Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Maureen Stone tells you all about color in data visualization
1.3k Plays1 year ago
Welcome back to a whole new season of the PolicyViz Podcast! I'm excited to bring you a whole new exciting slate of guests this year covering a huge array of data visualization and data communication strategies, technologies, and techniques.
Maureen Stone (Tableau Research) has been involved with Tableau since 2004, when she was asked to design the initial data colors for Tableau 1.5. She joined the company in late 2011 and became a founding member of the Research Team in 2012. As a member of Tableau Research, she continued her work on optimizing the use of color in visualization. She served as research director (2017-2021), and has recently retired (June, 2022). While best known for her expertise in digital color, she has a broad experience in information visualization, interactive graphics and user interface design. She is a member of the IEEE VGTC Visualization Academy and the author of A Field Guide to Digital Color.
Sponsor: Nom Nom
Nom Nom delivers fresh food made with whole ingredients, backed by veterinary science. And science tells us that dog health starts in the bowl so improving their diet is one of the best ways to help them live a long, happy life. All you have to do is order, pour and serve.
Check out more links, notes, transcript, and more at the PolicyViz website.
Recommended
Transcript
Nom Dog Food Advertisement
00:00:00
Speaker
This episode of the Policy Viz Podcast is brought to you by Nom Nom. Nom Nom is fresh dog food made by real people who really care. They're purposeful and enthusiastic about giving your dog the best food for them and their health so they can thrive and be their best. Nom Nom's food is full of fresh proteins your dog loves and the vitamins and nutrients they need to thrive. You can actually see the proteins and vegetables like beef, chicken, pork, peas, carrots, kale, and more right in the food.
00:00:29
Speaker
There's no fillers or weird ingredients with names you can't pronounce. Isn't it time to feel good about the food you're feeding your dog? Order Nom Nom today by going to nomnomnow.com and using the coupon code POLICYVIS to get 50% off your first order plus free shipping.
00:00:47
Speaker
Plus, Nom Nom comes with a money-back guarantee. That means if your dog doesn't love fresh, delicious meals, Nom Nom will refund your first order. No fillers, no nonsense, just Nom Nom.
00:01:12
Speaker
Welcome
Season Introduction by John Schwabisch
00:01:13
Speaker
back to the Policy Viz Podcast. I'm your host, John Schwabisch. Well, welcome to a new season of the show. I hope you had a delightful summer and a restful summer and a safe and healthy summer. I spent a lot of time just relaxing, working from home, going into the office a little bit more. Very nice summer here in Virginia. Unfortunately, while a lot of the country was under
00:01:35
Speaker
terrible heat extremes, terrible weather. It was actually quite nice here in Virginia. So I don't know what to tell you about that, but I am ready to come back for another season of the podcast. Very excited for the guests I have for you today and going on into the fall and of course into next spring. Lots
Data Visualization Developments
00:01:52
Speaker
of stuff
00:01:52
Speaker
going on here at PolicyVis, I've been writing a lot of new blog posts, thinking about a lot of different types of things in the area of data visualization, including qualitative data visualization, how we should convince our managers and our bosses to think about data visualization. I also just finished a new virtual data visualization course.
00:02:12
Speaker
So if you or your organization has ever thought about trying to learn more about data visualization and didn't want to bring in a trainer for a full day thing, I just partnered up with Skillwave to offer a new course, The Art and Science of Data Visualization. I'll put a link to that in the show notes. So that might be something that you would be interested in learning more about. If you listen to this show, you probably would like to learn more about data visualization. So I hope that will be a class that will be useful to you.
00:02:38
Speaker
I'm also thinking about doing lots more stuff with Excel.
Book and Social Media Impact Discussion
00:02:41
Speaker
As you probably know, I published a new book, Data Visualization in Excel last May, May of 2023, and that book has been doing great. And so I'm trying to do more video tutorials on YouTube and on my Instagram feed so that you can learn more about how to do better with data visualization in Excel. Lots of stuff coming up, lots of changes in the field, particularly around social media as Twitter.
00:03:05
Speaker
Now, Axe has sort of changed the feeling of the field, but I'm still out there trying to do my best to convince people to think harder about communicating their data through better charts, graphs, diagrams, and more.
00:03:20
Speaker
All right, having said all that, let's turn to the first episode of the 10th season of the Policyviz podcast.
Guest Introduction: Maureen Stone
00:03:27
Speaker
I am so delighted to chat with Maureen Stone, formerly of Tableau, now retired, about color. As you surely know, color is one of the more important
00:03:38
Speaker
aspects of our data visualizations, it can help encode values, it can help direct people's attention, and there's lots of ways that people think and argue about colors, whether they're good, whether they're bad, whether we like them, whether we don't, it's subjective, it's subjective, it's semantic, it's so difficult, and Maureen is one of the leaders in the field specifically, and most importantly for us, on how to use color in our data visualizations.
