Today’s question is all about how much explanation should be included with a visualisation. In terms of your overall look, should you clarify, or simplify? It’s a topic that often comes up in visualisations, and I think it’s one I don’t always get right.
Earlier this week I revisited an idea I had following the Tableau Conference last year. As part of his talk, Chris DeMartini presented the Jump Plot, a new visualisation type he has devised and documented. Many examples and instructions can be seen in www.jumpplot.com. After a few attempts, I got the hang of this particular chart, and game up with my own version, for Premier League Goalscorers this season (up to the end of week 19 – end of 2016).
I can take little credit for the design or clever technical work behind the chart, only for adapting what was already out there to fit my dataset, but I’m pretty pleased with the way it has ended up. The jump plot does a good job of showing the range of time/matches between each goal for all the division’s top goal scorers. Click through to the online version and you can see a lot more if you explore. Diego Costa’s prolific and consistent season (all low “jumps”), the locations of hat-tricks (three goals in one game, resulting in two horizontal flat lines in a row) for Romelu Lukaku and Salomon Rondon, etc.
But when I published it, I left out the “number of goals” text, before I got feedback that the person viewing thought the numbers referred to games, not goals. I can concede that adding text makes that clearer, so it’s in. Then I was contacted by Chris, the man behind jump plots, who explained that I should really include a y-axis and faint axis lines to make the chart easier to read and understand.
I think he’s absolutely right, but I haven’t done it! This is an unfamiliar chart type, so it makes sense to give the viewer more guidance than I’ve given. Show them the axis and grid lines, so there is more meaning to the size of each jump, without relying on the viewer to find out for themselves. But, I’ve designed it so that my information is on the tooltip when the viewer explores the chart. And I want the viewer to explore the chart, that’s the only way they’ll activate the highlighting and/or see which bump refers to which player. I haven’t overloaded the chart with instructions, but just made sure to include the instruction “hover over any line to learn more”. I think in best practice terms I’m probably wrong to do this without offering more viewer guidance. But I don’t always follow best practice, and I know that I err on the side of exploratory (viewer find out for yourself!) over explanatory (viewer, this is what I’m showing you). There’s personal style behind my stubbornness. My jump chart above hasn’t been active for very long, so I’d welcome opinions/comments/feedback as to whether it needs more explanation!
Moving on, in the last couple of days I’ve been compiling a great dataset from all the top 75 singles charts in the UK over the 1980s. I couldn’t quite get the data scraping process to work, so have been doing much clicking, copying, pasting and data wrangling to get my dataset. Of course, in doing this, the names of many 80s songs and artists have passed through my field of vision and I’m looking forward to the data visualisation possibilities lying ahead. It’s an exciting dataset, almost 40000 rows, each of which is describing an 80s song and artist, and should be a source for a wide range of possible visualisations.
I’ve loaded the data into Tableau, set up a simple visualisation pretty much out of the box, to show line charts of every song released, coloured by artist, showing a mass of lines to cover ten years, positions 1 to 75 on the charts. It was only done as a validation exercise at first to see if I had any duplicates or blanks showing as outliers (I did – so a quick correction and reload sorted the issue). Instantly, with a simple title added, I had this:
You can’t get much more exploratory than this. No instructions, nothing visible, just two axes and a mass of colour. It’s a mess. As a work of art, some might say it resembles Pollock’s (careful …). But I love it – I didn’t get any further, and as I blog about it, I almost don’t want to go. As it stands at the moment, hover or click anywhere, and a song emerges, along with its individual path and position within the timeline of the 80s. The tooltip will tell you the song, chart position, artist and date. As simple as that.
Above is an example (which shows better in full-screen – click above to see in a new window) – the trace of Soft Cell / Tainted Love shows it rising up the charts to number 1 in 1981, but unusually shows a couple of resurgences in 1982 and what looks like a possible re-release in 1985. But the whole screen is like a dense jungle waiting to be explored. Find “Relax” by Frankie Goes to Hollywood and you can see its trace shows 58 successive weeks in the charts (over a year). Find “Merry Xmas Everyone” by Slade and you can see regular pulses every twelve months. Use the Tableau lasso function to select a chunk covering one year and see the number of songs that had surprise re-entries later, or which were re-releases themselves. Click almost randomly and you’ll find songs you’d forgotten all about, or which had much shorter but individual imprints in the DNA of the 80s pop timeline!
Now this is the ultimate in exploratory visualisations, and I couldn’t possibly release this for everyone. It even looks like a dense jungle – the ultimate in exploratory environments! Data nerds like me who loved the 80s and have time on their hands don’t need more than this. We don’t need to be told where to find songs and bands, but are happier stumbling across them or honing in to them using our memory of where and when it charted first time round. Because this is a project with data that I’m enjoying, I’ve done enough for myself. Now, I will do something more with this – make it more appealing and accessible in some way, because in considering exploratory versus explanatory visualisations, we should always consider who our audience is. I haven’t yet decided who the audience is, and currently the audience is just me (or people like me). But, if I don’t like the finished and more polished visualisation I come up with, it won’t see the light of day.
Recently, Andy Cotgreave (I keep mentioning him – he doesn’t really need the publicity!) wrote an article here about the place for punk in data visualisation: I don’t think he meant it literally (given that this starts only just post-punk, with the likes of Stranglers, Clash and Boomtown Rats still featuring in this chart), but was referring in general to quickly-produced messy and ad-hoc visualisation. The upshot of the argument is that it doesn’t need skill, presentation and flair to tell a story or present data. I think we can safely say the above falls into those categories.
I don’t conclude from this that exploratory visualisations are better than explanatory ones. In many cases it’s a weakness of mine that I don’t explain enough, either to guide the user to navigate the visualisation, or to spell out the most pertinent points that the visualisation shows. As a style and preference, I must admit I lean towards exploratory visualisations for personal projects, as they more closely mirror my own taste/preference as a user. But part of my learning journey as a practitioner of visualisations is using all kinds of approaches and knowing when certain approaches are acceptable and when they’re not.