Well look at that, a new blog post! This site has been dormant for a while, but now it's ready to come back.

The short version is that I've created a new static site for it that's more modern and hopefully easier to navigate. It also makes it a lot easier to build new sections and add interactive pages and little apps.

I'm working on bringing back some of my old interactive pieces (which I'm now calling apps), most of which I broke in the transition to my previous static site version and never ended up fixing.

In particular, the ZIPScribble Map is back and much better than before – though US-only for now. It has a new navigation bar along the bottom that lets you hover over ZIP code zones or states to highlight and zoom them. It's a little finicky and needs some refinement, but it's pretty interesting to explore.

New static site

For anybody interested, the new site is built using Svelte and SvelteKit, and hosted on Github Pages. The original transition from WordPress was to a site I made with VitePress, but that turned out to not be such a great fit after all. VitePress is works well for documentation sites, but it lacks flexibility and I found debugging issues with Vue components to be quite frustrating.

With Svelte, I built all the logic and structure myself, so it's easy to change and figure out where things might be going wrong. I also recently started using Claude Code, and as you can see in the code repo, it has helped tremendously to finally get the site over the finish line.

I continue to be impressed by how much Claude Code can do (and I'm not even really pushing the boundaries, I think). I had it update the code for the ZIPScribble map from Svelte 3 to 5, fix some issues, and then integrate it into this site, all with just a few prompts. It also found DOIs for some of my papers where they were missing or wrong, and it has also already built a new photo gallery section for me. The latter isn't live on the production site quite yet, but it's only waiting on me now to pick photos and create the galleries.

More to come

Stay tuned for more! You can subscribe to the RSS/Atom feed or follow me on BlueSky or LinkedIn. I might also bring back email subscriptions, but that's a bigger project.

There will be more updates on apps as I bring them back, new stuff like photo galleries, and a few other things I've been wanting to build. And of course there will even be the odd blog post on various topics like data visualization and related topics, and maybe even more videos, who knows!

The words business and data visualization probably put you to sleep before you even reach the end of this sentence. But wake up! There's actually a lot of interesting work to be done in this area, if only we give it a chance. I think research has ignored this space for too long, and is missing out not only on more ways to create better applications, but to find really interesting research problems to work on.

In this new paper in CG&A Visualization Viewpoints, I cover a range of preconceived ideas that I think keep researchers from wanting to work on business data vis, and point out some interesting research directions.

One of the issues I see is a lack of understanding of who the users really are. There has been some interesting work in recent years, but I think most researchers still chase the idea of a subject matter expert, when most people using data visualization are wearing many hats and need to accomplish many tasks that don't neatly fit into a specific area of expertise.

There are also many things people do in the business world that don't follow the common rules or wisdom of the data vis world, but work despite (or because of) that. Numbers are one example, they're everywhere on business dashboards, and for good reason. How can they be integrated better, rather than be ignored? How can visual preferences, which lead to interesting (and sometimes bad) choices, be incorporated into visualization software? And why are the charts created by modern software so similar and boring? What can we do to make them richer and more attractive at the same time?

I hope that this paper will serve as a bit of a provocation, and to get more people interested in doing research in this area. I point out a number of concrete areas to work on, but obviously there's a lot more to be explored and done. But if data vis research can just tackle a few of these topics, that's a great start.

Robert Kosara, Business Data Visualization, Beyond the Boring, Computer Graphics & Applications (CG&A), vol. 44, no. 5, pp. 153-158, 2024.

Before we had D3 and all this fancy web technology, interactive news pieces on the web were usually built using Adobe Flash. Some of my favorite news graphics are from that era, and when I talk about them I like to joke that they're so old they were done in Flash (a joke that dates me as much as it dates these news graphics)!

The problem is that Flash, while ubiquitous in the early 2000s, was deprecated in 2017. That unfortunately means that they are all but inaccessible now, because no current browser includes Flash, and the workarounds I've tried have not been successful either. I wrote about the problem of bitrot on the web many years ago, and Flash pieces were a big part of that.

Now however, it appears that the New York Times has added a web-based Flash player to their archive website that can run these old pieces! Now you can experience them in all their interactive glory. Check these out:

• A Year Of Heavy Losses from 2008
• Turning A Corner? from 2009
• The Jobless Rate for People Like You also from 2009
• Why Is Her Paycheck Smaller? from 2010

Edit: a few more, suggested by people on Bluesky:

• The Ebb and Flow of Movies: Box Office Receipts 1986 — 2008 from 2008, via Derek Willis
• How Different Groups Spend Their Day from 2009, via Amelia McNamara
• A Peek Into Netflix Queues from 2010, also via Amelia McNamara
• Fractions of a Second: An Olympic Musical from 2010, via Jer Thorp

And there are many more I can't think of right now. What's kind of funny is that now that the Flash pieces are working again, it's even more painful to see some other pieces be broken that don't even use Flash (like this one, where a few of the steps don't work)

It's a great move, and I really hope that more news outlets will follow suit – looking at you, Washington Post!

