Conflicting standards for designing data displays: following, flouting, and reconciling them.
Individually, each type of standard can be illuminating and extremely persuasive, but when we begin comparing them, they can be perplexing and contradictory because they apply different criteria, focus on different aspects of design, and are founded on universal principles that foster a one-size-fits-all approach to information design. How then can these varying standards help us create data displays? How do we resolve the inevitable conflicts? If we view the standards as means rather than ends, as available strategies that we can adapt to a given rhetorical situation rather than as nonnegotiable absolutes, they can be extremely useful design tools. By allowing the rhetorical situation to guide us through the design process, we can more intelligently decide when to follow - and when to juggle or to flout - the standards.
To show how this rhetorical approach works, I will first outline the four standards and explain the principles and assumptions that underpin them. Then I will show how the conflicts among the standards can be reconciled by allowing the rhetorical situation to guide the design process. In critiquing the standards, I don't aim to discredit any of them - they all bring something important to the design process - but rather to demonstrate how to apply them intelligently and effectively.
CLASSIFYING THE STANDARDS
Standards for data-displays can be classified into four principal types:
These four types of standards have distinguishing characteristics because they emphasize different design criteria and derive their authority from disparate sources, ranging from information theory to empirical research to cultural knowledge. Although the four types of standards overlap some, their underlying principles vary considerably.
Like other forms of design, data displays are governed by conventions, the accepted practices that designers imitate to meet reader expectations. Prescriptive and rule-bound by nature, conventions can be divided into two general types: those driven by genres and those driven by organizations and by professional disciplines.
Perhaps the most widespread and pervasive conventions are associated with genres, several of which appear in Figure 1. These visual genres include, for example, bar charts (a) and related subgenres such as divided and multiple bar charts (b, c), high-low charts (d), percent charts (e), and Gantt charts (f); the related genres of line graphs and mountain graphs (g, h); and other genres such as pie charts (i), polar charts (j), scatter plots (k), and maps (1).
These only scratch the surface of data display genres; in Information graphics, Robert L. Harris (1996) compiles an enormous lexicon of genres - many familiar, others exotic to the uninitiated. Choosing among even the basic genres can be daunting for novice designers, and most technical writing textbooks provide guidelines for making those decisions. For example, bar charts and line graphs are good for charting data over time; pie graphs for percentages of a whole; scatter plots for nontemporal data; horizontal bar graphs for data sets that require lengthy labels; and so on. Conventional standards also define the rules for representing data in a given genre. Pie charts, for example, begin at the 12 o'clock position and contain no more than six or seven slices; the y-axis on line graphs begins at zero or has break lines; multiple bars are clustered together for each time unit; data values on scatter plots are shown with dots; and so on.
Technology strongly reinforces conventions by giving us a limited range of genre templates to choose from, a fact that makes our design work more efficient but which can also discourage innovation. In a study of computer-generated and hand-made charts, Lee Brasseur (1994) found that designers using computers settled more quickly for the computer-generated form, whereas those who designed charts by hand explored a richer range of visual alternatives. On the other hand, Harris (1996) points out that computers also expand the range of available genres by enabling virtually anyone to create what were once labor-intensive genres such as 3-d wire frame graphs and flow-charts for sequencing activities (p. 5). Technology has a paradoxical effect on data design, both ossifying and democratizing conventional genres.
Conventions often derive their authority from visual discourse communities, groups of readers with similar experiences and expectations about information design (see Kostelnick and Roberts 1998, pp. 36-37). Visual discourse communities can be large (a culture, a country) or small (a university, a company); in professional communication, they frequently coalesce around disciplines and organizations. Within disciplines conventional codes evolve for modeling data visually - for example, high/low charts in finance - and often such codes are sanctioned by journals and professional style sheets. Organizations also dictate data display conventions through in-house style guides, administrative edict, or longstanding practices. Designers can resist using the conventional code, but they may pay dearly in lost clarity and credibility.
Perceptual standards are based on the relationship between the eye, brain, and image, which if studied through observation and testing, can predict human performance and thus serve to underpin universal design principles. Perceptual standards often derive from empirical research, which has been conducted over the past six or seven decades and which has been summarized by Macdonald-Ross (1977); Cleveland and McGill (1984); and Cochran, Albrecht, and Green (1989). The goal of empirical research is to identify graphical forms that impede and that facilitate accurate judgments of data relationships.