00:04:03
Speaker
So Maureen was so nice to chat with me last spring about a style guide question I had when it came to data visualizations and color. And so we reconnected in August, just a month ago, to chat about her work and her experience and some of the challenges that she has seen and tried to resolve when it comes to color. And
Maureen Stone's Career Journey
00:04:25
Speaker
I think for those of you who are using Tableau, how the early days of choosing and developing color palettes in Tableau worked.
00:04:32
Speaker
So Maureen was gracious enough to come on the show. And so now you get to listen to my conversation with Maureen Stone on the first episode of the policy of it podcast for season number 10. Hey Maureen, welcome to the show. Great to see you. Thank you. It's good to see you too. How are things? How's the end of your summer going?
00:04:53
Speaker
Oh, it's been surprisingly busy between doing some music stuff and doing some went down to the SIGGRAPH 50th conference. This is the ACM SIGGRAPH graphics conference. And and then getting ready to talk to you. So I think like my kids started school yesterday, so it kind of feels like like my life kind of revolves around the school year. But I'm guessing now that you're retired, you don't really have that that kind of pace anymore.
00:05:21
Speaker
I don't have that kind of pace around school, have it for years. My kids are well out of the house. But by being an amateur community orchestra player, it seems like, and my husband is too, it seems like the structure in our life is coming from the rehearsal schedules.
00:05:40
Speaker
But I'm also trying to get some more things in structure that are around tech and color and the like. So your podcast is a very welcome deadline. All right, great, great. So I think most people, I mean, certainly everyone who uses Tableau, but most people in the data visualization
Color Palettes at Tableau
00:05:57
Speaker
field know you from your work on color, your book, a field guide to digital color. And so I wanted to get into that, but I think it'd be really interesting for people to hear a little bit about your background and how you
00:06:09
Speaker
ended up at Tableau, like at the beginning, working on the colors. And we can talk about that in a little bit. But maybe I could just tell folks a little bit about your background and how you ended up there. Yeah, thanks.
00:06:23
Speaker
I've spent the first 20 years of my career at Xerox Palo Alto Research Center, Xerox PARC. And I was there in the late 70s, starting in the late 70s, doing illustration and design systems under John Warnock and Chuck Geschke, who are the Adobe founders.
00:06:42
Speaker
And we were really early into the whole world of graphic design and imaging going digital. And so there was this big transition from stuff that had been craft by experts done with film to it's all just bits now. Being part of the computer graphics community, of course, there was this revolution in how people were creating things. You didn't have to paint it. You didn't have to go point a camera at it.
00:07:07
Speaker
you just mushed bits around. And they were creating these wonderful images, kind of things that we now see in movies. We are all digital photographers now. And so in that context, I was working specifically on illustration and design tools, but was very interested in color and how it affected how you created it and then how you got from the display out to the print.
00:07:35
Speaker
And that task, which is actually really hard. So anybody who's been around long enough, I mean, it's still a problem. You make a beautiful thing on your screen, you press print, you go, oh, my gosh, what is it? Back then, it was really hard. And what came out of the printer was really bad.
00:07:51
Speaker
So that
Challenges in Color Perception
00:07:52
Speaker
got me pulled into the image reproduction world and to solve that problem, we had to get away from thinking about color as just RGB values or produce CMYK values and think about what they meant perceptually. So we had to go in and out of a perceptual color space.
00:08:11
Speaker
and understand how people thought about color and could see color. So that's what really got me kicked off into the color world. And that was the mid 80s and did that work with some folks from the University of Waterloo. So from there, I kind of said, wow, I like this color stuff and got engaged in the people who do practical color of various forms, started bringing ideas to the computer graphics community.
00:08:39
Speaker
So I decided after 20 years, I wanted a more flexible life. And I started a one person consulting firm called Stone Soup Consulting. And that was in the was 1999. And as part of that, I did a couple of different interesting things. One is to write the book, The Field Guide to Digital Color, because I learned all this stuff and I wanted to write it down.
00:09:02
Speaker
And the other was get engaged with Pat Hanrahan and the computer graphics lab over at Stanford University. So we were living in the Bay Area at the time. And I'd met Pat back in the SIGGRAPH world. And Pat was getting, if you think about it in the early 2000s, he's getting really interested in database. Stolte is there working on his thesis.
00:09:26
Speaker
And so Pat and I got together and he says, you know, I want to think about how best to use color and visualization. Let's work out some principles I'm giving this course.
Developing Tableau's Color Palettes
00:09:37
Speaker
And together we came up with some very key ideas about using color for function and how you would use it in viz. How to take this massive world of color and kind of pracy out what we really needed. So cool, cool.
00:09:54
Speaker
And suddenly we moved to Seattle because of my husband's job. I'm still contracting. Right. And a little later that year I get a call from Stolte and Jock McKinley and Christian Chabot. And they say, you know, we've got this thing called Tableau, which I actually had known about. And I've known Jock forever as well.
