How to…do data visualisation

The coronavirus pandemic has been defined by data. How that data has been communicated and presented has really mattered. There have been some great examples of how to do it – think the early representations of ‘flattening the curve’ and charts from the likes of the Financial Times, Our World in Data and others – but also some great ones of how not to do it, not least from the UK government.

Of course, data visualisation – a broad term covering charts and graphs, tables, and infographics, often shortened to dataviz or data vis – wasn’t invented during the pandemic. William Playfair, an early pioneer, stated why it matters: a number in a table might be unmemorable, ‘like a figure imprinted on sand… soon totally erased and defaced’, but charts make ‘a sufficiently distinct impression… to remain unimpaired for a considerable time, and the idea which does remain will be simple and complete’. Last year’s bicentenary of Florence Nightingale’s birth reminded us of her dataviz innovation and the impact her charts had on government policy – they can convey key information to busy policymakers eye-catchingly, concisely and quickly.

The interest in ‘data journalism’ over the last decade and the ability of think tanks to communicate directly to their audiences thanks to the internet and social media means many research organisations are thinking about how to use data visualisation in their work. I ran the Institute for Government’s Whitehall Monitor project – which analyses and visualises the size, shape and performance of government – for seven years, and helped develop data and dataviz capability across the Institute. Here are some of my top tips for visualising data as a think tank.

Be clear about the story you’re trying to tell, and work out the best way to tell it

If your starting point is thinking at the end of a project, ‘we want a fancy chart that will look great on social’, let me humbly suggest you don’t start from there. Good charts allow you to communicate important analysis from your work in compelling ways, and that should be your starting point. Readers will often only spend a short amount of time on a chart, so you should be careful not to overload it – what’s the key thing you want them to take away from the chart? Be clear about this – if you’re not, throwing charts at the problem isn’t going to help.

Once you know what story it is you’re trying to tell, think about the best way to tell it. You wouldn’t throw letters randomly onto a page and expect them to land in crisp paragraphs and perfectly formed sentences. Similarly, don’t throw numbers into chart software and expect the first thing that comes out to be the best way of conveying what you’ve found in the data.

Think about the right chart type

There are a lot of different chart types – you may already be familiar with bar charts, pie charts, line charts, scatter plots, tables and maps, but there are many, many more and many, many variations within types. There is a wealth of online resources to help you navigate your way through and understand what chart types work in particular circumstances (try the Data Visualisation Catalogue, the FT’s Visual Vocabulary, or the Graphic Continuum by Jon Schwabish and Severino Ribecca for a start).

There is often nothing wrong with keeping it simple – sometimes, a simple line chart going up or down can be the most effective way of making your point, and your audience is likely to be familiar with how the chart works. (Though beware the false promise of pie charts, which are often not the best way to present data. And just because data is about a location doesn’t mean a map is the best way of displaying it.) But don’t be afraid to experiment – sometimes more unusual charts which tell a number of stories can draw the reader in (this was one of our signatures at the IfG).

That said…

Don’t distort the data, or let bad design distort it

Obviously, when we talk about telling a ‘story’ with data we’re not talking fiction. It should go without saying that you shouldn’t be bending the data to fit a preferred narrative. You should make sure that any caveats that would fundamentally affect the reading of the chart are clear, as is the source of the data (so others can go and dig into it if they wish). You will always be making choices and trade-offs in designing dataviz – but be careful with them.

Bad design can distort the data too. The chart may be so cluttered it lacks any clarity in what it was trying to convey. You might compress the range of the axes in a way that makes things look more dramatic than they are. You may get so caught up in the aesthetics it makes the analysis difficult to grasp (this upside down chart is a classic of the genre). In delivering dataviz training, I sometimes use the Star Wars prequel trilogy as a cautionary tale – in getting so caught up in what you are able to do with the tools at your disposal (computer graphics, in the case of a galaxy far, far away) you can lose sight of what really matters (narrative, dialogue, characterisation).

When it comes to clarity, attractiveness and accuracy, you may want to think about two Twitter tests: would someone seeing the chart be able to understand its key message in 15 seconds (or be sufficiently drawn in by it that they’ll study it for longer), and how would you feel if your chart image were shared without any further context?

