Conversations with Data: #31
Do you want to receive Conversations with Data? Subscribe
Data visualisation can be dangerous -- people inherently trust a map or chart. But, as we know, all maps lie and charts can be just as tricky. While that doesn’t mean that the truthful journalist should avoid visualisation, there are some common mistakes to watch out for.
Welcome to our 31st edition of Conversations with Data, featuring visual faux pas that other data journalists made, so you won’t have to.
What you said
Perhaps it’s fitting to start with some sound advice from Alberto Cairo, whose new book How Charts Lie deals directly with this issue: “...paying attention to charts is critical to understanding them. Many readers -- and this includes many journalists -- believe that charts can and should be understood easily and quickly, as if they were mere drawings. This is a dangerous assumption. Charts need to be read carefully, and this includes the descriptions and documentation their creators provide -- or should provide.”
Alberto provided us with the above graphic, by Our World In Data's Hannah Ritchie, as an example of what he means. It's well designed, but it can be easily misinterpreted if we don't read the story where it is inserted and the numerous caveats it contains.
- Avoid junk charts: “Chart junk refers to visual elements in charts that are not necessary to comprehend the information, or that even distract the viewer, or skews the depiction and makes it difficult to understand.”
- Remember the rules of data visualisation: “It is great to come up with unique and creative solutions, but in most cases it is best if you go with the established rules. The audience is used to certain ways of reading data visualisations, it is your job to decide if there is space for you to challenge these ways.”
Now that we’ve got the basics down, let’s turn to some more specific advice on particular chart types. Like many of us, Cole Nussbaumer Knaflic is a fan of bar charts: “they are familiar and easy to read. However, they must have a zero-baseline! Because we read the length of the bars relative to each other and the axis, we need the full bar to be present in order to make an accurate visual comparison.”
“This bar chart appeared on Fox News in 2013 and violates a data visualisation rule that nearly everyone agrees on: Bar charts must start at zero! Here, the top marginal tax rate in the United States was potentially slated to increase from 35% in 2012 to 39.6% in 2013. By starting the chart at something other than zero (here at 34%), the difference between the two numbers is overemphasised.”
When it comes to maps, journalists need to be just as wary. Let’s take a look at the following maps provided by Highsoft’s Mustapha Mekhatria.
“The first map displays Canadian provinces where members of the First Nations make up the majority of the population (in green). Given the large geographical area of these provinces, one might infer that a sizable part of the Canadian population is First Nations. However, the fact is that First Nations members make up only 5% of the total population. Unless the focus of the data story you are trying to tell is the geographical size of these provinces, you should complement your chart with an explanation, ideally also in chart form (see the second map), or don’t use the map,” he explained.
Likewise, Kaiser Fung of Junk Charts fame, also critiqued a map of China’s ‘digital silkroad’, which plotted the wealth of each province (measured by GDP per capita) and the level of internet penetration in geographic form.
“Irrelevant text, such as names of surrounding countries and key cities, expands the processing load without adding insights,” he warned.
As an alternative, he suggests conveying the correlation between the two variables in a grid format, with limited text to avoid overcomplicating the visual’s message.
“This pictogram chart on colour blindness is confusing because makes it looks like more women have the condition because of the use of more female icons than male ones. However, it's tricking you with statistics as 1 in 12 is way more probable than 1 in 200. If we calculate the percentages you get: (1 ÷ 12) x 100 = 8.33% men have colour blindness; (1 ÷ 200) x 100 = 0.5% women have colour blindness,” he told us.
Want to learn more? Jonathon Berlin covers data visualisation and mistakes he’s made in our DataJournalism.com course here.
Our next conversation
It’s been a while since we last hosted an AMA! So, in our next edition, we’ll have The Economist’s data team with us to answer your burning questions. Ask them about building statistical models, how they’re trying to make their data more open, or anything at all!
As always, comment to let us know what you’d like us to feature in our future editions.
Until next time,
Madolyn from the EJC Data team