A discussion of visual inference


Graphical methods are commonly used for exploratory data analysis and model checking in, however, graphics are often criticized due to the subjectivity involved. Recently, a protocol that puts graphics into an inferential framework has been developed, allowing analysts to understand the extent to which perceived structure in a plot occurs by chance. This talk will review the development and implementation of this protocol and discuss it's two most compelling applications: estimating the power of competing visual designs; and diagnosing models when asymptotic results are not available.

Invited Statistical Sciences Seminar Series
Los Alamos National Laboratory