The Visual Display of Quantitative Information
Author
Edward R. Tufte, born 1942, American statistician and Professor Emeritus of Political Science, Statistics, and Computer Science at Yale. He took his PhD in political science from Yale in 1968 and taught at Princeton before returning to Yale.
Graphics Press, Cheshire, Connecticut, 1983. 197 pages. Second edition 2001. Tufte self-published the book after publishers declined the production standards he wanted, taking a second mortgage on his house to fund it. He controlled the typography, paper, and printing himself, and the physical object is part of the argument.
The first of four books, followed by Envisioning Information (1990), Visual Explanations (1997), and Beautiful Evidence (2006).
Description
Tufte's subject is the display of measured quantities. His position is that most such displays fail because they spend their ink on decoration, not data.
Graphical integrity. A graphic should represent numbers in direct proportion to the quantities they stand for. Tufte introduces the Lie Factor, defined as the size of the effect shown in the graphic divided by the size of the effect in the data. A value of 1 is honest; his examples include published charts with Lie Factors above 14. The common causes are truncated axes, area used to encode a one-dimensional quantity, and design elements scaled in two or three dimensions when only one is varying.
Data-ink ratio. Tufte divides the ink used to present data by the total ink in the graphic, and argues for maximizing that fraction within reason. The practical instruction is to erase non-data ink and to erase redundant data ink. Applied to a bar chart, this produces bars with the gridlines cut through them rather than drawn over them, or removed entirely.
Chartjunk. Tufte's term for graphical decoration that carries no information: moiré patterns from mechanical shading, heavy grids, and what he calls the duck, a graphic in which the design has swallowed the data entirely. The name comes from the Big Duck, a duck-shaped building on Long Island.
Data density. The number of data points per unit area of the graphic. Tufte's argument is that readers can absorb far more than designers assume, and that small multiples, repeated small graphics varying one dimension, exploit this well.
The book also introduces two forms of Tufte's own devising. The range-frame replaces the standard axis with a line spanning only the actual range of the data, so the frame itself reports the minimum and maximum. The dot-dash plot marks individual observations along the axes, turning the frame into a marginal distribution.
Two historical graphics are well-known from the book. Charles Joseph Minard's 1869 chart of Napoleon's Russian campaign carries six variables, army size, two-dimensional position, direction of travel, date, and temperature during the retreat, in a single flat image, and Tufte calls it the best statistical graphic ever drawn. John Snow's 1854 map of cholera deaths around the Broad Street pump demonstrates a graphic used to reach a conclusion rather than to report one.
Relevance
The data-ink argument is the same problem addressed in Low Resolution Indicators from the opposite direction. Tufte is concerned with dense displays a reader studies; peripheral indicators are sparse displays a person glances at. Both rest on the claim that the designer's decoration competes with the reader's perception, and both treat the removal of material as the primary design operation.
Tufte's essay The Cognitive Style of PowerPoint (2003) extends the argument to presentation software, using the 2003 Columbia accident investigation to show how a slide format's enforced hierarchy and low information density obscured an engineering finding. The Columbia Accident Investigation Board's report cites the problem directly.
Criticism
The empirical status of chartjunk has been contested. Scott Bateman and colleagues, in a 2010 CHI paper, found that participants recalled Nigel Holmes-style illustrated charts better over a two to three-week interval than plain equivalents, with no loss of accuracy in reading values. Tufte's response has been that memorability is not the same as clarity and that the studies test recall rather than comprehension.
A second objection is that Tufte prescribes without testing. The rules are stated as principles supported by examples he selected, and the data-ink ratio in particular has no experimental derivation. Stephen Few and others working in dashboard design have argued for retaining gridlines and reference marks that Tufte's method would delete, on the grounds that comparison tasks need them.
The books are also expensive and are sold primarily by Tufte himself, an arrangement that has drawn comment given the subject.
Related Reading
- Envisioning Information
- Low Resolution Indicators
- The Design of Everyday Things
- Donald Norman
- Information Anxiety 2
- The Power of Maps
- The Natures of Maps: Cartographic Constructions of the Natural World
- Else/Where: Mapping - New Cartographies of Networks and Territories
- Ambient Findability
- Information Architecture for the World Wide Web: Designing Large-Scale Web Sites
- Everything Is Miscellaneous: The Power of the New Digital Disorder
- Designing Interfaces: Patterns for Effective Interaction Design
- Ways of Seeing: Based on the BBC Television Series
