Isn’t it fascinating how charts can dissolve divides and turn disconnect into discoveries? It’s a purposeful medium where even the fragments and imbalance in the world are always part of the story. By unifying or restructuring grand and small narratives in a more distilled expression, charts make complex truths accessible to a wider audience. Consequently, the audience become not just spectators but also collaborators, peeling meaning from their respective perspectives or contributing to a common understanding. However, producing charts with this level of effectiveness doesn’t just happen through the mere union of analysis and aesthetics.
It starts with choosing the right chart for your data. That choice should be driven by intuitive wisdom and structured control, and both are anchored on purpose. By defining what you need to bring out of the data, you eliminate the unessential and get your message across as quickly and as clearly as possible.
What’s Your Purpose?
Comparison is a common purpose for data visualization, and the bar chart is the tool to make this happen. Bar charts use one categorical variable and one numerical variable like income data across departments. For more nuanced comparisons, clustered bar charts can be used to introduce another categorical variable, like seniority levels within departments. Bar charts, as shown in Figure 1, have horizontal bars, while another type, the column chart, uses vertical bars.
Figure 1. A bar chart (left) and a column chart (right) | Image Source: Financial Edge
Another purpose for data visualization is to show composition, specifically the contribution of parts to a whole. This type of visualization generally involves one categorical variable and one numerical variable. Composition charts in a matched pair usually take the form of a pie chart or donut chart (see Figure 2). An example of what they can demonstrate is the percentage of customers who prefer one product over another.
Figure 1: A pie chart (left) and a donut chart (right) | Image Source: Medium
More complex data with multiple categorical variables will benefit more from using stackedbar charts, which are similar to bar charts butwith multiple categorical variables represented as additional parts stacked alongside each other (see Figure 3). Alternatively, complex data can be illustrated using a treemap, which involves dividing a rectangular area into smaller rectangles to represent hierarchical data such as income distribution across regions and cities (see Figure 4).
Figure 3. A stacked bar chart | Image Source: Medium
If your purpose is to show change over time in charts, it can be done with line charts or area charts (see Figure 5). The line chart works better when it comes to a numerical variable with a time series, such as monthly revenue. The continuity of the line allows the viewer’s eye to easily notice trends and changes over time. The area chart goes one step further by shading the area under the line, emphasizing both the change and its magnitude. For more compact visualizations within tables, sparklines can be used. These are small charts that may be placed in a sheet’s individual cells and can highlight trends in big data sets or point out maximum and minimum values (see Figure 6).
Figure 5. A line chart (left) and an area chart (right) | Image Source: Edraw
As for visualizing relationships between variables, go with a scatter plot or bubble chart for numerical data and a heat table for categorical data. A scatter plot displays data points to show the correlationof two numerical variables, while a bubble chart is similar to a scatter plot, in which the x- and y-axes consist of two numerical variables, but the bubbles (or circles) in it vary in size to encode a third numerical variable (see Figure 7).
Figure 7. A scatter plot (left) | Image Source: Medium | and a bubble chart (right) | Image Source: Medium
For categorical data, a heat table has one categorical variable placed in rows and another placed in columns. The cells of the table are then coded with numerical values, often by way of different intensities of color. This is a particularly useful way to identify patterns or relationships between categorical variables, such as countries and performance scores (see Figure 8).
When you’re working with geospatial data, you might find a choropleth map more suitable (see Figure 9). It plots a numeric variable like population density over a geospatial variable such as regions or countries. This type of map illuminates the perception and realization of the spatial pattern by shading particular regions with differing tones.
The right chart for your data isn’t always immediately obvious. By establishing your purpose first, you narrow down your choices. You avoid overcomplicating or underrepresenting information. Another layer of factors to consider is the type of data you have and its size. And beneath all of these is your audience. From their familiarity with charts to the complexity of your data, decision-making always involves the people you are creating a chart for in the first place.
To learn more about data visualization, check out more articles here.
**********
Editor’s Note: This article was written in collaboration with Islam Salahuddin, data consultant at Systaems.
