Amanda Cox, Bloomberg, about making better simple charts

"Simple"?

[00:00:04] Amanda Cox: Hi, thank you for having me. So Lisa asked me to talk about better simple charts today, or when she asked, I thought that is an interesting choice, in that if you know the work in my heyday when I was making charts, as opposed to managing or whatever else it is I do now, I don't think simple is the first adjective that comes to mind, right?

We're making string graphs, and we're making squishy Voronoi tree maps, we're making connected scatter plots and connected scatter plots of second derivatives of things. This one came back from the Flash Graveyard recently, and I was like, my goodness, what was going on over there on that stage?

We were making network diagrams that were not just network diagrams, but I was encoding the error in like little loops in the connections between people. I was really obsessed with error for a while. So Simple, I was like, Simple, that's an interesting choice, but it is also in some ways a choice in my career that does make some sense.

About two years ago, I left the Times. I joined this little startup in Seattle, where the design ethos is these like single little kind of pink sparklines, single line charts. And about six months ago, I joined Bloomberg and Bloomberg is an organization that makes all kinds of charts. But I think the ones that actually make the money are like high frequency line charts. I don't... that's not an official statement. I don't know that for sure.

It's also true that in my personal life, the charts that I stare at now are pretty simple charts. The ones I spend the most time with are these ones out of Google Sheets about my net worth and my weight that I've taken the labels off to be polite, but very simple charts.

And so I thought, okay, maybe I have some things to say about better simple charts. So there's four pieces of advice that apply to charts that you make in Datawrapper or elsewhere for simple things that I think can make them actually better.

Use your words

[00:02:00] Amanda Cox: The first one I've been preaching for more than a decade, which is to just use your words.

My favorite example of that is not something I did myself. It's an example I've stolen from Lee Byron and David McCandless back when Facebook was more popular. They made a chart of peak breakup times based on Facebook messages. Except they didn't make the chart that looked like this, they made it actually with an annotation layer.

And so if you look at the chart this way, you can see that there's a little Valentine's Day bump, there's a spring break bump. Two weeks before the holidays things really pick up and Christmas slows down. And I like this chart because in some ways the topic is so simple, right? It's not an advanced thing to recognize that there's some seasonality in this. Or if you know when Christmas is, or December is, to guess when springtime is. But the annotation layer on it, the words on the chart, make it a completely different experience, and Datawrapper happens to be one of the few products that actually take annotation seriously and think about ways that can make charts better.

My classic other example of talking about this is an old Jobs Day chart example, where normally you can just make a bar chart on Jobs Day that says ‘The economy added this many jobs, and the unemployment rate is whatever percent it is’. This one was before a presidential election, so we were just having some fun, and we said, look, you can put on your Democrat goggles, and say, there have been 31 months of job growth, and the unemployment rate has fallen more than two points since its recent peak.

Or, you can put on your Republican goggles, and you can say, Look, the economy's at 150,000 jobs a month, just to keep up with population growth, and the unemployment rate has been above 8 percent for 43 months now. That's the sort of example that... These are complicated examples about annotation, about how using the words to actually pretend your chart has a message changes things a lot. Or can change things a lot.

I realized that in particular some of the words on charts that are most interesting to me are verbs and some of the art that is most interesting to me are verbs too. So if you know the artist Richard Serra, you may think of things like this. This is in Bilbao, I think. He makes these giant steel sculptures, and it's always been a puzzle to me why they are so impressive to me. Because ships are the same, and ships actually sail in the same way, but I find them stunning. But earlier in Richard Serra's career, when he's trying to figure out what sculpture is, or what it is that he's trying to do, he made this list of verbs.

What am I actually trying to do here? I'm trying to rotate, to swirl, to support. To hook, to suspend. And I find that verb list really inspiring to think about in our own work. That the verbs are the thing that drive something. That the verbs are the words you might use in your chart, in your headline, in the annotation to really change things. When I was first getting started in my career, I spent a couple of years at the Federal Reserve Board. And so you say okay, that's fine for artists, but it doesn't apply to the kinds of charts that I am making. At the Fed, there was a list of verbs that people passed around there too, about describing the sizes of increases or decreases, in increasing order.

So something could tick up, inch up, edge up, firm, or it could collapse, plunge, plummet. And there's also an accompanying list of adverbs that you are encouraged to use sparingly. Tad, touch, bit, slightly, those kinds of things. So I think that the idea of verbs changing things, being serious about using the verb with your chart instead of just saying, this is a chart of x. So tell me what the chart is actually showing changes things a lot. The example that has inspired me most on that recently has been Matt Conlen, who's been playing with aggregating public data and presenting it and visualizing it. Automatically, right? Without human involvement. He's pushing on like some of the AI sentences.

And so this week, he said he was proud of a chart, proud of the system that he's been building. And they tackled an inflation report. Automatically, this code was able to say, the consumer price index rose whatever percent, half a percent last month. And it surpasses expectations for the fourth consecutive month.