Finger-wagging about rainbow colormaps is a pretty common pastime in visualization, I've done it too! And it's not like there aren't good reasons. Look at this map of maximum temperatures in the US, published by NOAA, for example:

The rainbow colormap used here has all the usual problems: it's not uniform in its luminance (brightness), different colors cover different ranges of the colormap (though it's usually green that is the worst, here it's purple and red), and of course the ordering is somewhat arbitrary.

And yet, these kinds of maps are seen by millions of people every day! And they're created by smart people every day. If they're so bad, how is that possible?

The authors of this paper point to research showing that rainbows turn out to be the most accurate when the task is reading a value off a map using a key (a common task when you're looking at a map of temperatures!). There's also evidence that rainbows draw attention to global structure and lead people to reason about relationships between data distributions.

This is not to say that rainbows don't have problems, they certainly do. Not all of them are all that critical for real uses, though. One commonly cited issue is that because luminance changes back and forth across the scale, it is impossible to read shape from a visualization that uses a rainbow colormap. The authors show that in this nice illustration:

But that is often not the actual task! Who cares if you can't recognize the precise shape of that apple when you're looking at a 2D map? While it is an issue, it's easy to avoid by knowing where shape perception is likely going to be relevant. This is basically a strawman that's largely there to pile on the criticism, not a real practical concern.

I'm not going to summarize the entire paper here, it's an easy read and not very long, and it's available in its entirety on the IEEE Computer Society website.

The authors of this paper represent both the old and new guard in perceptual and color research. Maureen Stone (who, I should mention, was my manager at Tableau Research for several years) and Colin Ware are part of the former, Danielle Albers Szafir the latter.

They end the paper with a call to improve rainbow colormaps instead of dismissing them as always wrong:

We understand well enough why rainbows can be bad; let us focus instead on finding out when and why they are good.

Ware, Stone, Albers Szafir, Rainbow Colormaps Are Not All Bad, Computer Graphics & Applications, 2023.

Computational notebooks offer an alternative to the common GUI-based tools used for data visualization and BI today. In this new paper, I talk about what they are, their pros and cons, and how research could fill in some important gaps.

Data visualization research has focused primarily on graphical user interfaces (GUIs) for creating data visualization, and for good reason. But notebooks have been used in data science for a while now, and they offer their own advantages over GUIs: reusability, integration of data analysis and modeling, and – especially – easy collaboration.

This is an invited piece for the Graphically Speaking column in CG&A, and I'm obviously biased because I work for Observable now. There are a fair number of computational notebook platforms out there though, like R Markdown in RStudio, Jupyter for Python, etc.

The paper talks about what notebooks are, where I see their strengths (and some weaknesses!), and in particular where I see opportunities for research. What I don't discuss in the paper, since it wasn't as big a hype when I wrote it as it is now, is that notebooks are also pretty ideal for exploring the current wave of AI tools, in particular ChatGPT and similar. But whether it's old-fashioned data analysis and visualization, financial or other modeling combined with analysis, or exploring AI models, I think there's a large research space here that is largely untapped.

Robert Kosara, Notebooks for Data Analysis and Visualization: Moving Beyond the Data, Computer Graphics & Applications (CG&A), vol. 43, no. 1, pp. 91-96, 2023.

I'm teaching a short course on data visualization for Observable. It's free, and you should join! Starts March 7.

The course will cover the fundamentals of data visualization, like the key chart types, how they work, when and how to use them. We'll also talk about transformations of data, like binning and smoothing data, and of course interaction.

We'll use Observable's Plot library for the most part, plus a bit of D3. The focus will not be on the tools however, but on the data visualization. If you don't know JavaScript, you can brush up a bit by reading our Just Enough JavaScript intro and following the quick guides on your Observable user home page. I'll introduce any JS and Plot features I'm using as needed in the course, too.

Here's a bit more in my little teaser video:

The course starts March 7 and consists of five sessions (every Tuesday and Thursday, 9am PST/noon EST). Sign up here to join!

They Rule connects companies through their boards of directors. You can follow these connections between seemingly unrelated companies, and sometimes even competitors.

By web standards, They Rule is quite ancient. It has been around since 2001, and since it was initially done in Flash, it was effectively offline for the last few years. On has recreated it in React and updated the data, which should make it future-proof for the next while. It's a great project and I'm very excited to see it back for people to explore.

Head over to YouTube to watch the video there, or right here!

This is my first video with a continuous music bed. Let me know what you think! And as always, I'd appreciate it if you could like the video and subscribe to the channel over on YouTube.

This site has been around for over 16 years now, and a lot has changed in the world during that time. I'm currently working on an overhaul and wanted to give everybody an idea of what I'm thinking about and why there has been little activity. In light of recent developments, here are also some good ways to follow good old-fashioned blogs and an alternative to Twitter.

Site Changes

When I first started writing this post to go out on the actual birthday of the site, October 1, I was planning on turning it into a static site with no comments, but better organization and still an RSS feed that would update when I added new things (see below).

I'm now rethinking this, but I am definitely going to reorganize the content, and change the layout and theme to make it easier to navigate. First, all current content will be moved into an archive (with redirects from the current URL of course). The idea is make clear what is old, and also turn off comments for those posts since all I get there is spam.

I will then turn a selection of useful content into little collections that will be easier to navigate and consume, and I will bring those up to date if necessary. I know people use my site for teaching, so I want to make sure it stays relevant and useful. The same thing is true for people finding it through search, etc.