Welcome to the first chart from The Uncharted: Chart of the Week!
This week, we look at how greenhouse gas (GHG) emissions from the heat and electricity sector—more than any other industry—are skyrocketing in China and India.
The electricity and heat sector, which consists of electricity plants and energy generators that use fossil fuels, is the main source of GHG emissions compared to other sectors, recording 15.2 billion tonnes of carbon dioxide-equivalents in 2020, up from just 8.7 billion tonnes in 1990.
By transitioning to renewable energy and improving energy production and consumption efficiency, 51 out of the examined 169 countries have managed to curb their GHG emissions over the past three decades, including the United States(US), Russia and the United Kingdom (UK).
However, the exact opposite happened in China and India. While China was the third most intensive emitter of GHG in 1990 with 0.7 billion tonnes, the emissions escalated to a staggering 5.7 billion tonnes—an almost seven-fold increase.
In India, the electricity and heat sector was the top eighth in GHG emissions in the world in 1990, with 0.2 billion tonnes. Emissions increased by almost four times as much in 2020 with 1.1 billion tonnes.
In both countries, the surge of GHG emissions is attributed to major economic developments, rapid industrialization, and substantial population growth, in addition to their dependency on coal and fossil fuels, despite their efforts to diversify their energy mix.
The figures show how maintaining ambitious economic development while curbing emissions for lasting sustainability can still practically be an unsolved challenge, even for some major developed countries.
Fortunately, curbing GHG emissions is possible, and it starts with your business by monitoring and managing its emissions using the relevant KPI. That is why The KPI Institute published a full documentation for the # Greenhouse gas emissions KPI, available for free here.
“If communication is more art than science, then it’s more sculpture than painting. While you’re adding to build your picture in painting, you’re chipping away at sculpting. And when you’re deciding on the insights to use, you’re chipping away everything you have to reveal the core key insights that will best achieve your purpose,” according to Craig Smith, McKinsey & Company’s client communication expert.
The same principle applies in the context of data visualization. Chipping away is important to not overdress data with complicated graphs, special effects, and excess colors. Data presentations with too many elements can confuse and overwhelm the audience.
Keep in mind that data must convey information. Allow data visualization elements to communicate and not to serve as a decoration. The simpler it is, the more accessible and understandable it is. “Less is more” as long as the visuals still convey the intended message.
Finding the parallel processes of exploratory and explanatory data visualization and the practice of sculpting could help improve how data visualization is done. How can chipping away truly add more clarity to data visualization?
Exploratory Visualization: Adding Lumps of Clay
Exploratory visualization is the phase where you are trying to understand the data yourself before deciding what interesting insights it might hold in its depths. You can hunt and polish these insights in the later stage before presenting them to your audience.
In this stage, you might end up creating maybe a hundred charts. You may create some of them to get a better sense of the statistical description of the data: means, medians, maximum and minimum values, and many more.
You can also recognize in exploratory if there are any interesting outliers and experience a few things to test relationships between different values. Out of the 100 hypotheses that you visually analyze to figure your way through the data in your hands, you may end up settling on two of them to work on and present to your audience.
In the parallel world of sculpting, artists do a similar thing. They start with an armature-like raw data in designing. Then, they continue to add up lumps of clay on it in exploratory visualizations.
Artists know for sure that a lot of this clay will end up out of the final sculpture. But they are aware that this accumulation of material is essential because it starts giving them a sense of ideal materialization. Also, adding enough material will ensure that they have plenty to work with when they begin shaping up their work.
In the exploratory stage, approaching data visualization as a form of sculpting may remind us to resist two common and fatal urges:
The urge to rush into the explanatory stage – Heading to the chipping away stage too early will lead to flawed results.
The urge to show all of what has been done in the exploratory stage to the audience, begrudging all the effort that we have put into it – When you feel that urge, remember that you don’t want to show your audience that big lump of clay; you want to show a beautified result.