What does that research tell us? Mostly, it identifies problems readers have processing graphical displays accurately. For example, readers have difficulty comparing areas - circles, squares, and odd-shaped forms. Although readers can easily see that the circles in Figure 2 vary in size, they won't be able to decipher exactly the extent of the variance (is the large circle on the left 10, 20, or 30 times as large as the smallest circle on the right?). Readers also experience difficulty comparing volumes: The cube on the left in Figure 2 is eight times the size of the one on the right, though few readers would judge those relationships accurately. Readers also have problems comparing gray scales: the one on the left in Figure 2 is 10%, the middle one 25%, the right one 40%; again, readers are unlikely to interpret these proportions accurately. Further, asking readers to compare data rendered in perspective will also diminish the accuracy of the display.
The optimal graphical configuration enables readers to compare data according to "position along a common scale" (Cleveland and McGill 1984, pp. 532-33, 545) - that is, lines, bars, or dots that plot data values from the same baseline, as shown in Figure 3.
Although perceptual standards are largely based on an objective, scientific examination of users' responses to data graphics, other perceptual phenomena such as gestalt principles also fall into the perceptual group. For example, the gestalt principle of figure-ground contrast maintains that a graph with gray lines plotted against a grayscale background would be far less legible than one with black lines against a white background [ILLUSTRATION FOR FIGURE 4 OMITTED].
Informational standards are related to the perceptual standards in that they emphasize the clear and efficient uptake of data; however, they derive their authority not from scientific testing or gestalt principles but from information theory - specifically, the conduit model of communication, originally posited by Shannon and Weaver (1949). Like those who advocate the conduit model of verbal language, advocates of informational standards view graphics as transparent, objective channels through which designers transmit data. As such, informational standards emphasize structural and stylistic transparency: Clarity and conciseness underpin this approach, and visual noise undermines it by clogging up data transmission.
The chief advocates of informational standards include Edward Tufte (1983) and Jacques Bertin (1981). Through his "data-ink" (pp. 93-105) and "data density" (pp. 161-69) ratios, Tufte explains how to cut dead wood from data displays, cleansing them of "chartjunk" (p. 107) and enabling them to function like fiberoptic wires. Not surprisingly, Tufte emulates the plain-style approach of Strunk and White (p. 81; see Barton and Barton 1990, pp. 266-267). Bertin also advocates a functional, minimalist approach, emphasizing compact displays that create pure channels between image and eye - any eye. In the visual realm of the informational standard, readers of data displays are generalized, universal beings who see through the same set of eyes.
We can see the informational approach in action in the displays in Figure 5. Let's say the displays show the unit sales figures for a computer company for the first 6 months of the year, January through June. Versions b and c progressively remove visual noise from version a, with version b deleting the 3-d effect, the grayscale background, half the labels and tick marks on the y-axis, and half the gridlines. Version c further economizes the display by transforming the bars into lines, deleting the remaining tick marks, and applying a grayscale to the gridlines, though adding the vertical gridlines forfeits some of the gain. With each successive revision, we broaden and purify the channel between display and reader.
Informational standards also give readers more than a single view of the data - from the top down and the bottom up, from the "macro" and the "micro" view, as Tufte puts it in Envisioning information (1990, pp. 37-51; see also Barton and Barton 1993b). In a display showing complex data, then, readers can decide whether they want to see the big patterns and relationships or to immerse themselves in just one segment of the data - in terms of a map, to speed across the interstates or to wander along the county roads. Readers can choose the level of information they want, deciding whether and where they want to go deeper. Bertin (1981, pp. 179-81) also examines this duality of that which we "see" (the big picture view) versus that which we "read" (the small local view), though he decidedly favors seeing over reading.
In addition to the other three standards, aesthetics powerfully influences data design, though many designers resist acknowledging its influence. Aesthetic elements appeal to the reader's taste, which is largely shaped by cultural knowledge, as well as to the reader's emotions and intuition, setting the tone of the display and giving it a visual voice. We might classify aesthetic approaches according to those that reflect movements in contemporary taste (modern/postmodern) and those that reflect the expressive intuition of the designer, though the distinction is problematic because no designer works in a cultural vacuum.
