Tips and tricks for qualitative data visualisation

Data visualisation is often seen as a numbers game - taking a data table of figures and turning it into shapes, icons and/or illustrations in order to make it more accessible. But what if your data is a series of interviews? Or a concept? Or perhaps a process - like the greenhouse effect or how to defeat Ganondorf in Zelda: Tears of the Kingdom (contains spoilers)?

This is known as qualitative data and it tends to be, in my view, a lot more interesting than its quantitative cousin. It’s at the heart of almost every news story, history book and documentary. Much academic research - particularly in the humanities and social sciences - will be qualitative. And because of this, it’s often where the real insights lie, because it helps us understand how humans think and feel, rather than simply observing and quantifying their behaviour from the outside.

The trouble is, because the data is more abstract, it is harder to get qualitative data viz right. When you visualise numbers, help is at hand for every stage of the process. Tools like OpenRefine can clean your data, Excel or R can help you organise and analyse the numbers, Raw, Datawrapper and Flourish allow you to make attractive-looking charts, and there is a vast library of books, blogposts and video explainers (including ours) dedicated to how to visualise quantitative data effectively.

Qualitative data viz doesn’t have the same support network. Its default charts (Venns, Word clouds, flow charts) can feel faintly comical compared to those quantitative titans - bars, lines, pies and scatters. Also qualitative data frequently requires something bespoke - as few concepts or systems are entirely identical - and therefore best practice examples aren’t always useful, even when they exist.

However, after 15 years of making qualitative data viz, I hope I can offer a handful of useful guiding principles. And (shamless plug klaxxon) I’m also giving a Guardian masterclass with Tobias Sturt next week, where I’ll explore all of this in more detail.

1. Can it be quantified?

Perhaps the easiest way of visualising qualitative data is figuring out if any of it is quantifiable. If you are able to count words or phrases, then you can chart them. David McCandless does this here, counting the number of times that novels are mentioned in prize lists, book polls and top 100s and creating a word cloud of Books Everyone Should Read.

But a word cloud is just one option. If you’ve turned your text into a data table (I use the free tool Voyant for this), then most charts will work, and in fact, it’s often easier for an audience to count the number of words or phrases in a more conventional chart. In this example that I made with Jim Kynvin, we counted the most common adjectives in Thomas Hardy novels and put them in a treemap.

In this example from a series on the Gothic novel that I worked on with Zhenia Vasiliev, we represented mentions of weather using a stylised bar chart.

2. Can you combine qual and quant?

Although they are treated as different types of data, it is rare for us to create a graphic that is purely qual or purely quant. The most effective data stories fuse both. Your quant data might show that something big has happened (e.g. the global temperature is rising), but you’ll need qual data to explain why. Or vice versa - you might be making an argument based on qual research (e.g. we should lower the voting age to 16) but you’ll use quant data to back it up (the percentage of 16 and 17 year olds that pay tax).

So if you’re staring at qual data, wondering how to create a distinctive visual, then it might be that you can keep the text as text (or text and icons), but use quant data - with its formidable charting arsenal - to lend visual support.

In the example below about the EU’s Digital Services act, we were asked to show how the law was going to change to give internet users more protection, but we used quant data to show why the law was needed, and to give us more visual options. For example, the law will crackdown on hate speech because 52% of young women have experienced abuse online.

3. Think like an editor (or hire one)

But sometimes all your data is qual - it is a case study or a process or a hierarchy - and quant data is unavailable or irrelevant. So you will need to think about which aspects of your qual data will benefit from visualisation, and which should stay as text in paragraphs. A clue here is the doodles and sketches you (probably) drew when you were trying to understand and organise the data yourself. Your audience will definitely feel the same way. If putting text into boxes and organising them by theme helped you, then it will help them too.

