Nick Desbarats breaks down the surprisingly common mistakes we make when visualizing data–and shares basic principles for communicating data more effectively.
- Why most charts are confusing or ineffective
- The top three mistakes people make with charts
- Why a “neutral” chart is an ineffective chart
As an independent educator and author, Nick Desbarats has taught data visualization and dashboard design to thousands of professionals in over a dozen countries at organizations like NASA, Visa, Bloomberg, Shopify, and the United Nations. He delivers main-stage talks at major data conferences and is a guest lecturer at Yale University, and his new book, Practical Charts, is an Amazon #1 Top New Release.
- Book: Practical Charts: The Essential Guide to Creating Clear, Compelling Charts for Reports and Presentations
- Website: PracticalReporting.com
- Article: “My favorite chart type”
- Article: “Connected Scatterplots Make Me Feel Dumb”
- Article: “I’ve Stopped Using Box Plots. Should You?”
- Article: “Why I Stopped Using Bullet Graphs (and What I Now Use Instead)”
- Chart type: Marimekko
- Chart type: Strip plot
- SlideShare: a16z – Adreessen Horowitz
- Book: The Elements of Style by William Strunk, Jr., E.B. White, Test Editor, and Roger Angell
- Book: Thinking, Fast and Slow by Daniel Kahneman
- Book: How to Measure Anything: Finding the Value of “Intangibles” in Business by Douglas Hubbard
- Book: The Comfort Crisis: Embrace Discomfort To Reclaim Your Wild, Happy, Healthy Self by Michael Easter
Thanks. I’m really delighted to be here. To be honest, I’ve been listening to the show, and it’s kind of an honor to be here. It’s a fantastic show. I’ve really been enjoying it.
Oh, thank you. Well, I’m honored to be chatting with you. I’ve been loving your book, Practical Charts. First, I just want to ask, you’re a very sharp guy, and I want to know. Of all the places you could be investing your energies to enrich the world, why have you decided to go deep on charts?
That’s a good question. My career path, can be summarized as circuitous, as in very indirect. I started out in software, and as a software developer, I got kind of bored of that, and then kind of moved around software organizations for a bit, doing some sales, marketing, that kind of thing, product management, product design.
And in my 30s, I kind of stumbled on to a lot of research around from the field of psychology, Daniel Kahneman, Amos Tversky, cognitive biases, psychology perception. I was just smitten. I just inhaled that information, which I figured was kind of a sideline interest to my kind of real job. But then I went to a workshop, a data visualization workshop from Stephen Few, who’s one of the big gurus in this field in 2013, and it’s just mind-blowing. It really just opened up a whole new kind of field for me that I really wasn’t aware of, and that combined my two major interests, which were basically psychology and data.
Because, as I think it will come out probably in our discussion, there’s a lot of psychology when it comes to designing charts. And so, I just went whole hog into that and I, actually, started teaching these workshops in 2014, and it was a huge privilege, and I did that for a number of years. Steve then retired in 2019, and then I used that as kind of an excuse to start developing my own courses and workshops. And I’m still extremely interested. I have not gotten bored yet, that’s for sure.
All right, Nick. So, I get the memo that you absolutely love this stuff deeply and dork out over it, as do I. Tell us, what’s really at stake for the professional in terms of whether they become Master Jedi-level with their charts versus can fumble their way through PowerPoint just fine like the rest of us?
Well, I’m not sure I would agree with that last part of your sentence, fumble through PowerPoint and sort of be okay just like the rest of us. I think, to be honest, if you haven’t had some formal training, basically, in this kind of thing, then you’re probably hitting a lot of problems and a lot of which you might not even be aware of. You might be leaving your audience, for example, with an incorrect understanding of the data, or they’re confused but they won’t say anything because they don’t want to look stupid.
Or the problems could be more obvious. They might actually be complaining about your charts, and saying, “This is just unnecessarily complicated,” or, “I don’t get it,” or, “What’s the point of this chart?” I like to compare it to sex and sales because these are two other things that people think you are kind of born knowing how to do, it’s like, “Well, what’s the big deal? Creating charts, how hard can it be? Select the data in Excel, and hit Make a chart, and Bob’s your uncle. There’s your chart.”
But like both of those other things, if you haven’t actually kind of learned the basics of how to do it, you’re probably not doing it very well. We’ve all had bad experiences with bad salespeople, for example, and it’s just because they just didn’t know how to sell very well. And it’s kind of the same thing with charts. There’s more to it than I think most people realize.
