I spent Monday at Edward Tufte’s Presenting Data and Information course. I’d received brochures in previous years and was tempted (especially by the books), but didn’t think I could get IBM to send me. This year, though, I could use IBM’s “retraining allowance” to pay for the class (and the books), so I signed up.
Monday morning found me at what used to be the San Jose Hyatt, near the airport, in a very large room, along with about 400 others. Each of us received a box with all four books and a handout with a pre-class reading assignment and the schedule for the day (the only opportunity for questions was during “office hours” at breaks, which was also the opportunity for autographs).
I unpacked my box, peeled the plastic off the books, and dove in. I had just about finished when 10am rolled around, and Tufte took the stage.
And away we went!
There wasn’t an introduction, though he did show a short video from Stephen Malinowski’s’ Music Animation Machine. But very quickly, we dove into the books, beginning with…
Tufte talked about its focus on causality, as shown by the use of linking lines as verbs and patients as nouns, with the annotations on the lines as adjectives (including the use of dotted lines for uncertain connections). He suggested never using a generic linking line — always annotate to show what you’re presenting.
What’s missing from this chart? Quantities. The grey box could have had a useful quantitative measure (number of victims or infection rate) for each location; what’s there is “chartjunk”, especially with the background and bold type, and all it does is repeat the information on the previous linking line.
Remember that the presentation has to convey both the story and your credibility; when you’re watching a presentation, use it to assess the credibility of the presenter. For example, if there is no link to the original data, you might suspect cherry-picking.
Both of these graphics put a huge amount of information into a relatively small space; in both cases, he had us spend time just exploring the graphics on our own rather than talking about them, though he did then point out facts he wanted us to see. He noted that everyone would find a different part of the data interesting.
Design combats information overload by making more data visible. The secret behind the supergraphics is that the design is simple and invisible, but the data is rich. In contrast, you know you’re being fed poor data if there are more colors on the chart than there are datapoints!
Don’t let the means of production control the work
This is a theme that came up a number of times during the day. Originally, text and graphics lived together (for example, Galileo’s drawings of Saturn in line with their description); then changes in technology required separating them (with a prime horrible example being a book with all of the photos on “plate” pages in the middle of the book, nowhere near the discussion), but now we can easily bring them back together, going as far as “sparklines” (which he talked about much later in the day).
Salman Rushdie’s “Stream of History” on pp. 120-121 of Visual Explanations
Sometimes, words are better than pictures. Tufte called his own diagram on p. 121 a weak attempt to illustrate Rushdie’s description.
This leads into another frequent theme: when you are making a presentation at a meeting, he suggests passing out a report and letting people read it during the first third of “your” time, then discussing it (perhaps using charts to illustrate points), rather than dragging people through a Powerpoint presentation. He is not a fan of Powerpoint, to say the least.
This is almost entirely done with words; the sheer size of the bill speaks for itself, but the annotations add impact to specific points. He would use lighter lines to connect the annotation to the item, or possibly a low-saturation colored dot — this is to reduce visual clutter and the “1+1=3” effect that arises when there’s a lot of figure/ground contrast.
Again, he urged us to clean out visual noise and chartjunk and referred us to p. 73 in Visual Explanations for a discussion of the “minimal effective difference” rule. As an example, he suggests using typography to make the structure of a table apparent rather than drawing lines.
Long-term Cancer Survival Rates, pp. 174-176 of Beautiful Evidence
He took the data as originally provided in the table on p. 174 and put it in a richer tabular format on p. 176; that table puts the items in performance order and adds trendlines, both of which assist in making comparisons. He also used Gill Sans as the font (suggesting Trebuchet as an alternative), because it is easy to read. This was one of the many times he used the sports section of the newspaper as an example to follow — there is a lot of data provided in a small space there, and people who are interested have no problem using that data.
Then he took us through the utterly hideous PowerPoint graphs on page 175, which were generated using the default charting options in PowerPoint. His claim is that the slides on that page are all about PowerPoint and not about the data (PowerPoint is not alone in this!), pointing out that people talk about sending around “PowerPoint decks” rather than saying something about the content of the deck. He pointed us at PowerPoint and Military Intelligence on his site for more discussion.