Once you’ve figured out which text will benefit from visualisation, the next job is an editorial one. You will have far far too many words. So keep the relevant text in a Word or Google doc, and cut it down, making it as clear and jargon-free as possible. Look at graphics with the same dimensions and objectives as yours to get a sense of the ideal word count. If your graphic needs to be 16:9 widescreen (e.g. it’s a PowerPoint slide), then find other 16:9 graphics that work.

If you insist on not cutting anything and keeping all the text, then design can’t save you.

Next, think about the information hierarchy. If you’ve got text in boxes, which order should they go in? Which ones are the most important? Which are connected? Are you mirroring two arguments? Dos and Don’ts? Myths and truths? Sketch your ideas first, using your final text (never lorem ipsum) to check everything fits and to check everything makes sense. It probably won’t, so you’ll have to sketch it again - and again. But that’s as it should be: ‘the first draft is just you telling the story to yourself’ (Terry Pratchett).

In this series of graphics I made about the most common first names for different professions (with design by Zhenia Vasiliev), I sketched constantly, trying out different professions (lawyers, footballers, teachers), experimenting with four circles, two circles, fifty names in each circle, twenty names in each circle. Only when I was happy with the choice of professions and the number of names did I hand my sketch to Zhenia for him to work his magic.

4. Find a designer to collaborate with

Whenever an academic researcher asks us how to design, I always feel like I’m in one of those films where an air traffic controller has to talk a passenger through landing a 747 because the pilot has just had a heart attack. In the films, the passenger always miraculously pulls it off, but in reality, it ends in tragedy. Always.

You need a pilot to fly a commercial airline, and you need a designer to design, because both are difficult, involving a combination of natural aptitude, years of study and a wealth of on-the-job experience. (Just like being a researcher, in fact. Or anything else).

If you’re an academic or an editor, I would focus on doing the part that you are qualified to do: getting the content right, ideally the content structure too. Then work with a designer to turn this wireframe into a compelling visual. That’s what most of the best examples of qual data viz have in common: look at this by David McCandless (text) and Stefanie Posavec (design), or this by Miraim Quick (text) and Stefanie Posavec (design).

If you don’t know any designers, or don’t have any working in your organisation, then yunojuno is a great place to find talented freelancers. Organisations like the Data Visualization Society are also fantastic at bringing analysts and designers together - designers are often looking for interesting self-initiated projects to work on.?

5. Can you go interactive?

Finally, going interactive can be another way out of the ‘too many words’ conundrum that I mentioned above. Because sometimes with qualitative stories, the wealth of content is the whole point. If I were creating a visual to represent all the testimonies to the UK Covid Enquiry, then I would want all the text to be included somehow because, as the enquiry has rightly stated, ‘every story matters.’?

This is where interactivity can help, as you can communicate main findings at the top-level (revealed on scroll), but then hide the rest of the content in optional click-throughs.

One good example is this from Bloomberg - exploring the ‘479 things that can give you cancer’. Notice how the designers walk us through the key story, but then leave us with the full list of nasty substances at the end (including, sadly, alcohol), and allow us to click through to the text of the relevant academic article in each case. Perfect for both casual readers and experts. Nothing is left out, but secondary content is only available if the user requests it.

As above, this requires collaboration - usually between an editor, a designer and a developer, although in the case of Bloomberg, they managed with just two (clearly very clever) people - John Tozzi and Jeremy Scott Diamond.

Conclusion

Of course there are many other information design guidelines we stick to: always use grids, use font consistently, one style of icons only, lots of white space. But the five principles above tend to take care of all these. A good designer (principle 4) will do all of these things anyway, particularly if an editor has given them a clear wireframe (principle 3) with a good mixture of qual and quant material (principle 2). If I were to be allowed a sixth principle, it would be this: show your work to other people. Regularly and often. Because even if you do all of the above, you can still end with too much information. Because you love this data, right? It hurts to cut it down. An audience will quickly tell you if they 'get it' or not. And remember that your audience is your final - and most important - collaborator.