00:10:20
Speaker
Can you come help us make the color really, really excellent? And I thought, sure, that sounds great. That sounds like fun color for bar charts. How hard can it be? Right, right. And that just turned into my second 20 year career.
00:10:37
Speaker
So like right at the beginning, they were thinking about color and kind of a innovative, different way. Like we're not just going to pick something random. Like we want to be very hard about it. Yeah. They wanted the appearance of their product. The visits you produced to be just excellent. Both Pat and all of them. Jock, Pat, Chris all cared tremendously about the presentation. They knew that good design was essential for people understanding them.
00:11:06
Speaker
And this is back in the world when, you know, people were using Excel and great backgrounds and black grid lines and screaming bright colors and everything created around data looked terrible. Yeah. Unless you were doing very bespoke sort of visualization. Right.
00:11:23
Speaker
So that is really fundamental. Tableau cares about color, but they also care about all aspects of the visual presentation of the viz. So you get into Tableau, I want to be able to make this clear for people because I'm guessing there are some people out in the data viz world or who are sort of on the periphery, they make graphs and they say,
00:11:44
Speaker
Like I said, how hard can it be? So you make a couple of color palettes. Like, what do you do with the rest of your time? And so that's like you were like director of the research is the research department under what they called it at Tableau. It's called the Tableau Research. But all of this predates that by a long time. This is 2004. Yeah.
00:12:04
Speaker
And I worked as a contractor getting a contract a year up until 2011 when I actually joined Tableau as a UX designer because that was the job Jock had. And Tableau research came shortly after that. So this was applied practice from what I already knew and sort of drove a bunch of research that I did even while I was still a consultant.
Research on Color Perception and Product Development
00:12:30
Speaker
Right. So tell me a little bit about those early days of developing those palettes, because now there's a lot that are built into the tool, but the very beginning. Yeah. Yeah. So the first thing we had to figure out was how are people to think about color? Right. If you do a bespoke sort of is you kind of create the vis and then you color it. Yeah.
00:12:53
Speaker
But in Tableau, what you're doing is you're moving your data, you're dragging and dropping and it makes vis after vis after vis. Right. Right. And so.
00:13:04
Speaker
We wanted color to be associated with the data and its function, not a, you know, oh, we've created an illustration you now paint. And in fact, we really didn't want to give people the freedom to just arbitrarily paint their views. We wanted them to, by default, create excellent color. So that meant that it functioned well, categorical colors, distinguish different categories, quantitative colors, try to give you a sense of the order and magnitude of numbers.
00:13:34
Speaker
And we wanted the vis to be front and forward and all the other stuff quiet behind it. The grid lines, the background, anything we called formatting.
00:13:44
Speaker
And we wanted it to be right for every view. So that makes the problem really, really hard, because when I assign the color to the data, to the categories themselves, not to the marks, then you start seeing a lot of interesting challenges about, well, that color looks good small, but it looks terrible large. I've got a lot of categories here. How do I remember what color goes with which? I can always
00:14:13
Speaker
produce a legend, but we discovered people kind of liked them to be semantic. Of course, there's issues of legibility and accessibility. I was just saying, then there's the difference between categorical and quantitative and sequential and all these different pieces that like
00:14:31
Speaker
I remember those early days of Excel with that dark gray background and like the bright green and the tableau colors are just much more relative to those are much more subtle. But but relative to those, everything was was more subtle, I guess. Yes. So was there a goal of saying we're going to have this many colors and this many separate palettes or was it more constrained than that, less constrained than that?
00:15:01
Speaker
The original goal was to have something that worked by default. Yeah. And so that was an open question. Well, just as was it palettes plus application to data, the right UX. And once we decided on that, there are a lot of details to work out. Yeah. Well, how many colors are in the default tablet palette? Right. People like to see a lot of them. But if there are too many of them, you have chaos and you can't distinguish them.
00:15:25
Speaker
So I did actually a bunch of very applied color research pushing colors around in a perceptual color space using a combination of Adobe Photoshop, there's that CIE lab space, and my own custom tools to kind of push and measure. Right.
00:15:43
Speaker
I mean, there are metrics you can use. They aren't perfect, but they're better than just having to eyeball everything and thinking about all these issues. And we came up with 10. We came up with the Tableau 10. And kind of an interesting thing about the Tableau 10 is that
00:16:01
Speaker
It lines up pretty well. You can name those colors. We knew that there's a red one and a green one and a purple one and a blue one. And this turns out to be really important for how people think about color and certainly talk about color. You know, if I'm pointing to the display and I say, you know, that red bar there as opposed to this reddish orange mauvish purplish thing, you know. Yeah. So think of your.
