The Visual Display of Quantitative Information

One of the most important books I've read and reread. Especially useful for those doing scientific data analysis, but I recommend it to anyone who has ever looked at a chart or graph (that means everyone).

I read this book for work. It's a pretty interesting text, though somewhat dated. Much of the theory and advice that Tufte gives here survives the more than 20 years since this edition's publication. But, not all of it.

For one, Tufte has rather lost the battle on data graphics being best serviced by artists. I would not argue that artists would be better suited to creating stunning graphics than someone doping around in Word and Excel, but I think the budgets of most of the folks producing data graphics would. The proliferation of easy-access office suites, and even slightly more advanced (but just as easy to access) chart builders, has made the standard format of many of these graphs concrete in our society.

A third edition of this book - or a technical accompaniment - would best focus on how to implement some of these changes within software that is accessible to users. Tufte's description of range frames is interesting - the use of negative space in bar charts to intuit the grid is interesting. There are a variety of things that I would like to take from this - but practically, I do not know how to accomplish them within the grammar of our data graphics life today.

There are a few parts of the book that have either aged poorly or were sour on writing. For one, I think much of Tufte's advice is not very accessible to folks with eyesight concerns, and also lives somewhat divorced from how data graphics are translated and proliferated over time. On pages 124 and 125, Tufte explores adaptation of the range bar and bar plot to consistency with his design principles. The results are illegible. The range plot works alright.

The box plot, however, is adapted by creating variance in line-weights, and then Tufte delights in his second adaptation - offsetting the box one pica above baseline. These are the sorts of design changes that look lovely in a picture book, but are not good for practical data use. For one, one pica (or one pixel on a screen), is not a sufficient offset - it could easily look like an error in printing (especially after numerous copies). This ultimately makes the chart communicate less and less fluently. I think if Tufte had combined the line-weight and offset, this would have accomplished his goal more fluently.

A very interesting reference text, but something that (for a reader today) is best taken as theory rather than practical advice. Unless the user is handcrafting charts by pencil and plane or digitally in photoshop or some such, some of the more innovative practices are perhaps lost to time.

I'm attaching some of my notes as a comment to this review for future reference.

A treasure trove of inspiration for better data visualization!