The internet is brimming with tidy lists of data visualization best practices, and their definitive confidence can be quite comforting. But at the end of the day, when we ask ourselves, “How can we make our data visualizations better?”, the answer is usually some derivative of, “Well, it depends on the visualization.” Our goal here is to offer an alternative to one-size-fits-all rules, and encourage a more nuanced strategy for improving our work. With cognitive load as our guide, we use twelve unique spectrums 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.
Cognitive Load + Data Visualization
Let’s start with a quick primer on cognitive load. The deeper you venture into the dataviz world, the more you’ll hear this term pop up, and for good reason. Cognitive load describes the amount of working memory someone uses when they take in new information and transform it into their long-term memory. In simpler terms, cognitive load helps us consider how easy or difficult it is for someone to make sense of something new.
Cognitive load has its origins in the psychology of instructional design: think teachers designing lesson plans for students. The concept comprises three different types: intrinsic, germane, and extraneous loads. Yep, we’re throwing around some technical jargon here. But stick with us, because these three types of cognitive load align beautifully with the three components underpinning every data visualization.
Intrinsic Load | The Data
Intrinsic load describes the new information’s inherent complexity. No matter how you slice it, learning calculus is more complex than learning basic addition. As instructors or analysts, intrinsic load is where we tend to have the least amount of control. The information or dataset we’re tasked with explaining comes with all its gnarly complexity (or lack thereof) baked right in.
Germane Load | The Audience
Germane load has to do with how familiar the audience already is with the new information they’re processing. Can they tether that new information to some preexisting framework, or are they starting with a blank slate? The same recipe will look different to a curious novice than to a professional chef. Likewise, the same visualization will look different to an average citizen than to a trained statistician. Here again our control can be relatively limited. Sometimes we pick our audience, but oftentimes we don’t.
Extraneous Load | The Visualization
Last but not least, extraneous load has to do with how the new information is actually presented. The goal is usually to match the content to the most intuitive presentation possible and therefore minimize the amount of extraneous load. Sure, you could verbally describe an equilateral triangle or the rhythm of a waltz, but it would be faster to just draw the shape or clap the beat. As a teacher or analyst, extraneous load is where we have the most control, and therefore, we should consider it last. That enables us to ask ourselves: “Given the existing intrinsic and germane loads, how much more cognitive load are we comfortable adding to the mix?”
When we add up these three components, the resulting cognitive load might be light, heavy, or somewhere in between.
simpler information + knowledgeable audience + pared-down delivery = lighter cognitive load
complex information + novice audience + detailed delivery = heavier cognitive load
(See how we just dialed back the extraneous load?)
12 Spectrums: It’s all about creating a good fit
So with that foundation, our goal becomes accurately assessing and calibrating the cognitive load in our own data visualizations. The following 12 spectrums are designed to help you do just that. Each spectrum below takes a different data visualization component and breaks it down, with lighter cognitive load on the left and heavier cognitive load on the right. The idea here is to take your visualizations and work through each spectrum to tally up its corresponding cognitive load. The answer may surprise you.
Just a quick note before we dive into the details of each spectrum. Remember that there’s no value judgement associated with either side of these spectrums. It’s not inherently right or wrong to strive for a low or high cognitive load. Both simplicity and complexity have an important place in the world. Once we’ve accounted for sound data collection, truth analysis, and accessible design, it all comes down to fit. You could have the most beautiful, information-rich, exploratory visualization, but if your target audience is a C-Suite commuter with 10 seconds and minimal context, you’re out of luck. Conversely, an aggregated dataset made into a succinct bar chart is wasted on a roomful of domain experts eager to dive into the details and gather their own insights. Visualizations are like a puzzle piece that connects your data and your audience: to be successful, your creation must respect the contours that already exist on either side.
Data Spectrums: How simple or complex is your data?
First up: the dataset. These four spectrums will help you evaluate whether your visualization’s intrinsic cognitive load is light or heavy. While this is typically where we have the least amount of control, a little data cleaning can tip the scales. As you tease apart the characteristics of your data, take note of potential opportunities to lighten some cognitive load by aggregating noisy categories or creating more intuitive calculations, like percentages or indexes.