00:16:22
Speaker
You're creating all the colors that you got. There's kind of eight of them. It turns out there's been research that says there's kind of 11 basic color names that we English speakers all use and recognize. And so the tableau colors are white and black. They're not those colors. So we had kind of a blue, greeny thing that doesn't really have a simple name. But to make a full 10,
00:16:49
Speaker
And as we talk later about research, that sort of research around the semantics of the color, how you hold the color in your head, turns out to be really interesting and important to make it easy to use.
00:17:05
Speaker
In terms of how screaming bright are they, we knew that good design principles did not make things that were screaming bright. If you also, back then even people were still printing those screaming bright colors don't print either. So we wanted something that really looked
00:17:22
Speaker
sophisticated, but legible. Every time I'd make pallets, I'd get feedback from Chris, who's got a very good eye, and from Jock, and I'd get, you know, Christian Cerbeau would have opinions about the UX and the general, you know, did you think he could sell it? You know, the customers love it. It was this really tight iterative sort of thing.
User Feedback and Color Palette Expansion
00:17:46
Speaker
And then when the tool started to kind of take off and there were more and more users, did you start to, did you receive feedback from customers and were you able to like, was any of it useful that you could implement?
00:18:00
Speaker
Well, certainly the first thing we got was more colors. We kept doing that. And so the Tableau 20, but beyond the default as well. Yeah. So the Tableau 20 is an interesting story because that was kind of a challenge from Chris. How far can you push it? You know, maybe we need more. I mean, at some point the colors, mentally they're not distinguishable. You can't kind of remember what they mean. Yeah.
00:18:29
Speaker
But if I'm taking a big hairball of a scatterplot, and I want to pull apart all of those different categories, even if they're not quickly and uniquely identifiable, I can at least see that they're different. So having a goodly number that you could plop on there that worked seemed important.
00:18:50
Speaker
So I pushed and I played and I really did push a lot of colors around and I stared at them and we had these great feedback loops going. I could just send RGB values to the devs and they would put it in the product because you can't really tell until you try it in the product if it's really going to work. And then I came upon this scheme of doing them in pairs, light, dark pairs. And that worked.
00:19:14
Speaker
The light colors, you know, they're not as great visually, but in there in that mix, and particularly since you can kind of identify them, well, there's a dark blue one, so this is probably the light blue one. That seemed to really help. And so that's how we got up to 20. And again, there's this, you know, what is your brain trying to do with these, not just your eyes aspect of it?
00:19:40
Speaker
But they wanted colors for, oh, they needed some sort of stoplight colors because business people need those. So I carefully designed them. So in fact, they vary enough in lightness that they are distinguishable by people with red-green color blindness. Not as great, but they're different. They wanted just different aesthetics.
00:20:03
Speaker
If you're going to do a dashboard and you had multiple views, you didn't want to use the Tableau 10 over and over again because the red one here might be different, mean different data. So we put these disjoint palettes in. So here's brown and yellow and gold and green and over here we've got purple and blue and gray. And then for the quantitative colors, those color gradients,
00:20:27
Speaker
People just had a lot of needs based on their conventions. There was kind of one color gradient for every color in the tableau palette, but there was always the arguments about finance. People want red, green, and I'm going, yeah, but you...
00:20:43
Speaker
You can't see that when it's quantitative. We really pushed actually an early decision people might appreciate is to make blue-orange, the first defaults both in the category and eventually in the quantitative. They really wanted red-green to start.
00:21:04
Speaker
So that's the kind of feedback. It was customers saying, I am using this. I have real functions. I have real standards. Yeah.
Color Semantics in Data Visualization
00:21:11
Speaker
And Chris would just come back and say, can you make more palettes both for marks, but also for things like the little bars and formatting and shading and everywhere there's a pixel, it's got to be some sort of color. Right. So I want to get to how then you were doing research, but then sort of later when it's sort of more formal at Tableau. But I'm curious in this this piece that you mentioned where you would
00:21:34
Speaker
develop a palette or a color and you'd send it to the developers and they would put it into the tool. So that would, because you had mentioned earlier about how we can perceive color differently if it's like a little mark versus a big thing on a big screen and even when it's printed out. So what do those experiments look like for you? Was that you bringing in data and Tableau and just trying as many things as you could?
00:21:57
Speaker
Well, so I call that problem color and size. And it's a really well-known color perception phenomenon.
00:22:07
Speaker
So I'm going to give people a little bit of a geek definition. Color is not how you create it. It's not the RGB value. It's how you perceive it. And how you perceive it depends on a lot of factors. We know it depends on background. And it turns out size is a huge factor. So if you ever think about, would you ever go paint a room and you got the little paint chips? So that's a good color. And then you spread a bunch on the wall and said,
00:22:34
Speaker
Yes. Had that experience several times, yes.
00:22:38
Speaker
Well, I didn't know that it would apply to dataviz, but in fact it does. And so the real problem is technically what happens is that as the stimulus is the thing you're looking at gets smaller, the color appears less vivid, less colorful. And as you shrink them way down, pretty soon you're just getting kind of warm, cool colors. Just two cases and you come up and anyway, it's interesting.
