What, exactly, makes one chart better than another?

This is probably the most important article about data visualization that I’ll ever write and many of my future articles will likely refer to this one. How’s that for setting the bar high? Here goes…

tl;dr: When people disagree on whether one chart design is better or worse than another, they often have quite different assumptions about what “better” actually means when it comes to charts. Depending on the person, “better” could variously mean more precise, more creative, more familiar, faster to visually process, more inspiring, more neutral, more versatile, more memorable, or any one of several other quite distinct definitions. People usually don’t realize that they have different definitions of “better” in mind, however, and that this is often at the core of their data viz disagreements.

While precision, creativity, memorability, etc. are important, they aren’t what ultimately make one chart design better or worse than another. Ultimately, charts are tools that we use to cause some desired change in the mind of the reader (answer a question in their mind, change their opinion on something, increase their awareness of something, etc.). The ultimate measure of how good any chart is, then, is how successfully it causes whatever change we wanted to cause in the mind of the reader, not how precise, memorable, creative, etc. it is.


A video version of this post is available for those who prefer watching to reading (17 mins.):


As anyone who’s been paying attention to the data visualization field knows, it has more than its fair share of best practice controversies and experts often disagree on whether one chart design is better than another. What, exactly, does “better” mean when it comes to charts, though? This might seem obvious but, over the years, I’ve noticed that different people often have very different understandings of what makes one chart better than another. In fact, I’ve noticed at least ten distinct definitions of “better” that people commonly have in mind when discussing charts (see list below), although they rarely seem to realize they have these different definitions in mind, and that this is often at the core of their data visualization disagreements.

So what?

If you and I have different ideas about what makes a chart “good” in the first place, we’re unlikely to agree on whether one chart design is better than another, or whether one data visualization best practice is better than another. IMHO, this has been hampering progress in the data visualization field for decades. Specifically, it has…

  • Created a lot of unnecessary best practice controversies among data visualization experts (pie charts, anyone?).

  • Resulted in a lot of unnecessary disagreements on what data visualization research findings mean and how to apply them to day-to-day practice.

  • Created a lot of confusion among beginners since they hear conflicting best practice recommendations and are usually told to sort these out by “just using their judgment” even though, as beginners, by definition, they haven’t developed that judgment yet.

What are some common definitions of “better” that people have in mind when discussing charts?

You’ve probably come across most of these already:

  • More precise/accurate. Many data visualization research studies measure how precisely or accurately people can estimate or compare values in different chart designs, implying that charts that allow people to do those types of visual tasks well are “better”.

  • More creative/beautiful. People with graphic design backgrounds and data visualization competition judges tend to consider that charts that represent data in novel, original, or artistic ways are “better”.

  • Simpler/more familiar. Others consider that charts that use simple, familiar chart types and techniques (i.e., not novel or creative) are “better”, arguing that such charts require less time and effort to interpret and, therefore, are more likely to be fully and correctly interpreted by readers.

  • More versatile. Yet others consider that charts that allow a wide variety of different questions to be answered about the underlying data or that make a wide variety of insights about the data obvious are “better” (even if such charts are more visually complex, i.e. not simple or familiar).

  • Faster to visually process. Some data visualization research studies measure how long it takes for people to interpret different chart designs, implying that charts that can be visually processed quickly are “better”.

  • Slower to visually process. Other research studies measure how long users linger on charts, suggesting that charts on which readers spend more time are “better” because they’re more engaging..

  • More memorable. Yet other research studies measure how much information people can recall about different chart designs after they’re concealed, implying that charts about which people can recall many details are “better”, presumably because a chart can only influence readers if they remember what was in it.

  • More obvious. Many people feel that charts that explicitly state key insights and takeaways in titles or callouts are “better”, presumably because they require less effort to interpret.

  • More objective/neutral. Others consider that charts that “just show the data” or “let the numbers speak for themselves” (i.e., that don’t explicitly state insights) are “better”.

  • More inspiring/evocative. Some argue that charts that provoke an emotional response among readers such as sympathy, curiosity, or outrage are “better”, since they’re more likely to prompt action.

Hopefully, collecting these different definitions into a single list makes it obvious that, if people have different understandings of what “better” actually means, they’re unlikely to agree on whether a given chart is better or worse than another, or on many data visualization best practices in general.

What should “better” mean when it comes to charts, then?

I use a different definition:

“A chart is ‘better’ if it more successfully accomplishes the purpose for which we decided to create that particular chart in the first place.”

Because there are many different reasons why we decide to create charts in the first place, this definition means that what makes one chart design better than another can be—and usually is—quite different from one situation to the next. For example, if we were creating a chart to persuade people to donate to our charity and we were considering two possible chart designs, the design that causes more people to make a donation would be the better chart by definition (assuming it isn’t misleading or otherwise harmful to readers). In that specific situation, the ultimate measure of how good any chart would be is the number of people who donate after seeing the chart, not how precise it is, how memorable it is, how fast it can be visually processed, how creative it is, etc.

Depending on the situation, then, one chart design could be better than another if it…

  • Answers a particular question more effectively.

  • Communicates a particular insight more effectively.

  • Persuades more readers to take a particular action.

  • Allows a more profitable business decision to be made.

  • Gets shared more on social media.

  • Convinces the hiring manager to offer you a job.

  • Etc.

Therefore, in order to have a productive discussion about whether one chart design is better or worse than another, we must first agree on what we wanted that particular chart to do. Until everyone involved agrees on that, any such discussion is literally pointless. In most data visualization discussions that I come across, though, people spend little or no time establishing what the chart in question is supposed to do. Instead, they tend to focus on what I call the “subordinate qualities” of charts, such as how precise they are, how creative they are, how memorable they are, how quickly they can be visually processed, etc.