The new structure will also have project pages of some sort that I'm still designing to have better landing pages for videos, papers, etc. This will also include moving the papers from my Kosara.net website over here (that site is purely a list of my papers these days).

As for the design, I have a few sketches and I've started building the new site using Svelte, but I'm now wondering if I should stick with WordPress after all. One of the motivations for this change had been to move the site to a serverless host and get rid of my virtual private server, but I guess that's not going to happen if I stick with WordPress.

So there are still decisions to be made, but I hope that by the end of the year, this site will look and feel quite a bit different, and next year I hope to have some more regular (though not very frequent) updates.

Following Blogs and People Without Social Media

It's kind of sad that social media ended up becoming the way to stay informed about things happening with people or any sort of "content." Before Twitter, there was RSS (and before RSS, there was email, I'll get to that). RSS feeds are great because they're extremely simple and you can follow as many as you want, and just the ones you want. All you need is a way to consume them, or a reader app. I like Reeder on macOS and iOS, but there are also free ones like Feedly.

And then there's good old-fashioned email. Email newsletters have become sort of thing over the last couple of years, though I'm a bit skeptical that they will be sustainable by themselves in the long run. You can subscribe to many websites and blogs, including this one, via email though (check the sidebar). I'm actually surprised that almost 3,000 people are doing this in my case, but it shows that this is something people want.

All of this is to say, we don't have to depend on centralized places like Twitter to share and stay up to date. Web 2.0 really wasn't so bad, and it's still around.

Twitter Alternative Mastodon

A good number of people I follow have recently set up accounts on a distributed social network called Mastodon. For better or worse, it's not a Twitter clone but rather a different approach to social networking (I've seen it described as the Linux of social networks, which strikes me as painfully accurate).

Since I follow mostly data vis and HCI folks, I see many of them on vis.social, and that's also where I am. There are many other Mastodon instances out there however, and you can follow people across them (this is called federation). It's all more tech-focused than it should be, but it's not very difficult to figure out.

There's also a great website called Debirdify that connects to your Twitter account and shows you Mastodon account names of people you follow on Twitter, and which instances they're on.

All that said, I'm not leaving Twitter just yet and I don't think it's necessarily doomed to fail (as many people seem to assume). The next months will be rough, but it might come back as a viable platform – we'll see.

Where I Am

I haven't been very active on Twitter for the better part of a year, and I'm probably not going to spend huge amounts of time on either Twitter or Mastodon going forward. Once I have worked my way through the site redesign, I do want to blog and keep things updated more often than I have, though. I also have bits and pieces for several videos that I need to finally put together and get out.

I've been spending a lot more time on Discord lately, specifically some servers dedicated to making music and synthesizer nerdery. I picked up music making as my pandemic hobby, and I've gotten more into it, which naturally eats into the time I have to spend on other things. I'm going to link to some of my efforts there from here and Twitter eventually, but this site will stay focused on data visualization.

It is nice to have this site to fall back on, even if it doesn't have the reach it once had. But it has been around for a while, and I hope to keep it around, and relevant, for a while longer.

Midjourney produces images from prompts, which are basically text descriptions. They don't have to be specific or concrete in any way. "Astronaut riding a horse in space" works just as well as abstract words. In fact, the latter can sometimes produce more interesting images.

Of course I had to try it on a variety of variations of "eager eyes," and it did not disappoint. Some of the prompts will seem a little random, but they're usually the result of a few attempts while adding and removing words.

It's kind of a fascinating setup. You interact with the MidJourney Bot through Discord (if you're not familiar, it's like a less refined version of Slack). This is especially addictive when you have the Discord app on your phone, because you can just send it prompts whenever the fancy strikes you, and it'll ping to let you know when your images are ready.

Other systems like DALL-E produce more photorealistic images, which in my experience MidJourney does not do very well at. But what really nails are styles like art nouveau, art deco, and 1960s futurism. The latter is particularly fun.

Clearly the most interesting use is to mix concepts and styles that are nonsensical or that would not have existed. What would harvest robots in ancient Egypt have looked like? How about musical robots during the height of art nouveau?

It's also funny to realize that while it can create images of animals that look good at first glance, it doesn't have a clue how animals work. You quickly realize that details don't make sense, there are extra legs, etc. But it does work for surrealist-style images at times.

It can also create some very surprising images, like Picasso-style paintings of UFOs. I was quite surprised by how well it did these, not just in terms of them looking like paintings (some even with frames!), but capturing this playful style.

Abstract terms can lead to some interesting surprises too, like the prompt "tonality." I have no idea where it is finding these shapes, but they

I also tried a few prompts I saw people use with DALL-E on Twitter, like "Pharaoh Darth Vader of Egypt." Midjourney did a better job here than DALL-E in my humble opinion, but I also can't find the DALL-E tweet anymore to show them side by side.

This is just a selection of what I've created with MidJourney so far. If there's any interest, I can put together a second post with some more examples. If you can find an invite, it's worth checking out for sure.

I really started thinking about this when I realized that gauges aren't just displays of numbers, but also show you where you are on the scale. That's quite useful when you're trying to understand things like how close you are to the maximum RPM in your car, or the range of empty to full on your fuel gauge.

