It feels a bit strange to discuss a book that you probably have heard about already if you cared about the topic, and if you haven’t heard of it, you might not care anyway. Still, I did not know about Edward Tufte until I started grad school, when I read a couple of his books and his short demolition of Powerpoint, and I imagine there’s some use in spreading the word a bit more.
Some months ago, Tufte published his new book, “Beautiful Evidence”. He also offered a package of his four books for sale at a slight discount, and I couldn’t resist the temptation, went ahead and bought the bundle. I’ll be talking about the books as I read them.
“The Visual Display of Quantitative Information”, which deals precisely with what its title suggests, has a huge reputation behind it. It’s been labelled “a visual Strunk and White” and a “timeless classic”. With some caveats, I think the praise is well deserved. Tufte has an unrelenting clinical eye for quality in graphics, and throughout the book he shares his judgments with the reader, explaining what is it that makes Minard’s graphic of Napoleon’s Russian campaign so effective, and what makes so many of our newspapers’ graphics so lame. His guided tour through the history of displays of quantitative information is entertaining and tremendously insightful.
I’ve heard people accuse Tufte of vagueness in his advice, of basically arguing that to create appealing displays you need to (a) become a genius like himself, and (b) then everything else will follow. That might be a fair criticism in other books of his –I’ll get there eventually-, but not this one. Here you get advice to choose, for example, between tables and charts, guidelines such as “Show data variation, not design variation”, and “The number of information-carrying (variable) dimensions depicted should not exceed the number of dimensions in the data”, which after being understood could easily be applied by us mortals.
I’m actually worried of the opposite: Tufte here has a set of principles that he follows until their last consequences, and I’m not sure this is always advisable. I’m particularly thinking of his “erase non-data ink” principle. Essentially, the advice boils down to “eliminate everything in a graphic that does not convey data”; and for most graphics that’s a desperately needed advice. But it seems you can go too far in this direction too. For example, Tufte transforms “range bars”, which show the four quartiles of a distribution, emphasizing the middle two and the median:
into “quartile plots”:
that eliminate non-data ink, but eliminate as well the ease to locate the median and the perception of the bulk at the middle of the distribution. Waste of ink should not be our priority – frankness and waste of human attention should be, and we will address this by maximizing comprehensibility, not necessarily by eliminating non-data ink.
The book feels slightly dated for one reason. In these past few years, the capabilities of devices for printing and displaying information that are available for the general public have increased dramatically. We can now use thinner lines and subtler colours to create graphics that, only a decade ago, required professional equipment, blunter lines, and sight-damaging primary colours. Whether we do use the capabilities that we now have available, however, is up to us.
It is actually disheartening, considering the exposure this book has had over the years, to see how little our commercial packages have progressed in following Tufte’s guidelines. I have a personal grudge against Microsoft Excel in this regard. Excel must be the most powerful enabler of graphic disasters in the world. Most people don’t have the time, or the dedication, or the skill, to improve Excel’s default graphic settings. So even the simplest data tables become hideous charts with a couple of clicks:
And if we’re feeling creative, Excel kindly allows even more terrible beasts with the same number of clicks: 3-D bar charts! 3-D pie-charts!
Tufte rightly calls these “chartjunk”, and we are being fed this junk constantly. If you want to help stop this, you should definitely check out this book and follow his advice. The price tag is slightly high, but it’s worth it.
Catenary 
It is very interesting that in Project Management class we studied the same Minard’s graphic of Napoleon’s Russian campaign as an example of the best graphic description ever made and how it deals with about six variables (longitude, latitude, size, time, direction, temperature, altitude) in a simple two-dimensional model. Indeed, eliminating non-data ink helps also to avoid “editorializing” a chart as it help to show the naked data. One of the main failures in data analysis via charts is not taking into account the scale, second only to not use logarithm scales instead of linear ones when appropiate…
Minard’s is really an amazing graph. It’s a technical achievement for the reasons you point out, but it’s also so powerful because it tells such a tragic story with an elegance that text, perhaps, couldn’t match. Seeing the dwindling line of *people*, the temperatures they were forced to march on, and that disastrous river crossing on the way back gives me the creeps.
“Editorializing” charts have one of the most entertaining chapters in Tufte’s book, where he attacks their lack of integrity.
And about log scales -I know sometimes they are appropriate, but I must say I find them very unintuitive too. Their designers must be especially careful not to mislead readers with them.