Description:

Department of Statistics Winter 2020 Seminar Series

Talk Title: Beyond Visualization: Improving Reasoning Under Uncertainty From Data

Speaker: Jessica Hullman, Computer Science & Engineering and Medill School of Journalism, Northwestern University 

Time: 11:00am

Abstract: Visualizations and other representations have become ubiquitous for communicating data-driven estimates in everyday decision making (around weather, finance, health, etc.) as well as science and policy. However, while guidelines dictate how to design visual encodings and interactions to support pattern finding, far less is known about how to develop visualizations that support reasoning about data under uncertainty, especially among non-scientists. Visualization authors and system developers must overcome difficult challenges, from users' aversion to and difficulties interpreting probability, to technical challenges in depicting uncertainty for complex data and encodings, to their own beliefs that transparent communication of uncertainty is irrelevant to users or likely to reduce trust.

I'll describe a vision for a more "uncertainty-aware" paradigm for designing interactive visualizations and data interfaces. Addressing representation challenges, I'll show how visualizing uncertainty via sample-oriented representations of probability can provide a general design pattern for communicating uncertainty even when visual encodings are already complex and improve inferences and decisions among non-experts. I'll describe how Bayesian approaches to interaction design and evaluation can simultaneously encourage better integration of user's prior knowledge in interacting with new data and deeper understanding of how visualizations "work" among authors and researchers. I'll conclude with thoughts on how these techniques can be applied more broadly to designing and evaluating human interactions with predictive models.

Location: Department of Statistics, Room B02, 2006 Sheridan Rd, Evanston 60208