Explanatory Visualization: Chipping Away the Unnecessary
Explanatory visualization is where you settle on the worth-reporting insights. You start polishing the visualizations to do what they are supposed to do, which is explaining or conveying the meaning at a glance.
The main goal of this stage is to ensure that there are no distractions in your visualization. Also, this stage makes sure that there are no unnecessary lumps of clay that hide the intended meaning or the envisioned shape.
In the explanatory stage, sculptors use various tools. But what they aim for is the same. They first begin furtherly shaping the basic form by taking away large amounts of material. It is to ensure they are on track. Then, they move to finer forming using more precise tools to carve in the shape features and others to add texture. The main question driving this stage for sculptors is, what uncovers the envisioned shape underneath?
In data visualization, you can try taking out each element in your visualization like titles, legends, labels, colors, and so on. Then, ask yourself the same question each time, does the visualization still convey its meaning?
If yes, keep that element out. If not, try to figure out what is missing and think of less distracting alternatives, if any. For example, do you have multiple categories that you need to name? Try using labels attached to data points instead of separate legends.
There are a lot of things that you can always take away to make your visualization less distracting and more oriented towards your goal. But to make the chipping away stage simpler, C there are five main things to consider according to Cole Nussbaumer Knaflic as cited in her well-known book, Storytelling with Data:
De-emphasize the chart title; to not drive more attention than it deserves
Remove chart border and gridlines
Send the x- and y-axis lines and labels to the background (Plus tip from me: Also consider completely taking them out)
Remove the variance in colors between the various data points
Label the data points directly
In the explanatory stage, approaching data visualization as a form of sculpting may remind us of how vital it is to keep chipping away the unnecessary parts to uncover what’s beneath, that what you intend to convey is not perfectly visible until you shape it up.
Overall, approaching data visualization as a form of sculpting may remind us of the true sole purpose of the practice and crystalize design in the best possible form.
Deepen your understanding of processing and designing data with our insightful articles on data visualization.
In 1983, American professor of statistics and computer science Edward Tufte introduced the concept of data-ink ratio in his famous book, “The Visual Display of Quantitative Information.” Data-ink ratio is a pioneering disciplinary theory in data visualization that has been highly influential in recommended practices and teachings, with its excellency reflecting a minimalistic style of design. The art movement of Minimalism is noticeable in Tufte’s theory, even though he did not deliberately mention it. However, academic research shows that data-ink ratio generated mixed responses, resulting in the need for more complex frameworks in data visualization.
What is Data-Ink Ratio?
Tufte’s Data-Ink Ratio is built on his preposition that “data graphics should draw the viewer’s attention to the sense and substance of the data, not to something else.” The conceptual formula is:
“Data ink” means what is absolutely necessary to show the data. The word “ink” is used because the theory was formulated in the age of print graphics dominance. The equivalent of “ink” in the digital world today is “pixels.” Data-ink ratio aims to find and maximize the share of informative elements out of the total elements used in a chart, raising a ratio of 1 as the highest goal. A visualization with such a ratio is the one that only contains elements that show the data, with no decorations or redundancies.
Tufte asserted this goal by articulating two erasing principles: “erase non-data-ink, within reason” and “erase redundant data-ink, within reason.” For these two types of ink, he coined the term “chartjunk.” The excessive elements to get rid of include decorations, background images, unnecessary colors, gridlines, axes, and tick marks. He said that the principles should be applied through editing and redesign. This led to the structuring of the process of data visualization as a cyclical one that ends when the most minimal output is reached.
Minimalism became one of the most important and influential movements during the professor’s young age in the 1960s. Minimalism is defined by Cambridge dictionary as “a style in art, design, and theater that uses the smallest range of materials and colors possible, and only very simple shapes or forms.” It is a kind of reductive abstract art that is characterized by “plain geometric configurations” that lack “any decorative… flourishes.”
That can be juxtaposed with what Tufte instructed under the data-ink ratio theory: to achieve a data visualization that is simple and, therefore, minimal. Tufte’s theory suggests that such simplicity can deliver a message that is clear, precise, and efficient. As a result, it could reduce the time required for the user to perceive the visualization.
