XXVIII.1 January - February 2021
Page: 58
Digital Citation

Potential bias in creative chart design: A review of nontraditional financial graphs

Wei-Cheng Shen, Chih-Chen Lee, Te-Wei Wang

back to top 

The cliché "a picture is worth a thousand words" highlights how graphical representation facilitates clear, concise, and impactful communication. As human eyes are more sensitive to colors and shapes than to words, data is processed and encoded in a manner that our eyes can discern and our brains can comprehend. The term data vizualization describes the process of displaying abstract information for communication and sense making. Often rich in colors, symbols, animations, or other visual cues, data visualization makes use of graphs, charts, or tables to convey information, evoke feelings, and provoke actions. Essentially, creating better data visualization is a design exercise in engaging users to consume boring data. It not only enhances effective and efficient information exchanges, but also alleviates information overload and supports data analytics, especially when the underlying patterns or trends are complicated or intertwined. However, using data visualization to enhance user experience can be a double-edged sword. Fancy graphics can appeal to users visually or aesthetically but might lead to both intended and unintended interpretations.


Recent advances in information technology enable the creation of various new tools and techniques for designing better visualizations. However, data visualization comes to fruition only when the story is best told graphically rather than verbally and when the picture is well designed. Poorly designed "chartjunks" might inadvertently lead users to stare at a graph but never understand the underlying messages or, even worse, draw misleading or inaccurate conclusions. To reap the benefits of data visualization, many guidelines or handbooks have been developed or published since the early 1980s. In 1983, Edward Tufte, a guru in the field of data visualization, advocated "above all else, show the data" to focus on the faithful representations of numbers. He also encouraged graph makers to take the minimalist approach and avoid distortion, ambiguity, and other construction deficiencies.

back to top  Insights


Annual reports are the official communication between corporate management and stakeholders. In addition to providing financial statements and various disclosures, annual reports frequently provide bar, line, pie, or other nontraditional charts to illustrate and enhance narrative or tabular representations of firm performance. To investigate how data visualization has been used and misused, we collected 10,965 financial graphs from corporate annual reports of Fortune 100 companies for the period of 2004 to 2017. Given the popularity of nontraditional graphs, it is imperative for creators to know when and how to use them to convey information effectively and, more important, for users to learn how to evaluate these novel graphical representations correctly. In this article, we review four of these nontraditional graphs, namely three-dimensional graphs, dual y-axis/combo charts, nontraditional bar charts, and waterfall charts. Specifically, we discuss the strengths and weaknesses of these graphs, as well as how inappropriate use can lead to distortion, ambiguity, and even biases in decision making. We also offer suggestions on when to create and how to interpret these graphs.

back to top  Three-Dimensional (3D) Graphs

When a third dimension is added to graphical representations on two-dimensional (2D) surfaces, the added dimension has the potential to carry more information, increase aesthetic values, and meet certain users' preferences or needs (e.g., users with naive realism believe 3D representations are always better than 2D). However, the third dimension may distort the proportions among graphical elements, create illusions due to rotation or tilting, lead to occlusions, and divert users' attention from the content.

Among all the graphs we collected, 312 (2.85 percent) of the financial graphs were 3D. Interestingly, none of these added dimensions in the 3D representations carries additional information. While these 3D graphs look fancy and high-tech, their seemingly superfluous third dimensions distort proportions, create illusions, and mislead users, as warned by the literature. For example, in Figure 1, a global machinery company used a bird's-eye view 3D representation to illustrate its upward earning trend, making the vertically stretched graph even more distorted. By rotating 3D objects, a large life insurance company beautified and exaggerated its upward sales trend by placing the current year closer to the readers in Figure 2. As shown in Figure 3, a multinational energy corporation put the current year in the back, counter to the viewing direction, making the decreases in debt-to-capital ratios look even more favorable. If presented without data values, these 3D visualizations are difficult if not impossible to decipher. Sadly, among all 3D graphs in our sample, only 235 graphs (75.32 percent) provided data values. In other words, users had to rely on their own eyes and judgment to interpret 24.68 percent of the 3D charts in corporate annual reports. For example, an aerospace company provided many 3D stacked bar charts in its 2006–2009 annual reports without providing data values (Figure 4a). Even if using gridlines in the background, users might still have difficulties interpreting the trends or segment performance, especially when the year-order was reversed.