That kind of context, I believe, makes this a much better chart than if you lose the verbs entirely. If you just say, this is a chart of the consumer price index. All of a sudden, if I don't know much about it... And machines are doing this now by themselves, right? So you have to be better than this, I think, to want to claim that you should still be in the craft, right? Like you have to do better than the baseline. And so I think, using the verbs, using the verbs is a really important thing. That's my number one tip, especially when I go talk to academics. And say, look, it doesn't need to be a sexy verb that you put in your headline. You don't have to push the bounds of meaning. You can say "is", "is doing" something. Just whatever I'm supposed to see in the chart, use a verb.

Eat your cereals

[00:07:03] Amanda Cox: The other, my other advice, especially for academics, especially if you've ever been to an academic poster session, and you see how they design their layouts to think about this.

There's no sense of hierarchy at all. And a tip for that is to remember what a cereal box looks like. If you've ever been to an American grocery store, you've seen an aisle like this, these things, these boxes, mostly marketed to people who can't read, honestly. So a giant font, maybe a cartoon character, maybe an advertisement of a toy.

But there's another side of the box, and then there's the nutrition information on the side of the box, where it will tell you this cereal is actually just like eating half a cup of sugar. It's really 50 percent of sugar, is what's in the box. But the thing to note is that information is not the same size as the other information, right? You have some control over size, and it's tempting to think, in the era of mobile graphics, none of this hierarchy stuff matters anymore. And I think that is completely, completely untrue, right? Like if you've seen some of the classic design hierarchy things. If you've ever seen this example, probably you read the biggest text first and then, if it actually works and if it actually feels like magic, if you read the text at the top last, you're like, huh, they have control over me. Even you have control over how someone else's eyes react to something, or how many people. It's not a hundred percent of people, but how many people encounter your work. And so where I think that shows up in charts, where I think that shows up in a charting tool like Datawrapper especially, is your choice of what the words you are using. Or what you're using in your headline, right?

I am a believer that the headline should tell you what the chart shows. This is a simple chart one of my colleagues made up at Bloomberg recently. And that headline, it just tells you what it is. Whatever the spank increased, failing, it made the penalties paid to something else increase, right?

That's a concept I don't know anything about, but the headline tells me what's going on in the chart. And both of the newsrooms that I have worked in, the New York Times and Bloomberg, have changed that style, right? At one point at the New York Times, the style for what the headline should be, it was just like a jokey little... You would, like, throw away words. You would say "a tale of two cities", or "getting warmer?" It didn't tell you actually what the chart was, it was basically throw-away words.

Bloomberg, same idea, but changed that style more recently about what the headline should be. My claim to you is if you remember the cereal aisle: we're gonna read the biggest thing first. So it's crazy to use that headline to not just tell me what it is. Tell me what the chart is showing. Use a verb in that headline. Use a verb in that space. That's the biggest thing you have. So it should be bigger than less important stuff. In particular, if the units on your chart are less important, that should be definitely smaller, right? And if you think you still have hierarchy, even if you're making a simple chart for mobile, in a tool out of the box.

So if we go back to cereal for a second, this is a chart made from one of my opinion colleagues, right? Breaking the headline rule a little bit. Like this is just a chart about how much sugar is in a cereal box. At Bloomberg, the official style right now is to describe what the chart is in a noun in this readout.

And that works fine for some charts. For a chart like this, that's like a pretty simple straightforward chart, I think it's fine, totally fine, to use that space to just say, this is a chart of sugars as a percentage of the weight of what's in your cereal. A little bit of cleverness going on here to introduce one package of cookies amongst the cereal. So that Honey Smacks, the box that we were looking at, it really is like 50%. You want to make that information, if you're selling Honey Smacks, smaller than the other information. That is the style. Using this text here, using the deck, using the readout to describe basically the units of the chart is a perfectly fine choice for very simple charts.

I think when you want to start to get just a tiny bit more complicated, it stops being such a good choice. For example, if you look at a chart like this, which is one made by one of my colleagues, which is about how much is sold on Black Friday. Black Friday is a very important retail day in America.

So at different stores this year. So Amazon, it's as important as about three average days. Crocs, it's like a much higher percentage of their whole sales over a year. So both Amazon and Crocs had a pretty good Black Friday. Their sales rose over the year. But, I'm thinking, this chart is pretty complicated. If I just look at it right now out of the screen, it doesn't work, right? I can't, because I need this space, I need this deck to lend me a hand. To tell me what it actually is, and to break Bloomberg style and use that space to actually use the words to tell me what it is, instead of just repeating the units.

