Analysing The Economist's excess deaths tracker
One way to account for the total number of fatalities caused by a pandemic is by measuring excess deaths. This is exactly what the British weekly magazine The Economist was set out to do in April last year when one of its data journalists James Tozer began tracking excess mortality statistics and, together with visual journalist Martín González, launched its Covid-19 Excess Deaths Tracker, which makes good use of the data in various charts. It has since been short-listed for Digital Initiative of the Year at the 2020 National Press Awards.
González wrote: “A better count of the casualties of the novel coronavirus can be achieved by calculating a figure that we call ‘excess deaths’, a comparison between the total number of deaths, the expected amount of deaths in a certain period and the official Covid-19 count.”
The tracker presents the data in a range of different graphics, but the visualisation I will be evaluating is the heatmap titled Excess deaths by country or city, which shows the percentage of deviation from expected deaths from January 2020 to February 2021 in 70 countries and one city around the world.
From my point of view, this visualisation efficiently represents the strength of excess mortality data as a way to capture the true extent of the pandemic. It allows for comparisons between countries and, either inadvertently or purposefully, also shows how patchy data on deaths around the world is, with some of it incomplete or reported late.
Statistician and data visualisation designer Nathan Yau wrote that “information graphics are technologies, means to fulfil purposes, devices whose aim is to help an audience complete certain tasks”. In this case, the graphic has a defined and clear objective: communicate that, in most countries, the number of excess deaths is greater than the number of covid-19 fatalities officially recorded by governments.
Would there be a better, more efficient way of presenting the data? According to Cleveland and McGill’s ranking of methods to represent data according to how accurately the human brain can detect differences and make comparisons between them, that would be the case. As per their ranking, journalist and designer Alberto Cairo wrote that a bar or line chart is always superior to a heatmap if the goal of the graphic is to facilitate precise comparisons.
However, given the number of variables presented, I believe a heatmap is an appropriate data visualisation technique in this instance. A quick glance at the visualisation allows the viewer to grasp its most important features, giving obvious visual cues about how excess mortality is clustered and varies in different countries over a period of time. Data scientist Paul Tulloch wrote that heatmaps are great at providing a simplistic view of multiple variables from a time series perspective, as it provides a very visually noticeable view that can illustrate patterns and replace a multitude of graphs.
Although the graphic does not necessarily respect Tufte’s data-ink ratio given it presents very little white space, it is not overwhelming, does not prevent readability and is still aesthetically attractive. The data is the most important part of the visualisation and there are no excessive additions that overshadow it.
Because there is less dependency on size or position, a lot of data can be encoded at once by using colour. The data of this graphic is decoded through a sequential colour scheme, representing a range from low to high. The more saturated the hue, the higher the deviation from expected deaths for a specific time period.
The colours are used intuitively for the phenomena that they are representing and make it easy for the user to distinguish differences. The visualisation accounts for tritanopia, protanopia, deuteranopia, but it does not look as good in black and white as the lighter colours are not easily discernible and harder to decode.
Negative values are represented in light grey, while values from 0 to +200% are rendered from a very faded pale orange to dark red. The usage of just six different colours helps readers decode the data faster. “The more colours in a chart represent your data, the harder it becomes to read it quickly,” wrote Lisa Charlotte Rost from Datawrapper.
The same pale orange is used for values ranging from 0 to +25%, which could be slightly problematic. In order to improve the visualisation, I would use a separate hue exclusively for 0% deviation of expected deaths to represent unchanged values. Rost recommends using a maximum of seven colours in the same graphic, so although tight, it could still work.
The graphic presents a colour scale legend at the top right corner to help readers understand what the visual encoding indicates. The challenge for readers is that they have to refer to the legend and look away from the actual chart to see what everything means. Luckily, the developers thought of a way to maintain the focus on the chart.
Even when a legend is provided, heatmaps can make deciphering the exact values represented quite hard. In this case, when hovering over a specific square, a tooltip with the actual figures for each time period pops up. While the visualisation itself shows the big picture of the data, the web interactivity allows the reader to see the details while preserving the overall message.
Overall, I believe this visualisation meets all the requirements that Yau proposed of what makes a good, readable graphic. It removes confusion, has a clear purpose, uses a visual encoding that makes sense for the context of a dataset, offers a clear direction for how to interpret and, most importantly, manages to provide clarity in the midst of a Covid-19 information overload.