ins03.gif Figure 1. Distorted 3D representation of upward earnings. (Source: Northrup Grumman 2005 Annual Report)
ins04.gif Figure 2. Exaggerated upward sales trend. (Source: New York Life 2006 Annual Report)
ins05.gif Figure 3. Current year placed counter to viewing direction to make decrease seem more dramatic. (Source: ConocoPhillips 2010 Annual Report)
ins06.gif Figure 4. 3D charts lack data values.

Superfluous third dimensions were added not only to bar charts, but also to line, area, and pie charts. Figure 4b exhibits how an oil and gas company provided a 3D line chart to show its increases in cumulative total return without providing data values. Even with shades or gridlines, the rotations and third dimensions made these visualizations misleading and nearly useless. A basic but essential principle in data visualization is to keep the areas of each object proportionate to its data values. Clear violations can be seen in Figure 5, where the added third dimension to the 3D pie charts artificially and misleadingly makes the largest wedges much more prominent.

ins07.gif Figure 5. Adding the third dimension artificially exaggerates the size of certain wedges. (Source: UPS 2015 Annual Report)

In sum, the addition of a third dimension to value-encoding objects adds ink but rarely data. The third dimension should be added to 2D graphical representations only when meaningful information can be presented and interpreted effectively and efficiently. Enhancing aesthetic values can hardly justify the drawbacks of distortion, illusions, occlusions, and the diversion of users' attention. For example, Google Maps or other geographic information systems can satisfy most users' daily needs to get directions to their destinations. However, when the decision making involves the shape of mountains or height of buildings, Google Earth or other 3D representations outperform contour or annotated maps. Due to their increasing popularity, 3D graphs must be interpreted with caution—don't become dazzled by the seemingly fancy third dimension. Look at the data values, if provided, and stay alert when 3D objects are rotated, tilted, distorted, or blocked.

back to top  Dual Y-Axis and Combo Chart

When presenting multiple series of data with different scales, using a dual y-axis can help demonstrate or contrast how directly related series move together. Sometimes the features of two chart types are combined to show combo charts. Dual y-axis or combo charts are typically more information-intensive than other graph types, so presenting and interpreting the complex information effectively and efficiently becomes more crucial to both creators and users.

As shown in Figure 6, a combined bar and area chart clearly contrasts the differences between energy intensity and cumulative savings. However, showing a secondary y-axis or combo chart generally requires users to spend more time interpreting and understanding which data should be read against which axis. In Figure 7, a telecommunications conglomerate showed its capital expenditure by using a 2D line in front of three partially obstructed 3D bars, whose association and underlying message are unclear. The third dimension adds little information value to the data visualization. Without displaying any legend for the line, readers had to guess what these percentages represent.

ins08.gif Figure 6. A combined bar and area chart that clearly contrasts the differences between two values. (Source: Dow 2012 Annual Report)
ins09.gif Figure 7. 2D line in front of unnecessary 3D bars connects ambiguous percentages. (Source: Comcast 2010 Annual Report)

In Figure 8, a multinational energy corporation showed one gray line and multiple blue bars with two different blue y-axes. Due to improper scaling, the changes of the gray line were suppressed to being almost unnoticeable, masking their associations. Figure 9 shows how a package-delivery company used yellow bars to show numbers of repurchased shares and green balls to show their expenditures, resembling toffee green apples. Those green apples are out of proportion and block the baseline of the yellow bars, clouding users' judgment on relative sizes and heights. The story and underlying messages of the data visualization are far from clear.