I never really thought that very much about it before, right? Because I was raised in a certain way about what should go there, I had beliefs. And sometimes, some of my friends send me, like, taunting examples. Now this is an example, it came out of the opinion section. It's an interesting idea, but you'll see this footnote that's "Note: Time on y-axis is a 24-hour clock expressed in decimals," and you're like, what is going on over there? Similar, not so dissimilar from that, what is going on in the flash era of the charts we started with? Why that is, is because as a tool, the charting tool that Bloomberg uses assumes you are putting your unit in this space. Whereas the person who made this chart decided instead to use that space to describe what it is.

In part, the reason I like Datawrapper as a product is it gives me control to play by reasonable rules within a reasonable set of defaults. So I said, okay, you have two minutes. What do you do, if you want to keep the units on this chart, but make them normal, right? So the way I was raised, you tuck the units in the axis when you can. So we just say here's a time in recognized units that I understand.

The other thing I noticed myself wanting to do when I made this chart was that if this chart is supposed to be about younger Americans eating dinner earlier, I should just make it be about younger Americans eating earlier, right? I have control over things like the color, and if I want the young one to be the most important, I can clearly make it the most important. And that is, tip number three, which can apply. So: Use your words, tip number one. Tip number two: remember your cereal box, that you have some control over hierarchy, even when you are doing pretty simple things.

Be brave

[00:14:04] Amanda Cox: Tip number three is just be brave. My friend Lena Groger, who now is the graphics director at ProPublica, I like some of her slides. She is the one who crystallized this idea in my own head, and her phrasing of this is don't be a wimp. So if it's going to be different, for goodness sake, make it actually different.

And the detail that she filled in for me on that is that you can't contrast 12 point font with 13 point font. And you can't contrast dark brown with black. So if you're going to be different, make it different. When we talk about this, we often talk about this in terms of design choices, right? So if you're gonna make your colors different, make them actually different. If one piece of text is clearly more important than the other piece of text, you want me to definitely read the headline first, and definitely read the units. Only if I'm confused and absolutely need them, make one a lot bigger and bolder than the other, make one a lot smaller than the other.

But I think it also applies to what is the data that you're putting in your chart. When I was working on this talk, I was thinking about, what are the charts that are my favorite, that are the sort of charts that you can make in Datawrapper, or you want to make in Datawrapper, you should make in Datawrapper or a tool like that. I thought about a chart like this. How effective is the Pfizer vaccine for COVID-19? And the reason that this chart is fun is because you don't have to squint at it. Like you don't need the confidence intervals to see that the lines are different.

I thought about the chart like this. This is what's the price of sardines you buy on the beach in India, before and after the introduction of mobile phones. Before mobile phones showed up in a place, the price was all over, right? It was crazy. When phones show up in a place, it becomes a true market and that does a lot to smooth out prices. This chart, I think, is a beautiful example of a chart that lets me do some thinking. So the idea of small multiples about, okay, if you just told me what happened in this first region, like, maybe it's because of phones, maybe it's not because of phones. But then when I start to add the second region and have to say, okay, this giant change, this change that I don't have to squint at, like a bold, non wimpy change happens over and over again. And now I can start to use this duck chart as a vehicle for thinking.

I also thought of a chart like this. This is a distribution of marathon times. I like this one because it's actually about a bigger idea. You'll notice, the chart, it's not smooth. If you think of most random histograms, they're very smooth. This one there, it's lumpy, right? There's way more marathon finishers at four hours than there are at four hours and one minute or four hours and two minutes. And that's because goals are important. I think the big idea around this... The data, hyper specific, how fast does it take people to run a marathon. But the idea behind it: "Are arbitrary goals important?" clearly shows, they are.

This chart. Structure is important, too. This is another chart that I love. It's about how many people apply to liberal arts colleges in America, depending on how well they're ranked in one ranking. And that there's this big drop off after 50 schools, right? The top ranked liberal arts school, controlling for some other stuff, gets the most applicants and then goes down and down, and there's a big drop off. And that happens to correspond to where there's a page break in the printed version of these college ranking guides. So there's actually no difference in the world between the 50th best school and the 51st best school.

I like these because, as data, they're structurally interesting, the patterns in them. There's also enough data that you have a chance of seeing if something's wrong or weird, right? I want to know what some of these dots are. Like, they've plotted me data that's actually rich enough that I have a chance of understanding the patterns in it. Which isn't true for all types of forms.

Similar kind of one where I just want to know more about the countries. This is someone on Reddit looking at, how does the tallest flagpole in a country... Does that have any association with the amount of freedom in that country and it turns out that countries with really tall flagpoles, they tend to play at places with really low freedom scores. It's an interesting idea, but I love seeing the dots. Like I just want to know more. And I think good charts do that. They say, here's one level of detail. Give me enough to think about more.

I contrast that, in some ways, to thinking about a bar chart. You may know that I'm not the biggest bar chart fan in the world. And that's because when I picture them in my head, I can remember so few of them that actually reveal structure. That's all, you should sort it, and then they all look the same. There are times, however, where even bar charts, I close my eyes, and I can remember. I remember this chart of the different prices of different liquids, in part because the comparisons are so interesting.

