Statistical Graphics for Exploring Data, Presenting Information, and Understanding Statistical Models
Graphical methods are being increasingly used for exploratory data
analysis. Some of the many graphical tools that are useful in this
setting are scatterplot matrices, nonparametric smoothers, and tree
diagrams. Statistical graphics for presenting information have been
used much longer, but most of the commonly used graphics used in
papers, presentations, and the popular media, such as bar charts and
pie charts, are either poor or misleading in communicating information
to the reader. This short course begins with a
series of graphical horror stories from the scientific and lay press.
Then elements of graphical perception and good graph construction,
many from the writings of Bill Cleveland, are covered. Practical
suggestions for choosing the best chart or graph type, making good and
clear graphics, and formatting are covered. Techniques for
simultaneous presentation of multiple variables are described.
Complex outcome or risk adjustment models are not easily grasped by
non-statisticians. Special graphics such as effect charts and
nomograms can assist physicians and other consumers of statistical
analysis in understanding statistical models and in using them for
obtaining predictions for individual subjects. Examples of model
presentation graphics will be given.
At the close of the short course some graphical marvels from
the literature (especially from Edward Tufte and Howard Wainer) are
presented.
--
FrankHarrell - 04 Feb 2004