Modernism values functional, streamlined design, tenets that were well established early this century and that parallel elements of the informational approach, especially Tufte's. The archetypal modernist data display can be found in the Isotype system of Otto Neurath (1936), where geometrical, high-contrast pictographs abound (see Lupton 1986). The modernist standard continues to value functional simplicity - like the clean, minimalist bar graph for computer sales shown in Figure 6.
Although modernism still pervades contemporary design practice, postmodernist theory has begun to make inroads by recognizing aesthetics as cultural knowledge and therefore encouraging a plurality of forms (see Barton and Barton 1990). Barton and Barton (1993a) have applied postmodern aesthetics to maps, advocating that the designer recognize alternative interpretations by externalizing the map's ideological underpinnings. How the emerging postmodern aesthetic will affect other forms of data displays remains to be seen.
The aesthetic standard also includes expressive approaches - advocated by Nigel Holmes (1984), Jan White (1984), and Roger Parker (1989) - which aim to capture the readers' attention, motivating them to examine the data. For example, using the computer screen to frame the line graph in Figure 7 draws attention to the display as well as announces its subject. These kinds of artful displays can radically alter visual tone - in Figure 7 making it more informal and conversational, maybe even playful - though the use of such artistry horrifies informational purists like Tufte (1983), who dismisses it as mere "chartjunk" (p. 107).
Overlaps among the standards
The development of different standards, or schools of data design, results in part because of the varying disciplines and cultures of their advocates - for example, Tufte, the Yale statistician and political economist; Bertin, the French cartographer/statistician; Holmes, the graphic artist for popular media. The tenets they advocate reflect the visual discourse communities of their respective disciplines. However, neither these discourse communities nor the design criteria and traditions that grow out of them exist in isolation. Even though the four standards may produce different data displays, their tenets overlap some.
For example, Tufte's informational maxims parallel modernist aesthetics: By advocating clean, minimalist forms and railing against "chartjunk," embellishments with no ostensible purpose, Tufte echoes functional modernism. Furthermore, Tufte's "Lie Factor" (1983, pp. 57-77) identifies graphing techniques that cause perceptual problems - for example, perspective and size variations among pictures - problems long identified by empirical research. Conventional genres also intersect with perceptual standards: With the exception of pie charts, genres that rely on area (for example, circles and squares) have lost credibility, largely because of perceptual deficiencies verified by practice and empirical research. None of the four standards, then, operates in a design vacuum.
RECONCILING CONFLICTS AMONG STANDARDS
Despite points of overlap, conflicts among the standards inevitably arise when we design data or evaluate an existing display. When we're faced with real audiences with compelling information needs, the standards push against each other; their invincible maxims start to break down, leaving us confused and uncertain which to follow. What do we do, for example, when the conventional approach produces a display that's perceptually problematic (a 3-d pie chart) or when informational standards produce one that's aesthetically unappealing, lacks interest, and projects the wrong tone?
Consider the conflicts that surface when we focus on a single issue like ethics. Because ethics has long been a factor in data displays, it's not surprising that each of the standards has an ethical component - or that each adds a different ethical twist. Advocates of aesthetic standards implicitly argue that designers have the ethical responsibility to empower readers by making the data interesting, inviting, and accessible. This impulse to democratize information stems from early twentieth-century modernism. From an informational perspective, Tufte (1983) argues that redundant, low-density displays cheat and patronize the reader, whereas a display with a high "Lie Factor" (p. 57) obscures the truth. Advocates of conventional genres proselytize against ethical faux pas like starting the y-axis above zero, which might deceive readers uninitiated in the nuances of the genre.
How do we resolve these inevitable conflicts that arise during the design process? Juggling these competing forces can be tricky because they push on us from different, often contradictory directions, as shown in Figure 8. Perceptual accuracy may conflict with conventional genres, conventional genres with informational aims, informational aims with the expressive urges of the designer to grab the reader's attention. Managing these conflicts can become easier and more productive if we view the standards as means to an end rather than as ends in themselves, as an a la carte menu of strategies that we can mix and match for each rhetorical situation. The rhetorical situation, not a given standard, ought to guide our design decisions.
When we design or evaluate data displays, then, we should ask the same rhetorical questions that we ask when we communicate with words:
* Who are my readers and what do they expect?
* What do I want my data display to accomplish?
* How will my readers actually use my data display?