Circular bar charts are kind of related, and not just because they're also round shapes. I think they work well on the Apple Watch to show you how well you're doing in your daily fitness goals, and they actually seem to be fairly common to show goal completion. That makes them part-to-whole charts, though with the twist that you can go beyond 100%. Different widgets and devices show you the part beyond 100% differently, and some don't actually bother to show you how far you got beyond 100%.

All of this is in my new video, which you can watch below or over on YouTube. As usual, I'm very curious about your thoughts! And if you like the video, please head over to YouTube to click the like button and maybe leave a comment – and subscribe to the channel, if you haven't already!

I gave a talk at the Outlier conference earlier this year, with the somewhat elaborate title, The Joys – and Dangers – of Bespoke and Unusual Chart Types. Though I eventually decided to go with the much shorter, This Should Have Been A Bar Chart! You can watch it on YouTube now.

In the talk, I go through a number of examples, some of them recent, some of them a little older, of charts that are unusual in one way or another. These aren't the kinds of charts that work for any old data, they're often quite specific (or, bespoke) to the context and situation. And they don't always show a lot of data, either. But that can be a strength when it makes a chart stand out, stick in your head, or makes you look in the first place.

I also believe that we tend to train people to make visualizations that are devoid of any fun (hence, this should have been a bar chart!). Sure, you can overdo the fun, but you can also make plain, boring charts that all look the same and are completely uninteresting. Throwing in a bit of fun can make a huge difference if you actually care to reach an audience.

I use a few quotes from the paper I published with Matt Brehmer at VIS last year, one of which got quite a few reactions right after the talk in February: to show ordinary data in extraordinary ways. I should probably have gone with that one as the title (I actually am for an upcoming talk at the New York Data Vis Meetup), because it summarizes what I think we should strive for in a really crisp and pithy way.

I end the talk with a bit of a taxonomy of more elaborate charts along two axes, chart familiarity and data specificity. They're somewhat independent, but not entirely. I think they are useful when considering unusual or bespoke charts, to decide how far you want to go away from the plain bar charts or similar, and what that might mean for your audience.

You can watch the talk below or over on YouTube. Like in my other videos, this one also has numbered references throughout, which are collected on this references page.

Encodings play a central role in visualization, but I believe our thinking about them is too simplistic. In a new paper, I argue that we need to distinguish between the encodings that specify how a visualization is drawn and the ones that are readable or actually read by an observer. While they largely or entirely overlap in some charts (like bar charts or scatterplots) they don't in others (pie charts, line charts, etc.). And what exactly do you even specify in more complex visualizations like treemaps?

Encodings are deceptively simple, once you've learned what they are. After all, they're what you see when you look at a chart, right? Well, it's not so easy. I keep harping on how we don't know how pie charts are read, and I've summarized the issue in this figure from the paper.

My point here is that while we clearly specify pie charts by angle, that doesn't mean that we also read them like that. And as I've shown, angle pretty clearly is not how we read pie charts. So we need to at least start to look at two different kinds of encodings: specified and observable encodings. In the pie chart, the specified encoding is the angle, and observable encodings also include arc length, area, and even chord length. Out of those, the observed encoding might be one of them or a combination.

In the paper (author copy here, official version in the IEEE CS Digital Library), I list a few more chart types, such as bar charts (specify length, observe area, aspect ratio, etc.), line charts (specify points, observe slopes, line segment lengths, etc.), and others, with their encodings. Think about what you really specify in a treemap, for example, and what you observe. I don't think we have a good model for encodings when the transformation is complex and algorithmic (as opposed to a straightforward mapping), like in a squarified treemap.

Line charts are especially interesting to me, because they also illustrate an even more complex encoding. What is the function of a line chart? It's specified and drawn as points that are connected, but you don't read a line chart like that. Instead, you look at shape, overall slopes/trends, etc. These are derived encodings, and they depend not only on the chart type but also the task.

I've clearly only scratched the surface here, but I think it's an interesting topic that needs to be understood more deeply than we have so far. Perhaps my little article will inspire some more work in this area.

Robert Kosara, More Than Meets the Eye: A Closer Look at Encodings in Visualization, Computer Graphics and Applications (CG&A), vol. 42, no. 2, pp. 110-114, 2022.

The Chart

Last week, the Biden White House published this chart showing that U.S. GDP had grown by 5.7% in 2021. That's the most in many years, with the chart showing numbers all the way back to 2001. Now one might argue that perhaps a new president won't immediately have a huge impact on GDP, but the data in the chart appears to be legit. What is unfortunate, however, is that the vertical axis is being distorted.

People on Twitter quickly picked up on the fact that the axis scale goes up in whole percentage points until it reaches 5.0, but then there's an additional 5.5 step before we get to the 6.0 value that should have been next. And since the growth number for 2021 sits between 5.5 and 6.0, that bar gets longer relative to the others.

A few days later, a correction was tweeted saying, "this is y you proofread." The new chart has a correct vertical axis and brings the 2021 bar down to where it should be relative to the others.

Was this an honest mistake or was it designed to get more attention? And was the idea to overstate the change or get attention for the stretched axis?