00:23:05
Speaker
So, we said, okay, what can we learn about this in a practical engineering sense? I mean, you can go off to the vision scientists and they'll tell you all sorts of reasons why this is, but they don't tell you how to fix it? Yeah. Right. Right. Easy to identify the problem. Exactly. Yeah.
00:23:21
Speaker
So we had a research intern named Daniel Zafer, who's now a professor at UNC. And she worked with me and Vidya, and we actually set up a bunch of experiments to try to model how size affected your perception of color. And so we put people through these horrible experiments where they're staring at a screen, and there's a big ball and a disk in the middle, and two little ones on the side, and we go, which ones on the side look the same?
00:23:47
Speaker
as the middle one. And just a lot of collecting data from people and just feeding them cupcakes to get them to come back and do more of these things. And from that, we did a bunch of data fitting and modeling and came up with some approximate linear models that gave us some sense of what would do if you could put it in the product to adapt the RGB value to the size.
00:24:16
Speaker
Right. So everybody says, oh, what color is it? And they mean the RGB value. Yeah. Right. Right. But we're saying what color is it? We mean, what does it look like? Yeah. And so we actually have to me, the most fascinating example of where people see this was in like the background washes that you want to use across lines or just to make the background a different color. Mm hmm.
00:24:43
Speaker
Those need to be really, really light. I mean, if you're working in a dark on light world, you want things that are just off the white. Yeah. So what is a typical formatting color picker look like? Oh, it's a little array of rectangles, you know, just tiny little squares. And if you make those colors light enough to be aesthetics and you make them little squares, you can't tell them apart.
00:25:04
Speaker
You just simply can't. So we had to do various tricks. Now, asking the Tableau engineers to put in my evolving model of color and size, they did not want to hear about that. But they were willing to do some little engineering hacks that we worked out for the early versions of Tableau.
00:25:25
Speaker
when we first started by having the colors and little triples. So if you look at the Tableau color picker, there's darker ones and then lighter and lighter ones underneath and little steps of three. And that's because those little light ones is probably what you really want on the background, but you can't tell what color it is, but you can tell what color the ones above it are, which are also useful for other things. So we're kind of labeling them.
00:25:50
Speaker
The other thing that we did when we redid the colors, 2014, I think, we just flat out lied in the color picker. We put colors there, but it's not lying, right? We put the right color there. This is just a different RGB value. Right. So we actually adapted what the user saw so that when it was applied, they got what they wanted.
00:26:17
Speaker
And so that, you know, for this whole, there's a lot of color issues, but the color and size in particular, I want to make a point people, it's not about the RGB value. And that
Rainbow Color Palettes Debate
00:26:27
Speaker
kind of thing is perfectly okay to, you know, think about what RGB triple do you need here versus what RGB triple do you need there? It isn't, it isn't that it's what it looks like.
00:26:37
Speaker
Yeah. Were the engineers, were they generally willing to make those changes? I'm guessing you had lots of conversations about the actual like user experience of how a person goes in and like makes that selection. Like what was the, what were those conversations like?
00:26:52
Speaker
Well, in the early days, we were kind of doing all of everything. The company was extremely small. By the time I actually got there, so this work is after I'm a Tableau employee and I'm after in Tableau research, we had a really good UX design team and a bunch of engineers that were competent UX implementers, shall I say. And they were very open to working with
00:27:14
Speaker
me and other researchers, if we could make it easy enough. And so the beautiful thing about a lot of color work is that it isn't, I don't have to transfer complex code to them. I just have to give them some RGB triples. Now, if I just give them a spreadsheet, that's a big pain because they have to transcribe them. So I would write code that would actually write the C++ tables they needed and let them drop it in.
00:27:40
Speaker
Oh, interesting. And if you do that, then they're like, OK, I'll just drop the new table in. It's easy peasy, right? Yeah. So I also worked a bunch with the Tableau formatting team, which we were working mostly on format, but it kind of affects Marx a little bit. But as we worked on the new design for the formatting system, I just found it
00:28:04
Speaker
a great experience to work closely with the visual designers and the UX designers and the engineers to say, how can we, you know, what is practical? What will really work here? But, you know, it's good practice, but it also gives us really interesting and excellent. It's doable. It's practical. Right. Right. But it's also interesting the way you describe that, because it is even within this team, this broader team, you are
00:28:33
Speaker
talking to the engineers and trying to talk in their language to make their lives easier, which is like a lot of what data visualization is about is kind of talk someone's language so that they can get something out of the graph. And you're doing that in the process. It's very meta, I guess, to build a data visualization tool where the whole point is to be able to communicate to people who may not actually speak that data language. Just kind of interesting.