It turns out that printer ink is about twice as expensive as human blood and way more expensive than vodka and Red Bull and crude oil, on a per liter basis. This is a chart that I remember because the choice of what went in it was an interesting choice. You don't have to stay so confined to data that comes from a single source, that comes from a single place.

You see that in the earlier Bloomberg opinion cereal example. It's fine if it's just a chart of cereals, but once you throw the cookie into it, now it becomes a more fun chart because I have different typesets of knowledge to link it to. If this was just a chart of printer ink, there's no chance I'd remember it. But as soon as it becomes a chart of printer ink and blood, now I remember.

It also happens to be a terrible chart. That's a chart that came out of the Excel default, in an era when the sort of weird 3D and legendy stuff... Like all bad ideas. But I remember it because the choices of the data are interesting.

Think if you should

[00:20:06] Amanda Cox: Which leads me to my fourth and penultimate point. It's: Think about whether you should. Think about what you're doing. What you're trying to do. Datawrapper as a tool mostly encourages you to make excellent choices. Mostly the guidelines are all there.

Not all tools do that. For example, the Bloomberg charting tool, I think, encourages people to make a lot of charts like this. This is a chart of how many migrants have new court cases in various parts of the United States. It's a chart that is hard to extract a pattern out of. I looked at this chart recently and said, what are we actually trying to say with it? Can we just tame this chart into something that has an actual point?

So this is a chart that the tool out of the box encourages you to make. You can show all the data. The question here is do you need to. Could we be closer to having a point with this chart for existing and extract any meaningful insight out of it at all? If instead of showing all of the data, we just reduced it, we'd reach something. Still not a great chart, but it's at least a chart with a message, which I think is one tiny step towards getting better.

That's the thing I type most now in my work Slack. I think there's some of these false hierarchies in the world that more charts are better in people's heads. And with some of my colleagues, too, they'll see these plans and it'll be like, and here are the seven different charts I can make. And constantly I'm typing: one good chart is better than seven okay charts.

And the other false hierarchy I hear a lot is, especially now that I'm not formally attached to a graphics department, it's that interactive is better than a static chart. Which is not true at all, right? At least not true in terms of some kind of insight. I covered this at my last job at USA Facts. I personally believe, and this is a radical belief, that it's one of the exceedingly rare places where Datawrapper lets people go wrong. Or not lets, but encourages people to go wrong some ways. It's the tutorial where it says, look, I can teach you how to make tab charts. Which are appropriate in some places. So, like, you can have this kind of false interactivity, where you can click on charts. Here's a chart of COVID cases. Like we made it for Spain and you could see it for Germany or Italy or whatever. This kind of false interactivity is appropriate for some places where you are just saying really, here I am to just share the data with you, right? This is really a vehicle for you to be responsible, you the reader, to be responsible for the insight. It's not really on me. I've made that slightly easier for you visualizing it.

For a long time I was showing this doctored screen grab from the New York Times about how silly that would look if on a banner headline day we titled the whole page like just here's some words. There's something in there. That is the equivalent of sometimes what we're doing when we make these dashboard type things. Which is okay if you know that is exactly what you are doing. That's my goal. The problem becomes when there's confusion about whether you're pretending to play in a space where you're offering insight and you're really just making a display of data.

Leave fingerprints

[00:23:08] Amanda Cox: Which leads me to my last and final point, which is, and this is a point that I've been playing with in my head a lot lately. This is about: my favorite art has fingerprints. Fingerprints left in it in some ways, right? The reason I have been thinking about this is I left The Times about two years ago now. I can still tell when an editor there named Jonathan Corum makes a bar chart. And you would think that there's nothing at all in a bar chart where there should be fingerprints of authorship, right? Where you can recognize the style. Where you know exactly... Partially, it's because Jonathan uses this blue in his palette a lot, like way more than anyone else. He often contrasts with this little green, so there's like easy kind of tip offs right away.

Partially, it's about the thoughtfulness with which he does things. Like he, Jonathan, is a thinker, and so the labels are always drawn to be exactly where you want them, nothing more, nothing less. He does a really good job of telling you what the chart is trying to show, so like this one from a few weeks ago was that people taking this drug have 30 times the odds of reporting abnormal behavior. That is a full sentence. That is not "here is a unit of some whatever", right?

But there's also a delicateness to Jonathan's work that I've never been able, even myself, to be able to quite mimic. And the thing when I think when I look at some of the style transfer work that is going on with AI now, there's part of me that wonders if we are not so far from anyone being able to pick up that delicateness, and some of that intentionality.

Matt Conlen has proven to us that the words part is already taken care of or very close to being taken care of and being spread. I have an easy time imagining a future where you're at least presented with options out of the box about what this chart is attempting to show? Or what is the pattern that you see? And often with charts, like we saw with that jobs example, there's not a single pattern, there's a few.