One of the many posts surfacing the internet to exemplify applying Tufte’s reductive principles on a bar chart. | Image Source: Darkhorse Analytics
Tufte’s priorities match the needs of the fast-paced business world. In business, saving time means saving cost and therefore maximizing profit. It was in the nineties, less than a decade after Tufte’s theory, when the accumulation of information started to skyrocket with the advancement of the World Wide Web. Businesspeople seemed to be lacking even more time with more information to consider, and a minimalistic approach seemed like a solution.
Minimalism did not only impact data visualization, but its effect also reached almost every corner of the human computer interaction (HCI) field. For example, as the internet became more widespread, search engines started to develop and became so powerful that they began to impose their own rules of the game. Because of the nature of how search engines work, the minimalistic structured design of web pages was more attainable for the engines to read and rank, and therefore more preferable and reachable for the audience.
For similar benefits, like reach in webpages and efficiency in data visualizations, minimalistic design became so invasive in all computer interfaces over the years. Today, minimalistic design is usually described as “modern design” and recommended to every designer building a user-interaction system, from mobile apps to webpages to data visualizations.
On the left: Donald Judd’s minimalist installation, Untitled, 1969 (photo from: Guggenheim Museum, New York) | On the right: Tufte’s minimalist variation of a double-bar chart with error lines. | Collage made by Islam Salahuddin
Reviews of Data-Ink Ratio’s Minimalism
Despite presenting his hypotheses as proven facts, Tufte had never empirically tested his promised achievements that can be reached by minimalistic design. It was not until the beginning of the nineties that academic research started to put the claims under the microscope.
The initial research findings struggled to prove the hypotheses. However, a major multi-experiment research paper in 1994 found that some non-data-ink and redundant data-ink, especially backgrounds and tick marks on the y-axis, may decrease accuracy and increase response time (which is proportionally bad). Meanwhile, the other ink that Tufte considered to be chartjunk and called for its removal whenever possible, like axis lines, was proved to increase performance in some cases. The negative effect was clear in some chartjunk types, like axis lines, but was less certain in others, like three-dimensional charts.
Such experiments were still built on the same proposition of Tufte that graphical excellency means clarity, precision, and efficiency but found that the relationship between data-ink ratio and excellency in that sense can hardly be linear as Tufte suggests. The paper states that “effects of ink are highly conditional on the features of the graph and task” and therefore “simple rules like Tufte’s will not suffice.” Instead of indicating that all non-data-ink and redundant data-ink should be erased, the authors call on data visualization designers to determine whether the use of any ink will facilitate or interfere with reading a graph, depending on its context.
Later research even questioned Tufte’s components of graphical excellency, especially the presumed all-cases importance of response time factor. An empirical paper in 2007 found that users may prefer non-minimalistic visualizations over Tufte’s minimalistic ones, partially because they may find the latter boring. This is a criticism that both Minimalism art and statistics face and a perception that Tufte tried to avert with his rule. Boredom should not be treated as a minor problem because it means less ability to induce attention. A visualization’s ability to generate attention is the gateway to the viewer’s perception in the first place.
Attention is one of the criteria that Tufte’s rule overlooks. Other significant factors are memorability and engagement. More advanced experiments in 2013 and 2015 re-asserted chartjunk as not always harmful. In some cases, it may even increase the memorability and engagement of a visualization. Attributes like color and human recognizable shapes, icons, and images can enhance memorability due to their ability to activate more parts of a viewer’s brain, leveraging its natural tendency towards what is familiar rather than what is just minimal.
Despite their popularity, chartjunk and similar terms also appear to be highly open to interpretation among practitioners. Interpretation can be affected by an individual’s circumstances that include culture, personal style, preferences, and views, as well as constraints of skills, tools, and user’s priorities, according to a discourse analysis that was published in 2022.
