… is a serious issue. That’s what Edward. R. Tufte says in is book The Visual Display of Quantitative Information.
Cover of Tufte's book. The image is an old French train schedule considered by Tufte has one of the best information visualization ever.
I’ve came to read this book after reading D. MacKay’s books (1–2), since I just was marvelled by their pretty uncommon presentation.
His books have a very large margin, to provide a lot of space for illustration or remarks. Moreover, his way of graphically visualizing things are always very inspired.
An neat visualization of "the probability distribution over the 27x27 possible bigrams xy in an English language document" – from D. MacKay "Information theory…"
So I decided to follow his example when I’d be writing my Ph.D. dissertation. That’s how I stumbled on the tufte-latex package, and discovered the work of Tufte. I discovered that a dissertation is not a book eventually, and that I should keep the good ol’ format.
Still, I was on and tried to learn more about data visualization.The idea is that it can be really hard to explain a situation with a graph, but it’s still the best way – when you really have quantitative information
Tufte gives a set of rule (such as the ink-to-data ratio) to present things clearly. He gives alot of historical and contemporary example, showing that most of the time poor graphics are just dishonest e.g.,
The worst chart ever according to Tufte. There really are only four data points
- trying to show a difference in scale between two object : for a double in quantity, double the size, but quadruple the area.
- if you have almost no information, use interpolation and lots of colors – your boss won’t notice how lame your result is.
- use optical illusions, to provoke confusion
- do not show the baseline
- make pie chart – especially if you make them scattered or 3D style
- (see here for more points)
Besides, the book is quite interesting, and quite beautiful : Taschen should have edited this book, not the crappy “Information Graphics“!