Just as with writing, a one-size-fits-all approach to design can't accommodate all audiences, purposes, and contexts any more than writing standards can. The standards should merely inform those decisions by providing a range of available options.
Allowing the rhetorical situation to guide data design opens up possibilities that the individual standards might conceal. However, that fact hardly means that anything goes - far from it. To be effective, data displays must closely match the situation in which readers will actually use them. Depending on the situation, then, even our computer graph "chartjunk" in Figure 7 might find a useful place in the universe of information design. Perhaps the computer chart appears on an invitation for a staff retreat that intends to get the reader's attention. Or perhaps it appears in a memo from the boss trying to take the edge off last month's downturn in sales, or in a newsletter where it competes for attention with other stories and visuals. Our decision to keep - or discard - this data display should be based on rhetorical grounds rather than adherence to a single standard.
THE CASE OF BIG FOOT ISLAND: HOW THE RHETORIC AL SITUATION SHAPES DATA Displays
To explore how the rhetorical situation enables us to control and integrate the standards, let's explore a little case in information design. Let's say that we have data about Big Foot Island, which sits atop a corral reef about a quarter mile long and which has been divided into five tracts. Trees on the island include three species - X, Y, and Z - which were planted as seedlings 30 years ago. We know the number and location of each of the three species that survived as well as the average height of each species on each of the five tracts. How do we design this data? How do we juggle the standards during the design process?
Let's say we have two different situations we need to design for:
1. Readers who might consider buying a lot on one of the five tracts, as an investment or on which to build their dream vacation home. These readers will see the data displays at a sales presentation while they're on a cruise in the area. For them, the information should be inviting and persuasive, drawing them into a subject that they may not initially have much interest in. Because prices for lots vary in each of the five tracts, buyers will carefully consider data about the trees (for privacy, shade, beauty).
2. Readers who are researchers in a government agency that oversees natural resources on islands in the area. Studying the climatic and geographic effects on the trees, the researchers will use the data displays to analyze the growth of the trees over long periods of time in the local climate. The data displays will appear in a monthly progress report circulated within the government agency.
These two rhetorical situations have very different audiences, purposes, and contexts, and if we allow these variables to guide our design decisions, they will inevitably pull us in different directions as we shape the information visually.
We could begin by plotting the location of the trees on a map of Big Foot Island, as in Figure 9. From an informational standpoint, the map yields lots of data about the distribution of the three species across the island, with each data point identifying the tree species as well as its location. This highly compact display (more than 500 pieces of data in about 5 square inches) meets the informational standards of a Tufte or a Bertin. And in doing so, the display partly satisfies the informational needs of both audiences: Potential buyers can see the big picture, the distribution of the trees relative to possible building sites, and researchers can pinpoint the location and density of each species on the five tracts. So in Tufte's terms, by providing both macro and micro access to the data, this single display might satisfy both audiences.
However, this display has some perceptual deficiencies: The tree species can't be distinguished from each other without meticulous reading, a task that the potential buyers won't find appealing at a sales presentation during their vacation. They won't be interested in that level of detail - not yet, at least. For them, the sheer density of the data that results from separately coding each of the three species has little informational value. Besides, the coding symbols project a formal, technical tone. The buyers might be better served by a map showing only the location of trees [ILLUSTRATION FOR FIGURE 10 OMITTED], regardless of the species. Adding a grayscale background, an aesthetic decision that subdues the display and makes the tone less formal, would enhance the map's credibility even as it erodes its perceptual integrity through loss of figure-ground contrast.
As the needs and expectations of our two audiences diverge, different designs take shape based on these differing needs and expectations: For the researchers, the data richness of the original map would serve them well because the details are accessible should they need that micro-level reading. The informational density of the display outweighs the perceptual problems of distinguishing the three species. For the researchers, deeper, richer data helps. For the audience of land buyers, however, the rhetorical cost of including all the data in Figure 9 exceeds its worth, given the purpose of the display - to engage readers in the information and help them decide whether they're remotely interested in buying.