In fact, I kind of think of data visualization, i.e., kind of the process, or the expertise of making charts, as kind of almost like its own language, and until you’ve learned the basic kind of spelling and vocabulary of that language, you’re probably not communicating very well, whether you realize it or not. And so, many of the charts that I see are full of these kinds of basic, what I call, kind of spelling and vocabulary problems with charts, which are things like poor chart type choices, scales that are too wide or too narrow, poor color choices, and just a whole host of other problems.
And so, reading a chart like that from the audiences’ perspective is kind of like reading a poorly written document, a document that’s full of spelling errors, and grammatical mistakes, and weird word choices. And so, it’s going to be really hard for them to read it, which means they, oftentimes, are just going to skip it, they’re not going to read it. Or, if they do, they could be very confused by it, or, worse of all, come away with an incorrect understanding of the data.
And this is something that happens a lot more often than people tend to realize. And we’ve all seen charts that deliberately misrepresent data, but what a lot of people don’t realize is that this also happens accidentally way more often than people realize. And so, if you had some training in the sort of spelling and vocabulary of data visualization, you’re going to avoid all these problems, and then you’re going to create charts that are just way easier and quicker to read and understand, and, ultimately, way more likely to sort of accomplish whatever purpose prompted you to create a chart in the first place.
We don’t create charts for no reason. There’s always a reason. We’re trying to explain something to somebody, we’re trying to persuade them to do something, or make them aware of something, and all of those things are much more likely when the basic kind of spelling and vocabulary of your chart is competent, is done well, just like a document that’s written well.
Well, well said, Nick. Okay. So, if you don’t have some formal training in charts, and you think you’re doing fine, you may very well be accidentally misleading people, and they could be murmuring behind your back about how bad your charts suck. Or, even then, maybe if the people you’re presenting…
Or, to your face.
Or, to your face. I guess, even if the people you’re presenting your charts to are not as sophisticated and able to discern what’s jacked up about your charts, I think I like that analogy to writing is it’s sort of like they might just meet your data with a shrug, like, “Yeah, okay.” Sort of like a piece of writing can be riveting like a page-turning novel, like, “Oh, my gosh, what’s going to happen next?”
Or, just like, “Okay, I guess,” and you’re just sort of tuned out, so it’s like folks aren’t even able to receive what can really be, and I guess I’m a bit of a dork here, but I don’t think it’s a stretch to say, if you understand the story some chart sequences are telling you, they can be heart-thumpingly thrilling. I mean that in all sincerity.
Hey, man, yeah. You’re in my tribe.
Okay. And so, if you’re thinking, “Yeah, Pete, I’ve never seen one like that,” I’m thinking about some of the folks who I think do it amazingly well that are available for view might be Andreessen Horowitz at A16Z. They’ve got a number of slide presentations that were on SlideShare and still, I believe, publicly available, which really do, they take you through a story, and you’re like, “Oh, wow, so that’s what’s going on with whatever tech sector, or investment, or whatever. And I really feel like I’ve read a novel, except in the artform that is a sequence of charts.”
Yeah. Well, I mean, data storytelling is a big buzz term right now. Like, over the last few years, it’s just really taken off. And I think it deservedly so. I think, though, what has gotten maybe less attention but is still really important is, like I said, that sort of basic spelling and vocabulary, because a lot of what I see is sort of data stories are kind of torpedoed because of really basic chart design problems.
Because, oftentimes, a data story essentially consists of a series of charts, just like you were describing, but the chart types are wrong, the scales are too wide, the colors are weird, the labeling isn’t precise enough, and so users don’t actually even understand what the numbers in the chart represent correctly. They’re just having to think too hard, having to read a 45-degree or a vertical text, and so the way I sort of look at it is, yeah, storytelling is great, and it is a skill that I think a lot of people should be developing but before you do that, learn the basic language first.
You can’t tell great stories, you can’t write great essays or great novels if you can’t spell. And I think that there’s a lot more awareness of that now than there was, like there is a spelling and vocabulary to this. And if you haven’t really mastered that, then your stories are going to flop. You’ve got to kind of walk before you can run, essentially.
And, unfortunately, a lot of people, well, a lot of people do realize that, but lots of people don’t. And so, they’ll jump straight into courses or books about storytelling and data storytelling without having really mastered the basics first, and then they wonder why their data stories aren’t working.
