Information Visualization
presented by
Edward Tufte

Edward Tufte, the author of the books "The Visual Display of Quantitative Information," "Envisioning Information," and "Visual Explanations", gave a one-day presentation on the concepts behind his works.

He covered four main areas: overall concepts related to information visualization, display of financial data, user interface design, and effective presentations. Each of these topics has a wider scope than their title implies.

• General Concepts - The main problems that need to be addressed by data presentation, and five rules of thumb that should be followed.
• Displaying Financial Data - Ten things to keep in mind when displaying financial information
• User Interface Design - Differences between a good and a bad user interface, with examples.
• How to Give Effective Presentations - some rules and laws that will aid in giving a talk or other public performance.

First he touched on general concepts relating to visualization of data. There are two main problems we have to deal with when representing information, be it on a computer screen, in a presentation, or in a printout. What we find is that the multivariate nature of much data we wish to represent conflicts with the two-dimensional nature of our displays; both computer screens and paper are 2D surfaces, less than optimal for representing data with more than 2 dimensions.

Secondly, we run into problems of resolution. At a later point in the presentation, he humorously described a scale of resolution of display media, ranging from the 'damned overhead projector' at the bottom of the scale, then to the current-day limited resolution computer screens, then printed paper, then film, and finally to the resolution of the real world, as perceived by our eyes. The problem of resolution is particularly acute for computers, with their limited size and number of pixels available.

After dealing with the two main problems, he touched on his 'five laws of representing information.' A 17th century ink drawing by Charles Joseph Minard gives an excellent example of how this can be done. (TVDoQI, p. 41).

1. The presented information should clearly answer the question 'Compared to what?'
   
   
   In C. J. Minard's depiction of the fate of Napoleon's army in Russia during the devastating winter of 1812. We can see see the comparison between the starting size of 422,000 men and the final size of the army, a mere 10,000 men.
   
   
2. The display should show process. The cause and effect should be clear to the viewer.
   
   
   Looking at a detail of C. J. Minard's drawing of Napoleon's army, we can see the cause and effect during the return from Moscow. In the previous picture we could see that the army was still 100,000 men strong when Napoleon left Moscow, and in fact was reinforced by two flanking armies totaling 39,000 men. Here, we can see that Napoleon returned to his starting point with only 10,000 men. Minard has linked the path of the return army to the date and temperature. We can see that while it was a balmy 0 degrees Fahrenheit at Moscow in October when he began the return trip, it had dropped to -20 degrees for the fateful crossing of the Berezina River, where nearly half the army perished. It dropped as low as -30 degrees, never rising above the -20 degree mark for the rest of the trip.
   
   
3. The display must effectively capture multivariate data.
   
   Minard's drawing shows no less than six variables: The size of the army, its location on a two-dimensional surface, the direction of the army's movement, and for the return trip, the date and temperature.
   
   
4. The information should be displayed in a single, unified display. Multiple displays forces the viewer to attempt comparisons while flipping pages, changing screens, or otherwise attempting to remember data from one view while looking at another.
   
   In this case, the six dimensions (size, 2D location, direction, date and temperature) are clearly combined in a single graphic. It is clear and uncluttered, and after a moment of familiarization, the viewer can clearly see the relationship between the variables.
   
   
5. The quality, integrity, and relevance is of the utmost importance.
   
   Charles Minard had the goal in mind of depicting the horrors of war. He did not state so anywhere in the pictoral; the information presented there could more than speak for itself. All of the information he presented was accurate and relevant, and this increased the quality of the presentation manyfold.

The analytic task should define the display. If something is not relevant to the analysis being performed, it should be left out. Something else important is avoiding 'one damned thing after another' syndrome. Galileo, in his 1613 work 'History and Demonstrations Concerning Sunspots and Their Phenomena,' fell prey to this; he had pages and pages of beautiful drawings of the sunspots on the sun, one per page:



It was up to Christopher Scheiner, a contemporary of Galileo's, to use what can be referred to as 'small multiples' to make a coherent picture of the lot.

Use of the 'small multiples' technique of information display has several advantages. First, it gives credibility. A number of displays gives the viewer the idea that the source has lots of information ("this author must know a lot about sunspots, he has lots of pictures of them"). In addition, it takes advantage of the viewer's investment in understanding: once the person viewing has figured out the first of the many duplicated images, there is no additional investment on their part to understand the remainder. They can instead concentrate on the information presented, rather than puzzling out what the picture represents. Another advantage of the small multiples technique of presenting information has to do with comparisons. A person can easily compare a number of images if visible at once; doing a similar comparison when the images are part of a sequence, with only one or two visible at a time is much more difficult.

There are some concepts that are relatively unique to displaying financial data. These can be summarized by the following 10 guidelines to displaying financial data. It is important to note that these can often be applied to other types of data as well, but are of particular interest in the realm of financial information.

