative interview data to assess audience members’ responses to the stimuli; and analyze and interpret the results. The team will then work together to develop visualizations that intentionally challenge visualization best practices and engage data visualization experts to make predictions about how audience members will respond to those visualizations. Team members will collect more data to test those predictions and will share results from both phases in a public-facing blog.
Goals of this project include studying how visual complexity and cognitive load impact financial and moral judgments about resource allocation for public emergencies and conducting pilot experiments to test five to six common “best practices” in data visualization. The project also aims to evaluate whether online eye-tracking tools could be used for follow-up work related to “Explainable AI” visualizations. The team will work with explainable AI expert Brinnae Bent to brainstorm visualizations to support explainable AI methods.
Anticipated Outputs
Blog for a general audience explaining the team’s studies; preliminary data; a symposium presentation to the Duke and broader Research Triangle data visualization community
Student Opportunities
Ideally, this project team will include 4 graduate students and 6 undergraduate students. Students may have interests or backgrounds in data visualization, cognitive or social sciences and visual arts or design. A graduate student will be selected to serve as project manager.
The team will be divided into 2-3 subteams, each focusing on one data visualization best practice or experiment, and each with 1-2 graduate/professional students and 2-3 undergraduates. It is intended that each group will have at least one undergraduate with interests or expertise in each of the following areas: 1) the technical skills required to create data visualizations; 2) cognitive or social sciences; and 3) visual arts or design. The graduate/professional students will ideally be able to cover the same three skill/interest areas among themselves and be willing to help other groups if they have skill sets other groups are lacking.
Team members will conduct literature searches, run cognitive experiments to test popular data visualization best practices, evaluate and discuss results, write blog posts to communicate findings to a general audience and present results to the local data visualization community with the support of the Duke Libraries Center for Data and Visualization Sciences.
Students will develop skills in creating and critiquing data visualizations, testing cognitive variables, experimental design, statistical analysis, popular science communication and teamwork.
The team will meet weekly to train in data visualization practices, discuss progress, plan for the following week and request and give feedback. In addition, each of the three subteams will meet on their own during the week to work on their assigned tasks together.
This project will include an optional summer component in 2025. The summer work will take place over eight weeks in June and July, and one to two students will be needed (with either 40 hours of work per week for one student or 20 hours per week for two students).
Timing
Summer 2025 – Spring 2026
- Summer 2025 (optional): Write up and conduct follow-up experiments to the 2024-2025 project; begin conversations and brainstorming with Explainable AI team contributors
- Fall 2025: Undergo training in data visualization principles; conduct eye-tracking and expressional analysis software training; design experimental stimuli, questionnaires and interviews; finalize experimental design; practice conducting qualitative interviews
- Spring 2026: Conduct experiments with in-person and online participants; analyze data; create blog posts describing each subteam’s results; solicit feedback from data visualization experts; give symposium presentation to local data visualization community; finalize poster for Bass Connections symposium
Crediting
Academic credit available for fall and spring semesters; summer funding available
See earlier related team, Improving Data Visualization With Cognitive Science (2024-2025).