The way the bars are spaced out and the choice of time window was clearly done to emphasize the 2021 value. While none of the bars even reach 4% in this chart, there were a number of years in the 1980s and 1990s with GDP growth over 4% (see this NY Times story with a different take on the same chart). Picking a timeline to fit a narrative is nothing new, and it's not necessarily nefarious. The Obama administration did it to produce a very interesting chart showing job losses that I also discussed in a video. But that one did not distort any of the axes.

A Mistake?

It's certainly possible to make mistakes when designing charts like these, but I don't believe that this was a mistake. This chart was probably not created in a visualization tool, but in Illustrator or a similar design program. If this were just a matter of the scale being wrong with an extra 5.5 gridline in there, that would be a reasonable mistake. But how did the final bar get stretched to fit into that scale correctly as well? That seems improbable.

It seems more likely that this was done on purpose, and there are two possible ways it might have been intended. The first and more obvious one is to simply stretch the bar to make the number look bigger and hope people won't notice. That seems rather pedestrian, but it's not like this hasn't been done many times before. It does seem kind of futile though, since chart nerds will be quick to point out the issue, and it seems that there's a low tolerance for bad and manipulated charts from official sources these days (remember "Sharpiegate?").

The other theory is more interesting: what if this was intended to be noticed, to give it more exposure as people make fun of the mistake? And more than that, they would tsk-tsk the chart while pointing out that it wasn't even necessary to distort the scale since the number was always going to be better than any in the 2000s so far anyway! That seems like a fun theory because it feels more like the kind of stratagem you'd expect from the kinds of people working in Washington.

Manipulating a chart is kind of boring, but the idea of somebody designing this as a sort of social engineering attack that would get people to discuss it in a particular way appeals to me. That doesn't mean it's true, of course, but it is a lot more fun than somebody trying to impress the boss by making a bar taller.

The piece in question is titled Here’s When We Expect Omicron to Peak. The opener image, which attracted a lot of attention on Twitter, shows the number of new COVID cases in the U.S. since the beginning of 2020.

The initial response was largely negative, which then led to a wave of people defending it. I don't want to get into any arguments about whether liking it or not means you're narrow-minded, etc. – but I do want to point out what bothered me from the beginning, even though it took me a while to put my finger on it.

Spirals aren't uncommon to show periodic data, and the seasonal nature of COVID infections was clearly part of the inspiration here. We could argue whether the width of a ribbon is a good idea, and the choice of color was probably a large reason why this piece got such a strong reaction. But what bothers me is the wobbly spiral.

I've overlaid an Archimedean spiral on the chart below. It's centered on the same center as the one in the chart on the left, and shifted over to better match the piece on the right.

Now granted, there are many possible spirals, but the Archimedean one fits very well. But look at the one on the left, where I've aligned the centers: the one in the NY Times piece is shifted over to the left, which is what makes it feel really off-kilter to me.

When you look at where my overlaid spiral crosses the vertical axis, you will notice that its tangent (direction) at that point isn't horizontal. That might seem odd in a chart, where you're expecting the beginning of the year to lead cleanly into the next year. So my guess is that they decided to shift it over, which is what I did in the image on the right above. It lines up better, but not entirely. But now the whole spiral is shifted, which is what makes it look so wobbly to me.

Even when it's shifted, it looks like the lower right quadrant in both 2020 and 2021 is flattened relative to my overlay. That adds to the impression of the shift and wobbliness, I think.

I don't think the spiral was a bad choice here, but I do question the decision to shift it over like that. Perhaps using a different spiral would have worked better, or just accepting the beginning of the year to not be on a horizontal part of the spiral.

If you've been following this blog, you've seen me talk about my pie chart papers. This video summarizes them in a way that I hope is interesting, informative, and entertaining. You can watch it below or over on YouTube (the latter is preferred so you can leave a comment, subscribe, etc.).

The video covers four different studies, which I've written about here before. In particular, I wrote a walk-through of the pie charts studies from the first couple of papers in 2016. I'm also covering the study I did using 3D pie charts as a way to distinguish between arc length and area in 2019. And if you think people hating pie charts is new, feast your eyes on the drama that unfolded after Eells' paper was published in 1926.

As always, I'm curious to hear what you think! If you want to support the channel, please head over to YouTube to like the video and subscribe if you haven't already. This is free of course, but it gives YouTube a signal to show the video to more people.

Also, my YouTube channel has been kindly shortlisted for a Data Literacy Award. If you enjoy my videos, please consider voting for me (and for other great folks in the other categories too!).

Who are the people who use data and visualization as part of their work every day? In particular, how do people use data as part of meetings, to present to others, to discuss their findings and recommendations, etc.? My colleague Matt Brehmer and I ran a pair of studies to find out.

I’ve written a little blog post about the paper over on the Tableau Engineering Blog, and I will give a brief summary here as well. But in this post, I want to focus on a few quotes and points that I found particularly interesting.

The study started with us mocking up ideas for features that would be useful as part of presentations with data, such as builds or progressive reveals with granular control, a secondary screen display with controls over what’s being shown to the audience, and a simple authoring system for syncing up video with an interactive visualization that would walk through a presentation but stay interactive for the viewer.