Beautiful. I want to hear, and so we talked about these basic fundamentals, and I think you did such a fine job of coming up with the nuanced distinctions in your book, Practical Charts. And starting from the very beginning, I think you say we even start with the wrong question, which is, “Okay, I got a bunch of data,” and you think, “Okay, what’s the best chart type.” And you say, “Hold up. That is not the ideal first question.” Set us straight, Nick, what should we be asking ourselves?
Well, you’re right, of course. Typically, when we sit down to create a new chart, we ask ourselves, “What is the best way to visualize this data?” And I think when you’re sort of maybe starting out, that is the question that people often have, but I think once you start to develop more experience, more expertise, you start to realize that, “Actually, the question I should be asking is, ‘Do I know why I’m creating this chart? Is there a problem that I’m trying to make the audience aware of? Am I trying to persuade them to do something? And if so, what is that thing? Am I just trying to explain something to them? And if so, what is that?’”
Because, until you’ve figured that out, you can’t really make any design choices. You can’t really choose chart types. It’s even hard to know even what data you should even be showing? Should you be showing the last six months of data or the last 12 months of data? All of these things depend on what I call the job – the job of the chart.
And so, really, I think that’s one of the things that I try to accomplish in the book, and also in the Practical Charts Course that I teach, is by the end of the book, you should be thinking of charts as graphics for doing a job, and not visual representations of data. Because if you think of charts just as visual representations of data, well, then even really bad charts would be fine because they are visual representations of data, but only good charts do their job. And so, you want to aim for a chart that does its job.
And so, at the end of the day, ultimately, that’s all that matters. People tend to get hung up on this sort of secondary characteristics, like how precisely people can estimate the values in a chart, or how fast they can read it, or how much information they can recall when the chart is hidden from view. I mean, they’re important but they’re not the thing that, ultimately, matters. What, ultimately, matters is, “Did the chart actually do whatever thing you wanted it to do? Did it do the job that prompted you to create that chart in the first place?”
And it might sound a little obvious but it requires a huge mental shift, and I can see it happening during my course just by the way that I’m teaching it in person. It lasts two full days, and it takes about that long to really fully make that leap to that sort of new way of thinking.
So, charts are graphics for doing a job and not mere visualizations of data. And so, I think when it comes to jobs, maybe you could lay out the menu for us. Because I think, sometimes, I find that the job is, “Okay, we’re being persuasive. We are trying to make a sales pitch, and the goal is that, whatever graphics we’re including will make the point that we are really awesome,” or, “This market opportunity is a big deal, so okay.”
But I think other times, in the course of day in, day out working with colleagues, they might say something like, “Hey, Nick, how about you put together a presentation to give us an update on where we stand?” So, it feels kind of vague or generic or broad or general, just like, “Show us what’s the state of things right now, or over the last month.” And so then, how do you think about choosing charts for that kind of a job?
Yeah, you’re right. There tends to be a lot of focus when you look at different books and courses on data visualization, articles. They tend to kind of assume that we’re always trying to persuade people or something. But you’re right, you’re absolutely right. That’s not always the case. In fact, very often. We’re just trying to explain something to somebody, make them aware of some interesting trend, for example.
And so, yeah, and I try and sort of address that in the book and in the course as well to say it’s important to understand that these charts can have a very wide variety of different kinds of jobs. And, in fact, sometimes we’re creating charts just for ourselves, like charts that no one is ever going to see. We’re just using the chart for analysis. We’re using it to discover new insights and patterns in the data. And that is just a completely different use case than something where you’re abusing storytelling, for example.
And so, one of the things that I find is a little bit sort of, maybe even a little frustrating, is that people think that, “Oh, every chart has to tell a story. Everything is a data story.” It’s like, “No, some things are data stories but not everything.” We’re not always trying to persuade people to do something. Sometimes we are but sometimes we’re not.
And, by the way, sometimes when we get those kinds of requests that you were talking about in terms of, like, oftentimes, we’re just not even asked for an update; we’re just asked for data, where, “I need to see a breakdown of expenses by department for the last 12 months.” And those are actually very tough situations because the obvious next question that I think you should be asking is, “Well, why? Like, are you worried that there are certain departments that are spending too much? Or, do you want to see how they compare to their budgets?” There could be all sorts of ways to respond to that request.
