The Ethics of Data Visualization

The world is flooded with data, and it needs workers who can interpret that data in ways that promote understanding and drive good decisions. But what happens when instead of clarifying the truth behind the numbers, data visualizations instead misrepresent it? Or worse, mislead the people relying on these visuals to enhance their understanding of a problem?

In an era of “fake news,” viral social media posts, and ideological divides, data visualization has become a key way to capture the public’s attention, says Alberto Cairo, the Knight Chair in Visual Journalism at the School of Communication at the University of Miami in Florida. Author of the book How Charts Lie, released in October 2019, Cairo has worked as a journalist, designed data graphics, and consulted for companies on the art of data visualization for more than two decades. In this time, he says, “I have observed the multiple ways that people misinterpret visualizations. This misinterpretation happens to journalists, graphic designers, business analytics people—it happens to everybody, including myself.”

In the chapters of his book, he discusses the many ways a chart might lie. It might be a product of poor design or use incorrect data. It might include an inappropriate amount of data, conceal data, or not clearly communicate uncertainty. Or, it might suggest misleading patterns or pander to pre-existing prejudices.

As an example of a visual that does not clearly communicate uncertainty, Cairo cites a map commonly used by the U.S. National Hurricane Center. The map shows what meteorologists call the “cone of uncertainty.” Its design is meant to indicate the likelihood that a hurricane might travel a particular path, Cairo explains—the wider the so-called “cone of uncertainty,” the less certain meteorologists are of where the hurricane will hit. However, many people mistakenly believe that the map shows how the hurricane will strengthen over time. Too often, people “believe they can understand charts with a quick look. They believe that ‘a picture is worth a thousand words,’” says Cairo. “Most of the time, that's not the case.”

We recently spoke to Cairo about the importance of training business students to become adept consumers of charts and graphs—by being aware of their own biases and learning to see what charts and graphs don’t necessarily show. This skill, he says, is the foundation for helping them become skilled and ethical creators of data visualizations.

“Charts—even those not designed with ill intent—can mislead us. However, they can also tell us the truth,” Cairo writes in his book. In other words, he adds, “Good charts make us smarter.” 

How often do you think people deliberately use charts and graphs to deceive?

There are two or three kinds of charts that are purposely designed to lie to people. But, more often, what happens is that we mislead ourselves when we misinterpret visualizations. 

You note the map used by the National Hurricane Center as an example of the potential for misinterpretation. Why does the NHC use a map that’s so prone to causing confusion—especially when people’s lives might depend on it?

Meteorologists are aware that it can cause confusion, but their challenge is that there are no viable alternatives. Other designs have too many shortcomings. I am part of several efforts to design alternative maps, but the problem is that what meteorologists are trying to convey is quite complex. It’s not just the uncertainty of where the hurricane may go, but also the time of its arrival, the watches and warnings, whether you need to be prepared 24 hours from now. None of the many alternatives that have been tried works along all these dimensions.

The “cone of uncertainty” isn’t going to go anywhere for the time being, not only because the National Hurricane Center will keep using it, but because journalists love it. They love it because it seems so clear cut; they can transform it into something a little flashier, a little bit livelier. So, while we should keep trying to redesign the map, I think it would be better to focus on educating those responsible for explaining that map to the public. It’s a real problem when journalists explain the map incorrectly or extract the wrong conclusions from it. 

In your book, you discuss ways that our subjectivity gets in the way of our reason when interpreting a chart or graph. How can business faculty teach students to manage their own subjectivity and become better consumers of visual information?

They should teach students that a chart isn’t “worth a thousand words”; it’s not just an illustration. Charts and graphs are visual arguments that convey ideas. In order to understand any argument, students need to be aware of their own biases. So, one key technique—one that I’ve had to learn myself—is to learn to become more mindful.

It’s completely impossible to control our own biases 100 percent of the time. Our brains naturally analyze the data around us and draw conclusions. We come up with assumptions and opinions, and then we unconsciously seek information to reinforce or persuade other people of the rightness of our opinions. But we can train ourselves to identify when a new opinion is bubbling up inside our brains. We can become mindful and say, “Wait a second. Let me stop and read the chart more carefully. What’s the source of this graphic? What lies beyond the chart? What could this chart be hiding? Is the chart oversimplifying the data? Is it displaying the uncertainty in the data, and if it isn’t, is that uncertainty important enough that it should have been displayed?”

Daniel Kahneman talks about [fast, emotion-driven] System 1 thinking and [slow, reason-driven] System 2 thinking. Professors want to teach students to use System 2 thinking when they read visual arguments.

It seems, then, that for students to become good creators of charts, they must learn to be good readers of charts.

Absolutely. And not only than that, they need to be good writers. To create a good visualization, they shouldn’t begin by launching the software tool, getting the data from Excel, and just saying, “Give me a pie chart. Give me a bar graph.” Instead, I teach my students to first analyze the data to see what the data shows or hides. Then, I tell them to write down a list of goals for what they want the visualization to communicate. What is the purpose? What do they want the visualization to accomplish? This activity will give them a script they can use later to make choices about what kind of headline, intro, or other notations to write; what graphic form to use; what encoding to use. By encoding, I mean design attributes such as length, type, position, shape, color, or any combination of those used to display the data.

Getting all of this right depends on first knowing your purpose. Yes, these visuals should look nice, but their main purpose is to promote understanding so other people can extract trends from the data.

You emphasize that another purpose of data visualization is to persuade—but there’s a difference between using data to persuade and manipulating the data to elicit a certain response.

