Science briefs: The Art and Psychological Science of Graphical Communication


Catherine M. Carswell, PhD
Graphs play an increasingly prominent role in the display of quantitative information. They have steadily found their way from census records, scientific journals, and technical reports into the popular press. Graphs have become commonplace in TV commercials and have even become an important adjunct to presidential campaign speeches.

Just as the use of graphs has moved from the technical realm into the mass media, the creation of graphs is no longer limited to technical artists and those with statistical training. Instead, anyone with a microcomputer and appropriate software can produce a graphic in minutes.

The variety of graphs has also grown rapidly in the last few decades. We have moved far beyond the old 'standbys'--the bar charts, line graphs, and pie charts--to exotic multidimensional formats that look like trees, castles, and even faces. In addition, even standard formats have been treated to a variety of purely decorative elaborations. For example, with a click of a mouse, one can turn a humble line graph into a cliff-like perspective image.

As judged by the number of 'how to' books for designing data graphics, some designers have clearly recognized the need for standards of graphic production or, at the least, for some restraint in graphic design. However, adequate guidelines or standards for graphical practice presupposes knowledge about what makes a data graphic truly effective. Unfortunately, an understanding of how we process the information in graphs has lagged far behind our use of them.

Design Conventions

It has only been in the last decade that a substantial amount of psychological research has focused on how people actually respond to different graph designs. Although there was a scattering of earlier research conducted by statisticians and educators, most of the suggestions found in today's graphic style manuals are the result of expert opinion and professional traditions. Some of these conventions evolved because of the difficulty of producing specific visual effects rather than because the intended audience might have difficulty understanding information presented in a particular way.

Still, many of these design conventions may ultimately prove to be very useful. However, some have been challenged by psychological researchers who argue that they ignore basic principles of visual perception. This has led to direct tests of the conventions, a few of which I will now describe.

Should We Graph Small Data Sets?

Probably the most influential graphic style manuals of the last few years have been those by Edward Tufte. Tufte leads the call for restraint in graphic design, presenting many convincing cases in which reducing the design detail in graphs increases their comprehensibility. The message, often endorsed by other designers, is that the value of a graph increases with the number of data points it contains and decreases with each bit of irrelevant ink added to the display.

Following this position, the smaller the data set, the less there is to be gained by representing the information graphically. With very small data sets, such as the highly summarized data sets displayed in the popular press, graphs may be altogether unnecessary. Tabular or even textual descriptions should suffice.

In my lab, my colleagues and I recently looked at the utility of graphs for presenting small data sets. Line graphs and bar charts showing 4,7, or 13 data points were compared with tables of the same data. Subjects were allowed to study all the data sets using one of the three formats. Their task was simply to describe what they had learned after studying each data set; that is, they were asked for their spontaneous interpretations.

Countering the claim that there is little reason to graph small data sets, our results revealed that subjects' interpretations were most sensitive to critical differences in patterns across data sets when the data were presented graphically. Somewhat more in keeping with the conventional wisdom, however, the benefit of graphical over tabular displays was highly dependent on individual subjects' graph-reading strategies in the 4-point condition.

Based on data from this experiment, then, display designers should not be discouraged from graphing small data sets. Work by Ian Spence and Stephen Lewandowsky, who measured the speed and accuracy with which subjects could answer specific questions about small data sets, further supports the idea that, even in this limited case, graphs enhance the communication of quantitative information.

Those 3-D Graphs

The 3-D look in graphs, noted earlier, is a design practice that a number of style manuals warn against. Nevertheless, these formats--with boxes for bars and cliffs for lines--abound in the popular press. This decorative perspective has even found its way back into the somewhat more conservative illustrations of scientific presentations and journals.

Some designers have argued that for many purposes, these decorative elements attract attention and detract little from the actual content of the graph. Most other designers, however, have argued that the 3-D techniques often distort the data and decrease the amount of information that can actually be presented in a single view. After all, the practice of adding perspective means adding more unnecessary detail.

Our own research has indicated that the addition of the 3-D look to line graphs and bar charts has little performance consequences when subjects are asked to detect global data trends (e.g., is there an increasing or decreasing trend across data points?). However, there are true reductions in performance when subjects are asked to read exact data values from 3-D graphs, particularly from 3-D lines. A somewhat surprising finding, at least to those of us who have used 3-D graphs because they look impressive, is that subjects view data more positively if they are presented in the more traditional 2-D format. This finding should be of particular interest to those who use graphic displays to market products such as pain killers and mutual funds.

Connecting v. Cluttering

A third design debate, related once again to the issue of graphic embellishment, focuses on one of the most basic of all formats, the line graph. From a minimalist position, a series of data points should stand alone, as a series of physical points on a page. A line graph, which inserts line segments to connect the isolated points, is simply another example of the addition of irrelevant and misleading and possibly distracting ink.

Yet, our research has indicated that adding line segments to connect separate points into a unified visual whole, for example, connecting the top of a series of bars to form a combination bar/line graph, may help subjects see overall relationships and patterns in the data.

This leads to the suggestion that line graphs should generally be preferred to formats that show each data value as a separate point or bar, at least when one wants to communicate information about overall patterns or trends. This also defies the common wisdom that line graphs should only be used for connecting data points along a continuum. My colleagues and I argue that lines should also be used to connect data from different categories. Again, the convention should be violated if the important information to be gleaned from the graph involves overall patterns.

Take, as an example familiar to most psychologists, the display of individual test scores from various personality inventories. Perhaps the most famous, the Minnesota Multiphasic Personality Inventory (MMPI), usually shows a person's score on each of its scales as a series of individual points connected by line segments. It is the overall shape of these connected points, the overall profile, that diagnosticians learn to classify. A clinician friend has noted his dismay at receiving computerized printouts of MMPI scores that simply provide tick marks along each of the individual scales. He solves the problem by manually drawing in the missing links.

The conclusion of this research, then, is that adding contour and connections among data points does not automatically equal clutter and confusion for the graph reader. Instead, the lines may help the visual system detect meaningful configurations.

Prospects for Data-Based Conventions

Currently, a number of cognitive and perceptual psychologists are not only evaluating current conventions, but are developing general models of graph perception and comprehension that may guide the development of future guidelines and graphical standards. Perhaps the most promising trend has been the formation of interdisciplinary research groups including psychologists, statisticians, and graphic artists. For example, the National Center for Health Statistics has promoted such a research program to develop standards for the presentation of its own data to scientists, policymakers, and the general public. In addition, empirically based books on graphic design have begun to appear, most notably a volume by William Cleveland and a recent book by Stephen Kosslyn.

Overall, these research programs have generally found that the conventions of graphical design are much like folk remedies. The more closely you study them, the more you find that they generally do work. However, there are notable exceptions, some of which I described, and it is through psychological research that we can identify the limitations of current conventions and devise guidelines general enough to be applied to formats and fads yet to come.




APA Home Page . Search . Site Map