Edward Tufte: Presenting Data and Information

Class given by Edward Tufte June 19, 2001, Palo Alto, CA

Below are my notes from this one-day class. Edward Tufte is one of the few very "rich" presenters I've encountered before -- there's no unnecessary repetition of content or other filler. I found myself really mentally involved with the class throughout the day. Tufte is a really gifted teacher/presenter; I left the class full of enthusiasm and excitement for the material he covered. But note: there's no way that my notes could do justice to this class, and they are done mainly for my own later reference.

These notes are over ten years old as I repost them to this site. There are very likely many broken links. I don't have plans to check and edit them at this time, but suggest you use Internet Archive's Wayback Machine to find archived copies.

There are two central issues in the general problem of the display of information:

  • The need to display three or more dimensions of information on two-dimensional displays (escaping flatland) -- the really interesting information is almost always multivariate.

  • The available resolutions for displays of information. This is evident in the fact that all display methods (and their improvements) tend to be evaluated on the basis of resolution.

General principles

During the class, Tufte showed us books from his own collection. My favorite was a first edition (published in 1613) of "History and Demonstrations Concerning Sunspots and Their Phenomena" by Galileo Galilei, who listed his name as "Galileo Galilei Linceo" to honor the "Linceo" (Lynx) society of which he was member. The title page includes the society's logo, a lynx surrounded by a crown of thorns.

It's difficult not to evaluate Tufte's class on the basis of his own teachings...not only were the books shown very interesting, but it really drove home the point that he is an expert in this field, because it illustrated the lengths he has gone to, enthusiastically, to study it. How to establish credibility was a recurring sub-theme during the class.

Should I also mention that the entire process of finding the conference area, registering for the class, getting an autograph, etc. was extremely smooth and enhanced by simple, direct signage?
 

"This is where God wants footnotes." - Tufte, joking while pointing out that his books feature side-notes, not footnotes that require one to jump to other parts of pages or (worse) other pages.

Information design should always be credited. Not only should the work be credited to give credit where it is due, but it should be done as documentation and to add to the credibility of the work; named work implies that someone stands behind it.

Use direct labeling; legends or keys usually force the reader to learn a system instead of studying the information they need.

There are two main ways to escape flatland: you either have to build a model, or you have to be very very smart. As an example of the very very smart, Tufte used the famous Napoleon's March chart to go on to illustrate the "5 grand principles of information design":

  • Enforce visual comparisons -- the width of the tan and black lines gives you an immediate comparison of the size of Napoleon's army at different times during the march.

  • Show causality -- the map shows the temperature records and some geographic locations that shows that the weather and terrain defeated Napoleon as much as his opponents.

  • Try to show multivariate data -- more than two dimensions. Napoleon's March shows six: army size, location (in 2 dimensions), direction, time, and temperature.

  • Completely integrate words, numbers and images -- don't make the user work to learn your "system". Related: the organization's internal bureaucracy shouldn't surface in a design meant for those outside the organization. (As an example of how to do it right, Galileo's use of tiny images drawn right between words that describe his observations -- see p. 120 of Envisioning Information).

  • The design should be content-driven -- Napoleon's March was designed as an anti-war poster...the designer was passionate about the information being presented. The point of the poster wasn't the design, it was the information.

Good information design can never salvage poor content. Good content must have these characteristics:

  • Quality (can't overcome bad data)

  • Relevance (has to be a need for presenting it)

  • Integrity

Additional principles of information design:

  • Whenever possible, show comparisons adjacent in spaces, not stacked in time.This is where resolution is most limiting; for example, computer monitors cannot really show much in the typical "eye span". Information presented on computers is often "stacked", causing users to ask the frequent question, "Where am I?".

  • Use small multiples -- the small multiple is a useful way of showing time, by showing several small images and how they are changing through time, within the eye span. This convention takes advantage of the investment the user has already made in learning other designs. Also, by showing a mastery of detail, using small multiples also often adds credibility. (Galileo was able to put 38 suns within the eyespan to present his observations).

The "meta-principle" of information design (or, "how are these principles derived?"):

Good information design is clear thinking made visible.

(unfortunately, the converse is also true).

Tufte mentioned that the next book, to be completed in a couple of years, will probably have a title something like "Beautiful Evidence".

The principles of information design tend to correlate with the established principles of analytical thinking. (see page 53 of "Visual Explanations" for a discussion of these principles).

"Chart junk" -- unneeded extras to be cute -- often indicates stupidity in the underlying statistical data.

Important: Always ask "What is the thinking task that this display is supposed to help with."

Scale: people can always differentiate between differing angles when they are close to 45 degrees (rather than being very steep or shallow). See p. 25 of Visual Explanations -- sunspot cycles. Determining scale needs to take into account what the thinking task is as well as the integrity of the data.

"Magic" is a very interesting information design problem, and it can be useful to think about since the point behind magic tricks is to hide information -- a systematic corruption of information design (magic=disinformation). To reveal a magic trick in an information design, five dimensions must be shown: the usual 3 dimensions of space, time, and what is hidden from the audience.