1. Displays of financial data want to assess change, i.e. 'compared to what?'
   
   This example is from an analysis of Connecticut Traffic Deaths; it shows that the 'compared to what?' question is of no mean importance.
   
   

   Here we have a display of traffic fatalities, supposedly linked to stricter enforcement by the police. The problem is, we have no context in which to evaluate the data. There are any number of contexts that might be present: The variation could be normal, and the more vigorous enforcement in fact had no effect. The variation could have been a single, atypical instance, in this circumstance, it is unlikely that the intervention of the police had any effect. Or, the enforcement could indeed have had the desired effect, namely to reduce the number of traffic fatalities.

   Now we see the data in context. Note that the 325 or so deaths reported in 1955 is an extrema; this means that it is most likely the number would go down, and it is hard to tell what role the enforcement played.
   
   
   
   Comparisons with other states give a still better context, revealing it was not only Connecticut that enjoyed a decline in traffic fatalities in the year of the crackdown on speeding. Perhaps it was an icy winter. Perhaps indeed the crackdown caused a ripple effect (chuckle). Not likely, however.
   
   
   
   
2. Variability and deviation is important, and should be represented.
   
   
   This weather chart from the New York Times shows 2,220 numbers. We can not only see the overall trend, but can see the variability. This lends credence and enables the viewer to double check against known information.
   
   
3. Often data need adjustment before being displayed. Money vs. time display must somehow reflect the effects of inflation.
   

   At the top we see raw data for total retail sales in the United States from 1960 through 1971, which appears to show a clear upward trend. These raw data are not useful without correction for a number of factors, which we can see in the lines for Holiday, Trading day, and Seasonal variation. Finally, at the very bottom, we see the figures adjusted for inflation. This is the relevant information to display.

   
   
4. Always footnote the information.
   
   Conversely, never trust information that is not footnoted. What correction for inflation was used? Where did the data come from? This is documentation of the figures, and allows the viewer to verify the information if desired.
   
   
5. Financial displays are mostly descriptive. One gentle way to add causal information is though annotation.
   

   Here we see how annotation can clearly add to the display without intruding on the information itself. A good design will allow transparent access to the data and yet provide information not available from the data.

   
   
6. A statement of error is necessary.
   
   Often information trades off timeliness vs. accuracy, or small variations may appear in the analysis. These should be noted.
   
   
7. Follow the New York Times and the Wall Street Journal.
   
   If you are displaying financial information, chances are one of these two publications already does so. Copy them. Not only have they been doing it for years, but viewers of the information will understand much quicker if it is in a familiar form.
   
   Do what the professionals do. Don't try to develop something from scratch; chances are what you're trying to represent has been done by someone else. Other good sources are the Government Statistical Abstract and the Bureau of Justice Statistics' "Report to the Nation."
   
   
8. Build a portfolio of excellent examples, and copy them.
   
   This analogy is present in software and architecture, and is just as valid here as in those fields.
   
   Going with this is use tools that fit the job. Use a spreadsheet to gather numbers. Use a professional graphics tool such as Adobe Illustrator of Corel Draw to create the final graphic; the spreadsheet tools' capabilities are not sophisticated enough.
   
   
9. Understand the issues relating to aesthetics and technique in designing a display.
   
   

   "Graphical elegance is often found in
   simplicity of design and complexity of data."
   
   - Edward Tufte, "The Visual Display of Quantitative Information"

   Attractive displays of statistical information:
   
   
   • have a properly chosen format and design
   • use words, numbers, and drawing together
   • reflect a balance, a proportion, a sense of relevant scale
   • display and accessible complexity of detail
   • often have a narrative quality, a story to tell about the data
   • are drawn in a professional manner, with the technical details of production done with care
   • avoid content-free decoration, including chartjunk (chartjunk is the gratuitous decoration often present in statistical displays, such as grids, cross hatching, and gratuitous decoration).
   
   From "The Visual Display of Quantitative Information"
   
   
10. Small multiples are as useful in financial information as in other types of information display.
    
    
    
    A display such as this allows a much faster comparison, as well as a determination of the context of recent changes. Across the board comparisons are much harder to do with tables of numbers.
    
    

There are two common problems with the user interface of software.

The first common problem is that software often replicates the bureaucracy of whatever organization created it; the organization is inflicting its own world-view and politics on the user. In this situation, the user is subjected to numerous unwanted and undesired screens before getting to where they want to be. Splash screens, introduction screens, and copyright notice screens are all examples of this imposition.

A second problem is that user interface designs often directly reflect the binary nature of computer design, resulting in the dreaded 'menu-tree'. The hapless user will descend down the tree, and then have to traverse up and down repeatedly to move to a desired location or get from one place to another. In addition, such a hierarchy is easy to get lost in, and results in a frustrated "how do I get out of here" response.

The goal is to have a FLAT interface. Forget the goodies, we want content. Compare the following two interfaces, one for an information system at an art museum, the other for a photographer index.