And, unfortunately, a fair amount of the time, if we try and get that information, we can’t, it’s like, “Well, it was the CEO who asked, and they’re really busy, and so just give me a chart.” And it’s like, “Oh, crap, now I’m in a position where I have to try and create a chart, and I don’t know why I’m creating it.” And so, I have a whole section in the book about how to deal with this, and I have a technique that I call spray and pray, where you, essentially, create multiple views of the data, and you make some guesses about what question they might have in mind, or what they might be wanting to know, and you build different visuals for those three, sometimes even four, different potentials reasons why they’ve might asked you for that information. And you hope that one of them is going to hit.
Nick, I think that’s so great, is that often it’s just a clarifying question away in terms of, “I want a status update of how things went over last month.” And if you ask a couple follow-up questions for clarification, it can be quite illuminating, it’s like, “Yes, I’m looking for cost savings opportunities within our operation.” “Oh, okay. Well, then I’m going to think about things differently,” versus “I want to see what looks weird, or different, or off, or broken so that I can allocate my energies to preventing a problem before it gets worse.”
Or, “I want to see what might be some of the most compelling opportunities that we need to go after in the subsequent months.” Any of those very different directions could spring forth from a, “Hey, just give me an update.”
Yeah, exactly. And so, we’re lucky in those situations where we can actually ask those follow-up questions and get that follow-up information. And the first step, of course, realizing that you need that information in order to design your charts in the first place. But there are times when we ask and we don’t get answers, it’s like, “No, just give me an update,” or, “Give me expenses for the departments, and don’t ask me any more questions.”
And so, I think it’s important to know how to address both of those, but, really, the key thing that I think is, the step that people miss, is that step of figuring out, “Okay, why am I creating this chart in first place?” And that’s a crucial step. Because if you don’t know, if you don’t have some kind of insight that you’re trying to communicate, or some kind of question that you’re trying to answer, most of your chart design choices will be random, and your chart will end up communicating random insights, which is not helpful.
Okay. Well, I’m going to get into some very particulars, but before I do that, I’d love to get some general principles in terms of what are perhaps your top tips, or principles, or mistakes you observed as folks are trying to do this kind of thing?
One misconception that I see a lot, especially amongst people who have more experience, who have more expertise, is that they believe that creating a chart or getting good at data visualization is just something that sort of requires experience, and trial and error, and intuition that’s developed over a long kind of period of time. And that’s what I believed for a long time as well.
But what I realized through teaching Steve’s courses, and now my own courses, is that it’s actually possible to distill a lot of these guidelines into surprisingly precise guidelines that don’t necessarily rely on having years of experience. And so, that was sort of the impetus, really, for me creating my course, and then the book that went along with it, is I was a little bit frustrated by the fact that people said, “Well, if you’re showing the breakdown of a total, sometimes it makes sense to use a pie chart, and other times a bar chart, and other times a stacked bar chart. Use your judgment. Do what feels right.”
And I was like, “Hold on a second. No, actually, these chart types are not interchangeable. There are specific circumstances under which it makes sense to use one or the other.” And so, really, that’s kind of, I think, a bit of a different approach that I brought to the field, and it is kind of, in some cases, it’s a bit controversial to say, “I think that we can actually sort of codify or formulate a lot of these guidelines in ways that can be applied by people who have even very little chart design experience.”
And they can follow steps and have a number of decision trees, in the course, and in the book as well, and you can just follow through the decision tree, and it will point you to the right chart for the right situation, the right design choice, or an expert-level design choice, anyways.
Oh, Nick, I love that so much, and that really conveys that same analogy I’m reminded of, of like writing, in that some folks are just like, “Well, it takes a lifetime to really refine your writing style and to make it excellent.” And then you got Strunk & White, The Elements of Style who just dropped, “Remove unnecessary words.” And, like, that’s really a pretty good rule almost all the time. And it’s like, “Oh, okay, just by doing that, my writing is better.
Yeah, exactly. And that’s actually, coming back to that language analogy, sometimes the way I describe sort of at least a lot of the books and courses that I’ve seen about data visualization is that imagine English as a second language, or you don’t speak English, and you’re trying to learn the difference between “they’re,” “their” and “there” the three ways to spell “there.” And your textbook says something like, “Well, this is actually kind of a nuance. It’s sort of ca omplex question. And over time, you’ll develop intuition which will sort of help you figure out what is the best spelling.”