Plotting the data as individual points on the map, with or without the species symbols, gives both audiences a sense of tree location and distribution. However, the maps don't enable readers to compare tree data across tracts very easily or accurately. In addition to the maps, we might consolidate the numbers of species per tract in the form of pie charts [ILLUSTRATION FOR FIGURE 11 OMITTED]. Although the pie charts lack perceptual accuracy because they require readers to compare areas (the slices of the pie), they will satisfy the needs of prospective buyers, who don't need optimum perceptual accuracy. Aesthetically, and as a conventional genre, the pie charts look like a presentation graphic, matching the expectations of the readers in this context. On the other hand, the audience of researchers will suffer from the perceptual deficiencies of the pie charts because they need to compare data accurately. Moreover, they may find pie charts too unconventional a genre for empirical research data: For them, the pie charts may appear too nontechnical and even patronizing, a dumbing down of the data.
Transforming the data into a multiple bar graph [ILLUSTRATION FOR FIGURE 12 OMITTED] will enable researchers to compare data more accurately and will also match their genre expectations, giving the display greater credibility.
What about the audience of prospective buyers? To give the presentation more variety and interest, as an alternative to the pie chart map, we might create a divided bar chart [ILLUSTRATION FOR FIGURE 13 OMITTED]. By segmenting the tree species within each tract bar, the divided bar graph swaps perceptual accuracy for economy, whereas at the same time it affords a more holistic view of the data (totals for each tract appear in a single bar), an appropriate trade-off for these readers. Still, the divided bar chart looks too stodgy and uninviting for a sales presentation. Readers seated around the room in their sunglasses, shorts, and sandals may simply nod off! So to draw the buyers into the data, we might add expressive elements - for example, in Figure 14, an abstract profile of the island and some leaves - to make the data more informal, conversational, and appealing.
As we design this information, then, we continuously adapt the standards to the needs of the two groups of readers: For the researchers, perceptual accuracy and detail, as well as the assurance that the data is credible and matches the conventions of their discipline; for the buyers, a broad view of the data in a form that motivates them to explore it, or even take the time to notice it. These two different rhetorical situations guide the two diverging design processes, enabling us to arbitrate the standards along the way.
The rhetorical situations continue to guide us as we design the data about the average heights of the tree species, A vertical bar chart [ILLUSTRATION FOR FIGURE 15 OMITTED] would make a good perceptual choice for both audiences. For the researchers, however, we now have two multiple bar graphs, which are perceptually effective but redundant based on informational standards. We can combine these two graphs by plotting the quantity for each tract (on the x-axis) against the average height (on the y-axis), condensing the data into a single display, a scatter plot [ILLUSTRATION FOR FIGURE 6 OMITTED]. The scatter plot also meets conventional standards because this genre and will match the researchers' experience and expectations as members of a professional discipline steeped in empirical observation. Labeling each of the five tracts may reduce the perceptual efficiency of the display, but that's easily offset in this situation by information density and the ability to see correlations among the number and height of the trees, the three species, and the five tracts.
These benefits don't accrue, however, for our audience of potential buyers. For them, the scatter plot will appear far too technical and off-putting, and will surely make them wish they were in the pool or playing shuffleboard. We could lighten up its visual language by filling the plot frame with a grayscale and inserting the island profile [ILLUSTRATION FOR FIGURE 17 OMITTED]. However, the conversational tone of the revised display creates a visual oxymoron - technical yet breezy and unbuttoned, undermining the ethos of the display and leaving readers confused and uneasy - if they even know how to read the genre in the first place. A simple bar chart with the island profile and leaves [ILLUSTRATION FOR FIGURE 18 OMITTED], which mirrors the divided bar chart in Figure 14, might work better.
Because of the differing rhetorical situations, the package of data displays we design for the buyers and the researchers will differ substantially. Our audience of buyers will get the map with the generic coding [ILLUSTRATION FOR FIGURE 10 OMITTED], the divided bar chart showing the quantities of trees [ILLUSTRATION FOR FIGURE 14 OMITTED], and the multiple bar chart showing variations in height [ILLUSTRATION FOR FIGURE 18 OMITTED]. The researchers will get the map distinguishing the three species [ILLUSTRATION FOR FIGURE 9 OMITTED] and the scatter plot [ILLUSTRATION FOR FIGURE 16 OMITTED]. Perceptual integrity, informational efficiency, conventional genres, and aesthetics all play a role in developing these designs, but the rhetorical situation acts as our guiding compass, arbitrating these roles along the way.