I also think that innovations are ready to be made in the design space about what it is that Jonathan Corum is doing with a bar chart that makes it so lovely, and the hierarchy is right, like you read the first things in the right way. I believe that is going to be easier by machine in the not so distant future. In the meantime, if you don't have access to any special technology or any special thoughts on that, I do think that the thing you can do no matter what, which is my final takeaway, is just use some verbs.

Use some verbs

[00:25:45] Amanda Cox: Use verbs in your Datawrapper chart headline. Take advantage of the fact that Datawrapper's annotation is serious and designed by people with serious charts in mind. And you do that and you think about if you actually have something to show. Think about if you combine data from a different place entirely. If what you had to show would become more revelatory, more interesting. And you will be well on your way to making better simple charts.

So we have a little time left and I'm happy to take questions if there are any and if not, we can all have a break.

Q: Tips for helping others create better charts?

[00:26:20] Lisa Muth (host): Thank you Amanda, that was wonderful. Thank you so much for your really great examples and for driving home so many points. I really like your heuristics. I do spend a lot of time thinking about what really helps people to hear when they create charts, what helps them to create better charts and, so I really liked your heuristics, like: If you want to make it different, make it different. Headlines should tell you what the chart shows. Yes, definitely my opinion, too. One good chart with a clear point is better than seven okay charts. Interactivity isn't necessarily better. I wonder, like, how do you bring these heuristics in this editing process that you're doing? As I understand it, you've been an editor for such a long time, helping others to create better charts. How do you go about it in practical terms? Do you have any tips for attendees here who edit other people's charts? What helps them?

[00:27:19] Amanda Cox: I think that editing charts and in particular simple charts, it depends, where they are, right? So I work with a team now at Bloomberg that is titled Data Journalists. So not focused so much on design, on data visualization, it's more in the reporting and the analysis.

That comes out of a field where I have been astounded by the number of bar charts that I see that are not sorted. And I'm like, good grief, what is your background? What is this coming from? And so some of it is that kind of really, just thinking about, what is it that I'm attempting to show in this thing?

My current test is do I want to talk about this at dinner? And that's partially more in the idea space than in the graphical design space. And then my other thing is asking yourself, do I need it all, right? There's a temptation sometimes, to be, like, here.

My advice for people editing really simple charts is to get good enough at a tool, and it doesn't matter what the tool is. My favorite, actually, truly, I'm not just saying this because we're at Datawrapper, it is Datawrapper. To be in a place where you can make the chart almost as quickly as you can think, right? For a relatively simple chart. So you want to be able to try on choices. You want to be able to iterate. And so for me, when it's sketching with data, like totally unprocessed, that used to be R, and really now it's often closer to Datawrapper, is that you just try it on and you're looking for the best one.

And then when you're editing, it's also thinking about What are the units on this chart? Do I actually understand it? Do I understand thinking a question in your head? Do I understand where this data came from? So there's sometimes a tendency to just be like, "source: Bloomberg", and I'm like, maybe that's true, but Bloomberg probably got it from somewhere else.

So are we actually using the space in the way that we could use this space. And so the editing of charts, I think really is mostly in my field, which is not true for everyone's field, but it's mostly about: does this have a sharp point? If it doesn't have a sharp point, is there data that we should throw away or acquire to make it have a sharp point? Or are the design choices that we can make to make it have a sharp point. So can we emphasize something like, if you have a sharp point, you should emphasize that sharp point. If you're like, no, it's all like an equal hierarchy, like maybe that's okay for what you're trying to do, but maybe you need to think more about what you're trying to do with this chart, as opposed to just, we have some data and we could make a chart.

So I think that's my number one editing tip is it's not about arranging the data in a space. It's arranging data such that some pattern or comparison to scale or some insight is clear. And maybe you have a chart, maybe you don't have a chart, maybe you need to do more reporting to get a different chart. When you're reporting the data, you want to be thinking about structure, if you want to make good charts, right? You want it to be deep enough so that if there is a true pattern, you can see the true pattern.

That's all messy. But the answer really is like, it depends. It depends what the chart is.

Q: Pitching only one great chart?

[00:30:19] Lisa Muth (host): That's some great advice. Thank you. We actually start to get some questions from the audience, too. Let's get this one up from Elliot. Great talk, Amanda. Thanks so much. In my experience, it's easier to sell seven mediocre charts to editors rather than one great chart. What's your advice for pitching the latter?

[00:30:39] Amanda Cox: No. If you work at an organization that has scroll depth, you can say no one's even gonna read the last four charts, right? And we know that. It's not "no one", but rounds to no one, in reality, if you are making public facing internet content. If you're working a different job, I cannot speak for you.