And so, it makes it really hard and slow to learn the language, but as native English speakers, we know, it’s like, “No, actually, I can give you very simple guidelines which you can learn in, like, 60 seconds, which will point you to the correct choice every time.” But I think the difference with data visualization is that formulating those sorts of simple-looking guidelines and decision trees was actually really hard. It’s like the hardest thing I have ever done. And so, it’s not surprising, I think, that it’s taken a long time for those kinds of simple-looking guidelines to emerge about data visualization.
Well, Nick, could you maybe give us top three guidelines in terms of this makes a huge difference, and mistakes happen all the time?
Yeah, probably number one is chart type choice. The most common problem I see in charts is something that was a line chart when it should’ve been a bar chart, or it was a stacked bar chart when it should’ve been a pie chart, for example. And I forgive people for making these kinds of, what I consider to be, mistakes, anyways, because there are a number of considerations that go into those kinds of choices.
But because it is so tricky, and there are so many factors to take into consideration, like, for example, if you’re just trying to figure out how to show the breakdown of a total, there are at least eight things you need to take into account in order to decide between sort of the five major chart types for doing that: your pie charts, bar charts, stacked bar charts, etc. And so, yeah, that’s probably the most common mistake that I see, and the solution is, well, you’ve just got to get a bit of training to know how to do this.
Probably the second most common type of problem that I see are problems with quantitative scales. So, these are the scales of numbers that you see in charts, like the number of employees, or dollars, or whatever. And I have a whole section in my book about that, and it kind of surprises people, because they’re like, “Isn’t that pretty straightforward? Like, why not just go with the default scales that come out of Excel, or Tableau, or whatever?”
And I’d say probably, maybe a third of the problems that I see with charts are related to quantitative scales, scales that are too wide, too narrow, start at zero when they shouldn’t, don’t start at zero when they should, have too many stops on them, for example, or not enough. And so, there’s a lot to learn about quantitative scales. And so, again, if you haven’t had that training, then this is a very common way that charts, essentially, misrepresent the underlying data.
So I guess the third most common would be labeling problems, usually insufficient labeling, or insufficiently precise labeling. And so, these are situations where you see a chart, it’s maybe a line chart, that says the quantitative scale is just labeled with transactions, and maybe it’s for over 12 months or something like that. Okay, is that like successful transactions, or successful and failed transactions? Is it accumulative total of transactions running throughout the year? There could be all sorts of ways of interpreting that.
And so, with inadequate labeling then, once again, the audience might assume that they’re looking at numbers that aren’t the actual numbers in the chart. And I would also kind of put in that labeling of key insights. This is sometimes controversial when I say I’m a big proponent of actually putting messages right in the chart, “We have a problem because transactions have been declining since July,” and actually putting that, like write it as a collar, or maybe even as the title of the chart. People tend to shy away from that but I think that there are good reasons to actually be really explicit about, “If I had a reason for showing you this chart, I might as well tell you what it’s for.”
Nick, I totally resonate with that. And it’s intriguing when I trained on this sort of thing, I’ve been accused of having sensational slide titles or headlines, I was like, “Wow, if these are sensational then you are accustomed to very, very boring…” I’m not swearing, I’m not using extreme language. I’m just saying things like, “Sales of this segment have dropped radically since this quarter.” It’s like, “Huh?”
And I guess that is sensational but I guess what’s really driving it, and my observation, is fear. And so, like, “You’re basically saying that the guy in charge of that thing over there is a screwup and a failure.” I was like, “No, I didn’t say. I’m just commenting on the most noteworthy thing that is to be gleaned from these data.” But it seems that folks are often, in many cultures, quite shy about calling a spade a spade because it has all sorts of emotional implications under the surface.
Yeah, I think that there’s really kind of two ways that that problem surfaces. The one is what you just described, where you’re basically saying something that’s kind of maybe politically sensitive. And that happens, unfortunately, a lot. As the people who handle the data, we’re often the first to see the bad news. We’re the canary in the coal mine. We’re the deliverer of bad news.
But I think that there’s another kind of knee jerk or inherent objection that people have to putting any kind of interpretation in the chart at all.
A lot of people think that that’s actually kind of unethical, that we’re biasing people’s interpretations of the data, and that charts should be these kinds of neutral interpretation-free, just the numbers kinds of representations of the data. And this all sounds great. It sounds perfectly noble. I don’t think it’s even theoretically possible though because this kind of relates to what we’re talking about before. When you create a chart, you have to have a reason for creating it in mind, a question you’re trying to answer, an insight you’re trying to communicate, an action that you want somebody to take. It’s baked into the chart.
