Communication has been quite a challenge during the COVID-19 pandemic, and data visualization hasn’t been the most helpful given the low quality of the data – see Amanda Makulec’s plea to think harder about making another coronavirus chart. A great example of how to do things right is the widely-circulated Flatten the Curve information graphic/cartoon. Here’s a look at the work it is built on and how that has evolved from a figure in an academic paper to one of the clearest pieces of visual communication in some time.
The information graphic was created by epidemiologist Dr. Siouxsie Wiles and illustrator Toby Morris for the New Zealand publication The Spinoff. I think it’s fair to call this a cartoon, and I mean that in the best possible way.
Why Does It Work?
I think what makes this infographic work is the combination of a few key elements: a clear and straightforward message, a foundation on science, a clear tagline that you might call actionable, and enough visual elements to be informative enough without getting overwhelming.
To be clear, this is not data or a visualization. It’s an illustration based on a conceptual drawing, which in turn is based on simulations. It’s also a cartoon, and I don’t mean just the two characters under the chart. The chart itself is basically a cartoon, and that’s a good thing. Real data would be much messier and spikier, so a smooth and simple drawing works much better to get a message across.
People have wondered if the flatter curve should have the same area under it, but I think that’s also missing the point. This is not data, it’s showing the idea of what would happen in two different scenarios while trying to not get into the weeds of small details of the simulation, its settings, assumptions, etc.
If you want to get technical, there are also annotations on the chart: the line showing the healthcare system’s capacity is a key to making the point here. And the little cartoon of the hospital nicely illustrates the idea of an overloaded system (and does it in a tasteful and subtle way – this might actually be my favorite part of the whole thing).
The Visual History
It’s quite interesting to trace the history of this chart through a number of different stages. One early version looked like this, from a paper by Kelso et al. in BMC Public Health, published in 2009. It shows the results of simulations for different vales of R0, which is the rate at which diseases spread between people. That value can be reduced by closing schools, etc., which is shown on the left here compared to not taking action on the right.
The rest happened very quickly. Fong et al. combined the two curves into one chart, which made the difference more apparent. This paper is still only pre-published in the May 2020 issue of the Emerging Infectious Diseases publication by the CDC (though as David Napoli points out on Twitter, a similar chart was published by the CDC in 2007).
Dalton et al. then added another crucial element, the conceptual line showing healthcare system capacity. This has a publication date of March 5, 2020, on SSRN (which appears to be a sort arXiv for social sciences research).
From here, it’s easy to see how this turned into the iconic graphic (there’s also some interesting backstory in this Fast Company piece). However, I don’t think it would haven been nearly as effective without the three simple words Flatten the Curve written across the top. I believe we’ll be looking back at this as a prime example for effective visual communication for a while.
Simulations Instead of Data?
Some of the papers mentioned above are based on simulations. Harry Stevens at the Washington Post has created a fantastic article with several simple simulations that show different strategies for containing an outbreak and what their effects are. The simulations are easy to follow and I especially love the charts of healthy, infected, and recovered people that build up as the simulations run.
Given the current state of testing, especially in the United States, the current data is extremely unreliable. I’m afraid visualizing this low-quality data is not going to be of much use. Simulations and information graphics or cartoons, especially done in a smart and tasteful way like the Flatten the Curve cartoon, are a great alternative.
4 responses to “The Visual Evolution of the “Flattening the Curve” Information Graphic”
Thanks for highlighting this.
I am sceptical about most forms of this “flatten the curve” graphic for a number of reasons, which touch on the role the visual plays in understanding, communication and behaviour change.
Firstly, we in the datavis field are always on the lookout for success stories that show the impact of visualization in “our” domain so will often amplify what we think might be good examples. I think it important to be aware of that to temper uncritical contexts in which we share visualizations (I’m not suggesting you are doing that Robert, but I think there are many who have with various flatten the curve examples).
You say that there has been discussion about the relative areas of the peaked and flattened curve but that is not the main point. I disagree and think there are details there that are critically important. In particular, versions that show approximately equal areas under both curves imply this is just about temporal spreading. To have an unskewed normal distribution for the unchecked curve implies herd immunity is responsible for the reduction. This in turn implies a vertical scale that encompasses the majority of a circulating population (somewhere in the range 50-80%). Combine this with a health capacity line that is somewhere around half the peak, is a major assumption that I struggle to find evidence for at the scale of the Covid-19 pandemic. It is quite possible that that dashed line for many countries (including the US and UK) could, in reality, be very near the bottom of the y-axis. Even though these are ‘schematic’ charts, we have to take care in what they imply. For example, Joscha Bach shows a scenario where to flatten the curve below a typical health capacity ceiling might take over a decade – https://medium.com/@joschabach/flattening-the-curve-is-a-deadly-delusion-eea324fe9727 Of course the assumptions he makes are open to challenge, but the point being, so are the assumptions made in the more widely circulated charts. I don’t think we can hide behind “it’s just a schematic” when the outcomes of different assumptions are so radically different.
I think there is also a wider issue about whether simply because a chart is widely circulated, it is evidence of a (positive) effect. Strategies to flatten the curve to reduce peak demand may well be important at the population/policy level, but can we say it is equally important at the individual level? I suspect there are much stronger motivations for individual behaviour change. I wonder if part of the appeal of the graphic is that people simply get the idea, one that they had not previously considered and that there is some satisfaction in doing so.
Interesting, thanks! I do think that “it’s just a schematic” is the reason why it’s fine the way it is, though. There are lots of questions you can ask, the area is only one of them (and is it really the same or should be more or less?). You mention the total duration, but there’s also the shape, it’ll likely not be symmetrical. And it won’t be nearly this smooth.
I think there is a place for something that’s simplified and clear, which is completely separate from an actual visualization of either data or more simulation runs with different assumptions, parameters, etc. Of course those should be done, but those will never make it to the general public.
I have no idea what the cause-effect is between the popularity of the graphic and people’s behavior. Maybe people just share it because it reinforces what they’re already doing (or are perceiving to be the right thing). And it’s not like it would be easy to test how much it influences people. But I’d be surprised if the reinforcement hadn’t helped some percentage of the population take the measures like washing hands, social distancing, etc., more serious.
Seems like Carl Bergstrom at UW has influenced this representation a bunch, e.g.,
Hi Robert, I am a long time follower of your blog posts, they have influenced my dataviz journey a lot. Thanks for sharing this post. Could you say a word about how you evaluated the data quality to be low?