1. Measurement (quantitative → qualitative)
Let’s start with that age-old question of making data: can it be counted? Is your data’s measurement quantitative, qualitative, or somewhere in between? Put another way, does the measurement have a predetermined or obvious unit like dollars, points, miles, or milligrams? Or is it squishier, like a rating scale from 1-5 or from “very dissatisfied” to “very satisfied”? Or is it a concept quantifiable only by proxy, like “happiness” or “democracy”?
2. Knowability (certain → uncertain)
Knowability has to do with our level of confidence that our data is true. Some things are easy for humans to know. Some things are downright impossible. Is your data easy or difficult to collect? Could the methodology skew the data? Does the data directly answer the question being asked? Can you reduce the level of uncertainty with more data? Does it come from a contained universe, like the teams in a sports league, or does the data come from an extrapolated sample, like the number of estimated swing voters in a given state?
3. Specificity (precise → ambiguous)
As humans create data, we inevitably create categories, the specificity of which varies widely. Are your categories clean cut, like the scientific classifications of blood type or species? Or are your categories blurrier, like the socially determined concepts of class, gender, or race? Are the categories fixed or in flux? Would most folks produce the same categories, like political parties, or would the groupings depend on who you asked, like categorizing wealth? Aggregation can introduce ambiguity as well, sometimes as an important tool to anonymize data. Does your data have a high-resolution granularity or a low-resolution aggregation?
4. Relatability (concrete → abstract)
Finally, let’s look at how relatable your data is. Does your data describe something super concrete or does it capture a more abstract concept? Do folks interact with these items or ideas in daily life? Are you dealing with quantities in common amounts, like coffee or rent, or in astronomical sums, like a billionaire’s wealth or GDP? Is your data easy to visualize, like a family of five, or impossible to bring to mind, like the entire population on earth? If your data is leaning toward the abstract, are there analogies you can offer to anchor it to something more familiar?
Audience Spectrums: How much bandwidth does your audience have?
Next up: the audience. Picture the folks who will actually consume your visualization, whether they’re a group of three stakeholders or a conference session of 3,000. As you work through the next four spectrums, remember: you are not your audience. That’s the biggest blunder we make as analysts, and we make it all the time. You and your audience will—and should—land in different places on these spectrums. Return to the cognitive load framework: your audience is always lugging around way more germane load than you are. After all, you’re the one knee-deep in the data, slogging through the details, actually building the visualization. That process matters. It’s why consuming your own work is infinitely easier than consuming anyone else’s.
5. Connection (intentional → coincidental)
Let’s start with first impressions. How will your audience end up looking at your visualization? Did they seek it out? Did they subscribe to your newsletter, get a ticket for your talk, search for your work by name? Or did they stumble across your piece? Did your visualization pop up in their social media feed or happen to be in the waiting room’s magazine? Keep in mind that the circumstance of the connection will often affect the audience’s headspace. Are they feeling interested, engaged, supportive? Or are they feeling ambivalent, frustrated, or skeptical?
6. Pace (slow → fast)
Ok. Now you’ve captured your audience’s attention. How long do you have it? Three seconds as they scroll through a morning news brief? Five minutes on the board meeting’s agenda? Or is your intended audience settling in for the next hour with a fresh cup of coffee and your annual report? Are you personally dictating the pace by leading a live presentation? Or will the audience skim through your work independently and close the browser the moment their interest wanes? Get honest here. We all want folks to take more time to engage with our work than they typically do.
7. Knowledge (expert → novice)
Now let’s evaluate your audience’s content knowledge. Are folks super familiar with the data? Did they help generate the data themselves, like sales reps examining their own progress? Or should you assume your audience has no familiarity with the data whatsoever, like a general audience walking in for a TED Talk or opening the morning paper? Or perhaps your audience’s knowledge base lies somewhere in between. They’ve got a basic foundation to build upon, but will need you to provide some additional scaffolding if you want them to follow you to that “ah-ha” finale.