Positive advice: Look for successful high-resolution techniques in the wild and copy them; for example, the sports page.
Advice on making non-fiction presentations
- Find good examples of information design and copy them; don’t engage in amateur design. Steal from the best.
- For “real science” presentations, the ideal model for a handout is an article in Nature. Never, never, never hand out a PowerPoint presentation or use it as the deliverable. If you’re creating a Web asset, model it after an article in Public Library of Science.
- Non-fiction reports have performance data; use tables, and model them after the sports section of a good paper.
- Have a supergraphic (in addition to earlier examples, he likes using a wall chart with smallish type (not a poster)).
- For “non-science” presentations, a good intellectual model is a good newspaper or magazine article, focusing on words, not pictures.
- Demonstrate your credibility by providing detail and sources.
- Use real objects or models if you can — it escapes the Flatland of the screen or page.
Advice on graphs and charts
- Letter codes and legends are bad, especially when they’re one-offs. Put the names of the data items right on the lines or in the areas, not off to the side.
- People come for the content, not the design.
Fundamental Principles of Analytic Design, pp. 136-137 of Beautiful Evidence
Minard’s Map of the French Invasion of Russia, pp. 124ff of Beautiful Evidence
Tufte really likes this map, and has called it the best graphic ever made. It illustrates Tufte’s principles of analytical design, which are:
- Show comparisons (throughout the map)
- Show causality (the temperatures during the retreat caused many deaths)
- Show multivariate data (the map shows six different variables)
- Integrate words, numbers, images, and diagrams
- Include documentation and credentials
- Content counts most of all
The Galileo show
Tufte spent a good bit of time on Galileo; he brought an original printed edition of Galileo’s book and showed it to us (he did not, of course, pass it around!).
Tufte pointed out that Galileo put his name on the document — not that of an anonymous organization. He says that taking ownership of what you write is important.
Galileo put the images of Saturn right in the text, as if they were words.
Galileo’s engravings of sunspots
Galileo tracked sunspots through a month and created an engraving for each day, showing their progress across the sun. One of Tufte’s students scanned in those images and turned them into a movie, which made it much easier to see what was happening — unless you can use animation, it is far easier for a reader to compare items which are adjacent in space than to remember them over time. This led to a brief discussion of…
John Gotti’s acquittal, based on a Small Multiple Chart, pp. 30-31 of Envisioning Information
The chart showed the convictions of the government’s witnesses against Gotti, leading the jury to discount their evidence and acquit Gotti. The original chart was done in Lotus 1-2-3!
Sparklines are word-sized graphics, integrated into text or tables, providing a lot of information in a very small space. They are especially useful when comparisons are needed. He took us through an example using sparklines to show the progress of a baseball season.
How to give a presentation
- Open the meeting with a high-resolution data dump – for example, a report for the audience to read right then (don’t assume people will do pre-reading). It’s much faster for them to read it than for you to drag them through a presentation. A wall chart is another possibility for the data dump, with discussions taking place right at the chart.
- Provide a short (200-word) summary with the problem, its relevance, and your proposed solution.
- Show up early and make sure all is in order; give early arrivers the material.
- Finally, finish early — no one ever says “I wish that meeting had gone longer”.
And that was it.
Was it worth it?
The books are, of course, easily available; Tufte sells them at list price, or you can get them at a discount on Amazon. I don’t know that I would have ever gotten around to ordering them, much less actually opening, all four of the books without the class, and hearing Tufte’s asides was interesting. And I took quite a few notes during the class!
On the other hand, there wasn’t any interaction with Tufte during the class; it could as easily have been a big-screen broadcast (except for the chance to see Real Very Old Books a few feet away).
My bottom line: If you can get someone else to pay for the class, whether that be your employer, your ex-employer, Uncle Sam (as a deduction), or a very good friend giving you a nice gift, go for it. Otherwise, realize that almost all of the content is in the books, and decide whether you’ll get enough added value from seeing and hearing Tufte to make the price worthwhile.