00:28:57
Speaker
Well, I mean, I started my life as a software engineer, so... Yeah, right, right. So you kind of knew that language. Okay, so now we're like 2014, 2016, Tableau Research has kind of started up as its own, kind of sounds like a formal group. And you've talked about the work you did with Danielle and with Vidya and some others. The other thing that you mentioned earlier that I know you've worked on is on color semantics.
00:29:23
Speaker
And you've mentioned a little bit, but I was thinking maybe you'd spend a little bit more time talking about that work.
00:29:29
Speaker
Yeah, happy to do it. Just to be specific, Tableau Research started in 2012. And I worked as a researcher, you know, throughout my career there, but towards the end of my career, I was more of a manager. So yeah. So a lot of the research really kind of got done the 2012 to 2017 kind of timeframe. Right. And you can put a link on your podcast to the Tableau Research website. Absolutely. Yeah, absolutely.
00:29:56
Speaker
So, Vidya comes to Tableau with a deep background in natural language and semantics, and she was carrying a vision for Tableau to engage semantics more in the way it helped people think about visualization and the tools for that. And so, we had observed an interesting phenomenon, and this is actually something I did some research on
00:30:22
Speaker
about color naming with Jeff Hare, who was then at Stanford, but now at the University of Washington, and his students. If you have a vis of, say, I don't know, fruits, let's pick something simple, and you apply the Tableau 10, and the colors go wherever,
00:30:42
Speaker
you may end up with purple bananas and orange cherries and the like. And you can kind of instantly say, this is wrong. This would be so much better if the colors of the vis match the colors of the object, or more generally, some color that I strongly associate with that data value. So things like brands.
00:31:06
Speaker
sports teams. And if it doesn't, it kind of makes your head hurt. And if it does, Jeff Herr and his students even did some research to prove that you will, in fact, remember better and be a bit faster at the task of selecting things and finding things. This all kind of makes common sense, but being in research, we have to kind of prove it all.
00:31:29
Speaker
So Vidya had noticed this, we had a whole bunch of examples from Tableau Public where people had gone to extraordinary measures to be sure that the semantics all lined up. And she said, can I do this automatically? And so there's a paper that we wrote together where she used her semantic language and scraped the web for Google tools to come up with automatic assignments based on the semantics of the categories. Right.
00:31:55
Speaker
And we blended that. Now, you can say, well, I can do that, but there'll be terrible colors. But we then use that to snap those two color palettes that I had already designed. We knew they were decent. They were decent colors.
00:32:10
Speaker
And we're really excited about that. And that would have been a great feature to add. But it, you know, it has its tendrils out into the web. And, you know, and also by that time, the company is not self-contained and involves, you know, potentially using tools and IP. Right. Right. Bigger and bigger and bigger. Right. Right. Right. So so you have that paper. So now you're a manager. But what's interesting. So you retired when in twenty twenty one.
00:32:37
Speaker
Well, it was just a bit over a year ago. So it's 2022. Is that right?
00:32:45
Speaker
But now you have a new paper with Danielle and with Colin Ware on the rainbow color palette. I want to give you a few minutes to talk about it. But it's to me, it's fascinating because, you know, in my relatively limited time in the data visualization field, you know, the first the first big fight you run into when you get into this field and you hear about Tufty and few is you should never ever use pie charts. And then and there's still that fight and then there's still the
00:33:13
Speaker
less of a fight, but bar charts just started zero. But it's like the two big ones that are like, or maybe three big ones, never, never, never. It's like no pie charts, no 3D, and no rainbow color palette. And you have a new paper out. I'm gonna bring it up because I wanna make sure I get the title exactly right because it's right here. So rainbow color maps are not all bad. So this is like, yeah, I don't know. I don't know how many people get excited about this. For me, this is really exciting, right? This is like shaking one of those like fundamental beliefs
00:33:41
Speaker
Yeah, right. Shaking the fundamental beliefs that people have, but maybe not based in certainly thorough research. So I thought we would talk about that before we wrap up because really kind of exciting, exciting stuff, I think.
00:33:56
Speaker
Thanks. Colin Ware was the instigator of that paper, but he reached out to me and Danielle. So Colin comes more from the scientific visualization community than the infoviz community. So there, the whole world of... We've talked in Tableau mostly about categories, but we too also map quantity to color. It's just we don't do it over images.
00:34:22
Speaker
And he had been asked to review yet one more paper, or no, actually found it published, one more paper saying the same old things about why the rainbow is bad. And we went back and the first paper that said the rainbow is terrible was in the mid 80s. And here it is, the 2020s.
Task-Influenced Palette Choices
00:34:43
Speaker
And you have to ask yourself a couple of things.
00:34:46
Speaker
Why do people still use them? Is it because people are idiots or is it because there's something we don't know? And furthermore, why do we keep publishing the same old rants over and over again? Because nobody's adding any new information about it. So it kind of breaks down into two kind of domains. What is a rainbow?