On the left: Frank Stella’s minimalist painting, title not known, 1967 (photo from Tate Modern) | On the right: Tufte’s minimalist variation of a multiple vertical box and whisker plot. | Collage made by Islam Salahuddin
How to Make Sense of the Previous Discussions
The growing body of research shows that data visualization is a task that can hardly be led by only a one-factor rule like data-ink ratio. It shows that even the simple task of choosing what elements to include or exclude in a visualization remains largely an uncharted territory and needs further examination.
However, one of the common underpinnings that all theoretical works share is a consideration for the importance of context in which a visualization is designed. To be fair, even Tufte himself did not ignore this consideration after all and emphasized that certain principles have to be adopted “within reason.” Asserting the “reasonability” factor, he deliberately mentions in the Data-Ink Maximization chapter of his book that maximizing data-ink “is but a single dimension of a complex and multivariate design task.” He recognized the possible existence of factors other than excellency that come into play, including “beauty,” even if he did not prioritize them.
Therefore, synthesizing all the critiques arising against Tufte’s rule of data-ink ratio appears to be possible by quoting Tufte himself. He said that determining to which extent the data-ink ratio should be maximized rests on statistical and aesthetic criteria.” This allows data visualization designers to figure out the sweet spot where a visualization delivers what it intends to and, at the same time, does not alienize itself for the sake of being minimal.
All in all, minimalism can be considered one of the means to design a great data visualization, but not a goal. After all, the goal will remain to deliver the intended message to the audience so they can perceive it best.
Edward Tufte’s principles of data-Ink ratio have prevailed in data visualization since they were introduced in the 1980s. His theory has imposed a tendency towards a minimalistic style, defining excellence as clarity, precision, and efficiency and reducing the time users perceive information.
Meanwhile, academic research that has put the American pioneering statistician’s teachings to test does not show the linear relationship between data-ink ratio and visualization’s excellency. Further research shed light on other important criteria that Tufte overlooked, like the ability of a visualization to induce attention, memorability, and engagement. Overall, the academic body of literature has strongly suggested that no simple rule like data-ink ratio can suffice in data designing.
Debates among practitioners have been ongoing about the repeated notion of “less is more,” which leans back on Tufte’s teachings. Some believe that simplicity and quick perceiving should be the goals of all visualizations at all times. Others support embracing complexity and slow viewing time in some circumstances.
As a response to these debates, two interesting frameworks have emerged to suggest more criteria that should be considered. The first is “Levers of Chart-Making” by Andy Cotgreave, a senior data evangelist at Tableau, and the second is “Cognitive Load as a Guide” by Eva Sibinga and Erin Waldron, data science and visualization specialists.
Cotgreave suggested this under-formulation framework in the November 2022 edition of his newsletter The Sweet Spot. He put forward five scales of levers that “chart producers can use to enlighten, not bamboozle.” They are as follows:
Speed to primary insight – How fast or slow insight is intended to be extracted from a graph According to him, “it is ok to make charts that take time to understand”.
Granularity – How sparse or granular is the data that a chart intends to show?
Explore or explain – whether a visualization is intended to give the users the opportunity to explore the data themselves (like self-service dashboards) or to be accompanied by an explaining presentation
Dry or emotional – refers to how serious the way of presenting the data is versus how informal and relevant it is to non-data people. According to Cotgreave, an example of the serious approach is a normal column chart and for the emotional, a necklace of which the bead’s size represents the same underlying data.
Ambiguity vs. accuracy – For Cotgreave, there can be intended ambiguity in chart-making instead of clear accuracy.
Cognitive load is a more detailed and rigid framework that takes its inspiration from the psychology of instructional design. Suggested by Sibinga and Waldron, the framework was published by the Journal of the Data Visualization Society (Nightingale) in September 2021.
Cognitive load proposes 12 spectra, offering “an alternative to one-size-fits-all rules” and aiming to “encourage a more nuanced strategy” for data visualization. Divided into three categories, the spectra are supposed to “gauge the complexity of our data on one side, identify the needs of our audience on the other, and then calibrate our visualization to successfully bridge the gap between the two.”
