Last November, Andy Cotgreave declared bar chart races the “fidget spinner of data viz.” They’re captivating to watch but they’ve hit their sell-by date, he argued.
Cotgreave, a senior technical evangelist at Tableau, voiced his criticism shortly after Data Is Beautiful announced a temporary moratorium on bar chart races — animated charts in which horizontal bars overtake each other over time — after the popular data visualization subreddit became inundated with the style.
It was, in Gartner hype cycle terms, the method’s trough of disillusionment.
Cut to nearly a year later and Cotgreave still has plenty of reservations. But as bar chart races continue to go viral and become even more prevalent in the public consciousness, he’s no longer sure the phenomenon is a fad.
“Maybe in 2020 it’s time to reassess that claim,” he told Built In.
One case in point: Dan Goodspeed’s bar chart race of COVID-19 cases by state partisanship. It was shared thousands of times on social media and garnered a wave of press attention last month.
Even Joe Biden’s campaign made a COVID-related one this year. Theirs visualized COVID-19 cases by country. It, too, took off.
Aiding the rise are companies like Flourish and Observable, which have made it easy to create these animations. The former offers a drag-and-drop template, and the latter has a customization-friendly, forkable notebook. Both companies offer far more visualization options than bar chart races, but few have captivated broader audiences like the race. (Goodspeed’s chart was made using Flourish; another seminal entry, a most-populous-cities race created by Financial Times senior data visualization journalist John Burn-Murdoch, was made with Observable.)
How Did We Get Here?
It’s tough to pinpoint the ur-bar chart race, but the first broadly shared example was a race that tracked which companies placed among the top 15 global brands between 2000 and 2018. A version posted in February last year by Montreal-based professor Arthur Charpentier was shared thousands of times. But it had also appeared a week earlier on TheRankings’ YouTube channel — an account that’s been creating and posting bar chart races since 2017.
Now there’s an entire ecosystem of YouTube accounts devoted to bar chart races. Sometimes they get meta, like this bar chart race of the most popular bar chart race channels.
From there, the aforementioned most-populous-cities race by Burn-Murdoch was the watershed. It was so popular that he made a video follow-up, adding contextual commentary that explained and pointed the viewer to notable chart shifts. By that point, the match was lit. Soon, the form’s cousin, the line chart race, arrived.
Within months, Cotgreave would be applauding Data Is Beautiful’s pause-button push, only to see the trend maintain steam into 2020.
So What’s the Beef?
Aside from ubiquity, Cotgreave’s chief complaint is that, although such races can be attention-grabbing, he finds himself not retaining much of the information being put forth — or even knowing what piece of information is worthy of note, as the bars vacillate up and down. In a nutshell, it’s information overload, with little to guide the eye, his argument goes.
“It should have been a line chart,” he argues, noting that a static line visualization would allow viewers to track durational trends across time and simultaneously pinpoint any specific moment-in-time information.
Flourish co-founder and former data journalist Duncan Clark pushed back on that criticism in an email interview with Built In. He told me bar chart races often visualize data sets that have very large numbers of entities, even as some fall in or out of rank, “whereas a chart with 50 or 100 lines would be very messy.”
“Line charts are great at efficiently showing the overall picture at a glance on a fixed scale, whereas a bar race typically has a dynamic axis, which makes it easier to focus on rankings over time, regardless of scale,” he explained. “They’re just different tools for different purposes.”
Clark also cautioned against universalizing retention concerns.
“I’d love to see some real research on [that] question,” he said. “My hunch would be that bar chart races score well on that metric, for the simple reason that they seem to be unusually good at holding attention — perhaps because they touch some deep human instinct to watch things that move. But I’ve never seen this measured scientifically.”