00:35:07
Speaker
The first one, the original rainbows were if you think of the like the HSV color space and you just kind of go around that circle. So what you're doing is you're going around to the corners of the RGB color cube and taking the brightest and most vivid colors. And when that was, you know, invented back in the mid 70s, that was really exciting because computers were really slow and that was really cheap. Yeah.
00:35:35
Speaker
And it was better than asking people to think in RGB, fine. But when people started doing digital imaging for visualization of this CyVis stuff, they grabbed that and they plopped it on their images and they created visual chaos. But we know why that particular rainbow creates visual chaos.
00:36:00
Speaker
And one of the reasons is because the colors and lightness are jumping up and down and up and down. Whereas if you think about an image of a 3D surface, you would want to look at it in grayscale and see the shading just exactly like it would be in nature with the black and light photography, why your eyes would work.
00:36:19
Speaker
And your color map needs to do something similar. Otherwise, you start losing shape from shading. But a lot of his tasks aren't shaped from shading dependent. And there are a lot of rainbows that can be made. I mean, a rainbow is just a multicolor
00:36:37
Speaker
color map, right? We use them all the time and people like them because they, the classic, I'll just start with one color and make it lighter. You can't see really very many levels. Yeah. Especially if the data is scrambled. Yeah. Okay. Um, and so people add some color to it. And now you can, the fact that your brain says, Oh, this is the red part. And we're back to semantics again. This is the blue part. This is the green part. Yeah.
00:37:03
Speaker
acts as kind of a ruler for where you are and makes it really easy to create categories, regions of similar data values. Now, the other problem is if those are arbitrary categories, they don't have anything to do with the data semantics, then they give you a false idea of what's important. But if they do line up with the data semantics, in fact, it's extremely powerful.
00:37:28
Speaker
So that is, you know, this is about the task and why people do it. And so our messages were twofold. If you're going to do rainbows, use good rainbows. We know enough you can stop beating up the HSV rainbow. I don't want to see another research paper that compares it to well-designed multicolor ones. And still, this one is terrible for all these reasons. I'm going, you know, that's in the literature.
00:37:57
Speaker
But then again, now we're back to, well, so why do people want multiple colors, right?
Future Directions in Color Research
00:38:04
Speaker
I think we're hearing something about how it is easier to see if you do get these categories. And even for some tasks, I'm less familiar with than Danielle and Collins, but especially Danielle was going, even the bad old rainbow,
00:38:22
Speaker
doesn't work too badly, you know, once you get used to it. Because in fact, what you do want is to color these things that are the same values in different part of the images, bright recognizable colors. Yeah. Yeah. And that's a very different, you know, so we just don't know enough about what people want, right? And what they're trying to do with with the color in these quantitative situations.
00:38:45
Speaker
But it's so interesting because it also ties back to what you said earlier about semantics and, you know, a yellow banana versus a purple banana. And also your conversation with the engineers, because I remember talking to, I think, like family and talking about the rainbow color map and, you know, in the hurricane, you know, the weather map saying it's not, you know, you should use something different. Like, why? Red means it's hot. Blue means it's cold. So, like,
00:39:11
Speaker
People understand it at a core level, but then we get into all these arguments. And the three of you have demonstrated, which to me, I don't want to say it was mind blowing. I'm not going to go that far. But the fact that, yeah, the rainbow palette doesn't have to be the one that you just see online somewhere. You can vary those segments. And that, I think, is just really interesting.
00:39:38
Speaker
You touch on a really other important factor in those weather maps. People are really familiar with this. Right. And in some sense, they're not broken, right? What you're doing is you're saying this is hot, this is cold, and you can set them up so that, yeah, the color name shifts on the tens boundaries, right? Right. 60, 70, 80, 90. That would be really easy to read. Right.
00:40:06
Speaker
Right. Right. And so why yell at some prince based on some principle, you have to think about what people are doing. Exactly. With the color maps is they're keeping having to extend them up in terms of temperature. Right. Right. So I want I want to close up by your thoughts on because you mentioned several times already that there's more to know. And I'm curious where you think the big areas are that people
00:40:33
Speaker
can and should and are exploring in the research world? Well, I would say both research and product. So one is simply to stop focusing on single vises and start thinking about dashboards and more complex displays, right? That I've seen some early research on people trying to normalize the colors across
00:40:58
Speaker
different, you patch together a bunch of views, and now you want anything that's the same data value to have the same color and anything that's a different data value to have a different color. Well, you know computers can help you do this. At the very least, they can tag it, and people are showing that, of course, you can set up optimization systems that will even set it. So let's stop thinking about color as just one vis.
00:41:27
Speaker
I'd like to encourage people based on the rainbow color map paper type work to say, you know, we stop just saying there's got to be simple hardcore rules that everybody has to follow. This is people trying to look at things and see them and understand their data underneath it. This is not an easy problem. It's not going to be a handful of rules. There's a handful of rules that keep you from see how to say this politely. You can get rid of the stuff that clearly is terrible. Yeah. Right.