It is the content person's constant responsibility to guard the integrity of the data.

Presenting financial data

To be useful, financial data needs to show an assessment of change (this is the thinking question in this case).

Using empty vertical space to show a zero point that never occurs does not help, (p. 74, Visual Display of Quantitative Information). Re-measuring will usually move data closer to the average ("regression toward the mean) -- more horizontal space can actually show more useful information in many cases.

The content should be rich, not the design. -- the variability of data around the mean/average should typically be the focus of it.

Standardization of data: any financial data shown over time for over 1-2 years must adjust for inflation to be accurate. However, seasonal influences can often be appropriately adjusted out.

Do not trust an information display if there are not notes, references, documentation, sourcing. The source of the data must be documented for the display to be credible (p. 38, Visual Display of Quantitative Information).

Bring in explanation of causality via annotation (p. 56-57, Envisioning Information - using the idea of gesture/pointing, small but effective).

"Don't get it original -- get it right." -- Tufte on copying the techniques of effective displays

Look for standard kinds of financial data, and do what they do -- these are effective techniques (for financial data specifically, see the NY Times or Morningstar). Use the conventional great designs -- the effort into finding out what works has already been made. Learn from it.

"Spark lines" -- very small displays of a stock's performance over a year -- puts a great deal of dense information in a very small space. This amount of information is only useful when it is made visual. It also helps overcome the bias inherent towards "recent" data.

Web site or kiosk design

Tufte commented that every time he hears something like, "We need a metaphor" in relation to a web design, his heart sinks.

(See Visual Explanations, p. 146; in general, this is the book that is probably most useful for web designers)

Metaphors are usually not effective (eg. trying to replicate the structure or visual design of a book on a web site). Often most used, and most ineffective, is the metaphor of the binary structure of computer data itself. (Get halfway, make a choice, get halfway there again, make another choice -- binary splits)

One of the biggest problems in web design is that "turf" between the different sub-divisions of an organization is revealed (and is irrelevant and distracting) to the user.

Tufte does not believe in deeply hierarchical web displays and believes as much as possible should be shown immediately -- eg. Excite has an effective display with 162 links, yet does not appear over-crowded.

"It is essential to let the user know the scope of the domain of information right away."

This was the single point that most struck home with me in terms of ways to improve oreillynet.com and related sites. Users need a way of knowing how much information is available so they are able to decide that this is a site they want to explore more. (Tufte uses "domain" in its traditional sense here, not as in "domain name.")
 

Tufte mentioned the point others have made: that the computer industry is only one of two types of businesses that refer to their customers as "users". Tufte suggests patrons as an alternative. (of course, he does not suggest this as a term to use to describe advertisers...in general, advertisers do not exist in Tufte's world -- which doesn't make his message any less effective, it's just not what he covers. -t)
 

Tufte discussed some of the book content, specifically the Swiss Alps cartography section. He admitted that the cartographer had great material to work with, "It's hard to do a brilliant topographic map of Kansas."
 

While waiting for people to return from break, Tufte mentioned the the deep discounts on his books being sold in the lobby, and suggested that they'd make thoughtful "laid-off" gifts.

It is essential to let the user know the scope of the domain of information right away.

Content is the only real thing that makes a site unique (and rarely is the content the design). Content must be delivered to be successful.

Tufte claimed that one study of web usage indicated that the time of average download exceed the time of the average visit -- users leave before the entire page is presented. He also claimed that there was a study showing that, on average, people spend more time waiting for Windows to load than they do making love. (We did not get a visual display of these statistics. -t)

The ten most widely hit sites have over 100 links on their opening page.

"Clutter" is a problem of bad design...you can't show enough information on a computer screen to have clutter unless it's something you can solve through design. (I would note that this is assuming the designer is allowed to solve it...advertising and branding can be a form of clutter that can be difficult to deal with. -t)

One type of way to evaluate: determine what proportion of the screen is dedicated to "real" content. If 80% or above, you're probably on the right track.

"If the design is the first thing visitors notice about a web site, you probably have a problem."

Search engines

(This discussion happened while waiting for people to return from break)

The best meta-search engine will only turn up about 40% of the related material available on the web. Domain search engines (domain=subject-specific, not domain name) are a better bet for finding information.

Tufte used SETI@home as an example of a site that is well-designed from a sociological perspective as well as for the presentation of technical data.

Recommended web sites:

Information displays and decision making

The annotated medical chart from Visual Explanations, p. 110-111 was discussed. Tufte also mentioned the problems of digitizing data and privacy (giving an example that Tonya Harding was admitted to a Portland hospital, and how records showed hospital employees were viewing the medical records unnecessarily because of her celebrity status).