8. Confidence (confident → anxious)
Finally, set aside the data and consider your visualization’s format. How much experience, if any, does your audience have consuming new information through this medium? Has your audience been primed on how to get the most out of this type of interactive dashboard, longform essay, or video format? Or will they need to learn how to engage from scratch? If you’ve got an audience full of format newcomers—maybe even visualization newcomers—take the time to show them how to learn, not just what to learn. Otherwise frustration or embarrassment will take root and your audience will tune out before you’ve communicated a thing.
Visualization Spectrums: What visualization best connects your audience to your data?
At long last, it’s (finally!) time to create some visualizations. You’ve taken stock of the data’s inherent complexity and your audience’s level of preparedness. With a sense of how light or heavy the combined intrinsic and germane cognitive loads are, you can start making some informed design decisions. Do you have the breathing room to lean into a more complex visualization and dial up the extraneous cognitive load? Or is your audience already facing an uphill battle with a complex dataset, and your visualization needs to keep any extraneous load to a minimum? These final four spectrums will guide you towards a presentation that enables your audience to best understand your data.
9. Chart Type (common → rare)
Let’s start with the chart type itself. Is your audience used to reading this kind of chart? Does it have standard x- and y-axes, like a bar chart or a line chart? Is it a map of a familiar place at a familiar scale? Or is the chart type more obscure? If you handed the chart to a colleague, would you feel the need to preface it with a quick verbal explanation? Are the axes unusual or missing altogether? Count up how many variables are encoded in your visualization. Will the audience understand the chart immediately, or will they have to slow down and build the chart’s logic from scratch?
10. Interpretation (accurate → approximate)
Humans are adept at reading some types of visualizations, and lousy at interpreting others. Is it important for your audience to take away exact values, or will a general understanding of basic ratios or relationships suffice? Are you asking your audience to compare highly legible components like length or position? Or are you asking your audience to draw an approximate interpretation like angle or area comparisons? Accurate color differentiation tends to be more limited than we often assume. Does your visualization leverage just two or three colors or will your audience need to reference an extensive color legend?
11. Composition (concise → detailed)
Next evaluate the overarching composition of your visualization. (We’re already assuming good foundational design here. The chart junk is gone, the data to ink ratio is up, etc.). How much information is on the page? Is your audience looking at a few aggregated bars with the key takeaways baked into the title? Or are they considering dozens of individual points in a scatterplot, hovering over each one for additional information? Are you providing the context for the visualization outside of the graph itself, through a presentation’s previous slides or an article’s surrounding paragraphs? Or is the visualization packaged up with its own contextual annotations that will need to be read?
12. Delivery (explanatory → exploratory)
Lastly, consider the final delivery of your visualization. Are you taking an explanatory or exploratory approach? Put differently, who’s in the driver’s seat: you or your audience? Will you walk the audience through your argument on a curated path? Or have you built an interactive environment in which the audience is expected to explore and make discoveries for themselves? Are you a domain expert who’s already analyzed the data and now wants to present your findings? Or are you a visualization expert democratizing a previously inaccessible dataset with a new approachable format?
Cognitive Load as a Guide: Immediacy Isn’t Everything
Our conversations around data visualization are heavily influenced by society’s broader pendulum swing towards immediacy. And given today’s onslaught of information and pervasive sense of overwhelm, who could really blame us? In fact, data visualization’s unique ability to expedite insights is one of its greatest strengths. But when we talk about cognitive load in data visualizations, the message is often a one-note insistence that cognitive load is fundamentally bad. We seem to have imported the term from psychology, but flattened its complexity along the way. By presenting these twelve spectrums, our hope is to offer a more nuanced perspective. We see cognitive load as an excellent guide: it can lead you to the visualization that will enable this particular audience to understand that particular dataset.
No visualization should have a heavy cognitive load because it’s poorly made. But plenty of well-made visualizations carry a heavier cognitive load. Just take a look at the visual essays by The Pudding or the portfolio of information designer Giorgia Lupi. These works aren’t instantaneously absorbed, nor are they intended to be. They remind us that many things worth knowing take time to unpack and a bit of elbow grease to understand. That’s not inherently a bad thing. In fact, if we slow down long enough to engage, we’re often reminded how much this human brain of ours enjoys untangling the world’s complexity.
We’ve created a summary print out you can download here.