In the process of interviewing people about their responses to our ideas, we realized that we weren’t spending enough time asking them about their own uses of data in their own organizations. These were all people working in organizations and regularly using data in presentations to others, a group of people we actually know very little about. So we ran a second study, which involved a few of the same people, to dive deeper into their actual uses, frustrations, and ideas.

In the paper, we use musical metaphors to describe three categories of presentations: jam sessions are highly interactive small-group meetings where the presentation tends to be just the opener to collaboration; semi-improvised performances are more formal presentations outside the presenter’s immediate team, often with more preparation, and delivered at regular intervals like weekly, monthly, or quarterly; and finally recitals are the kinds of formal presentation many of us associate with data presentation, such as TED Talks, big productions on stage, etc. In organizations, that latter group also includes presentations to senior management or the company board, company owners, as well as customers.

Bar Charts Suck

One thing that was fun to see was that people really dislike the workhorse analytical chart types like bar charts for presentation. Especially in the more formal kinds of presentation, they want to use more interesting chart types to get people’s attention. “Some people are half asleep” was one interesting quote, and the same person told us that he “hates bar charts, [they’re] very 2001.” Other participant insert uses lollipop charts to “break up an otherwise boring series of charts”.

Presentations to owners and boards are sometimes created by design agencies that make them much more visually interesting without (hopefully) changing any of the data. I think one person really summed it up nicely, saying that he wanted to “present ordinary data in extraordinary ways.”

Regular Meeting Cadence

I don’t think I’ve ever seen anybody discuss recurring uses of visualization in the literature. But meetings where data gets presented aren’t one-offs, they happen on some kind of schedule, like every week, month, quarter, etc. In fact, one of our interviewees sometimes meets with her team multiple times per day, saying “you don’t want to see my calendar!”

Presentations for recurring meetings aren’t built from scratch and don’t start from zero. The people there know the data and context, and they often will have expectations for what they will hear. This is completely different from the common idea of data analysis, and in particular data exploration, in visualization. What tools do we have for these people? How do you show change from the previous presentation, how do you incorporate expectations?

Poor Tool Coverage

What also struck me as we were writing the paper was just how poorly presentation with data is supported by tools today. Visualizations and business intelligence tools are used a lot in jam session-style meetings, but lack most of even the basic features you’d expect from PowerPoint or Google Slides. And presentation tools, which are much more common in semi-improvised performances and recitals, lose the direct connection to the data. So when the data changes, often the entire presentation has to be updated – by hand! – to make sure nothing is missed. This isn’t uncommon either, sales and other data that gets reported externally goes through many revisions before it’s final, so having to update data in presentations is as common as it is painful.

We believe that there is a lot of opportunity here, both for research and for tool development. Visualization research has mostly ignored business users and the presentation use case so far, and even publishing work about them is challenging – I know of work that we actually used as inspiration for our studies that is still in revision limbo. Given how many people this kind of work might affect, it seems like an enormous missed opportunity.

There’s a lot more in the paper, including about our mock-ups, as well as many more quotes and other insights into this very common use of data.

Matthew Brehmer and Robert Kosara, From Jam Session to Recital: Synchronous Communication and Collaboration Around Data in Organizations. Transactions on Visualization and Computer Graphics (Proceedings Vis), 2021.

EagerEyes is 15 years old today! Rather than look back at 15 years of visualization and blogging (though I will do a little of that too), I want to reflect a bit on what blogging means today and where things are going.

Blogging Is Over

It would be easy to think that blogs are old-school and nobody pays attention to them anymore. I don't think that's true, but they certainly have a lot more competition today. There's the well-established social media like Facebook and Twitter, there's the newer ones like TikTok, SnapChat, Clubhouse, there's YouTube, there are podcasts, etc.

Back in 2006, when I started doing this, blogging was still a thing. People had blogs. Wikipedia still calls Twitter (which had only launched earlier in 2006!) microblogging. Academics had only just discovered blogging (hey, I was an academic back then!), though they still haven't really embraced it.

It seems that email newsletters are finally really happening, with tinyletter and especially Substack. But blogging hasn't really gone away. Medium is blogging, it's just a specific service. The Data Visualization Society's Nightingale blog seems to be going well, and there are many others. The line between long-winded Twitter threads and blogs can be blurred at times.

Is Video Next?

I used to value being able to write about data visualization and getting people interested in the more or less meta discussions about the subject. I think Moritz Stefaner made a comment once about how he hadn't thought that you could write so many words about visualization without pictures. Of course the joke's on him, because he went on to launch the Data Stories podcast about visualization with Enrico Bertini (or maybe I'm completely wrong about who said it and the joke's really on me in the end).

I launched my YouTube channel a little over two years ago, and I'm very actively working on more videos (the next one should land next week). While I still think that writing about visualization is important, I also think that it's silly to neglect the more visual media. It's called visualization, after all. So why are there so few videos about it? And most of what exists are tutorials about how to make things in various tools. Where's the fun? Where's the visual exploration? Where are the visual explanations of visual representations?

That is what I set out to do with my videos, and I think I really hit my stride with the last couple of them. If you stopped paying attention because you didn't care to watch my face talking at you, check out the video I made about Iraq's Bloody Toll and the one about linear vs. quadratic change before that. This new style with all motion graphics is clearly the way to go, and I've finally figured out how to make that work. All my future videos will be like this, no more talking head!