Details matter

What you do with labels, axes, gridlines, ordering the data and everything else on a chart area also helps with your storytelling (it’s instructive that one dataviz expert has an entire series on ‘the little’ of dataviz design). Axis labels taking up so much of the chart area that the data gets squeezed? Gridlines so bold they distract you from the data? Then you have a problem. A decent rule of thumb is that you want most of the ink on the chart to be spilt on the data itself (or things which actively help your reader understand what it’s showing) – minimise the ‘chart junk’, as it can sometimes be called.

Think about your audience, what you’re trying to achieve, and how you’re presenting to them

You’ll have been in the weeds of a particular subject, and engaged on the finer design points of the chart displaying information about it, for a while. Your audience won’t. Will they understand the abbreviations you’ve used and all the assumptions you’ve made?

Who do you want to reach and what do you want them to do? Social media can get very excited about whizz bang interactives which may help increase your reach, but simple static charts that the right civil servants can easily download and drop into presentation slides for ministers may be what you want. And you’ll be able to do different things with a single graphic for social media, versus a chart in a longer form piece, versus an interactive, versus presenting in person where you can take more time to talk your audience through what you’re showing them and build up the story.

There are lots of great dataviz tools. Microsoft Excel is one of them

There are some great online tools – Flourish, Datawrapper, Highcharts, Tableau (many others are available) – and lots of visualisation software packages. If you have the right expertise, you might be coding things directly. But any tool is only as good as what you’re doing with it (see all of the other advice above).

Don’t write off Microsoft Excel. Nearly every IfG chart to date has been made in Excel. It’s a powerful tool, you’ll be surprised at just how much you can do with it, there are digital reams of advice on how to get the most from it, and (crucially) it’s something that most people are likely to have some familiarity with, so it’s a brilliant starting point if nothing else. But move away from the default colours and fonts; use your corporate ones instead, and don’t be afraid to play around with the other chart junk. It’s amazing how much more professional your charts will look with a little tweaking.

Build on what you’ve done before

Every new dataviz created, every new dataset explored is a learning experience. You don’t want to have to reinvent the wheel every time you publish a new chart – build on what you’ve done before. Develop a style guide, a portfolio of charts and a process of getting from the dataset to a finished product (at the IfG that was getting charts from Excel into a template in some free graphics software) that you can go back to. Your existing knowledge is what will help you innovate and is good for branding, understanding (you’ll be going back to things that work) and ease (save time by using something that works).

Are there particular data releases that your think tank can ‘own’, and datasets you want to keep going back to? For example, I live-blogged government reshuffles for 6 years at the IfG and there’s no way we could have produced numerous data-informed annual reports without these building blocks. The Resolution Foundation and IFS have a strong focus on particular government fiscal events. These are worth investing in – both in terms of the plumbing as well as the finished product.

You’re not running a dataviz project. You’re running an organisational change one

One of your most important audiences is going to be your internal one. To do dataviz right is to do data right – to build it into your research and comms processes and think about it throughout a project, not as an afterthought as the press release is about to go out. ‘Data’ can still be intimidating and/or boring to many – you will need to bring people with you in supporting them to develop the right skills, in showing how better use of data and dataviz can help them in their work, and in making it interesting (and less intimidating) and fun as well as functional. (My weekly newsletter started as an internal effort to do some of this.)

You have to start somewhere

The best thing to do is give it a go. What starts with a simple dataset in Excel today could lead to an all-singing, all-dancing interactive dashboard with APIs spitting out data in time. You’ll never know if you don’t try.

You’re going to make mistakes – that’s not to be cavalier with accuracy, merely to say that you’ll publish the odd chart that doesn’t quite sing the way you want it to– but you’ll learn from them, and from the mistakes other think tanks, data journalists and others have made along the way. The freedom to experiment (take some time to get things wrong internally) is vital, and another reason to get others in your organisation on board: they’ll be your most useful critics.

The dataviz expert Alberto Cairo has described data visualisation as a ‘functional art’. Both words are important. Don’t get so carried away with the artistic possibilities that you forget its function: to communicate stories you’ve discovered in data. But equally, if it is so functional as to become dull and have no art, it’ll cease to be functional as no one will want to look at it.

By Gavin Freeguard

17 February 2021