Graham Wills, SPSS Inc

Robust Graphics and Graphical Robustness

Popular statistical graphical packages often have a wealth of
features for displaying different varieties of data. For example,
a scatterplot can be adorned with a trend line, can show clusters
as ellipses and so on. For real-life data sets, and especially for
large "warehouse" data sets, naive choices of statistical procedure
are greatly to be preferred. On the other side of the coin some
innovative types of graphical display need robust procedures to
optimize for appearance or performance. In this talk examples of
the two-way street between graphics and robustness will be given
and discussed.