Improving Graphic Communication
The MIT Report, March 1986

In September 1981, the Nuclear Regulatory Commission granted the Diablo Canyon nuclear power plant permission to begin testing at one of its two reactor units. The plant, located at San Louis Obisbo within several miles of the San Andreas Fault, was designed with several reinforcing structures to help its cooling system withstand earthquakes. Shortly after the NRC's decision, it was discovered that someone had read the construction diagram backwards, and as a result some of the reinforcing structures had been installed on the wrong side of the reactor building. The error caused substantial delays as workers devised a way to correct the problem.

Most graphics will not have such dramatic consequences if they fail to convey information properly. But according to Steven Pinker, a professor in MIT's Psychology Department and Co-director of the Center for Cognitive Science, graphics too often tend to obscure information rather than to convey it. And as graphics programs and personal computers make creating a graphic as easy as pushing a button, Pinker asserts, the number of bad graphs has drastically increased.

In today's age of information, graphics hold the promise for conveying the meaning of large amounts of data quickly and efficiently. In recent years color has added new dimension to such displays, and motion adds still another. But Pinker warns that these elements also hold great potential for misrepresenting data and confusing the observer unless used with care.

Pinker is one of a growing number of researchers involved in applied cognitive psychology or "cognitive ergonomics," a discipline which studies how humans perceive and interact with external objects. The design of some computer systems, most notably the Apple MacIntosh, have already drawn on the insights of cognitive ergonomics. Pinker is applying the methods of this emerging field to the principles of graphic design.

Transforming data into pictures is a relatively recent invention in human history. One of the first uses of graphical displays of data was during a cholera epidemic in London in the late 1700's. A researcher at that time took a map of the city and put a dot where every death from cholera had occurred. The dots fell into clusters which corresponded to the locations of water pumps, showing that cholera could be spread from household to household by people touching a common pump handle. A list of addresses, from which the map was compiled, could never have provided the information which saved hundreds of lives.

Pinker's interest in graphics was sparked by a request from the U.S. Department of Education to investigate why schoolchildren were not learning to read graphs. The Department funded three years of research involving tests of human subjects. When graphics programs and personal computers made graphic design and comprehension a concern of industry, Pinker found a new audience for his research results and drew on his knowledge of human perception to develop a theory of graphic comprehension.

According to Pinker, bad graphs are those which do not take the inherent characteristics of human visual and cognitive systems into account. The part of the brain which interprets what we see, the visual system, didn't evolve to perceive graphs, he explains. Instead, it evolved to perceive objects, and their relative sizes, locations, shapes and orientations.

Pinker hypothesized that no single type of graph is especially effective or ineffective; it all depends on the message that the reader is supposed to extract from the information. If the graph displays the information as a single geometric pattern designed to be perceived by the visual system as a "gestalt" or configuration, a skilled graph reader will absorb the message rapidly. But if the crucial information is displayed in a set of individual pieces scattered over the graph, the reader will have to scan over the display and consciously deduce the message that the graph designer wanted to convey. Extra effort, and possibly extra errors, will result.

For example, line graphs and bar graphs, even if they display the same data, do not convey the same kind of information to readers easily. Individual values (the price of gold in January, for example) are easier to grasp in bar graphs, while trends (whether the price of gold increased or decreased) are easier to see in line graphs. Pinker has shown that this is because the visual system perceives lines and bars in different ways. Clusters of bars are automatically seen in terms of how large each bar is, so differences in the heights of adjacent bars can only be deduced by consciously comparing them two at a time.

Lines, on the other hand, are seen in terms of their overall orientations. Individual values on them are less noticeable unless measures are taken to call attention to them. Since trends come out as slopes in line graphs and comparative heights in bar graphs, line graphs will have the perceptual advantage in conveying trends. Since single values come out as bar heights in bar graphs and the heights of isolated points in line graphs, bar graphs have the perceptual ad vantage in conveying single values. By considering the geometric patterns that the visual system easily picks out, therefore, one can predict what kind of information each type of graph is best suited for.

In his book, Graphs for People and Computers (co-authored by Harvard Professor of Psychology Stephen Kosslyn), which will be published by Holt, Rinehart and Winston later this year, Pinker discusses different types of graphic format, and explains why some formats do not easily convey some information. He also describes other ways in which good graphics differ from bad ones.

For example, one simple way to make any color graphic easier to see is to make its background color a different brightness from the color of its component figures. The part of the human brain which perceives edges is activated by differences in brightness, not hue. It is difficult to perceive a sharp edge between two different colors unless their brightness is also different.

Regardless of background color, the use of different colors to indicate graduations can make charts and tables harder to understand. Color is a "metathetic" perceptual continuum: as wavelength changes, the human brain does not perceive an increase or decrease within a color continuum; it perceives qualitatively distinct colors. Lightness, on the other hand, is a "prothetic" continuum; it is perceived as values which can be added, and which fall into a natural hierarchy. Lightness is therefore a much better physical attribute to associate with quantitative change than is color.

A more complicated problem arises with charts drawn to represent three dimensions, which are increasing in popularity. Humans subconsciously figure out the size of objects in three dimensions by considering the size of the object's image on the back of the eye ball in the light of other information obtained from the environment about how far away the subject is, including gradients of detail in visual texture, relative velocity, disparities in the images in the two eyes, and linear perspective. A graphic which only uses perspective to indicate how far away an object is does not provide enough information to the brain for the brain to accurately figure out the depth, and hence the real size, of that object. A chart where 3-D is used only to make a picture more interesting can thus mislead the reader by providing false information which interferes with an accurate estimate of the amounts being compared within the chart.

In addition to describing these and other pitfalls for the graph designer, Pinker's book teaches the reader how to decipher graphs. It also reviews some of the new microcomputer graphic software, evaluating such characteristics as ease of use and ease of understanding the sorts of graphics each can produce.

Pinker's work on graphic communication is actually a spinoff from his main line of work: fundamental research in cognitive science. Cognitive science is the study of memory, reasoning, skills, and the other higher operations of the brain. In particular, he is interested in psycholinguistics - how children learn to speak language - and visual cognition - how the brain processes visual information.