ins10.gif Figure 8. Improper scaling obscures changes represented by the gray line. (Source: Chevron 2012 Annual Report
ins11.gif Figure 9. Out-of-proportion green circles obscure the baseline of the yellow bars, making the story unclear. (Source: UPS 2012 Annual Report)

In addition to the selection of scales and chart types, determining the quantity of presented information in a dual y-axis or combo chart involves a trade-off decision—too little information within a graph fails to tell the story properly, but too much information within a graph makes it too busy or even incomprehensible. For example, Figure 10 showed two lines above two stacked bars with multiple data values and their percentage changes over time. Without exerting extra time and effort, users can hardly interpret their underlying messages. Perhaps Figures 11a and 11b show the best illustration of the trade-off dilemma, where similar information was presented by the same insurance company in two consecutive years, but in a different manner. In 2016, three data series were displayed by using one blue line, one green line, and three blue bars against two y-axes in green or black. In contrast, the same data series in 2017 was displayed much more clearly by using two combo charts, albeit taking more space.

ins12.gif Figure 10. Too much information. Stacked bars and lines showing percentage changes over time. (Source: Archer Daniels Midland 2014 Annual Report)
ins13.gif Figure 11. The same data series presented over two years. The graph for 2016 presents too much data in one graph. The 2017 graph is much clearer but requires more space.

Displaying multiple data series against the same axis or in one chart implies a relationship that may or may not exist. If companies believe a relationship exists and a story can or should be told, appropriate chart types and y-axis scales must be carefully chosen. Jamming too much or overly complicated information into one chart inevitably requires users' time and cognitive effort for interpretation, and thus rarely achieves its intended goal.

back to top  Nontraditional Bar Charts

Bar charts present categorical data in rectangular bars with heights and lengths proportional to the values that they represent, facilitating comparisons among the categories. Probably due to its ease to prepare and interpret, the bar chart is the most popular chart type in corporate annual reports, representing 64.72 percent of all surveyed financial graphs. Of these bar charts, 86.33 percent displayed values vertically, 12.58 percent displayed them horizontally, and 1.08 percent neither. Other variations, such as stacked or grouped bar charts, can also be used for more complicated comparisons of data.

Traditional bar charts keep the thickness of the bars constant, so readers can focus on the height and length of the bars for quantitative meaning and comparison. In other words, the areas of each object in the data visualization are proportionate to their data values. In contrast, our review of financial graphs shows several nontraditional presentations of firm performance that distort these values. For example, in Figures 12a and 12b, the height and width of golden buildings or colorful hearts of the bar charts were changed simultaneously, exaggerating their increases over time. In addition to issues in proportion change, Figure 12a shows only three golden buildings to represent the insurance amount for five years, without a y-axis. Figure 12b displays five hearts along an unequal-distance x-axis without providing data values or gridlines. Users have to guess the number of volunteer hours in each year or just blindly accept the message encoded by its creator: Volunteer hours increased dramatically.

ins14.gif Figure 12. Changes in the height and width of icons exaggerate changes over time.

In addition to using rectangular bars, companies have started using curls, radiants, parabolas, circles, boxes, and other shapes to show changes over time (Figure 13). Without data values, users can hardly determine changes among data bars and the underlying messages of such highly decorated charts. For example, Figure 13a shows wireless use over time by using curls to resemble a race on an athletic field, but the curls are not proportionate to their data values. Figure 13b shows U.S. sales to various regions by using four concentric circles. Other bar charts from Figures 13c to 13f exhibit similar problems. Due to visual distortions, users of these nontraditional graphs must exercise due care when extracting their data values, mentally plotting them and re-creating the underlying messages.

ins15.gif Figure 13. Some companies use curls, radiants, parabolas, circles, boxes, and other shapes to show changes over time.