Here we have a screen with flat access to our data. About 10% of the space is taken up with administrative controls, leaving 90% of the space free to show content.


On the other hand, this screen uses only 18% of the space for relevant information, namely the photographers and their work, with 82% of the space taken up by administrative controls or by nothing at all.

The amount of content should always be maximized; measure the number of characters of data versus control, and the area devoted to control buttons and the like compared to the area devoted to data display.

Presentations themselves are a form of information presentation, and while they should not consist solely of charts and other statistical exhibits, many of the qualities of an effective data display are also present in an effective presentation. Making a presentation has a lot in common with teaching as well. There are 13 basic aspects to making an effective presentation:

1. Show up early.
   
   Not only will this give you time in case something unexpected happens, but it will give you a chance to speak to the people individually as they arrive. If there are attendees you wish to sit together (or apart), you'll have an opportunity to place them as desired. Something good is bound to happen.
   
   
2. Keep the audience's attention.
   
   One good way to do this is by giving them a tiny overview. Tell them what the problem is you are addressing, who cares, and what your solution is. This gives them a context in which to evaluate the information you present.
   
   Another somewhat more risky way is the 'stumblebum' approach. You make an intentional but obvious error at the beginning of your presentation, and let the audience catch it. The hope is that the audience will watch intently for future errors, the idea being that you do not make any other errors.
   
   
3. To explain a complicated concept, use PGP (Particular, General, Particular).
   
   Choosing one instance, you explain what the significance of that particular point is. Then you give an overview of what all of the information signifies, then you describe another specific point.
   
   
   
   For example, we can see in this display that the high temperature at the end of February was less than the normal low for this time of year - about 20 degrees below the normal high, and the low was almost 20 degrees colder than the normal low.
   
   Overall, however, we can see that such variation is not unusual, and in fact variations of about 20 degrees outside the norm happen about six times a year. Even with this variation, we can see that the average temperature for the year was pretty much as expected, and that the overall trend followed the normal curve of highs and lows.
   
   December 25th, the low for the year, was another instance where the temperature fell 20 degrees or more outside the norm. In fact, were it not for this particularly cold Christmas day, the previous low would have held the record for the year. It is also interesting to note that the low for the year falls over a month away from the average low for the year.
   
   
4. One law of presentations: ALWAYS give the audience something tangible to take with them.
   
   A piece of paper is something they can take away with them, it gives them something to refer to later. If they don't come away with something tangible, it's almost as if the presentation never happened.
   
   
5. Find out what the audience reads.
   
   Not only does this get some response, but it gives you an idea of what they are interested in, and may let you focus your presentation or avoid explaining information they already understand.
   
   
6. If you're thinking of using an overhead, think again.
   
   And if you're still thinking of using an overhead, read this again.
   
   
7. Audiences are precious. Act that way.
   
   Nothing will turn off an audience faster than an arrogant or patronizing attitude.
   
   
8. Humor is important, but you have to be careful.
   
   Humor is a two-edged sword. Some level of humor will relax the audience and get their attention. Inappropriate humor is worse than no humor at all, and too much humor can turn the presentation into a carnival.
   
   
9. Avoid masculine pronouns.
   
   "So he goes to the menu, and he chooses from his selections, and his next window appears." Mixing singular and plural in speech is formally correct, as in "When the user goes to the menu, and makes a selection, their window will appear." The user/their combination is proper.
   
   
10. Questions are important.
    
    People will often judge your entire presentation on how you answer their question. If you are evasive, people will notice, so if you do not know, say so. Also, it is acceptable to handle long or involved questions offline "Yes, I have that information here, if you would like to come up afterwards, I can show you."
    
    Also, it's important to avoid having your presentation taken over by a persistent questioner. One good way to handle this is with a statement at the beginning along the lines of "I have about 20 minutes of material here. We can get through this, and then I can answer questions at the end." Then, when the interrupting questioner raises his ugly head, you can appeal to the agreement "Well, I have about 15 minutes of material here..."
    
    This will avoid the issue of being led by questions from a particular attendee. While this is also rather poor on the part of the questioner, since a presentation is not a place to show your knowledge, it also indicates that the presenter has let things get out of hand.
    
    On the other hand, if you have a question and answer period and don't get any responses, consider having a plant to get things moving.
    
    
11. If you believe something, make sure your audience knows you believe it.
    
    Not only will it give you credibility, but it will avoid the opposite problem, where your audience does not think you believe what you are presenting.
    
    
12. Finish early.
    
    As with arriving early, this allows something good to happen. Perhaps you can answer some questions. Perhaps someone who did not want to speak up in front of everyone will approach you privately. At the very least, keep in mind all those times you walked away from a presentation going "Gee, I wish he went on for another half an hour."
    
    
13. Practice, practice, practice.
    
    Not only will you sound more relaxed and confident, but it will allow you to pay attention to the audience and their responses, rather than to your material.