Yes, we can manipulate data, but we shouldn’t. Ethical thinking is about not only intentions, but consequences. That said, most of the mistakes that I highlight in How Charts Lie are due to oversight, not conscious decisions. This means that even people who are very well-intentioned may end up designing graphics that mislead the public. 

How can students guard against unintentionally creating graphics that mislead?

We must teach them to test their own graphics. That is, before they make a graphic public, they should show it to a focus group and ask, “What do you learn from this?” This can be done informally with friends, with family, with people in their companies. This kind of testing is essential, because in many cases, what you design is not what people see. To bridge that gap, you need to listen to the people who are seeing your graphic and learn what they’re getting from it.

Have you been asked to evaluate a graphic that was misleading in a way its creator did not intend?

Yes. I see it happen most often with the dual-axis charts commonly used in business and economics. Dual-axis charts plot data for one value along one axis and data for another value along the other, so that we can see whether two variables are running parallel or inversely. These charts work well if the person reading them is able to detach one line from the other and not extract the wrong inferences. But the problem is that someone without some knowledge of the subject matter might automatically infer that the two variables correlate, when that’s not necessarily the case. For that reason, dual-axis charts can be extremely dangerous, depending on who the audience is. They are the perfect example to show that what we design is not necessarily what people see.

“Testing Your charts is essential, because in many cases, what you design is not what people see.”

Can you give an example of a misleading dual-axis chart doing real-world damage?

I’ve seen people in the anti-vaccine movement use a dual-axis chart that shows that an increase in the prevalence of autism runs parallel to an increase in vaccination rates. From that, they infer that the introduction of new vaccines causes autism. But correlation does not mean there’s a causal connection.

In this case, while the prevalence of autism has certainly increased, scientists explain that it's due to reasons that have nothing to do with vaccines. For example, the public is much more aware of autism today than it was decades ago, so we may be better prepared to detect it. We’re having children later in life, on average, and having older parents is a risk factor. The technical definition of what constitutes autism has broadened, which necessarily leads to an increase in diagnosed cases. And diagnoses have become more accurate: in the past, people with autism were often misdiagnosed.

How can students become aware of and overcome the tendency to “see what they want to see” in the data?

In his book For Argument’s Sake, British psychologist Thomas Stafford conveys that research shows one of the best ways to persuade people is to first sow the seed of doubt. So, if someone tells you you’re wrong, you don’t confront that person with facts. You just say, “That’s so interesting. Why do you believe that? Can you explain, without appealing to arguments of an authority, what led you to that opinion?” And then let the other person explain.

Students can use this idea to train themselves to challenge their own opinions. For example, they might strongly believe that it’s a great idea or a bad idea for the U.S. to set the minimum wage at $15 an hour. They can learn how to explain their opinion to someone who doesn’t agree with them, without appealing to arguments from an authority, by sticking to the data and creating a rational chain of reasoning. This exercise helps them realize that most of their opinions are on very shaky ground, which makes them become less sure of their own opinions. It’s extremely important to teach young people to doubt themselves.

Are there any other approaches they can use to guard against this tendency to hold too tightly to their own opinions when creating data visuals?

Yes, they can start with the assumption that their project team wants to believe in the chart it has created. Then, they can seek out alternative explanations for the phenomena that they’re seeing displayed in the chart—they can look for counterfactuals. It’s similar to reading a text; if they’re trained in critical thinking, they don’t take everything they read a book at face value. In this way, they can develop their internal detectors of possible bad ideas.

We are all resistant to changing our minds, because our ideas become almost part of our bodies. This makes it painful to change our minds or acknowledge that we’re wrong. However, we can learn. We can all become persuadable, and we can persuade others, if we apply certain techniques. And I’m not talking about using persuasion to sell a product, but using meaningful persuasion to change somebody’s mind for the better. 

How would you teach students to take an ethical approach to data visualization?

At the simplest level, I believe that most people don’t want to lie and that most people don’t like being lied to. So, we can teach students to think more systematically about moral reasoning, which begins with a version of the Golden Rule: Don’t do to others what you don’t want done to you. Then, we can ask them to explore other schools of ethical thinking. For example, deontologists focus on what comes before you make a decision, and consequentialists focus on what happens after you make a decision. If we’re talking about the consequences that graphics have on the public, that’s consequentialist ethics. These discussions can help us teach students how and why they should come up with visualizations that doesn’t mislead people, either on purpose or unintentionally.

Do you have any final message for faculty on the importance of teaching students to view data through the lens of moral responsibility?

Going back to what I mentioned before, it’s important to teach students that a visualization is meant to communicate an idea, to enlighten people. It’s fine to create a graphic that helps them make more money, but only if they also apply the Golden Rule: If they don’t like to be tricked, don’t trick other people. It’s so obvious, so elementary, but it needs to be repeated.

After that, faculty should expose students to tons of examples that show students how easy it is to be misled by graphics, and why it’s important to look at the primary source of a chart’s data. Faculty also should use examples to emphasize how powerful a well-designed and correctly interpreted graphic can be. I talk in the book about the negative side of data visualization, but I have a positive message as well. We should teach students that data visualization is very powerful, persuasive, attractive, and wonderful—they just need to be aware of how to use this tool responsibly.

One last thing I want to emphasize is that we are all responsible for creating a better informational environment. In today’s society, we’re talking past each other, we’re throwing graphics at each other. We need to recover the art of having meaningful conversations. Thoughtful, well-designed charts can help us have more meaningful conversations.