Also discussed here was the very interesting Challenger space shuttle story (Visual Explanations, p. 38). Essentially, the engineers and rocket manufacturers knew that disaster was probable and tried to convince NASA to cancel the launch. Tufte approaches it from the perspective of information design, and shows how much more compelling the data was when presented a different way.

Demands to make of a presenter:

  • Show me causality

  • Show me all of your data (omissions are suspect; Tufte says that if a presenter makes the claim that "it would be too much to print out", your response should be "that's what 6pt type is for, let's see it."

  • Demand the right analysis (very difficult to do in a group setting). View everything as selective evidence out of thousands of possibilities. Tufte recommends getting physically away from the presentation setting to take a break to ask, "What do I really need to know?". Hard, blunt, direct questions may need to be asked for important decisions.

Principles for making presentations

Show up early...you can fix any problems with the presentation setting, and it's also a chance to meet audience members (they will often leave immediately after).

Never start the presentation by apologizing.

Early in the presentation, give an overview of what the problem is, who cares about it, and what the solution is.

Stumble-bum technique is very daring, but some have used with success -- eg. high school math teacher presenting to mathematics professionals, makes intentional mistake during first proof -- audience follows carefully thru rest of presentation, hoping to catch other mistakes.

See how long you can go without using first person language.

Particular - General - Particular (PGP - see Visual Explanations p. 68-71) -- when explaining an information display, first point out something specific the audience can learn from it; then give a general overview of what is shown, and follow it with another (different) specific example.

The "Law" of presenting -- Always give everyone at least once piece of paper to show some responsibility for the data (implies that you stand behind it -- they can refer to and ask about it later). Paper is the highest resolution display device there is. (Tufte describes the evening news -- 21 minutes that manages to get about halfway thru only the first section of the New York Times. In 21 minutes, most people can pretty much read thru the entire New York Times...or 2 USA Todays...or many SF Chronicles...). Also: "A paper record tells your audience that you are serious, responsible, exact, credible." (quote from his web site).

Ask what your audience already reads -- what do they already know, what visual displays are they already familiar with? Use similar materials.

Tufte went thru the first couple of slides of Peter Norvig's Gettysburg Power Point Presentation as if he were a typical presenter giving it -- hilarious and instructive.

Assume that the audience is worthy of your respect and that they are intelligent. Don't insult their intelligence by "dumbing down".

Use humor to reinforce points -- but it must be very specific, not just a random joke thrown out. And, of course, avoid offensive humor. "Alienate the audience on the merits of your content, not by an inappropriate joke." :-)

Don't use male pronouns exclusively...per the Oxford English Dictionary, plural pronouns are fine to use and he recommends this.

Affect is carried by body language -- let people see your honest enthusiasm...don't hide behind podiums, etc.

Finish early -- you'll be an instant hit. Tufte said that no one ever left a presentation thinking, "Gee, I wish he would have gone on for another 20 minutes."

Practice, rehearse -- this is difficult, requires effort...so what. Use video tape and watch for distracting gestures and filler talk (uh, uhm)

The best way to improve the presentation: improve the content

Closing

Animation presentations were shown:

  • The "severe storm/severe grid" animation redesign (available on the web here; another version here; and a paper about the process can be found here -- the PDF link at the top right corner of the page is probably the easiest version to deal with)

  • The Viz-O-Matic: The Dangers of Glitziness and Other Visualization Faux Pas (hilarious animation, vailable on the web as a Windows Media file here)

  • The Music Animation Machine by Stephen Malinowski

Thanks to Patrick Geille for the updated links to the storm and Viz-O-Matic animations!

Tufte closed the class with a reading from "Haroun and the Sea of Stories" by Salman Rushdie (appears on page 120 of Visual Explanations).

I found myself wishing he'd gone on for at least another 20 minutes.


Tufte's Suggested Reading List

Fundamentals

Interaction of Color

The Elements of Typographic Style

Mastering the Techniques of Teaching (good for information on making in-person presentations)

Understanding Comics

Light and Color in the Outdoors (this one recommended because of the quality of the photos, though he said there was a Dover edition that is less expensive that would also be ok).

The Design of Everyday Things

Elements of Style (he recommends reading the last chapter on style once a year)

Maps and Civilization

And of course the three classics by Edward Tufte himself:

Advanced readings

Visual Thinking

The Elements of Graphing Data

Visualizing Data

Cartographic Relief Presentation -- by Euard Imhoff, responsible for the Swiss mountain maps discussed in Tufte books; recommended for "escaping flatland" problems. (Not listed on amazon.com).

Designing the User Interface

About Face

Philip and Alex's Guide to Web Publishing (especially for issues regarding web sites with massive amounts of content. Also available in online format.

Web Style Guide

The Humane Interface: New Directions for Designing Interactive Systems

The Clock of the Long Now: Time and Responsibility (especially for two chapters on the problem of digital archivability)

Strategic Stories: How 3M Is Rewriting Business Planning (article from Harvard Business Review, how "narratives replace the dreaded bullet list" and the problems of the typical Power Point presentations for strategic planning)