Because if you don’t have that in your mind, you don’t have some specific job or thing that you’re trying to accomplish with the chart in mind, then you don’t get a neutral or unbiased chart. You get a chart that produces random insights, essentially. And so, because our interpretation of the data, and why the audience needs to see that data is baked into the chart, anyway, it’s in all of our decisions, it’s what we based the chart type choice on, it’s what we based our color choices on, it’s what we based our scale ranges on, and a whole slew of other choices, we might as well just tell them, “This is why I think you needed to see this data.”
They may disagree with it. They may have a different interpretation of the data, and that’s fine. Then you sit down and talk about it, and say, “Okay, we seem to have different views of reality in this situation. Let’s figure it out together then.” But the solution is not to try and produce these sorts of very generic “interpretation-free” charts because, like I said, that’s not even theoretically possible.
Your charts will always have your interpretation of the data built into them anyways, so you might as well kind of save the audience a bit of brain cells and just tell them, “This is why I think you needed to see this data.” And, like I said, if they disagree with it, that’s fine. Then you talk about it and try and get on the same page.
All right. Well, Nick, I want to challenge you if we could have the rapid version of thinking about chart type and axis scale matters. Could I have the two-minute version of when is the absolute best and absolute worst conditions for using a pie chart?
Yes. So, if you’re not very familiar with the data visualization field, you might be surprised to learn that pie charts, they are very controversial. The community is split. You have people who are violently opposed to pie charts, and those who think that they are just fine. And I’ve been in both camps. I used to be an anti-pie charter but then I sat down a couple of years, I had a long hard look at my reasons, and realized, “You know, there are valid use cases for pie charts.”
Pie charts have a couple of unique properties. The first is that they allow us to perceive fractions of the total much more quickly and precisely than any other chart type. Compared to a regular bar chart or stacked bar chart, I can immediately see, “This is about a quarter of the total,” “This is about two-thirds of the quarter,” “These two parts together represent about three-quarters of the total.” This discussion can get a lot more nuanced though, and, in fact, I just wrote a 3500-word article in the journal of the Data Visualization Society last month about this, and it went pretty viral because it is a big question.
But ultimately, that’s what I think is a major point that people miss around pie charts, and people who don’t like pie charts, is that they allow people to perceive fractions of the total much more quickly and precisely than any other chart type. Plus, the fact that it’s a pie chart immediately tells the audience that they’re looking at the breakdown of a total before they would’ve read anything. They don’t have to read the chart title or the labels or anything. They immediately know they’re looking at the breakdown of a total.
Whereas, with a regular bar chart, for example, they actually have to read the chart title, and the labels, and figure out, “Oh, these parts of a total. They’re not, for example, values over time” Whereas, the pie chart, it’s like, “Bang!” It’s instantaneous. So, they do have some unique properties that make them, I think, the best choice in specific situations. But knowing what those specific situations are requires a bit of training.
Okay. All right. So, that’s the thing. It’s like if what we are all about is quickly and intuitively conveying the proportion of one segment relative to the whole, the pie chart can do that pretty intuitively. But if we’re venturing into other territories, like, “Let’s see how these proportions have shifted over time,” then maybe the pie chart is not going to be our friend.
Yeah, or if, for example, you want to compare the parts very precisely, say, “Okay, here’s a breakdown of our sales by region.” But the main point of the chart is to show that we sold more in the West than we did in the South. Well, especially if those values are very close to one another, you should use a bar chart because one of the weaknesses of pie charts is that they don’t allow the parts to be compared very precisely to one another. But if your main insight is that, for example, the West plus the East accounted for more than a third of our sales, well, that’s going to be a lot more obvious in a pie chart than it will in a bar chart.
Okay. And what’s the top thing we should never do with our axis scale?
That’s a tricky question because, as I mentioned earlier on, there are a lot of mistakes that people make all the time with quantitative scales. If I had to pick just one, I’d probably say that it’s starting the scale at zero when it shouldn’t be started at zero, or vice versa, not starting at zero when it should’ve been started at zero.