Intrinsic load – This is the first group of spectra that is concerned with the data itself. It considers the inherent level of complexity in the data that a designer is tasked to explain with a visualization. The included spectra are:
Measurement (quantitative vs. qualitative) – According to the authors, quantitative data has less cognitive load (easier to perceive) than qualitative data. That is because the former usually has obvious measuring units, like dollars or miles, while the latter usually needs a conceptual rating scale, like satisfaction rate from 1 to 5.
Knowability (certain vs. uncertain) – Data collected from the whole population is easier to perceive than data estimated depending on a sample or predicted for the future. This is because the former usually has a high level of certainty that is easier to perceive than the uncertainty that comes with the latter, intertwined with its inevitable statistical margins of error.
Specificity (precise vs. ambiguous) – Undebated data categories, like blood type or zip codes, tend to be easier to perceive than socially determined concepts, like gender, race, and social class.
Relatability (concrete vs. abstract) – How relatable is the data to what humans see in everyday life? Concrete data would be small numbers like the cost of lunch and one’s age, while abstract data would be conceptual ones like GDP and the age of the earth.
Germane load – The second group of spectra is concerned with the audience and how ready they are to process the new information shown by a visualization. The included spectra are:
Connection (intentional vs. coincidental) – How will the audience have the first look at the visualization? Intentional viewers are likely better propped to perceive the visualization than viewers who stumble upon it by accident.
Pace (slow vs. fast) – Slow viewers are the ones that have more time in hand and therefore +have more ability to perceive a visualization (interpreting into lighter cognitive load).
Knowledge (expert vs. novice) – Expert viewers are the ones who are already familiar with the subject and therefore will have to afford lighter cognitive load when viewing a visualization.
Confidence (confident vs. anxious) – This spectrum addresses the intersection of the audience and the data reporting format. The cognitive load that is required from an audience familiar with the data reporting format, such as an interactive dashboard and a data-based report, will require lighter cognitive load than the one that is encountering such a channel for the first time.
Extraneous load – The final group addresses how new information is presented. The authors believe that these are the criteria where a designer has the most control and should therefore be considered last. The advice to determine a visualization’s place on the following spectra is by answering the question: “Given the existing intrinsic and germane loads, how much more cognitive load are we comfortable adding to the mix?”
Chart type (common vs. rare) – Chart types like bar charts need lighter cognitive load than uncommon ones, like violin charts or rose diagrams and the more innovative ones.
Interpretation (accurate vs. approximate) – Does the chart aim to deliver precise values or paint a wide picture? According to the authors, charts delivering specific values tend to take a lighter cognitive load than the ones dealing with overall objectives.
Composition (concise vs. detailed) – This spectrum assumes a high data-ink ratio and no chartjunk (from Tufte’s concepts) are already in place and then asks, how dense is the information on the page? Less dense visualizations require lighter cognitive load.
Delivery (explanatory vs. exploratory) – Does the data report explain itself, or is built to be explored? Exploration, naturally, takes more cognitive load than a self-explaining visualization.
How to make sense of all the previous discussions
Levers of chart-making and cognitive load as a guide are two of the recently suggested frameworks that offer a more complex approach to the task of data visualization. The two have similarities, like their consideration of complexity, granularity, and way of delivery. They differ from Tufte’s approach mainly through their acceptance of the need to slowly perceive designs in some circumstances. Cognitive load still deliberately pre-assumes applying data-ink ratio principles beforehand.
Therefore, no framework is likely to totally replace the others. At best, they tend to complement each other to cover the vast territory of the data visualization domain.
Data-ink ratio principles remain a good point to start as it best fits most business contexts. It can also help designers keep in mind the point of their design and avoid getting distracted amidst all the available software tools today. However, considering the emerging frameworks can make the practice more nuanced for tackling different needs, messages, and audiences.
The final determinant of how to incorporate the three frameworks -and any other emerging ones- in practice will remain to be the context of the visualization. A better understanding of the audience, the message, and the medium is key before using the different frameworks to decide on how information should be delivered.
