Update (August 6, 2016): Based on some feedback on reddit and elsewhere, I’ve added to the post. Changes are at the bottom of the post, and are clearly marked.
But sometimes they don’t quite get it right. Here’s an example from an article about the political divide in rich countries shifting from a left-right to an open-closed dichotomy.
In one section, The Economist writes about the different popular attitudes toward newcomers of different races and ethnicities. To illustrate this point, they use a graph that shows survey results from the Pew Research Centre.
As far as graphs go, it’s good enough. It does its job of communicating how attitudes are different among countries. I can tell, for example, that Americans seem significantly more welcoming of newcomers than Greeks or Poles.
This graph could be so much better, though. Inspired by this post from storytelling with data, I decided to try and improve on “good enough”. Here are the shortcomings I noticed:
- The stacked bar chart. Stacked bars are easy to make and efficient with space, but tough to read. Look at the graph again. It’s easy to compare the respondents who replied “Better”, but tough to do anything else. What’s bigger: Spain or France’s “Doesn’t make much difference” percentage? By how much? You can look closely to figure it out, but you shouldn’t have to look closely. It should be obvious.
- Ugly colours. Dark blue, light blue, and peach. Maybe it’s just me, but that’s not an appealing colour scheme. Let’s say it is just me, though. Beyond being ugly, it’s ineffective. The three groups being compared have a positive, negative, and neutral sentiment. To me, neutral = grey and negative = red. Using dark blue for positive and light blue for neutral seems non-intuitive.
- Non-obvious takeaway. What am I supposed to take away from this graph? After studying it for a few moments, conclusions like “the United States is more positive than Europe about newcomers” and “Italy, Poland, and Greece are particularly negative about newcomers” come to mind. These conclusions shouldn’t come after a few moments of study; they should immediately hit me in the face like a bag of cement.
I went through a few iterations of improvements. Here’s the progression. First, I separated the stacked bars into individual bar charts and changed the colours. These changes fix issues (1) and (2), which I think automatically fixes issue (3).
Then I thought more about issue (3), the non-obvious takeaway. I thought, “Do we really need the ‘doesn’t make a difference’ category? Do we lose the message if we get rid of it?” I don’t think we do.
The real takeaway here is the difference between people who think their country is better or worse because of newcomers. Including the neutral categories just completes the data, but it doesn’t advance the message.
Here’s what it looks like when we get rid of it. I decided to flip the “Worse” bar chart so that it went from right-to-left off a shared axis. I also put the country name on the side of how the plurality of people responded. On countries with narrow margins, like France, this creates an extra visual cue for the reader.
That looks pretty good!
One criticism is that the improved graph is huge, whereas the original is nice and small. For print publications like The Economist, compact graphs are probably important. We can shrink the improved graph down.
I’ll let you be the judge, though. Here’s the Before and After, side by side. What do you think is better?
Everything from this point onward is part of the update to the post.
Wow, lots of feedback. Some people like the changes and a lot of people don’t. The most common criticism is that removing the “Doesn’t make much difference” group distorts the data and I shouldn’t have done it.
That’s a fair point. So I created two alternatives that include the neutral group.
First, we can simply unstack the bar chart, which gives us something like this:
I don’t love this version. Like the original graph, I think it’s “good enough”, but in different ways. You can compare individual values more easily, but the new graph is messy. It’s hard to discern an overall pattern because there’s so much stuff competing for your attention. Try focusing on just the “Worse” group. You can do it, but it’s not easy.
So this is an option, but let’s try something else. What if we take the original revision and address the main concern, which was that I omitted the neutral group? We get this:
That’s not bad, but it’s not as hit-you-in-the-face-with-a-bag-of-cement intuitive as my first revision. Beyond that, it’s best suited to making comparisons among countries. When it comes to making comparisons within countries, it’s not very good.
There are tons of possibilities, but it really depends what aspect of the data you’re trying to highlight. The lesson here, for me at least, is that you’ll always have to make a tradeoff. If you go for completeness, you may have to sacrifice clarity. If you want to make easy comparisons among countries, you may have to settle for hard comparisons within countries.
Ultimately, the most important thing is to choose what matters most, and let that guide your design decisions.