And this, again, is one of those questions that people tend to think has a really simple answer but it doesn’t. Like, I have a whole section in the book on how to make that decision. It’s not as straightforward as a lot of people think. And, by the way, that’s kind of the case with a lot of these design choices. People tend to think that they can be made very simply. Like, for example, when it comes to choosing chart types, a lot of people think, “Oh, well, if you’re showing data over time, always use a line chart. Or, if you’re showing the breakdown of a total, always use a pie chart.” But unfortunately, those are simple, yeah, but you’ll often make bad design choices.
And so, whenever I see very simplistic rules, like, “Always start the scale at zero,” or, “Never start the scale at zero,” unfortunately, they’re just too simple. You’re going to end up making bad design choices all the time if you rely on those very, very simple rules of thumb. It doesn’t have to be really complicated but it can’t be that simple. It needs to be a little bit more complicated.
All right, tell us, what would be a horrible context situation for us to start an axis at zero?
So, I guess the classic example here would be body temperature. Let’s say we’re in a hospital and we’re tracking the temperature of a patient over time, and whether you’re working in Celsius or Fahrenheit, if you start the scale at zero, well, first of all, it’s going to be hard to see small shifts that could be very meaningful. If you’re going from, I don’t know…
Ninety-eight point six Fahrenheit to 102.
Okay, yes. Or, 37° Celsius, choose your methods there. But if it just goes up two or three degrees, of course, that’s often very meaningful from a medical perspective, but you’re not going to see it very well if the chart starts at zero. And there’s another wrinkle in that situation as well, which is when you’re talking about something like temperature, at least on the Fahrenheit or Celsius scale, zero is kind of a meaningless number.
Zero degrees, Fahrenheit, for example, is not the absence of heat energy. That would be zero degrees Kelvin, which is something that’s totally different. And so, I would say that, yeah, in a situation like that, starting the scales at zero would be a huge mistake.
Okay. Thank you. Well, now you mentioned earlier, before we’re recording, that you have a bit of a reputation as a chart type killer, which feels like that needs to be a lyric in a rap song or something.
The nerdiest rap song ever. I would listen to it.
And so, I noticed in your book that there was nothing, there was no mention of the Marimekko, or Mekko, and when I was in consulting, we were utterly infatuated with the Marimekko or Mekko. And we’ll link to this in the show notes if you all never heard of it. Sometimes it’s used as, for example, a market map. So, we might have on the X-axis, maybe we’re talking about different kinds of computing processors.
And so, on the X-axis, we might have phones, tablets, laptops, desktops, servers, and so we see, “Okay, so that’s the relative proportion of different segments of different devices that use processors.” On the Y-axis, we might see how much penetration percentage a given player in that field has, like, Apple versus AMD versus Nvidia.
So, we like to use that in consulting to show, “Hey, Nvidia, look how you’re doing nothing in tablets,” for example, “But all your competitors are. Maybe you should, too.” That’s often how that goes on but you’ve got a different point of view. Let’s hear it.
Well, yeah, so I don’t discuss Marimekko Charts in the book because I tend to find that…well, maybe sort of coming back to one of the basic principles that I have. I’m a big advocate, of course, of showing the data in the simplest way possible that still communicates whatever it is that the chart needs to communicate.
And in my experience, it’s very, very rare that the simplest way to say what you need to say about the data is with a Marimekko Chart, which is kind of a complicated chart type. There’s a good chance you’re going to be needing to explain it to people, and there’s a lot of kind of moving parts to it. You have the heights of the bar segments, as well as the widths of the bar segments, and so it tends to be kind of hard on our working memory, the part of our minds where we do all of our thinking, which is actually very small. We can only think of a very small number of things simultaneously.
And so, if I’m thinking of using a chart like that, I always look for “Are there simpler chart types?” It might even be a combination of charts. I might have two or three charts but that are going to be sort of simpler to consume, and yet that say the same thing about the data. And so, I’m not saying never. It is possible where the simplest way to say what you need to say is a Marimekko Chart. It’s just in my experience it’s usually not. Usually, there are simpler alternatives.
I do mention chart types, though, like box plots, for example, and connected scatterplots, which I think are virtually never the simplest way to say what you need to say about the data. And this has generated a certain amount of sort of response when I’ve published articles about why I don’t use box plots anymore, for example.
But I’ve just found that things like strip plots, jittered strip plots, stacked histograms are virtually always much easier for audiences to understand because box plots are pretty abstract, if you even know what it is. A lot of your audience probably won’t even know what a box plot is, and they require lots of time to explain, and there are virtually always simpler ways of saying what you need to say about the data.