The ultimate reason why we create charts isn’t to show data precisely, quickly, or memorably, though. Ultimately, we create charts for other people to cause a desired change in their mind, such as an increase in their comprehension of something, a change in their opinion on something, an increase in their awareness of something, etc. If one chart is more successful than another at causing the change that we wanted to cause in the mind of the reader, then it is, by definition, the better chart (assuming it’s not misleading or otherwise harming the reader), regardless of how precise, creative, memorable, etc. it is.

Yes, but is there a fancy Greek word for this way of thinking?

Philosophers sometimes refer to this as “teleological” thinking, i.e., thinking about things in terms of what they’re for instead of what they are. For example, if we’re thinking of a saw teleologically, we don’t think of it as “a tool with a handle and a blade” (what it is), we think of it as “a tool for cutting” (what it’s for). If we’re thinking about a chart teleologically, we don’t think of it as “a visual representation of data” (what it is) but, instead, as “a tool for causing a particular change in the mind of the reader” (what it’s for). Philosophers sometimes refer to a thing’s specific purpose as its “telos”, which roughly translates from Greek as “reason for being” or “ultimate purpose”. I mainly mention this because, when discussing charts, telos is a handy shorthand for “the specific change that we want a given chart to cause in the mind of the reader” or, in other words, “the reason why we decided to create that chart in the first place”.

Are you saying that subordinate qualities like precision, creativity, and memorability don’t matter?

No, those qualities still matter because they tend to improve a chart’s chances of causing whatever change we wanted to cause in the mind of the reader. The importance of each quality can vary widely from one situation to the next, though. For example, sometimes, a chart must allow values to be visually estimated with high precision but, in other situations, low precision is just fine. Or, a creative, eye-catching visual design might be very helpful in one situation, but a time-wasting distraction in another. One of the keys to learning how to create effective charts, then, is learning how to determine which subordinate qualities are important and which ones aren’t, based on the situation at hand and the specific reason why we’re creating that chart in the first place.

I suspect that people tend to focus on subordinate qualities in data visualization discussions because they’re generally easier to use as arguments. For example, it’s easier to argue that one chart is better than another because readers are able to recall 27% more of the information in it afterward (an argument based on a subordinate quality) rather than arguing that it does a better job of answering a given question in the mind of the reader (an argument based on the chart’s telos). Just because subordinate qualities are easier to use as arguments in debates doesn’t mean that they’re the best way to evaluate charts, though.

Well, duh…

While all of this may sound obvious to some, it requires a fundamental shift in thinking that relatively few people seem to have made. Most of the data viz debates and discussions that I come across reflect “non-teleological” thinking, for example:

  • Critiquing a chart’s design without knowing what that chart was supposed to do.

  • Debating “the best way to visualize this data”, instead of, e.g., “the chart design that’s most likely to convince the audience to adopt this point of view”.

  • Arguing that one chart design is better than another because, e.g., “people are able to visually estimate values in this type of chart more precisely”, as if that alone makes that design better.

  • Articles and books that offer “universal” data visualization best practices that, in fact, only apply to charts that are intended to serve certain types of purposes.

  • Research studies that measure one or two subordinate qualities and imply that charts that score higher on those qualities are better overall.

What now?

It’s not hard to learn how to design charts that serve their telos well, but it does take a certain amount of time and practice to “retrain your brain” so that the specific purpose of the chart that you’re designing is at the center of all of your design decisions (selecting chart types, selecting colors, formulating titles, formatting scales, etc.). It also requires “relearning” many data visualization best practices to know when they apply and when they don’t, based on the specific purpose of the chart being designed.

While I’m certainly not the only person who thinks about charts in this way, most of the data visualization courses and books that I come across don’t reflect this teleological way of thinking, which is why I developed the Practical Charts course. When I teach that course, I’m constantly referring to different reasons for creating charts when discussing best practices. For example, in our discussion of when to use a pie chart, the purpose of the chart figures prominently in that decision-making process:

IMHO, thinking about data visualization teleologically has several important implications for the field at large, which I’ll be exploring in future blog posts:

  • Many longstanding, controversial data visualization questions (e.g., “Are pie charts ever the best choice?”, “Must quantitative scales always include zero?”) have answers with which I think most people would agree, it’s just that those answers are more nuanced than simple “always/never” edicts and usually begin with “If the purpose of a chart is to…”.

  • Even though those nuanced best practice answers aren’t as simple as “always/never” edicts, they make it easier for beginners to learn how to create truly useful charts.

  • There’s never a “best way to visualize a given type of data” (e.g., time series, breakdown of total, etc.), there’s only a “best way to visualize a given type data for a given purpose”.

  • There’s no such thing as a “neutral” chart that “just shows the data”. Every chart design makes certain types of insights more or less obvious, i.e., serves different teloses.

  • It’s possible to prove that one chart design is objectively better than another; it’s not just a matter of personal opinion or preference.

I hope that the ideas in this article and in the writings of others who think about charts teleologically will spread more widely among data visualization students, practitioners, and researchers. If this happens, I believe that it will allow us to get past some longstanding debates and formulate better best practices that make it easier for everyone to learn how to design truly useful charts that serve their telos well.

By the way…

By the way, if you’re interested in attending my Practical Charts or Practical Dashboards course, here’s a list of my upcoming open-registration workshops.

Thanks to Enrico Bertini and Chris Tauber for providing valuable feedback on an early draft of this article.