There's an enormous opportunity here I think, both in terms of using video as a medium for visualization (again, this seems so obvious I can barely even type it, but who does it?) and to explore what motion can do for visualization. Most people seem to have written it off based on a handful of studies, but I think those all miss the point. Motion can help people understand what's going on, there's a reason there are so many instructional videos on YouTube. We just need to figure out how to do this for visualization.

How EagerEyes Has Changed

This site has changed quite a bit over the years. On my About page, I still maintain that it's not a blog. Of course, virtually all postings over the last couple of years have been in the blog category. And who am I kidding, this is clearly a blog.

The reason for my insistence that it's not is that I always wanted to build more tools. As I mentioned in the riveting EagerEyes Origin Story, I initially wanted to build a social data visualization website, similar to Many Eyes (if anybody remembers that, it came and went too quickly). I did build a few, but they mostly became obsolete or stopped worked (remember Java? Yeah, I built a couple things here in Java). Perhaps the only thing that really survives, and only because I rebuilt it a couple of times, is the ZIPScribble Map.

I won't dwell on the past here, but it's fun to look back at how dorky this thing looked a while ago. This is also the logo that was so cleverly parodied by the excellent eagerpies website, which is online again.

At the 10-year blogoversary five years ago, I launched a big redesign of the site. That was the first time I paid for design for this site (other than buying WordPress themes for a few dollars), both to get a custom-designed logo and theme. I'm still happy with the logo, but I'm increasingly unhappy with the look of the site. Not sure when I'll get around to updating it, but once I do it will look more like a blog again. It also likely will be a static site rather than WordPress, but that's also one of the things slowing down progress on this project considerably.

How I Have Changed

One category of post here used to be making fun of bad visualizations. I haven't done this in years now, and I have no intention of doing it ever again. It was fun for a while, but it has really run its course. It just doesn't seem productive. I see people do it on Twitter, and I mostly want to defend the targets of the outrage. In fact, this might turn into a post category here once I feel like writing more again, and I did include a bit of that in my most recent videos.

Partly this change is clearly due to age, since I've gotten 15 years older in the time this blog has been around – though not everybody seems to be getting mellower or wiser with age. I don't intend to end up being the old man yelling at clouds. Life is too short for that.

What I want to do instead is figure out what to do here that is of value. I do think that making videos has value, even though getting people to actually watch them has been a challenge. But this isn't unlike the early days of this blog when nobody would read it. It's just not a topic people look for on YouTube. The next few videos will be on broadly interesting topics in the hope that I can get more people interested in.

Just like the frequency of posts here has slowed down, I also haven't felt the need to tweet nearly as much as I used to. I think this is again a change in what I feel is valuable, though also a reflection of how much time I spend on Twitter these days (a lot less than I used to).

Where to Next

So where is this thing going? I've been asking this question almost continuously over the years. And I still don't think I have an answer. I do have a long list of topics I want to write about, little apps I want to build, and certainly videos I want to make. It's hard to figure out where to best spend time, and the videos are a lot of work.

But this blog has been incredibly helpful to get word out about my work and that of others in the visualization research community. And I think that's a worthwhile thing to keep doing.

To another 15?

This chart was posted by Cal Fire a few days ago on Twitter. It shows the acres burned by the 20 largest wildfires in California.

Yes it's a pie chart. We're looking at ranked data here, even though the chart isn't sorted by size of the fire. It's sorted by time, when the fire started. The color scheme is chosen to group fires before 2000, from 2000-2019, and then 2020 and 2021.

The pie chart is clearly the wrong chart here because it's not a part-to-whole comparison. Or is it? Their tweet says, “all but 3 of the top 20 largest wildfires have occurred since 2000." That's a part-to-whole statement. The whole here being the 20 largest wildfires in California's (recorded) history, and the parts the fires before and after 2000. What the pie chart also shows is that these three fires before 2000 made up only a very small fraction of the acres burned. Sure, the top 20 is kind of arbitrary, but it's what they chose for this comparison.

I figured I'd try the usual, this should have been a bar chart. I'm not copying all the colors here, just roughly following their color scheme for the years. You can see which of the fires happened last year and this (red and orange), and also the three that occurred before 2000 (dark gray).

I don't think it works all that well, though. I'm sorting by acres burned, so I'm losing the time grouping (other than through color). But even so, the part-to-whole comparison here is much harder than with the pie. Comparing the sizes of fires is certainly much easier, but that's not the point of the chart.

It's also interesting to see how much more the bars for the largest two fires stick out here. I made a pie chart in Tableau as well just to check that my numbers were correct and that they didn't do something odd with scaling (they didn't, it looks the same). But the difference seems smaller in the pie chart than in the bar chart. It's an interesting effect. And it makes for a chart that lets you focus a lot more on the message, which is the comparison of different time periods, rather than comparing individual fires.

In my opinion, the pie chart is successful here. A bar chart would let you do other things, but it wouldn't work nearly as well to get the point across about the vast majority of the largest wildfires having happened in the last 20 years. And since that is their message, I'd say this is the correct choice of chart.

