This chart of population density across Europe by Henrik Lindberg has been very popular online this last week.
Long-standing readers will recall my stab at this but nowadays everybody just does it in ggplot2. It’s good to have options. While you’re at his Gist page, checkout his other stuff too.
This is just the greatest thing I’ve seen in a while, and definitely in the running for dataviz o’ the year already. Emoji scatterplot:
There’s also a randomisation test which I’ll leave you to discover for yourself.
This week, a chart with some Bayesian polemic behind it. Alexander Etz put this on Twitter:
He is working on an R package to provide easy Bayesian adjustments for reporting bias with a method by Guan & Vandekerchhove. Imagine a study reporting three p-values, all just under the threshold of significance, and with small-ish sample sizes. Sound suspicious?
Sounds like someone’s been sniffing around after any pattern they could find. Trouble is, if they don’t tell you about the other results they threw away (reporting bias), you don’t know whether to believe them or not. Or there are a thousand similar studies but this is the (un)lucky one and this author didn’t do anything wrong in their own study (publication bias).
Well, you have to make some assumptions to do the adjustment, but at least being Bayesian, you don’t have to assume one number for the bias, you can have a distribution. Here, the orange distribution is the posterior for the true effect once the bias has been added (in this case, p>0.05 has a 0% chance of getting published, which is not unrealistic in some circles!) This is standard probabilistic stuff but it doesn’t get done because the programming seems so daunting to a lot of people. The more easy tools – with nice helpful visualisations – the better.
It’s the world’s first ever hot air balloon plot. With the emphasis on the hot air.
This is from “The Global Epidemiology and Contribution of Cannabis Use and Dependence to the Global Burden of Disease: Results from the GBD 2010 Study” by Louisa Degenhardt and colleagues. I don’t really get the rationale for the mash up of pie chart with expanded stacked bars. Each of those is common enough by itself, although pies are pretty much never acceptable to the discerning datavizzer, but they map proportions to different visual parameters: angle and length. If you combine them, they just don’t work. Maybe the reason is that the focus is cannabis, and if they included alcohol in the bars it would swamp everything else. But… the stacked bars are actually smaller in height than the pie slice they supposedly expand on! Still, they could use a waffle plot, which is always a favourite of mine:
The other worry is that to make a mash-up, you have to get some bits out of analytics / spreadsheet packages and then get creative in some graphics package. That might just distort things a little. The pie is in tell-tale Stata default colors, while the bars are not; I suppose they were bolted together from different sources.
Here’s a graphic of a really deep oil well by Fuel Fighter via Visual Capitalist. This is rather reminiscent (ahem) of the long, tall graphics by the Washington Post (and the eerily similar one from the Guardian a few days later which they had to admit they had nicked) about flight MH370 at the bottom of the ocean. The WP graphic works because you have to scroll down, and down, and down, and down, and down (wow, that’s deep!), and down, and down (no way), and down before you get to the sea bed. Yes, all the usual references are there, hot air balloons and Burj Khalifas and Barad-Dûrs and what have you, but they don’t matter because it’s the scrolling that does it, giving you GU2 (“Conveying the sense of the scale and complexity of a dataset”) and GU6 (“Attracting attention and stimulating interest.”) The references don’t mean anything to me (or probably you); I may have seen the Burj Khalifa and thought it was amazingly tall, but I have no grasp of how tall and that is what matters: I’d have to have an intuitive feel for what 3 BKs are compared to the height of a jet aircraft, and I don’t have that, so why should I care about the references?
My problem with the Fuel Fighter graphic is that it doesn’t have that same sense of depth. The image file is 796 x 4554 pixels, which is an aspect ratio of 1:17. The WP image (SVG FTW) is 539 x 16030 or 1:30, which is pretty extreme! It feels to me like you’d have to get past 1:20 before it started to have enough impact.
The Washington Post have an article about the US budget out by Kim Soffen and Denise Lu. It’s not long, but brings in four different graphical formats to tell different aspects of the data story. A bar showing parts of the whole (see, you don’t need a pie for this!)
then a line/dot/whatever-you-want-to-call-it chart of the change in relative terms
then a waffle of that change in absolute terms, plus a sparkline of the past.
there’s also a link to full department-specific stories under each graphic. I think this is really good stuff, though I can image some design-heads wanting to reduce it further. It shows how you can make a good data-driven story out of not many numbers.
Corinne Riddell posted this on Twitter. It’s one version of multiple time series that she tried out, one for each USA state. It’s not the finished article, but is really nice for its combination of that recognisable shape (I suppose if your country has a dull shape like Portugal — no offence — then readers won’t immediately recognise the meaning of the arrangement) and the clean, simple small multiples. Admittedly, the time series has enough signal:noise to make this possible, and only a few unusual states, and without that it might start to get like spaghetti, but it’s always worth sketching options like this out to see how they work.
Could the state names get dropped? Probably not, but they could turn into two-letter abbreviations. The whole idea of the small multiple is that it’s not a precise x-y mapping but a general impression, so the y-axis labels could go (just as there are no x-axis labels).
Overall, I really like it and would like to write up a UK function for this to add to my dataviz toolbox.
Filed under R, Visualization