Building effective uncertainty visualizations with Stan/brms, tidybayes, and ggdist
Bayesian modeling and effective uncertainty visualization are a natural pair: Bayesian modeling techniques produce samples from joint probability distributions that describe the uncertainty in estimates and predictions, and a growing body of research suggests that sampling-based visualizations of uncertainty can lead to better estimates and better decisions from users. In this talk I will tour a variety of modern uncertainty visualization techniques, discussing systematic principles for matching uncertainty encodings to the communication and decision goals of an uncertainty visualization, grounded in perceptual and cognitive aspects of uncertainty visualization understanding. Along the way, I will demonstrate programming techniques for easily constructing complex uncertainty visualizations using Stan/brms with the tidybayes (http://mjskay.github.io/tidybayes/) and ggdist (https://mjskay.github.io/ggdist/) R packages, which are designed specifically for creating uncertainty visualizations. Examples will be drawn from existing vignettes (e.g. and my repository of uncertainty visualization examples (https://github.com/mjskay/uncertainty-examples), and the uncertainty visualization literature (e.g. medical risk communication, hurricane path prediction, and real-time transit arrival prediction). The goal of the talk will be to give the audience some grounding in the basic principles of effective uncertainty visualization design, and then demonstrate how these principles can be applied to visualizing uncertainty from Stan models using APIs expressly designed for that purpose.
Matthew Kay is an Assistant Professor at the University of Michigan School of Information working in human-computer interaction and information visualization. His research areas include uncertainty visualization and the design of human-centered tools for data analysis. He is intrigued by domains where complex information, like uncertainty, must be communicated to broad audiences, as in health risks, transit prediction, or weather forecasting. He co-directs the Midwest Uncertainty Collective (http://mucollective.co) with Jessica Hullman, and is the author of the tidybayes (https://mjskay.github.io/tidybayes/) and ggdist (https://mjskay.github.io/ggdist/) R packages for visualizing Bayesian models and uncertainty.