So is anyone actually going to read it? I think it's also helpful sometimes to have the comparison side by side. And so here's the one good one. We could also tack eight things on the bottom. But go home and sleep on it and tell me tomorrow what you remember. And by the time tomorrow you remember that's oh, yeah, there were just like charts I could scroll through like I can, it's not a thing.

So my advice is like one, except that the way the internet works now..., not entirely, like there are people actually thinking in the thread of seven charts. So I'm not opposed to seven charts in general. If you think of, John Burn-Murdoch is a master of this, right?

Like he makes these Twitter threads where the charts actually build on each other. So if you got to make seven, I think, some of you just have a boss who's just a mediocre person and just thinks that seven is better. You can say okay, I can rise to that occasion and think about how these build on each other and end up as tools for thinking as opposed to like miscellaneous things strung together.

And then I would say I bet if you look at the analytics of whatever your organization is, I bet it will be clear that the top chart gets at least twice and possibly ten times as many eyeballs as the bottom chart. And so that is one argument that is possible to make too.

Q: When's interactivity a good idea?

[00:32:18] Lisa Muth (host): Thanks. Perfect. And we have another one from Catherine Smith. In which cases do you think interactivity is a good call?

[00:32:27] Amanda Cox: Yeah, I am pro interactivity. Interactivity is definitely a good call when people want to find themselves in a chart, right? So there are certain types of data, that... I remember, this was many years ago, but one of my colleagues, John Carter at first had children and his editing philosophy changed, thinking about schools, because he was like, I just want all the data. I want all of it possible. And I want to know everything possible about my situation. There are also cases, where you expect people to bring a lot of knowledge to the thing. So I'm thinking about in the United States, there's things like census data and neighborhoods and block level data, where it's really me thinking about the context that I know as a human being and can bring to this chart is going to make it such that such a personalized vocal is the thing that is really helpful.

I think there's other cases where the detail is interesting or helps you understand things that are wrong, like some of the examples that we looked at today. Like that Reddit one, I do want to know what the countries are that have the shortest flagpole, or the highest flagpole, and the lowest amount of freedom, right? And you can get those on with labels, but if you start to think instead of, 30 countries or whatever that is, if I had 200, at that point, where the detail is actually helpful, when you want to drill into the detail, that's another case for it.

Gregor has made this argument super well. He has a lovely talk on it, pushing back many years ago, about the anti interactivity space, saying no, it's useful for thought. It's useful for catching errors. And so if you want to think more about this, I would definitely encourage you to... I think it's called "in defense of interactivity" or something. And it's very thoughtful and very good. And I co- sign everything that Gregor says in that space.

Q: Balance between too much and too few words?

[00:34:18] Lisa Muth (host): Yeah, agreed. Me too. We have two more questions about text and data visualization. You made lots of points about that: use your words, use verbs. Leon Hohmann is asking: I see many colleagues using only one or two keywords for a chart and don't tell what the chart is about. On the other hand, you can have way too much text in a visualization. How can I find the right balance?

[00:34:44] Amanda Cox: Yeah, I mean it's gonna depend probably on what the chart is and who your users are. Like how much hand holding they actually need. That is in some ways a different case sometimes for some of the interactive stuff. Like, I'm not a personal fan of the scrollytelling era, but there are times where you can build it up, right? Where you can say, look, first you want to understand this and now you want to understand something a little bit more. And that's a way that you don't have to dump all of the text at once, but you can build on it, and make it more palatable. I think the context matters for the right amount of text, there's not an exact answer.

Because for some data, that's super easy. How much sugar is in a box of cereal? I don't need much help. I don't need much hand holding. I don't need many words, right? I got it. Okay, let's move on. For some things, like that chart we looked at about penalties paid in the federal home loan banking system, I'm like, I don't know what's going on. I need more words. I need you to help me understand. The right amount of text is the amount that you need to understand it. I think if you have too much, you can think about, Should this be two charts? I don't have to do everything in one chart. Can I do some of it this way and then maybe highlight something else and do it in a different way?

The "too few text" is probably easier to solve in that, just add more text, right? And the balance is going to depend on how complicated the thing that you're trying to show is. How familiar is your audience with the thing that you're trying to show? And then: are you trying to do too much? There's a simple answer: If that is true, then you can just try less hard, right? Like when at a smaller problem, if you have the opportunity to change the problem.

Q: How to write a title for a chart?

[00:36:31] Lisa Muth (host): Great. Okay. We have another one from Jovi Dai, wondering how you write a title or subtitle for the chart. I guess you already shared a lot of advice for that one, but maybe it's more in practical terms. Maybe there's something you want to add to that.

[00:36:48] Amanda Cox: Yeah, I do think it is interesting to think about. One kind of practical tip is to just turn to the person sitting next to you and tell them what the chart is about, and then type that into the box in whatever software you are using to make the chart, right? That is better than thinking too deeply about what it...

There are bonus points for chart headlines. You get a bonus point if you keep it to one line. One mobile line. So keep it very short if you can. Or two lines, if you need to, at most, I think, in terms of some of the practical "how".

