Graphs in Statistical Analysis: Is the Medium the Message?

Abstract

In an influential article in the February 1973 issue of The American Statistician Frank Anscombe remarked that “[m]ost textbooks on statistical methods, and most statistical computer programs, pay too little attention to graphs.” This general observation no longer holds. Graphs of data are everywhere. Our children learn to draw and use bar charts in kindergarten and boxplots in the elementary grades; most newspapers and magazines regularly use graphical representation of data, and nearly all books on statistical methods use graphs. Graphs can be drawn in statistical packages, in spreadsheet programs, and in stand-alone graphics packages. Statisticians often bemoan the poor construction of the graphs produced, but even when the construction is adequate, graphs can be of little value. In this article, we argue that useful graphs must have a context induced by associated theory, and that a graph without the well-understood statistical context is hardly worth drawing.

Key Words:

• Dimension-reduction subspaces
• Regression
• Regression graphics
• Residual plots
• Scatterplot matrices