And so, I wouldn’t necessarily put Marimekko Charts in that category in terms of, like, they’re never the simplest way, but there often are simpler ways of communicating the information.
So, yeah, there are a couple of chart types that I think fall into that category of never the simplest way. Like I said, box plots, connected scatterplots, and bullet graphs, for those who know what those chart types are. There are virtually always simpler alternatives. And I have articles about all of these. Maybe we can link to those in the show notes as well.
Yeah, that’s really resonating on a couple dimensions when it comes to box plots. I think the first time I encountered the concept of a box plot, I had to think about it for, like, 20 minutes and look at the box plot and then the percentiles. But then once I did, I was like, “Oh, okay. Cool.” It’s sort of like I had to do that hard work of understanding the concept of a box plot. And then when I saw them later, I appreciated them. But if you haven’t done that, then it’s going to not resonate. It’s, like, you’re in a different language.
And, likewise, with the Marimekko’s, I remember I was on a consulting project, and we were sort of showing a number of employees by country on the X-axis, and by function on the Y-axis, so we were using these Marimekko’s. And we had a client who hated the charts so much, he forbade us to make another one. And then I had a colleague who made one, nonetheless, and shared it with the manager, who said, “Didn’t you hear the guy? He said no more Marimekko’s.” And the consultant passionately pleaded, “It’s the best way to show it.” And so, he was shot down because the client tends to win these sorts of debates.
So, yeah, point taken. We can fall in love with a thing, and in so doing, lose connection to the audience and where they’re at.
Yeah, that’s a great point, too. I wrote a blogpost with a very clickbait-y title called “My favorite chart type.” It was I guess about two years ago, where I basically argued that “This is actually something we should try to move away from, like having favorite chart types,” because, really, that can only make our chart type choices worse. We’re going to be biased towards using certain charts, even when they’re not necessarily the best choice.
It’s kind of like people who have their favorite words, and they tend to use those words all the time even in situations where it’s really not the right word. And so, I think one of the marks of somebody who’s gotten really good at this is that they don’t have favorites. They just use whatever chart type is most appropriate for the situation.
The catch is that it just takes some time to learn when to choose from these various chart types. In my course, I cover 50 chart types because I think that all of those are needed in kind of everyday, when you’re making everyday charts for reports and presentations. And it takes a while to learn when to use them all.
All right. Well, tell me, Nick, anything else you want to make sure before we hear about your favorite things?
I think, really, the point that I was hoping that was going to come out in the discussion, and I think it really did, is to encourage people to really start thinking about charts as graphics for doing a job rather than visual representations of data.
All right. Well, now could you share with us a favorite quote, something you find inspiring?
There’s one that I really like from an American journalist called HL Mencken, who’s active in the 1920s. And he said that, “For every complex problem, there’s an answer that’s clear, simple, and wrong.”
All right. And a favorite study or experiment or bit of research?
I think pretty much anything by Daniel Kahneman and Amos Tversky, the sort of godfathers of the study of cognitive biases. That has just informed my thinking in innumerable ways since I first came across it. It’s well-summarized in Kahneman’s book Thinking, Fast and Slow.
All right. And maybe that is a favorite book. Any other favorite books you want to highlight?
Yeah, actually, but I’m going to cheat. I’m going to give you two. One work-related, which is How to Measure Anything by Doug Hubbard. It’s an absolutely brilliant book.
And then in terms of kind of general kind of books about living a good life, there’s The Comfort Crisis by Michael Easter.
All right. And if folks want to learn more or get in touch, where would you point them?
My website is PracticalReporting.com, all one word. And if you go on the top nav to the Contact/Follow page, then there’s my email form and where to follow me on LinkedIn. And I invite people to do that.
Okay. And do you have a final challenge or call to action for folks seeking to be awesome at their jobs?
I think it’s really important to develop this basic skill, like the spelling and vocabulary of data visualization. There’s a very rapidly growing awareness that this is something that a lot of people probably need to learn because, of course, so many of us are now handling data as part of our job. And, really, to me, that’s kind of the starting point. Before you start learning about data storytelling, or anything like that, learn the basics of the language first.
All right. Nick, thank you. This is fun. I wish you many fun charts.
Thank you. Yeah, I really appreciate it. Fantastic discussion.