And then the subtitle, it's going to depend on what your style is. I said earlier, the Bloomberg official house style: we're just describing it. Just describing what's the unit. It's more of a unit-y kind of thing. Or like exactly a noun phrase about what you're trying to say. Works super well for very straightforward, simple things. What I was trying to say with that Black Friday example, doesn't work well at all when the pattern is more complicated. And maybe wherever you work has an official style. I am a rule breaker. You can break the rules if it's not working. If you understand the rules, you can break them as soon as you understand them. As long as you understand, am I doing this with intentionality?

And the how is again related to the same questions of ‘what do I want to see in this chart’. Or what do I think this chart actually reveals? Not what you want to see, you don't have control over what it shows. But: You're going to spend more time with your chart if you're working in most contexts, than any of your individual readers. And so to respect them, you spend 10 minutes staring at it and saying, here are the most interesting things that this chart shows, or the one most interesting thing. It's a very simple chart. You're dedicated to mobile. You don't have room to do them all. This is the point that this chart supports, or reveals, or raises questions about, or whatever it is.

[00:38:53] Lisa Muth (host): Thanks. We have another very practical question, from Pupul Chatterjee: Hi Amanda, how do you face a map versus chart situation? Would you make a map where a chart could tell the story quicker?

[00:39:07] Amanda Cox: Yeah, people do like maps, right? It's interesting. I forget... I think it's Kim Reese, this is actually her idea about, like, why are maps so attractive? It's because they're space filling. They actually do a good job in the design. They fill all the pixels of the shape. Or not all the pixels, because some countries have better aspect ratios for phones than other countries.

The real answer, though, is Are you trying to show geographical data? Is geographical patterns the most important thing? In a map, already out of the gate, you've given up your two most important dimensions. You've given up your x dimension and your y dimension by just showing the geography. So in making that choice, you are declaring geography is the most important part of this.

I think sometimes it's the wrong choice. So, for example, the work that has the most eyeballs on it, it's election night work in a newsroom. The question, I think, that traditionally graphics were answering, was, where are candidates winning?

And I think that is not the question I actually care about on election night. The question I care about is: Who will win. And where did the person who won, won? That's a different question than who will win. And so, if the question is really about where something happened, you should make a map.

It's also true that maps in some cases can be like semi efficient displays of lots of data. For example, there are 50 states in the United States. A 50 state bar chart, it gets deep. It gets boring, right? If I don't really care that much about precision, if it's just about, it's a lot in these places, and it's a little in these places, I can probably do that in tighter space with a map, but I lose a ton of precision. If you don't really care about precision, a map might be a more space efficient solution, depending on the units that you're showing.

Or, more appropriately, is it about geography? Because you've given up your best dimensions out of the gate to geography and not about the data that you're showing. So the thing that you're showing with the data better really actually be about geography if you're making a map.

Q: Changing your mind?

[00:41:24] Lisa Muth (host): Thank you so much. I don't think I've ever heard that expressed so clearly. That's so powerful. You do have a lot of strong opinions, which is great. So Jeff is wondering, what have you changed your mind about?

[00:41:38] Amanda Cox: Oh, I changed my mind about all kinds of things. Interactivity is one of them. If you remember that example from the beginning, I didn't describe it, but it was a crazy one that looks like a scribble. That was about econ data. It was about recessions. It was about what stage of the business cycle we are in. I remember Steve Duenes, who was running the graphics department at the time... You know, I was never like the world's best interactive developer, but I would dabble a little bit occasionally, and it was really important to me that you have a slider on the bottom of that so you could change the time range to any arbitrary thing, like any arbitrary thing that you wanted. And Steve was like: "just cut the slider, just make it like basically it's just a tour of this."

No, I remember fighting with him. I was like, "This is the most important thing. If we cut this, it's pointless." And now I look back and I was like, what a moron I was about that. A lot of us, and a lot of the choices in that era too... The parts of that era that hold up are the mobile ones where we had to be really strict about the editing choices that we made and really edit things down instead of being bad interface designers. We shouldn't have just been mobile first all the time. And part of that is just about the change of what the era is. I think some of me now is really drawn to more insight stuff. And so that's just a change of taste, and a change of era a little bit too.

Q: Most difficult lesson learned?

[00:43:05] Lisa Muth (host): Thanks. And related to that, Yanika Borg is wondering which was the most difficult lesson for you to learn in the world of data viz? If any.

[00:43:16] Amanda Cox: I'm sure there have been many difficult ones. I think... What's the most difficult lesson,

Because I've spent so much of my career in news, the lesson there is just that, news always wins, right? That you don't control what the most important thing is today. And so, being drawn more to breaking news than these kinds of weird thought experiments is still a lesson that I struggle with a little bit. But news always wins is a lesson. It's not a "data viz at broad" question, but it's "data viz in the context of my career".