When the rhetorical situation guides our design, then, none of the standards dominates the process, and none has veto power over the others. The audience, the purpose, and the context in which readers use the display shape decisions about conventions and genres, perceptual thresholds, density of information, and aesthetic texture. The standards inform the design process rather than drive it. As we move through that process, we need to be flexible and vigilant, recognizing that a bar chart with a profile of an island inside its plot frame [ILLUSTRATION FOR FIGURE 18 OMITTED] may be appropriate in one situation but a disaster in another. Depending on the rhetorical variables, we need to remain open to the possibility that Tufte or Bertin, Parker or White may be dead wrong, right on the mark, or somewhere in between.
Designing information should be a process of give and take, of critically examining standards and alternatives within the context of a specific design problem. While we might admire the work of Tufte, Bertin, Cleveland, White, Parker, and the others, we should read them critically, realizing the unique strengths, as well as the limitations, of their advice. Reconciling the standards means exerting control over them, of realizing that every mark on the data display does some rhetorical work, which can't be evaluated in a rhetorical vacuum. The rhetorical situation tells us when to follow, blend, or flout the standards. To exert this kind of control over the standards, however, we must first recognize that, like writing, designing information is a rhetorical act.
Barton, Ben F., and Marthalee S. Barton. 1993a. "Ideology and the map: Toward a postmodern visual design practice." Professional communication: The social perspective. Ed. Nancy Roundy Blyler and Charlotte Thralls. Newbury Park, CA: Sage, pp. 49-78.
Barton, Ben F., and Marthalee S. Barton. 1993b. "Modes of power in technical and professional visuals." Journal of business and technical communication 7:138-62.
Barton, Ben F., and Marthalee S, Barton. 1990. "Postmodernism and the relation of word and image in professional discourse." The technical writing teacher 17:256-270.
Bertin, Jacques. 1981. Graphics and graphic information-processing. Trans. William J. Berg and Paul Scott. New York, NY: Walter de Gruyter.
Brasseur, Lee E. 1994. "How computer graphing programs change the graphic design process: Results of research on the fill pattern feature." Journal of computer documentation 18, no. 4:4-20.
Cleveland, William S. and Robert McGill. 1984. "Graphic perception: Theory, experimentation, and application to the development of graphical methods." Journal of the American Statistical Association 79, no. 387:531-554.
Cochran, Jeffrey K., Sheri A, Albrecht, and Yvonne A. Green. 1989. "Guidelines for evaluating graphical designs: A framework based on human perception skills." Technical communication 36:25-32.
Funkhouser, H. G. 1937. "Historical development of the graphical representation of statistical data." Osiris 3:269-404.
Harris, Robert L. 1996. Information graphics: A comprehensive illustrated reference. Atlanta, GA: Management Graphics.
Holmes, Nigel. 1984. Designer's guide to creating charts & diagrams. New York, NY: Watson-Guptill.
Kostelnick, Charles, and David D. Roberts. 1998. Designing visual language: Strategies for professional communicators. Needham Heights, MA: Allyn and Bacon.
Lupton, Ellen. 1986. "Reading isotype." Design issues 3, no. 2: 47-58. Also in Design discourse: History, theory, criticism. 1989. Ed. Victor Margolin. Chicago, IL: The University of Chicago Press, pp. 145-156.
Macdonald-Ross, Michael. 1977. "How numbers are shown: A review of research on the presentation of quantitative data in texts." Audio-visual communication review 25:359-409.
Neurath, Otto. 1936. International picture language: The first rules of isotype. London, UK: Kegan Paul, Trench, Trubner.
Parker, Roger C. 1989. The makeover book: 101 design solutions for desktop publishers. Chapel Hill, NC: Ventana.
Shannon, Claude E., and Warren Weaver. 1949. The mathematical theory of communication. Urbana, IL: The University of Illinois Press.
Tufte, Edward R. 1990. Envisioning information. Cheshire, CT: Graphics Press.
Tufte, Edward R. 1983. The visual display of quantitative information. Cheshire, CT: Graphics Press.
White, Jan V. 1984. Using charts and graphs: 1000 ideas for visual persuasion. New York, NY: R. R. Bowker.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Visualizing Information|
|Date:||Nov 1, 1998|
|Previous Article:||Bridging the gap between SGML and HTML: the potential of XML for technical communicators.|
|Next Article:||Information graphics at the 'Boston Globe': from concept to execution.|