back to top  Waterfall Chart

Popularized by the strategic consulting firm McKinsey & Company, waterfall charts pull apart the pieces of a stacked bar chart to focus on one shift at a time and demonstrate how a running total changes sequentially or chronologically. For example, in Figure 14, a company provided "Cash Waterfalls" in its 2007–2014 annual reports. On the left-hand side, we see the ending cash balance of2010 in plum color. As we move to the right, its cash balance increases due to Cash from Operations and Asset Sales (in magenta) and then decreases due to Net Change in Borrowing and Acquisition (in dark red). With other cash inflows and outflows changing the "water level" from 2010–2014, the final bar shows the ending cash balance of2014 (in crimson).

ins16.gif Figure 14. A waterfall chart pulls apart pieces from a stacked bar chart to show how a running total changes. (Source: Dupont 2014 Annual Report]

Proponents of waterfall charts believe such representation visually decomposes the upward and downward changes to an initial value and reconciles the differences between two numbers. More and more companies adopt waterfall charts and explore new ways to present reconciliation information (Figure 15). For instance, since 2013, Target has used waterfall charts to show the gross margin and SG&A expenses changes over time. Caterpillar visualized its consolidated financial performance since 2016. Starting in 2016, Intel used "revenue walk" to show its revenue changes over time, which is similar to Pfizer's 2017 annual report. In addition to visualizing the reconciliations between numbers, Ford added annotations to provide more contextual details, such as volume/mix, contribution costs, and structural costs.

ins17.gif Figure 15. Examples of recent waterfall charts. From top: Target, Caterpillar, Intel, Pfizer, and Ford.

In response to market demand, Microsoft Excel 2016 simplified the process of creating histograms and box and whisker charts, as well as providing innovative visualizations like waterfall charts, treemaps, and sunburst charts. While the contributions of these new charting techniques remain unexplored, there is no doubt is that these new grasses are spreading quickly—regardless of whether they can efficiently and effectively support decision makers.

back to top  Concluding Remarks

Visual communications utilize narratives, tables, graphs, and other elements. Information encoders must find the best mixture of these elements to help decoders interpret the underlying messages. As the official communication between corporate management and stakeholders, annual reports must effectively and efficiently convey information about a company's activities throughout the preceding year. While adding graphical representations has the potential to increase aesthetic values and improve communication, misusing data visualization techniques may lead to unexpected outcomes, especially when the graphs deviate from the traditions. As suggested by Edward Tufte, whether a picture is indeed worth a thousand words depends on the correct selection of communication medium (e.g., narratives, tables, graphs) and then the appropriate design of the visual components (e.g., colors, dimensionality). Merely using nontraditional graphs does not automatically guarantee effective communications.

back to top  Authors

Wei-Cheng Milton Shen is an associate professor at University of Alabama, Huntsville. His research focuses on data visualization, gamification, and graphical financial reporting. milton.shen@uah.edu

Chih-Chen Lee is the William and Dian Taylor Professor of Accountancy in the Department of Accountancy, Northern Illinois University. She serves as the associate editor of Journal of Forensic Accounting Research and is a member of the editorial board of Journal of Forensic & Investigative Accounting. cclee@niu.edu

Te-Wei Wang is an educator and a practitioner in the area of information systems. Currently, he is an associate professor at the University of Illinois Springfield (UIS). He is also an independent consultant in the area of business intelligence. twang22@uis.edu


Storytelling with Data: A Data Visualization Guide for Business Professionals
by Cole Nussbaumer Knaflic. (ISBN: 978-1119002253)

The Wall Street Journal Guide to Information Graphics: The Dos and Don'ts of Presenting Data, Facts, and Figures
by Dona M. Wong (ISBN: 978-0393072952)

Visualize This!
By Nathan Yay (ISBN: 978-0470944882)

Show Me the Numbers: Designing Tables and Graphs to Enlighten
by Stephen Few (ISBN: 978-0970601971)

back to top 

©2021 ACM  1072-5520/21/01  $15.00

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.

The Digital Library is published by the Association for Computing Machinery. Copyright © 2021 ACM, Inc.

Post Comment

No Comments Found