And I totally acknowledge that some of these points may be the wrong points for people using data viz in other contexts. It is context dependent. So that is one for me that I don't know that I fully learned it yet, but it's still a struggle in my heart.

Q: Double-y-axis charts?

[00:44:22] Lisa Muth (host): All right. Let's see, we have another question. Maybe a quick one. What are your thoughts on using double axis charts? Datawrapper doesn't allow them. Do you think there are good use cases for it, especially given the complexity of economics data?

[00:44:37] Amanda Cox: This is actually a super interesting question in my current job. Bloomberg is a bunch of a lot of different teams doing a lot of different things, a lot of different norms and cultures, like they're all related, right? People at Bloomberg officially use, I think, two main charting tools and four auxiliary charting tools.

So It's a series of mergers that never really finished, right? In the terminal, designed especially for clients, especially for sophisticated clients, the dual axis charts are an option. People use it. People love it. People think that way.

In the more newsroom-oriented public-web-facing charting tool, they are not allowed. That option is not happening. Those tools, they are merging to be more design consistent, which Bloomberg has worked to do.

And so as I've made clear, my job is not really even about charts right now. It's more about the analysis and reporting data driven side of things. I visited the Hong Kong Bureau. They're hiring in Hong Kong. I was like, "it would be good to step foot in that country before you hire someone there". I visited there and my colleagues there were like..., not even my colleagues, but people on these other teams, all they wanted to talk about was whether the new charting tools would allow dual axes. And I was like, It's not my job. It's so weird, but like people really deeply care about this.

So my answer to that is: it depends on the context of who you are showing the charts to. There are fields where these are normal. People understand the ability to cheat with them. There's also places in the world where it might be harder.

My own view is that like most of the time stacking a chart is not going to be that much worse, right? You just make one chart, you stack it on top of the other chart, like it'll show you the same basic thing. There are examples where I think it's cleaner and the actual access is more helpful.

My answer to them was: "If you want it, you can probably still cheat". Because if you're making a dual axis chart, you probably don't care that much about what the actual units are.

So just index both of your series to something, and then plot it as an index. This is an index of whatever benchmark to 2007. I think there are ways. And if your tool of choice does not allow dual axis charts, and you want it as a user, I believe most of the time, you can cheat to get it for yourself. By just changing the data such that it's something indexed to something and if you need to rebase something into something then or divide it by whatever you can get there.

So depends on the context, depends on who your audience is. I acknowledge that there are audiences in the world for which this is a normal thing in their field and they feel like they've learned how to read it and so for them it probably makes sense.

Q: Impact of AI on graphics?

[00:47:35] Lisa Muth (host): Thanks. And then I'm going to ask you two more questions, if that's okay for you. One of them is a quick one. This one isn't it: you have always been an inspiration to me. Do you think artificial intelligence will design better graphics than us?

[00:47:50] Amanda Cox: That is my hope, right? That's in part because I've never been an amazing designer, right? I say, the New York Times was my journalism education, and it was my design education. Not a bad place for either. But I was lucky enough to have colleagues who are much better designers than me. What I was trying to say with that Jonathan Corum bar chart, that blue bar chart at the end: I think no artificial intelligence will ever design a more graceful, thoughtful... Not ever: in my lifetime. I'm getting old now. In my lifetime, artificial intelligence will not design a more delicate, lovely, thoughtful bar chart than Jonathan Corum.

I believe it's possible that artificial intelligence is able to capture 80 percent of Jonathan's magic and therefore will develop a better bar chart than me, is my actual belief. I've talked to Lisa a tiny bit about this too. I believe Datawrapper is one of the companies best positioned to do this because of the log of charts that go from, "Here's something I pasted in from my Excel spreadsheet" or whatever, to actual publications. That actual data, I believe, is gold in thinking about... So, when you say better than us, I think, not better than the great masters, but better than the 80th percentile people by now.

I don't think we're even that far away from that. I think we are three years, five years, seven years away from that. That is my own belief. So yes, if "us" means "me", no, if "us" means "Jonathan Corum", is my answer to that.

Q: Best data viz book?

[00:49:20] Lisa Muth (host): Perfect. Thanks. And here's the last one, a quick one. Bill Rapp wants to know: If you could only recommend one data viz book, what would it be?

[00:49:29] Amanda Cox: One data vis book, what would it be? I have a cheaty answer to this, that it's a subject matter specific book. It's not about data viz, it's about whatever you are charting in the data viz that you are making, to actually understand. It's domain knowledge. Domain knowledge is the book that I would recommend as opposed to charting knowledge. It's domain knowledge.

[00:49:51] Lisa Muth (host): That's a nice diplomatic answer. All right. Thank you so much Amanda for coming on here. Thanks for all the wisdom you shared with us. It was great having you here.

[00:50:01] Amanda Cox: Thank you for having me.