00:41:56
Speaker
And then you can leave people with a core of things that are good enough. Yeah. Right. So stop trying to optimize for perfection. Oh, is it six colors or is it five? And say, how do we make sure you get rid of the really terrible cases? And then let people have the flexibility to play within the OK space. Right. Right. Right.
00:42:22
Speaker
And some people have proposed that AI can help with some of this, that the AI can also play within this reasonable space. Given the success of AI for images, I think that's quite possible. And then remember that not everybody sees the same. And so instead of saying, oh, we have to have these hardcore rules so you never make a vis that someone with extreme color vision problems or
00:42:49
Speaker
spatial vision problems can't use. Can we use our computers, our tools and smarts to make things adaptable?
00:42:59
Speaker
let people give you feedback about what's useful and what's not. If color can be really important and it can be really not important, you can make perfectly good visualizations in grayscale. In fact,
Podcast Conclusion and Engagement
00:43:16
Speaker
my primary recommendation for accessibility is that all the important stuff should be visible in a grayscale version of your vis. If you can't, get it right in black and white is what designers have been saying forever.
00:43:29
Speaker
Yeah. If you have to distinguish the marks, double encode them. But then there are certain aspects of color that it doesn't matter if it's really exactly the color that I see and you see. Right.
00:43:44
Speaker
Does that come to a good conclusion? Yeah, I think so. I just think there's a lot more to learn and you have moved the field so far ahead. And it's just to me interesting how all that's kind of like this narrative of how everything kind of links together.
00:44:00
Speaker
they're all separate, in some ways separate projects or separate tasks or separate tools or separate this, but they all sort of link together and how people think, view, talk about color, and then how we talk to each other both clearly within your team, within Tableau Research and the other groups, but also how we talk to people who are using the visualization via the dashboard.
00:44:23
Speaker
you know, a PDF graph or something like that. So it's, it's just, it's just really interesting. So just to kind of finish up, if you were back at Tableau, so they, they reached out, they're desperate. They need you back.
00:44:36
Speaker
And they say, we just need you to go back into the color tool. Would there be anything that you would go in and change? Is there any kind of like last big modification you'd make either to the user experience or to the palettes or how many there are or anything like that? Or do you feel like it's one of the better ones out there at this point in terms of all the, you know, there's a lot of tools.
00:45:01
Speaker
So that's a really complicated question. At this point, you'd have to say it's not Tableau, it's Salesforce. Sure. That's a good point. Yeah, that's right. And when I look at Colorize now often, it's given to the user experience team and the visual designers. It's a big company now. It has experts. And most of those experts, while they don't use exactly the same tools and terminology, I do can get equally good results. Yeah.
00:45:28
Speaker
Um, and they have user researchers and they have accessibility officers. They should. And so my coming back and saying, Oh, I wanted to innovate in some researchy sort of way is kind of irrelevant. Yeah. Right. For the, for the bigger product. Right.
00:45:46
Speaker
I would hope that they would use and Salesforce is a very AI forward kind of company. Yeah. And Vidya is still there heading Tableau Research. Yeah. That we will see some of these semantic ideas get in even to the color world as well as other parts of his. Yeah. And that would be very cool. Yeah, very cool. Well, Maureen, thanks so much for coming on the show. I appreciate it. This was really interesting. My pleasure. Great story.
00:46:15
Speaker
All right, well, enjoy the kickoff to fall and to whatever your next concert is. I appreciate you coming on the show. I appreciate being invited. Thank you very much.
00:46:27
Speaker
And thanks to everyone for tuning into this week's episode of the show. I hope you enjoyed that conversation with Maureen, and I hope you check out a lot of the links that I've put on the show notes page on the website. I hope you'll also consider checking out my new asynchronous video course, The Art and Science of Data Visualization with Skillwave.
00:46:45
Speaker
I think this is a new way for me to connect with more people in the data visualization field and help people do a better job of communicating their data visually. You can also connect with me on Twitter. Yes, I'm still there. You can also connect with me on Instagram or on the website, or you can subscribe to my sub stack newsletter that comes out every other week with the podcast. So until next time, this has been the policy of his podcast. Thanks so much for listening.
00:47:12
Speaker
A number of people help bring you the PolicyViz podcast. Music is provided by the NRIs, audio editing is provided by Ken Skaggs, design and promotion is created with assistance from Sharon Satsuki-Ramirez, and each episode is transcribed by Jenny Transcription Services. If you'd like to help support the podcast, please share it and review it on iTunes, Stitcher, Spotify, YouTube, or wherever you get your podcasts.
00:47:34
Speaker
The PolicyViz podcast is ad-free and supported by listeners. If you'd like to help support the show financially, please visit our PayPal page or our Patreon page at patreon.com slash PolicyViz.