My Day With Edward Tufte

Friday, June 25, 2004

I spent most of Thursday in a seminar given by Edward Tufte, the author of several seminal books in the area of data analysis and presentation (with another, Beautiful Evidence, in the works). I enjoyed it thoroughly and here's what I was able to capture:

Tufte's Grand Principles

The goal of good information design is to minimize the amount of time spent figuring out the design of the presentation of information (or even admiring it) and maximize the time spent reasoning about the information. Good designs should be as nearly invisible as possible.

Tufte: "The point of an information display is to assist thinking. Good design is clear thinking made visible. Bad design is stupidity in action. Chart junk is a good sign of statistical stupidity."

My favorite Tufte term is what he calls designers brought in to make boring numbers look interesting: "chartoonists."

Ed's Grand Principles of Analytical Design

1. Show comparisons (this is big theme of Tufte's)
2. Show causality
3. Show more than 2 variables, as the world we're trying to understand is multivariate
4. Integrate word, number and image (this is another big theme)
5. Say from where the data came
6. The value of an information presentation comes from the quality, the relevance and the integrity of the content, i.e. good design won't fix bad content.
7. Use small multiples, e.g. showing multiple pictures of sunspots on the sun over time
8. Show and embed scales of measurement because it's an essential part of the context
9. Annotate everything and if the audience reads ahead great, they're reading your material
10. Use proven design templates (like those in Tufte's books)

Tufte: "Don't let the marketing people corrupt your presentations by eliminating detail. Chances are, the people looking at the information know more than the marketing people. People haven't suddenly gotten stupid just because they come to hear you talk. If people can read the financial tables and the sports tables from the paper, they can get read data from you." (The last point assumes a good design.)

In my own work, I often remove detail for clarity, but I'm very careful to leave enough details to keep context. I admit, it's a fine like to walk and hard to get right.

As a way to pack more data into a small space (and in an attempt to secure his immortality, imo), Tufte was inspired by Galileo's work to create a new kind of graphic meant to be integrated into a sentence as yet another word. He calls them "sparklines" (the sparkline is the graphic before each word):

Sparklines pack a lot of information into a small space and are a cool alternative to graphics that break the flow of thought and need to be called out with a phrase like "as seen in Figure 3." I have a mind to implement them in Avalon when I get a chance.

90% Content

Tufte hates most kinds of user interfaces because they spend too much of the already low resolution screen on "fluff," e.g. menu bars, captions, navigation, etc, and too little on the content itself. As an example, he showed a screenshot of a photo display app that used only 18% of the screen for the actual photos. He wants nearly invisible operating systems and apps and 90% of the screen used for the content. He's a big fan of Google News as a way to pack in a lot of information and he's a big fan of scrolling. For web sites, Tufte recommends only a small set of links at the top of each page for navigation and nothing else. His own site is like this and I have to admit, I like it.

Design For High Resolution

Ed hates low resolution. He wants everyone to design for high resolution to be viewed at 22 inches so you can pack a lot of information in, showing comparisons and causality. He hates PowerPoint because it's made to be viewed from 22 feet and doesn't provide a lot of room for showing dense data, comparisons or causality. It's far easier for humans to make judgments when they can see everything at once instead of sequenced one slide after another (or "one damn thing after another," as Tufte says : ). In general, he believes everything should be printed and that PowerPoint should only be used for showing pictures to an art class. I see his point, but ironically, when he wrote a paper on the problem of PowerPoint to be read at 22", after 3 pages of him blaming PowerPoint for sloppy thinking, I put it down in disgust. If he had had his thoughts put together into PowerPoint-style format, I could have skipped ahead to find his real point, i.e. low resolution makes it hard to do serious analysis.

As a point for Tufte, when producing this summary of my day listening to him, I'm mixing prose, images and references, but I'm not producing it in PowerPoint. According to Tufte, 4 pages of prose is equivalent to between 50-250 PowerPoint slides and I have to admit, it would be hard to convey what I want to say about him in a PowerPoint deck without using a lot of slides and throwing away the subtleties. In general, I believe that PowerPoint is best for reminding the presenter what they want to say and is very poor for the audience. Instead of a slide presentation, Tufte recommends providing the audience 4 pages of prose, which they read 3-5x as fast as you could say it, and then having a conversation. Interestingly, he allowed no conversation during his own presentation but instead talked to various parts of his books and hand-outs.

 On the other hand, has anyone even read this far? Would you have gotten farther if I'd have summarized everything in slides, skipping ahead to the juicy bits when you got bored?

Giving Presentations

Tufte has a great deal of advice about how to prepare for and give presentations. To prepare, you must have great content (no amount of design or practice will fix bad content) and spend a great deal of time practicing (especially important if you're not using PPT to remind you what to say). When presenting, follow these rules (these aren't all of my rules, but I agree with them):

1. Show up early to avoid problems and greet everyone at the door to learn their names, introduce yourself and begin advancing your cause
2. Start talks with The Problem, Why They Should Care and The Solution
3. Never start your talk with an apology, as it just make you look pitiful
4. Use Particular/General/Particular when presenting information, e.g. start with a particular example of something, generalize it and then show another particular example that shows the same general characteristics (I use this technique all the time and it's very effective)
5. Use first person singular only to express opinion, i.e. this isn't a talk about you but about a serious topic
6. Give everyone at least one piece of paper (Ed prefers a single 11" x 17" folded in half with 4 sides of content)
7. Knowing your content and respecting your audience is more important than knowing your audience if by "knowing" them, you "dumb down" your talk for them
8. Avoid project slides (aka PowerPoint or the equivalent)
9. Use humor as appropriate, e.g. hyperbole to make a point, but don't tell jokes. "Don't alienate your audience because of a joke. Alienate your audience on the merits on your content!"
10. Avoid the use of "he" or "she," e.g. "The user, he takes his mouse and moves it to his menu bar where he makes his manly choice." Use "they" instead.
11. Make sure folks know you believe what you're saying by getting out from behind the podium, getting close to people and use appropriate gestures. If you don't believe what you're saying, they're not going to believe it.
12. Finish early and something good will happen. No one will ever look at their colleague, commenting on how they wish you had gone on for another 20+ minutes.

I think that Ed has enough good things to say about how to approach the presentation and analysis of data that I would recommend that pretty much anyone involved in the presentation and analysis of information take the time and attend his day-long seminar.

Discuss