The Visual Display of Quantitative Information
by Edward R. Tufte
Graphics Press, Cheshire, Conneticut (1983)
Tufte's book is an extraordinary look at creating what he calls "data graphics": graphical views of data. His emphasis is on creating graphics which are devoid of clutter, densely filled with information, extremely easy to read, and help formulate insights about the data. Tufte values the use of as little ink on the page as possible, while still displaying the data clearly. Since, as he shows, in most cases the excess ink hides the data, these 2 objectives are not contradictory.
The first part (GRAPHICAL PRACTICE), starts with a chapter giving a breathtaking look at some of the best graphics ever produced. The most famous of Tufte's examples is probably Charles Joseph Minard's astounding graph of Napoleon's invasion of Russia, drawn in 1861. Tufte's analysis shows how the graphic plots six different variables (plus some additional information, such as splits of auxiliary troops to protect the flanks).
Then comes a chapter on "Graphical Integrity", which shows how often-used graphics give a misleading view of data. Tufte analyses specific examples, and calculates "Lie Factors" for each; his record is 59.4 (the ratio between the apparent and real values). Some conclusions about what should be avoided in data graphics are presented. The most important of these is that data graphics should never compare areas or volumes, or cause a misleading impression that these should be compared. This is usually done wrong, and is extremely hard to judge correctly by eye.
This part helps create a sense of what good and bad graphics are, and how to tell them apart.
Part II (THEORY OF DATA GRAPHICS) gives a more structured look at what makes good graphics good and bad graphics bad. As everywhere, the emphasis is on using the minimal amount of ink required (but no less). Some of his conclusions may be controversial. For instance, he recommends replacing the box plot, with its bilateral symmetry and wide presence, with just 2 line segments and a dot. But his "redesigns" of traditional figures are always interesting and impressive.
A book about graphics design should be graphically pleasing. Tufte's book is much more than that; it is graphically astounding. Tufte not only wrote the book and drew the graphics, he was also solely responsible for the design (and I believe for Graphics Press, too!).
On a first reading, one notices very little of this, except perhaps that the book is very easy to read. Later, the astonishing facts emerge.
The book has an abundance of figures (almost every page has a figure or a table), but no figure numbers! Even more impressively, the text rarely makes explicit reference to a figure; the figure "just happens" to be where you're reading, so it is immediately apparent which figure Tufte is writing about. When a figure is needed a few pages later, a small version appears again, so you don't have to flip pages.
Some of the figures are drawn from various publications. Conventionally, bibliography for these would appear in a footnote, or in endnotes at the end of the chapter. But these would imply the use of a reference, which Tufte doesn't like. Instead, the reference appears in the margin next to the figure (or sometimes in the text itself, where relevant).
This is exactly what Tufte's book is all about: how to create extremely elegant and informative layouts, which require no training to read and give a real view of the data.
In short, this is one book you MUST read, if only for your enjoyment. It may be a bit difficult to get a hold of, so just steal a copy from your local library (Sylvar can be help explain how to do this without getting caught). Failing that, you can always poke around in the used books shops; I got mine for ~$10 (and you can't borrow it). Maybe even your local library still has a copy which you can borrow...