No such research exists yet, but when Burn-Murdoch spoke with PolicyViz last year about the bar chart race trend he helped launch, he referenced the seminal 2008 animation study Effectiveness of Animation in Trend Visualization, led by George Robertson. The paper remains widely cited and was required reading in data visualization classes taught by Jeffery Heer, who helped develop the visualization library D3 alongside future Observable co-founder Mike Bostock.
Participants in the study who viewed animated visualizations read values less accurately than those who viewed static visualizations, but not dramatically worse. At the same time, participants enjoyed the animation more.
“[U]sers repeatedly reported that the animation condition was ‘fun’ or ‘exciting’ or, in one case, ‘emotionally touching,’” the authors wrote. “Yet they also found it confusing: one user complained that ‘the dots flew everywhere,’ and shook her head in frustration.”
Such tension always exists in animation. “That seems pretty intuitive: when things are moving around, it’s harder to get a measure on them,” Burn-Murdoch said on PolicyViz. “But if you’ve also got people saying they really enjoyed the experience of consuming a visualization, they found it fun, exciting, emotionally touching, and they gave it more focus, then I think that’s noteworthy as well.”
That emotional resonance is the driving force behind the appeal of bar chart races, Bostock told Built In. As in any race, there’s an element of mystery and anticipation. “Not everything is visible upfront, and the waiting gives you a bit of an emotional investment.... If you knew the winner ahead of time, you probably wouldn’t watch it,” he said.
“The bar chart races you see on social media are typically entertainment, not analysis.”
Even a skeptic like Cotgreave acknowledges that particular appeal. “Animation creates drama — moments of crisis and moments of conclusion,” he said. Although for him, the drama isn’t worth the time investment. “Why is a bar chart race popular in a social media, fast-moving world with short attention spans, even though the information is given in a really slow way? It’s a contradiction that baffles me,” he said.
Cotgreave was also quick to note that it is sometimes worth pushing back against data viz best-practices purity, as he called it, including the idea that “it should’ve been a line chart.” He echoed a sentiment he made in 2016: “[G]etting people to engage is sometimes as important as building the cognitively most valid method.”
He offered another olive branch to the pro-racing crowd: If fluctuations in a bar chart race happen to prompt unanswered questions within a viewer, yet the chart nonetheless illustrates an intended overall point, that’s not the visualization’s fault. “Those subsequent questions are you demanding more of the chart,” he said.
Still, Cotgreave finds many bar chart races wanting. As an example of a better approach, he pointed to ProPublica’s map of coronavirus positivity trends. The top map — which, like bar chart races, has an interactive time-lapse element — provides trend information in a quick, compelling way. But below is a host of state-level visualizations with more granular information.
That idea of ramping up complexity can be key to more effective visual communication, Bostock said. He referenced a scrollytelling-style article published by the New York Times in 2018 about phony Twitter followers. “You don’t necessarily have to worry about understanding everything up front,” he said. “It was a more gradual onramp into understanding this fairly abstract but interesting and detailed visual analysis.”
There’s More to Love. And More to Maybe Not Love.
As one of the minds behind both D3 and Observable, Bostock has done a lot to democratize bar chart races. He’s keenly aware of why they’ve proven so captivating. “There are good things about bar chart races that are perhaps not always considered if you’re taking a purely rational perspective,” he said.
One reason for their success? They move! Movement naturally grabs the eye. Such grabbiness “encourages it as a popular form, for better and for worse,” Bostock said.
Also, they’re intuitive. Information visualization often means mapping spatial orientation onto some abstract concept. Here, “time is mapped to time,” he said. That in turn cuts down the intimidation factor. Bump charts, for instance, would lend themselves better to careful analysis, but they’re not exactly approachable.
“[Bump charts] are really good for seeing parallel activity across different parts of some system,” Bostock said. “They’re also pretty intimidating.”
It’s easy to wonder if the social media-ization of bar chart races has driven the backlash. Both Observable and Flourish allow visualization designers to include a scrubber bar, which lets users toggle to different points along the time spectrum rather than only experiencing the chart as a whooshing rush of action. That element of control and clarity gets compromised if the chart is ripped to a Twitter or Reddit video. (Or, as above, a GIF; apologies for our own embedding limitations.)