I’ve written about this chart and other downward-pointing bar charts before, but I’ve put a lot of effort into explaining why this is such a great chart in this video. I also talk about recent (accepted but not yet technically published) research that shows that aligning the chart direction with the emotional valence (good or bad) of the chart topic actually helps people read them.

Part of the reason for making these videos has always been that I wanted to show things the way I saw them. This piece in particular is extremely vivid to me, and I hope that my little animations bring it to life for people who might not otherwise be as impressed as I am.

I’m embedding the entire video (about 7.5 minutes) here. I don’t think my strategy of embedding the teaser worked, too many people seem to have just watched the teaser but not the full video. If you want to comment, please do that over on YouTube, though!

Also, if you like the video please give it a thumbs-up over there. That really helps tell YouTube that people think this video is worthwhile.

As in my previous video, there are numbered references in the video with the corresponding links in the notes below it. If you want to see more chart appreciations, I now have four that you can conveniently watch in this playlist.

Running Pace

This discussion started with this tweet by Andy Kirk making fun of a chart made by Under Armour, which shows pace for kilometer splits of somebody's run:

On the face of it, this looks completely bonkers. Here’s a bar chart where the vertical axis is upside down, and it’s not even starting at 0 at either end. How can this possibly make any sense?

If you’re a runner, you probably understand what’s going on here. Pace is measured in units of time per distance, so usually minutes and seconds per mile or kilometer. That makes for a much more relatable way of thinking about speed at human scale where a difference of, say, 10 seconds per mile is meaningful. Translated into km/h or mph, that would be miniscule.

The issue with this measurement is that unlike speed, pace gets smaller as you get faster. Running a mile in eight minutes is faster than doing it in ten minutes. So a numerically smaller 8 min/mi pace is going to make you quite a bit faster than a numerically larger 10 min/mi.

In a chart, that’s kind of weird. You could just do the “correct” thing in the visualization sense and ignore the meaning. But you’d end up having to explain that the shorter bars mean faster pace. That’s not very logical, and it’s not motivating – you want to see larger bars for faster miles. Performance benchmarks that measure completion time of a task (like these ones at Ars Technica) have a similar problem, so they usually include an annotation to say whether higher is better or lower is better (especially when that changes between different benchmarks).

Running apps have long shown pace data upside down. Here’s a quick screenshot I posted from Strava where they’re using a line chart. Note the scale on the right.

Line charts tend to get a pass when it comes to zero baselines (though that isn't actually warranted). Strava also shows splits as bars though, with the same inverted axis.

Other apps tend to do the same, as far as I’m aware. If you’re a runner, I think you’re used to this representation and easily understand what it means: smaller pace number means faster means better, which should make for a taller bar – visualization rules be damned!

We did discuss this a bit further in my twitter thread responding to Andy and started talking about how to represent this data differently (or better, if you will). Andy proposed this alternative, using bars but anchoring them on the average instead of a completely arbitrary baseline:

I agree that this is a good idea. It would actually be quite useful to see negative splits in a race (meaning your pace is faster in the second half than the first) more easily. Instead of the average, you might also pick a target pace to show the bars against. Then all the bars might point up or down, showing you how you were doing relative to your goal (again, most important for a race). But that doesn't change the scale being upside down so faster points up.

Audio Software and Interfaces

There are other places where axes can be scaled in unusual ways and make sense. The dB scale on the right edge in this screenshot from audio software iZotope RX goes from -∞ to 0, both up and down. Yes, that infinity. You're looking at not just one infinite scale, but two!

Sure, dB is a logarithmic scale, but the spacing between -20 and -∞ is not to scale here. The range between -20 dB and 0 dB is the most important for finding and fixing problems, so they gave it as much space as possible. And since the signal extends both above and below the zero crossing, they did that for both halves.

A similar reason is likely behind the uneven scaling of the level indicators on this MOTU audio interface. Note that the top three labels on the vertical axis (on the left, this time) are spaced 6 dB apart, and the distance then triples to 18 dB for the lower four. That again gives a lot more space to the range between -12 dB and 0 dB, where you want more precision to avoid overdriving the mic preamps or other input and outputs.

Here's one more example, this one from Logic's compressor. This meter shows the reduction in level by the compressor, so it starts on the right. It's kind of logical actually since the reduction is a negative number, but the spacing is still quite creative: note the even subdivisions on the right side of -10 dB and how they differ from the left. This is not a logarithmic scale either, it's really two different scales that meet at -10.

Visualization dogma of course does not allow such unevenly-spaced axes. But many decades of audio devices show that they work and serve a purpose, just like they do with the (seemingly) weird running pace charts.

Expertise vs. Visualization Rules

I think this little discussion highlighted something important about how we need to think about charts in the context of expertise and use. We aren’t chart-reading machines that mechanically follow supposed chart rules when working with charts. We can adapt. We can use our intelligence. We can read axis labels. We can understand context.

Distorted axes can be a problem, no doubt. But distortion isn’t always wrong, and in fact sometimes is the better way to go. Use case, convention, and expertise trump visualization rules any day. And I think these two examples, in particular the one about running, is one that many people can relate to who otherwise would scoff at the outrageously bad charts used by experts in some field they don’t understand.