Improving Visual Statistics
Statistics as a field provides powerful methods for dealing with data. We can summarize data, build models, and make inferences. However, statistics can be complex, esoteric, and rely on many assumptions — aspects that make it difficult to communicate statistical information. Visualization offers another set of powerful techniques for data, relying on the human perceptual system to summarize and structure visual information. I believe that these two approaches can operate in harmony, and that particularly we can trust viewers to perform visual analysis that mirrors statistical analysis.
In this work, I explore visual statistics — the interaction between statistical and visual techniques for dealing with data. I present experiments and theoretical work showing that, with careful design, humans have a robust capability to extract and make use of statistical information from visualizations. I also explore the limits of these abilities, and how, in order to support different styles of argumentation, and to overcome certain perceptual and cognitive biases, designers may need to radically alter visualizations. Lastly, I present deployed systems which are mindful of the capabilities and limitations of visual statistics in order to support thoughtful and complicated data analysis of statistical patterns.
Images and movies
BibTex references
@PhdThesis{Cor15, author = "Correll, Michael", title = "Improving Visual Statistics", school = "University of Wisconsin-Madison", month = "aug", year = "2015", address = "1210 W. Dayton, Madison, WI.", url = "http://graphics.cs.wisc.edu/Papers/2015/Cor15" }