“As with any visualization type, race charts can be done very well or very badly.”
Flourish has also developed a pop-up caption feature that can annotate and explain notable shifts within the chart and also guide attention. (“Watch China and India rocket from the mid-70s!” reads a pop-up on the most-populous countries race in the link above.) In addition, the company has built an audio-layer feature, so that contextual voiceover can be added to a visualization, bar chart race or otherwise.
Still, Bostock wondered if such “heavy-handed” intervention might be a signal that a more direct, static visualization is in order. No matter how much annotation or voiceover, a bar chart race still only shows a single point in time at a given moment — unlike those intimidating but useful bump charts, he added.
“If we’re honest, the bar chart races you see on social media are typically entertainment, not analysis,” Bostock said. “If you want to analyze a data set, you’re going to want to see the time dimension mapped into space, so that you can see those fluctuations, trends and patterns.”
He continued: “If something is critical to your job, and you need to understand what’s going on, you’ll happily spend much more time with a more rationally optimal encoding, because you’re willing to spend the time to learn what that means and interpret.”
Alternatives that deliver some of that animated storytelling kick of racing bar charts while also boasting more analysis-friendly aspects of static choices include animated treemaps and this popular approach to visualizing the effects of climate change, Bostock added.
Maybe They’re Neither Good Nor Bad, but Just 1 Option Among Many
Bostock is direct about the appeals and limitations of bar chart races, but he also doesn’t make an up-or-down value judgment on them. Similarly, Clark stressed that it’s really a question of situational applicability and designer capability, rather than the good-or-bad binary that sometimes overtakes data viz debates.
“The bar chart race is just one of hundreds of visualization styles, each of which is suited to different situations,” Clark said. “As with any visualization type, race charts can be done very well or very badly, so any grand sweeping statement about them is unlikely to hold true.”
“You don’t need to be ‘for’ or ‘against’ any one visualization style, or think one will always be better than the other,” he added. “The skill is picking the right option for your data and story, and then using it well.”
To that end, Burn-Murdoch last year put forth three criteria that might signal that a data set is a good candidate for a race visualization. Along with the above perspective, these pointers might guide curious parties:
- If the values in a data set change substantially over a given period of time, as in his population visualization.
- If those changes include entities moving in or dropping out over time.
- If the data set naturally lends itself to a race framework. “So anything where it’s ranking,” he said.
However, if you make your bar chart race with Flourish, you can’t take it to Data Is Beautiful. “Visuals made with this tool have become too prolific and are no longer allowed,” say the subreddit’s updated rules. The latest ban, issued in July, stems from a spamming issue that, in a cruel irony, may have been facilitated by the racing bar chart template’s ease of use.
“It’s a bummer because it looks like Flourish can be used to make some great data viz,” Data Is Beautiful community leader Randy Olson wrote on Twitter. “However, at some point a spammer ring discovered that it was dead easy to make racing bar charts on there and to use those charts to promote spam.”
The latest ban was met with both cheer and disappointment. “hallelujah I despise racing bar charts,” wrote one user. Another user replied, “I didn’t mind some of them if they were actually interesting. But like the mods say lots of them were low quality and of dubious interest.”
On the other hand, one user thought “it seems harsh to ban use of a tool, particularly one that has elegant viz at its heart.”
“Slow them down and you get rid of 90 percent of the objections,” posted another account.
Love them or loathe them or embrace the nuance in between, the latest deluge again points to their endurance — unlike those faddish fidget spinners.
In fact, Cotgreave’s employer, Tableau, now also supports bar chart race designs.
“I’ve got a blog post decrying them, followed by a blog post on how to build them in Tableau,” said the style’s most vocal critic. “Which one got the most hits? Of course it was one on how to build them.”